Natural Selection for Disease Resistance in Hybrid Poplars Targets Stomatal Patterning Traits and Regulatory Genes

Home , Eucalyptus carnei, Eucalyptus tricarpa, Salix eriocephala

University of Vermont ScholarWorks @ UVM

Graduate College Dissertations and Theses Dissertations and Theses

2019

Natural Selection For Disease Resistance In Hybrid Poplars Targets Stomatal Patterning Traits And Regulatory Genes.

Karl Christian Fetter University of Vermont

Follow this and additional works at: https://scholarworks.uvm.edu/graddis

Part of the Plant Pathology Commons

Recommended Citation Fetter, Karl Christian, "Natural Selection For Disease Resistance In Hybrid Poplars Targets Stomatal Patterning Traits And Regulatory Genes." (2019). Graduate College Dissertations and Theses. 1162. https://scholarworks.uvm.edu/graddis/1162

This Dissertation is brought to you for free and open access by the Dissertations and Theses at ScholarWorks @ UVM. It has been accepted for inclusion in Graduate College Dissertations and Theses by an authorized administrator of ScholarWorks @ UVM. For more information, please contact [email protected]. NATURAL SELECTION FOR DISEASE RESISTANCE IN HYBRID POPLARS TARGETS STOMATAL PATTERNING TRAITS AND REGULATORY GENES.

A Dissertation Presented

by Karl Christian Fetter to The Faculty of the Graduate College of The University of Vermont

In Partial Fulﬁllment of the Requirements for the Degree of Doctor of Philosophy Specializing in Plant Biology

October, 2019

Defense Date: August 19th,2019 Dissertation Examination Committee:

Stephen R. Keller, Ph.D., Advisor Melissa Pespeni, Ph.D., Chairperson Terrence Delaney, Ph.D. Charles Goodnight, Ph.D. Jeanne Harris, Ph.D. Cynthia J. Forehand, Ph.D., Dean of Graduate College Abstract

The evolution of disease resistance in plants occurs within a framework of interacting phenotypes, balancing natural selection for life-history traits along a continuum of fast-growing and poorly defended, or slow-growing and well-defended lifestyles. Plant populations connected by gene flow are physiologically limited to evolving along a single axis of the spectrum of the growth-defense trade-o, and strong local selection can purge phenotypic variance from a population or species, making it dicult to detect variation linked to the trade-o. Hybridization between two species that have evolved dierent growth-defense trade-ooptima can reveal trade-os hidden in either species by introducing phenotypic and genetic variance. Here, I investigated the phenotypic and genetic basis for variation of disease resistance in a set of naturally formed hybrid poplars. The focal species of this dissertation were the balsam poplar (Populus balsamifera), black balsam poplar (P. trichocarpa), narrowleaf cottonwood (P. angustifolia), and eastern cottonwood (P. deltoides). Vegetative cuttings of samples were collected from natural populations and clonally replicated in a common garden. Ecophysiology and stomata traits, and the severity of poplar leaf rust disease (Melampsora medusae) were collected. To overcome the methodological bottleneck of manually phenotyping stomata density for thousands of cuticle micrographs, I developed a publicly available tool to automatically identify and count stomata. To identify stomata, a deep convolutional neural network was trained on over 4,000 cuticle images of over 700 plant species. The neural network had an accuracy of 94.2% when applied to new cuticle images and phenotyped hundreds of micrographs in a matter of minutes. To understand how disease severity, stomata, and ecophysiology traits changed as a result of hybridization, statistical models were fit that included the expected proportion of the genome from either parental species in a hybrid. These models indicated that the ratio of stomata on the upper surface of the leaf to the total number of stomata was strongly linked to disease, was highly heritable, and wass sensitive to hybridization. I further investigated the genomic basis of stomata-linked disease variation by performing an association genetic analysis that explicitly incorporated admixture. Positive selection in genes involved in guard cell regulation, immune system negative regulation, detoxification, lipid biosynthesis, and cell wall homeostasis were identified. Together, my dissertation incorporated advances in image-based phenotyping with evolutionary theory, directed at understanding how disease frequency changes when hybridization alters the genomes of a population.

i Citations

Material from this dissertation has been published in the following form:

Fetter, K. C., Gugger, P. F., and Keller, S. R.. (2017). Landscape Genomics of Angiosperm Trees: From Historic Roots to Discovering New Branches of Adaptive Evolution. In: Groover A., Cronk Q. (eds) Comparative and Evolutionary Genomics of Angiosperm Trees. Plant Genetics and Genomics: Crops and Models, vol 21. Springer, Cham.

Fetter, K. C., Eberhardt, S., Barclay, R. S., Wing, S. and Keller, S. R.., (2019). StomataCounter: a neural network for automatic stomata identiﬁcation and counting, New Phytologist, vol 223, no. 3, 1671-1681.

ii You picked the right time to accept this diploma,

As the whole family is thankful for an August in Vermont.

And after sleepless nights and years in snow capped mountains,

The family can unite for the pride of progress.

Take the time to enjoy it.

For your gluttony for punishment awaits in the Canadian prairies.

CHURCH STREET - //

iii Acknowledgements

The philosophiae doctor is an achievement one can obtain only with the guidance and encouragement of a cadre of professors, post-doctoral researchers, fellow graduate students, and peers. I have been fortunate to ﬁnd myself surrounded by people whose desire to increase and diuse knowledge inspired me to do the same. The journey I took to write the words in this dissertation was characterized, or one could say beset, by learning to listen carefully, embracing doubt and questions, and developing a healthy relationship with failure. Some of the traits required to complete the philosophiae doctor I brought with me to Vermont, but most were taught to me by those in my scientiﬁc network in Burlington. I want to thank all of you for being there for me, for challenging me, and for believing in me. In particular, I would like to thank my Vermont family: Dr. Stephen Keller, Brittany Verrico, Dr. Vikram Chhattre, Ethan Thibaut, Jamie Waterman, Dr. Kattia Palacio-Lopez, Susan Fawcett, Dr. Michael Sundue, Dr. Dominik Thom, Dr. Jeanne Harris, Dr. David Barrington, Dr. Mary Tierney, Dr. Deb Neher, Tom Weicht, Dr. Yolanda Chen, and Dr. Andrea Lini. Porky Reade, Sarah Goodrich, and Karyn McGovern were the best administrative staI have ever worked with! The Smithsonian Institution was where the spark of intellectual curiosity inside me was turned into a welding torch. Unequivocally, I would not be the scientist I am today without Dr. Scott Wing, Dr. Carrol Hotton, Dr. Rich Barclay, Dr. Bill DiMichele, Dr. Nathan Jud, Dr. Jun Wen, and Dr. Michael Gates. I must also acknowledge the important contributions to my education by professors at the University of North Carolina: Dr. Alan Weakley, Dr. Pat Gensel, Dr. Paul Gabrielson, and Dr. Haven Wiley. Christiane Hawkins, Debbie Barbot, and Gordon

iv Warburton inspired me in the classroom and in the music room as a high school student, and I thank them for that. Finally, my family are the roots I have used for decades to stay up right. They have supported me in thousands of ways, inspired me, told me to keep going when the times were tough. There is no way to write how thankful I am to call as family Maureen Mallon, Harry Blair, Caitlin Brennan, Evan Brennan, Hudson Brennan, Poppy Brennan, Maegen Novak, Dr. David Novak, Reynolds Novak, Annie-Lee Wed- dle, Eileen Mallon, Charles Fetter, and Dr. Sydney Conrad.

Chaos, Horror, & Hospitality

v Contents

Citations ...... ii Dedication ...... iii Acknowledgements ...... iv List of Figures ...... xiii List of Tables ...... xv

1 Landscape genomics of angiosperm trees: from historic roots to discovering new branches of adaptive evolution 1 1.1 Introduction ...... 3 1.1.1 Historical Review ...... 6 1.1.2 Challenges and Solutions for Landscape Studies in Trees ... 19 1.1.3 Future Research Directions for Forest Tree Landscape Genomics 26 1.2 Conclusions ...... 35 1.3 Acknowledgements ...... 35 1.3.1 Author Contributions ...... 36 1.4 Tables ...... 37 1.5 Figures ...... 38

2 StomataCounter: a neural network for automatic stomata identiﬁ- cation and counting. 67 2.1 Introduction ...... 69 2.2 Materials and Methods ...... 72 2.2.1 Biological Material ...... 72 2.2.2 Deep Convolutional Neural Network ...... 73 2.2.3 Statistical Analyses ...... 76 2.3 Results ...... 79 2.3.1 Stomata detection ...... 79 2.3.2 Classiﬁcation accuracy ...... 81 2.4 Discussion ...... 82 2.5 Acknowledgements ...... 87 2.6 Data availability ...... 87 2.7 Author contributions ...... 88 2.8 Tables ...... 89 2.9 Figures ...... 92

3 Trade-os and selection conﬂicts in hybrid poplars indicate the stomatal ratio as an important trait regulating disease resistance. 105 3.1 Introduction ...... 107

vi 3.2 Materials & Methods ...... 110 3.2.1 Study system ...... 110 3.2.2 Materials ...... 111 3.2.3 Molecular analyses ...... 112 3.2.4 Trait measurements ...... 114 3.2.5 Quantitative genetic analyses ...... 117 3.3 Results ...... 120 3.3.1 Genetic ancestry and trait variation ...... 120 3.3.2 Selection trade-os and conﬂicts ...... 123 3.4 Discussion ...... 124 3.5 Acknowledgements ...... 132 3.6 Data Accessibility ...... 132 3.7 Author Contributions ...... 132 3.8 Tables ...... 134 3.9 Figures ...... 137

4 Guard cell regulation, the plant immune system, and growth transcription factors are under selection and associated with fungal disease variation in advanced generation P. balsamifera x trichocarpa hybrids. 159 4.1 Introduction ...... 162 4.2 Materials & Methods ...... 165 4.2.1 Plant collections, traits, and sequencing ...... 165 4.2.2 Admixture mapping ...... 166 4.2.3 Selection Scans ...... 168 4.2.4 Gene annotation and candidate gene ﬁltering ...... 169 4.3 Results ...... 169 4.3.1 Global and local ancestry of hybrids ...... 169 4.3.2 Trait variation ...... 170 4.3.3 Admixture mapping ...... 171 4.3.4 Selection scans ...... 172 4.3.5 Candidate genes and local ancestry class phenotypic distributions172 4.4 Discussion ...... 173 4.4.1 Candidate genes ...... 175 4.4.2 Limitations of our study ...... 180 4.4.3 Conclusion ...... 181 4.5 Acknowledgements ...... 183 4.6 Data Accessibility ...... 183 4.7 Author contributions ...... 183 4.8 Tables ...... 184

vii 4.9 Figures ...... 192

viii List of Figures

1.1 Populations for landscape genomic studies are frequently sampled out of a context of spatially varying genomic structure and environmental change. The genetic structure of P. balsamifera is represented from a sample from 412 reference SNPs genotyped by (Keller et al. 2010) with three demes observed: low diversity central and western demes, and a higher diversity eastern deme (a). Environmental gradients abound in nature, and greenness onset is a classic example of a biological phenomenon, controlled by both photoperiod and temperature, that changes across large geographic areas (b). Eliminating false positives in selection scans is always challenging; however, fewer false discoveries are observed from the tail of empirical p-value distributions derived from putatively neutral reference SNPs (c). This approach is demonstrated with the results of a BayeScan analysis (Foll and Gag- giotti, 2008) of 412 reference SNPs and 335 candidate SNPs from 27 flowering-time genes that were sampled from the localities in (a). Set- ting a threshold conditioned with empirical p-values identifies local selection in SNPs from GIGANTEA 5 (GI5 ), EARLY BOLTING IN SHORT DAYS (EBS), PHYTOCHROME-INTERACTING TRAN- SCRIPTION FACTOR 3 (PIF3 ), andCASEIN KINASE II BETA CHAIN 3 (CKB3.4 ). Allele frequency variation in four SNPs from GI5 are associated with latitude (d). GI5 is involved in light signaling to the circadian clock and local adaptation of these alleles may allow precise timing of phenology. Figures reproduced, with permission, from M. Fitzpatrick, (a); A. Elmore (b); S. R. Keller, (c, d)...... 39 1.2 Overlap of candidate genes for local adaptation inPopulus trichocarpa identified in selection scans by Geraldes et al. (2014), McKown et al. (2014)c, Holliday et al. (2016), and Evans et al. (2014). No overlap was detected by all studies, but six genes overlapped among three studies (underlined). Version 2.2 Gene annotations from Geraldes et al. (2014) were updated to 3.0 for comparisons...... 40

ix 2.1 Architecture of the deep convolutional neural network (DCNN) and classification tasks. Left: Training and testing procedure. First column: Target patches were extracted and centered around human- labeled stomata center positions; distractor patches were extracted in all other regions. A binary image classification network was trained. Second column: The image classification network was applied fully convolutional to the test image to produce a prediction heatmap. On the thresholded heatmap, peaks were detected and counted...... 92 2.2 Sample sizes of images for the top 20 families represented in the training (yellow) and test (blue) datasets (a). Examples of pre- and post- analysis images. A probability heatmap map is overlain onto the input image in the red channel. Detected stomata marked with circles with peak values given in green (b). Species in image pairs: Adiantum peru- vianum, Begonia loranthoides (top row); Chaemaerauthum gadacardii, Echeandia texensis (second row); Ginkgo biloba, Populus balsamifera (third row); Trillium luteum, Pilea libanensis (bottom row). Scale bars in row one and two equal 20 µm, scale bars in row three equal 25 µm, and scale bars in row four equal 50 µm...... 93 2.3 Results of human and automatic counts and precision for each dataset organized by collection source (a,d), taxonomic group (b), and magnification (c), and imaging method (e). Precision and image quality scores have non-linear relationships, and changes in variance correlate to image quality (g-i). Positive values of precision indicate undercounting of stomata relative to human counts, while negative values indicate overcounting. The dotted black line is the 1:1 line. See Table 2.2 for model summaries. Abbreviations: tEntropy, image entropy ;fMean, mean power-frequency tail distribution; fSTD, standard deviation of power-frequency tail distribution...... 94 2.4 Accuracy by training image count and the eect of increasing the training set size on classification accuracies. Since classification is a binary task, chance level is at 50%. Training image count “all” includes all 4,618 annotated training images...... 95

x 2.5 Samples that were mislabeled with high confidence by either the machine or human. False negatives (left panel) from stomata that were detected by the human, but not by the machine. Image features typically generating false negatives are blurry images, artifacts obstruction the stomata, low contrast between epidermal and guard cells, and very small scale stomata. False positives (right panel) are image features that are labeled by the machine as stomata, but not by the human counter, or are missed by the human counter (e.g. top and bottom right images in the panel). Errors typically generating false positives are image artifacts that superficially resemble stomata, particularly shapes mimicking interior guard cell structure...... 95 2.6 Accuracies for models trained on dierent training datasets (vertical), tested on dierent test datasets. Combined is a union of all training and test sets. For precision and recall values, see Supporting Information Fig. S2.3...... 96 S2.1 Stomata identification test results ...... 97 S2.2 Stomata identification in Arabidopsis cell-patterning mutants. .... 97 S2.3 Precision and recall of transferred training and testing datasets. ... 98

3.1 Summary of sample localities, phenotypes, and expected ancestry proportions for the trees under study. Sample locality map with the geographic ranges of four species of poplar: P. balsamifera in blue, P. trichocarpa in green, P. angustifolia in orange, and P. deltoides in yellow (a). See Table S1 for population coordinates and samples sizes. The phylogenetic relationships of the four species under study are presented as a cladogram (b). The expected proportion of ancestry of hybrid genotypes was calculated from NewHybrid (Anderson and Thompson, 2002) estimates of filial generation (c). Histograms and box plots of three disease traits summarized by PC1 of a principal components analysis (DPC1), relative growth rate (G), and stomatal ratio (SR). Tukey tests indicate significant dierences at –<0.05. PCA of stomatal patterning and ecophysiology trait data. The presence or absence of disease and the hybrid cross type is mapped onto the points, but was not used to decompose variance of the traits in principal component space (e). Disease presence/absence calls in e were made from the eigenvalues for DPC1, where PC1 eigenvalues greater than -0.8 were classified as diseased. Images of the uredospore (f) and disease sign of orange-yellow uredinia (g) of Melampsora medusae on an BxD hybrid. 138

xi 3.2 Results of partial least squares regression (PLS) of ancestry proportions and traits. B,D,T, and A represent vectors of the expected ancestry proportions for each individual estimated by NewHybrids (Table 3.2). The color ramp is the scalar product of variable loadings U’ and V’ from the X and Y matrices, respectively, indicating the correlation between pairs of vectors. Broad-sense heritability (H 2) estimates for each trait are given from data sets with hybrids present (black) and absent (red). 95% credible intervals indicated as whiskers. See text for PLS and H 2 estimation details and Table S3.2 for H2 estimates. ... 139 3.3 Backcrossing into P. balsamifera decreases disease (DPC1) (a) and the stomatal ratio (SR) (b). When accounting for the eect of SR on disease, residual variance decreases from F1 to Fn generations and unadmixed P. balsamifera (c). Growth is uniform across the filial generations (d). Dierences in Tukey’s post-hoc tests for significance (p-value < 0.05) indicated by letter...... 140 3.4 MANOVA regression coecient heat map...... 141 3.5 Regression of stomatal ratio onto principal components 1 and 2 of disease severity and growth. Regression coecients and significance indicated in table...... 142 S3.1 Distribution of clonal values. See Table 3.1 for trait abbreviations. .. 147 S3.2 Variation of traits by cross type. See Table 3.1 for trait abbreviations. 148 S3.3 Variation of traits by filial generation. See Table 3.1 for trait abbreviations...... 149 S3.4 Permutation tests for significance of selection conflict products. The product of the regression coecients for each pair of traits was estimated 1000 times with permuted data (grey distribution). The observed product is considered significant if its value is more extreme than the permuted distribution at –/2=0.025. See Table 3.4 for results.150

4.1 Map of collection localities in western North America with the range of P. balsamifera and P. trichocarpa colored in blue and green, respectively (a). Ranges from (Little, 1971). Traits distributions fall into three categories: approximately normal, left-skewed, and zero-inﬂated distributions (b). Global ancestry estimates from locus-speciﬁc ancestries estimated by RASPberry (c). Local ancestry for each individual (in rows) and for every site (in columns) colored as gray if the locus is homozygous for the P. balsamifera allele, blue for heterozygous loci, and green for homozygous for the P. trichocrapa allele (d)...... 192

xii 4.2 Summary of logistic model and trait variation. Forest plot of standardized regression coecients from a logistic model with disease presence as the response. Blue and red points are positive and negative odds ratios, respectively, and red line is an odds ratio of 1 (a). See Table S4.1 for logistic model output. Predicted values of disease presence for stomata ratio (SR) (b) and global ancestry (c). Biplot of SR and global ancestry (d). Global ancestry of 1 indicates an unadmixed P. balsamifera individual...... 193 4.3 Output of BMIX models represented in manhattan plots. All traits are plotted in the same space in a. Disease and stomata traits grouped in b, ecophysiology traits in c, stomata traits in d. Red points in b, c, and d indicate significant loci at –=0.05/237.5...... 194 4.4 Intersection plot of significant genes from BMIX results...... 195 4.5 Plot of µ statistics for three sets of individuals by chromosome. The overlap of BMIX candidate genes and the top 1% of µ statistic outliers are indicated by the solid and dashed boxes for the disease + stomata gene set, and disease only gene set, respectively. The density distribution of µ-statistics is presented by set, and the vertical colored ticks on x-axis indicate the top 1% of values for each set...... 196 4.6 Patterns of stomata ratio (SR), disease severity (D1’), and relative growth rate (G) by local ancestry for genomic regions identified to be associated with stomata patterning and disease traits, in the top 1% of RAiSD selection windows, polymorphic, and with phenotypic distributions indicative of true positive results. Local ancestry genotype 0 indicates site is homozygous for P. balsamfiera ancestry, and 1 indicates the site is heterozygous with P. trichocarpa contributing the other copy of the locus...... 197 S4.1 Global ancestry estimated from ADMIXTURE (Alexander et al., 2009) at K = 2 (top) and K = 7 (bottom)...... 200 S4.2 Individual manhattan plots of p-values from BMIX tests. See Table 4.2 for trait abbreviation definitions...... 201 S4.3 Individual manhattan plots of p-values from BMIX tests. See Table 4.2 for trait abbreviation definitions...... 202 S4.4 Individual manhattan plots of p-values from BMIX tests. See Table 4.2 for trait abbreviation definitions...... 203

xiii List of Tables

1.1 Annotations for six genes implicated in local adaptation from range- wide samples of Populus trichocarpa by four separate studies. (M) McKown et al. (2014c), (G) Geraldes et al. (2014), (H) Holliday et al. (2016), (E) Evans et al. (2014). Annotations retrieved from www.popgenie.org...... 37

2.1 Description of the datasets used for training and testing the network. Abbreviations: DIC, Dierential interference contrast; SEM, scanning electron microscopy...... 89 2.2 Summary of linear model fit parameters in Figure 2.3 for dierent test datasets. Dataset definitions given in text. Abbreviations: y-intercept, a; standard error, SE. 200x images removed from this USNM/USBG † set. Significance indicated by: ú: P<0.05; úú: P<0.01; úúú: P<0.001. 90 2.3 Root mean square error (RMSE) of the model residuals. Lower RMSE values suggest a better fit of the model. The response in each model was precision. Abbreviations: fMean, mean power-frequency tail distribution; fSTD, standard deviation of power-frequency tail distribution. 91

3.1 Trait definitions, abbreviations, and the mean (µ) standard deviation ± (‡) of traits with hybrids present and absent...... 134 3.2 NewHybrids results. The Ref. non-Pb (0) and Ref. Pb (1) columns refer to the size of the reference populations from which segregating loci were selected. The remaning columns describe the sizes of the filial generations in unknown samples. Abbreviations: Ref., reference population; Pb., Populus balsamifera...... 134 3.3 Output of MANOVA with three disease severity traits jointly modelled as the response. Significance indicated as *** P < 0.001; ** P < 0.01; * P < 0.05; . < 0.1...... 135

xiv 3.4 Trade-os and selection conflicts inferred from linear and multiple regression models of three traits: D, disease resistance; G, relative growth rate; and SR, stomatal ratio. Negative regression coecients (—n)or path analysis (RSR,G) indicate a trade-os is present. The path analysis sums each independent path of the eect of SR on G: path 1, —2; path 2, — — . Selection conflicts are inferred when the product of 3 ú 1 regression coecients are negative (schluter1991conflicting). The hybrids absent set includes only unadmixed P. balsamifera, and the Tacamahaca and Aigeiros data sets add hybrid individuals. —1 was taken from model 1. Significance testing for selection conflicts was performed with permutation tests (Fig. S3.4). Significance indicated as *** P < 0.001; ** P < 0.01; * P < 0.05...... 136 S3.1 Summary of populations, samples sizes and geographic coordinates. . 144 S3.2 Broad-sense heritability estimates and credible intervals for the hybrids present and absent datasets...... 145 S3.3 Model output for trade-oregressions. Significance indicated as *** P < 0.001; ** P < 0.01; * P < 0.05...... 146

4.1 Population locality and sample size summaries for individuals used in admixture mapping. ref-Pb. and ref-Pt. are abbreviations for Populus balsamifera and P. trichocarpa that formed the reference sets. .... 185 4.2 Trait deﬁnitions and abbreviations...... 186 4.3 Summary of genes associated to stomatal patterning and disease traits also under positive selection. The number of genes found in each chromosomal region under selection is given beneath each chromosome. . 186 4.4 Average allelic eects for substituting a P. balsamifera ancestry site (0) for P. trichocarpa (1). Phenotypes were scaled and standardized before calculating the allelic eect. The allelic eects for each site in the chromosomal region under consideration was identical. The phenotypic distributions for these sites are in Fig. 4.6...... 187 4.5 Descriptions of candidate genes. Arabidopsis orthologs were identiﬁed by the best BLAST hit from www.popgenie.org or through BLAST results from The Arabidopsis Information Resource (www.arabidopsis.org). See text for details regarding candidate gene selection...... 188 S4.1 Output of logistic model of disease presence and traits...... 198 S4.2 Summary of populations used in the RAiSD selection scan analysis. Pb. is an abbreviation for Populus balsamifera...... 199

xv Chapter 1

Landscape genomics of angiosperm trees: from historic roots to discovering new branches of adaptive evolution

Karl C. Fettera,1, Paul F. Guggerb, and Stephen R. Kellera aDepartment of Plant Biology, University of Vermont bAppalachian Laboratory, University of Maryland Center for Environmental Science 1Corresponding author: Karl Fetter (e-mail: [email protected], tel: +1 802 656 2930)

1 Abstract

Landscape genomic studies analyze spatial patterns of genetic variation to test hypotheses about how demographic history, gene flow, and natural selection have shaped populations. For decades, angiosperm trees have served as outstanding model systems for landscape-scale genetic studies due to their extensive geographic ranges, large eective population sizes, abundant genetic diversity, and high gene flow. These characteristics were recognized early in the landscape genetics literature, and studies on angiosperm trees, particularly Populus and Quercus, tested hypotheses about how landscape features shaped neutral patterns of gene flow and population divergence. More recently, advances in sequencing and analysis methodologies have allowed for greater opportunities to directly test how natural selection acting locally across the landscape has shaped the genome-wide diversity of populations, often in the context of broad climatic gradients in growing season length. Despite, the methodological gains and successes of the last decade, landscape genomics studies face new challenges of study design, hypothesis testing, and validation. Here, we explore the development of landscape genetics and genomics in angiosperm trees and what we have learned from investigating the evolutionary consequences of life as a tree in heterogeneous landscapes. We outline the past, present, and potential future of landscape genomic studies in angiosperm trees, highlighting successes of the field, challenges to overcome, and ideas that scientists from all backgrounds engaged in landscape genomics should consider. 1.1 Introduction

The biological characteristics of some organisms lend themselves almost perfectly to the fields of study that have developed around them: Drosophila and population genetics; Caenorhabditis elegans and development; maize and plant architecture. Trees are, at first glance, unlikely model organisms, especially given their longevity and long generation time (Taylor, 2002); however, as models for ecological and landscape genomics, they combine a unique set of features (Petit and Hampe, 2006). Trees have large eective population sizes (Petit and Hampe, 2006; Evans, Allan, et al., 2015), disperse genes widely through pollen and seed over long distances (Nathan et al., 2002), produce millions of progeny that encounter strong selection (Augspurger, 1984), can clonally propagate and persist for millennia (Ally et al., 2008), extensively hybridize with congeners (Moran et al., 2012), and have large lifetime exposures to pathogens and herbivores that can change annually (Floate et al., 2016). These varied characteristics result in complex evolutionary processes that play out in landscapes as small as a few hectares (Strei, 1999), to entire biomes (Keller, Olson, et al., 2010) and continents (Callahan et al., 2013), making trees excellent candidates for the study of how ecological selection and landscape features structure genetic diversity. There is a rich history of ecological genetic studies in forest trees beginning with provenance trials of the 18th century (Langlet, 1971). Foresters would collect range- wide samples of seeds or cuttings from diverse populations and propagate these into common gardens, often replicated in dierent climatic regions of the landscape, with the objective of assessing how genetics and environment influence growth and yield, especially patterns of vegetative bud phenology and the timing of adaptive seasonal

3 dormancy (Savolainen et al., 2007). While these eorts began well before the age of genomics, researchers were already recognizing the importance of climate, disease, and other sources of selection in structuring forest tree adaptive diversity. With the rise of the genomic era, ecological geneticists working on trees have been increasingly in a position to evaluate the genetic architecture of local adaptation, identify regions of the genome associated with climate, and accelerate the selection and breeding process of tree domestication.

The expansion of genomic resources in poplars (Populus, Tuskan et al., 2006), Eucalyptus (Myburg et al., 2014), oaks (Quercus , Plomion et al., 2016), and other forest-trees (Liang et al., 2011; Fang et al., 2013) was critically important for opening new avenues of research in ecological and landscape genomics, and in many ways, studies on trees drove this ﬁeld forward. Populus emerged as the model for tree biology in the 2000’s due to the ease of clonal propagation, manageable genome size (~480Mb), ecological diversity, extensive hybridization, the availability of genomic resources (such as its transformation by Agrobacterium), and its candidacy as a source of cellulosic bioethanol production (DiFazio, Leonardi, et al., 2012). The accumulation of population genetic and genomic studies in Populus has resulted in large gains in our understanding of the genetic architecture of phenology (Keller, N Levsen, Ol- son, et al., 2012; Olson et al., 2013; McKown, Guy, Quamme, et al., 2014; McKown, Guy, Klápöt, et al., 2014; McKown, Klápöt, et al., 2014), wood and biomass growth (Porth, Klapöte, et al., 2013), disease resistance (Porth, Klápöt, et al., 2015;Foster et al., 2015), and many other physiological processes. The Populus community is global and continues to grow in the ﬁelds of biomass production (Zhang et al., 2015), phytoremediation (Di Lonardo et al., 2011), the plant microbiome (Hacquard and

4 Schadt, 2015), and of course, among other ﬁelds, landscape genomics (Evans, GT Slavov, et al., 2014; Geraldes, Farzaneh, et al., 2014; Porth, Klápöt, et al., 2015).

More recently, Quercus sp. are emerging as models for the study of ecological and landscape genomics (Petit, Carlson, et al., 2013). Consisting of over 400 species, oaks are dominant in many Northern Hemisphere forests, and represent a large carbon sink in terrestrial landscapes (Myneni et al., 2001). Many oaks are foundational species that maintain biodiversity by providing habitat and food for numerous plants and animals (Block et al., 1990; Burns and Honkala, 1990; Fralish, 2004; McShea et al., 2007). They are also of high economic value for their lumber (Luppold and Bumgardner, 2008; Espinoza et al., 2011) and ecosystem services that include carbon sequestration, riparian buering and water quality enhancement (Dosskey et al., 1997; Kroeger et al., 2010), improvement of hunting and rangelands (Standiford and Howitt, 1993; Kroeger et al., 2010), improved nutrient cycling (Dahlgren et al., 1997; Herman et al., 2003), and cultural signiﬁcance (Pavlik et al., 1991; Anderson, 2013). In this chapter, we explore the development of landscape genetics and genomics in angiosperm trees and what we have learned from investigating the evolutionary consequences of life as a tree in a heterogeneous landscape. We focus especially on

Populus and Quercus,which have made the greatest strides to date among landscape studies on angiosperm trees. We then highlight several areas where researchers in tree landscape genomics face challenges, or historically have been narrowly focused. Lastly, we oer recommendations for investigating overlooked aspects of tree biology and evolution in landscape genomics studies, and suggest new inquires for future research directions.

5 1.1.1 Historical Review

Landscape genetics: gene ﬂow and population structure

Before the mid-2000’s and the onset of the next generation sequencing (NGS) revolu- tion, landscape genetic studies focused primarily on how genetic diversity of anony- mous markers were shaped by neutral demographic processes, including the eects of gene flow, clonality, isolation-by-distance, range expansion, and population structure (Manel, Schwartz, et al., 2003). Polymorphic loci were rarely associated with adaptive signatures in allele frequency driven by environmental agents of selection (Gonzalez- Martinez et al., 2006), a fact largely due to the type of molecular markers employed (e.g. amplified-length polymorphisms, AFLPs; random amplified polymorphic DNA, RAPDs; and simple sequence repeats, SSRs/microsatellites) and the typically unknown location of these markers in a genome. Nevertheless, these loci were sucient to answer questions regarding demographic processes shaping genetic diversity. Early landscape genetic studies in angiosperm trees focused on both contemporary gene dispersal (VL Sork and Smouse, 2006) as well as larger-scale phylogeographic structure (Petit, Brewer, et al., 2002). Pollen and seed dier by the distances they are dispersed, their dispersal vectors, the genomes they carry, and their ploidy level; traits that profoundly influence spatial patterns of diversity (McCauley, 1995; Petit and Hampe, 2006). Elegant early studies of landscape genetics in white oaks by Petit and colleagues laid the foundation for our understanding of how long-distance seed dispersal creates legacies of genetic structure during range expansion (Le Corre et al., 1997; Petit, Pineau, et al., 1997). In Quercus robur and Q. petraea, Petit et al. (1997) discovered homogenous patches of chloroplast (cpDNA) haplotypes

6 on the landscape, suggesting that rare long-distance dispersal events followed by local population expansion, rather than a continuous wave of migration, structures landscape diversity in cpDNA. Finer-scale understanding of pollen and seed movement came from Sork and colleagues who demonstrated how landscape configuration and land use history shaped the eective number of pollen donors (Nep). Fragmentation and habitat loss lowered Nep in wind-pollinated California valley oaks (Q. lobata) (VL Sork, Davis, et al., 2002), while fragmentation actually facilitated gene flow in the insect-pollinated flowering dogwood (Cornus florida) (VL Sork, Smouse, et al., 2005). Studies in Q. robur and Q. petraea demonstrated that pollen-mediated gene flow occurring over long distances (>200 m) is common, accounting for 67% of matings

(Strei, 1999). Studies in Populus have also informed our understanding of gene flow distances. Dispersal of exotic or transgenic alleles is a concern in hybrid Populus plantations. Carefully designed gene flow studies demonstrated that most pollination and seed establishment occurs near the source tree (within 450 m) (DiFazio, Leonardi, et al., 2012), and pollination occurs less frequently when native tree populations are large. Establishment of exotic or transgenic genes in populations of native species is more likely when native population sizes are small. Removing native female trees near exotic or transgenic plantations may be an eective means of eliminating gene flow in managed landscapes (Meirmans et al., 2010).

Some angiosperm trees (especially in Populus ) are known for extensive clonal reproduction that can strongly inﬂuence local genetic structure (Mock, Rowe, et al., 2008), and once established, a set of clones can dominate an area for long periods

(Imbert and Lefévre, 2003). Early studies in black poplar (Populus nigra) assessed the extent of clonal diversity remaining among two remnant populations along the Rhine

7 River in the Netherlands (Arens et al., 1998). Using SSRs, they identiﬁed 13 distinct clones among the 71 ramets sampled, suggesting extensive clonality. Recent studies of the highly clonal quaking aspen (P. tremuloides) in the western U.S. also used SSRs to identify and map the spatial extent of clones on the landscape (Mock, Rowe, et al., 2008). Mock, Callahan, et al. (2012) was also able to document the widespread existence of triploid aspen clones in western U.S. landscapes, which represent a unique source of genetic diversity relative to the more recently colonized populations in the glaciated northern regions of the continent (Callahan et al., 2013). The molecular estimates of clone age in P. tremuloides , ranging from 175 to 10,000 years (Ally et al., 2008), suggest that once established, Populus clones may shape the genetic diversity of landscapes for millennia. At larger spatial and temporal scales, the Pliocene and Pleistocene glacial cycles had a substantial impact on the genetic diversity of many North American (VL Sork and Smouse, 2006) and European trees (Hewitt, 1999) through waves of range expansion and contraction, although not all species experienced population size changes (Grivet et al., 2006). Post-glacial migration rates from pollen cores predicted fast tree migration rates out of refugia (100-1000 m/yr), but estimates from genetic data are contradictory (<100 m/yr) (McLachlan et al., 2005; Feurdean et al., 2013), and the increasing recognition of high latitude glacial refugia may reﬁne our estimates of tree migration rates downward (Petit, Hu, et al., 2008; Breen et al., 2012). Studies in trees revealed population size changes and genetic isolation in allopatric refugia create the conditions for speciation (Sorrie and AS Weakley, 2001), as well as opportunities for adaptive introgression upon secondary contact (Bragg et al., 2015; Suarez-Gonzalez et al., 2016). For example, range shifts associated with Pleistocene climate change

8 have been implicated in speciation between Populus trichocarpa and P. balsamifera (ND Levsen et al., 2012), as well as driving adaptive introgression during admixture of divergent European populations of P. tremula (De Carvalho et al., 2010). The development of genomic resources in Populus (Tuskan et al., 2006) allowed for genotyping large numbers of individuals at hundreds to thousands of physically mapped loci (Ingvarsson, 2008; Keller, Olson, et al., 2010; GT Slavov et al., 2012). Applications of coalescent demographic modeling and Bayesian clustering to population genomic datasets revealed a general pattern of strong latitudinal gradients of population structure and geographic gradients in genetic diversity (Keller, Olson, et al., 2010; GT Slavov et al., 2012). For example, in P. angustifolia, STRUC- TURE analyses revealed southern, mid-latitude, and northern demes (Evans, Allan, et al., 2015). These confounding eects of population structure (Fig. 4.1a) introduce challenges for ﬁnding adapted loci under environmental selection from variables that covary with latitude (Fig. 4.1b). Hence, understanding the demographic history of a sample became essential for any association genetic study using natural diversity.

Candidate gene studies of local adaptation

Heritable phenotypic clines in phenology, growth, and other physiological traits are widespread in nature (Haldane, 1948; Ingvarsson, García, et al., 2006; Hall, Luquez, et al., 2007; Kempes et al., 2011), suggesting an intimate relationship between adaptive evolution and the landscape (Linhart and Grant, 1996). In temperate and boreal forests trees, locally adapted clines of bud phenology (spring bud ﬂush and late summer bud set) allow trees to coordinate their growth and dormancy with the available growing season, and achieve cold hardiness before low temperatures or drought

9 become damaging (Ingvarsson, García, et al., 2006). These phenotypic clines are the hallmark signatures of adaptation to climate, exhibiting strong heritability and genetically-based population dierentiation, these traits are attractive targets for investigating the genetic basis of local adaptation (Savolainen et al., 2007).

The increased knowledge of gene function from model plants, especially Arabidop- sis, coupled with the development of growing genomic resources in emerging angiosperm model trees (e.g. Populus, Quercus, Eucalyptus, and Castanea;(Neale and Kremer, 2011) facilitated a wave of local adaptation studies in tree bud phenology primarily focused on candidate genes from the ﬂowering time and circadian clock associated pathways. Early eorts to associate adaptive clines in bud phenology with candidate genes were conducted in P. tremula by Ingvarsson and colleagues. PHYB2 senses red to far-red light and regulates several important traits in Arabidopsis, including hypocotyl elongation, meristem development, and ﬂowering-time (Sheehan et al., 2007). It was hypothesized that changes in allele frequency in PHYB2 would associate with clines in bud set driven by latitudinal variation in critical photoperiod. Latitudinal clines in PHYB2 SNPs were detected, and, consistent with local adaptation, paralleled bud set clines; however, analytical methods were not capable of disentangling whether this was due to local selection or neutral processes (Ingvarsson,

García, et al., 2006). Follow-up studies including SNPs ﬂanking the PHYB2 region found allele frequencies did not deviate from neutral expectations, even though two nonsynonymous SNPs in PHYB2 explained 1.5 - 5% of clinal variation in bud set (Ingvarsson, Garcia, et al., 2008). Hall, Luquez, et al. (2007) leveraged the release of the P. trichocarpa genome (Tuskan et al. 2006) to design 18 neutral SSRs (7 near candidate genes) and sequenced 50 SNPs from ﬁve candidate genes (PHYB2,

10 COL2B, GA20OX1, ABI1B, and HYPO312 ) for phenology responses. While phenology traits showed strong quantitative genetic dierentiation among populations

(QST ), candidate SNPs and SSRs near candidate genes did not show elevated FST , nor did the genetic data covary strongly with latitude (Hall, Luquez, et al., 2007).

Ma et al. (2010) broadened the focus to include SNPs from 25 genes across the P. tremula photoperiod pathway. Using FST outlier tests for local adaptation, they were unable to dierentiate adaptive from neutral patterns of population dierentiation, even though several candidate SNPs were associated with phenotypes and showed allele frequency clines with latitude. Using alternative tests for positive selection based on comparing synonymous vs. nonsynonymous mutations (MK test) and intraspe- cific polymorphism vs. interspecific divergence (HKA test), Hall, MA, et al. (2011) did find evidence for adaptive evolution in several photoperiod and circadian clock associated genes, including PHYB2. These early studies in aspen revealed a complex reality that the genetic basis of local adaptation would be dicult to unravel for highly polygenic traits, and helped usher in two important realizations for the field of adaptive landscape genomics: (1) finding genes involved in local adaptation requires testing candidate genes against samples of loci from across the genome using statistical models that account for the confounding eects of demographic history (Platt, Vilhjálmsson, et al., 2010); and (2) deducing the genetic architecture of local adaptation would be a dicult road, as the average eect of a locus on a polygenic trait is small (Rockman, 2012), and selection at the phenotypic level is distributed across many QTLs in the genome (Tin and Ross- Ibarra, 2014). An enduring lesson for landscape genomics thus became limiting false discoveries while searching for local adaptation by developing, testing, and applying

11 methods that explicitly modeled the demographic history of the sample (Lotterhos and Whitlock, 2014). These new analytical methods included tests for clinal gene- environment associations (GEA), supported by programs such as Bayenv (Günther and Coop, 2013) and latent-factor mixed models (LFMM) (Frichot et al., 2013), as well as FST outlier programs like BayeScan (Foll and Gaggiotti, 2008) that detected diversifying selection against the background of neutral population structure.

In a study of local adaptation for phenology in P. balsamifera, Keller, N Levsen, Olson, et al. (2012) used 412 putatively neutral reference SNPs to model eects of demographic history and tested for local adaptation in 27 candidate genes in the

flowering-time pathway (Fig. 4.1c, d). FST outlier tests and GEA convergently identified 11 candidate SNPs that had signatures of excessive population structure or association with climate. Interestingly, phytochromes (e.g. PHYA, PHYB1, PHYB2 ) showed no signs of local adaptation in P. balsamifera despite their signature of adaptation along a latitudinal gradient in the related P. tremula (Ingvarsson, García, et al., 2006; Ma et al., 2010), nor did phytochromes show evidence of species-wide positive selection, as in P. tremula (Hall, MA, et al., 2011; Keller, N Levsen, Ingvars- son, et al., 2011). Instead, local selection in P. balsamifera was primarily targeted at the circadian clock controlled gene GIGANTEA (GI ) (Oliverio et al., 2007), a master regulator of multiple developmental processes such as germination, flowering, and osmotic stress (Mishra and Panigrahi, 2015). GI is known to mediate day length signaling from the phytochromes to regulate expression of the genes CONSTANS and FLOWERING LOCUS T involved in flowering and vegetative phenology in Populus (Böhlenius et al., 2006). Thus, an important comparative insight from these candidate gene studies was that convergent phenotypic clines in related species may result

12 from selection targeting the same overall functional pathway (e.g. light-regulated circadian responses) but involving dierent genes within the pathway. Indeed, this trend also appears to be true for the evolution of sex-determination in Populus section Populus and Populus section Tacamahaca despite the ancient origin of dioecy in the genus (Geraldes, Hefer, et al., 2015).

More recently, Menon et al. (2015) took advantage of P. balsamifera’s extensive distribution across boreal and arctic latitudes to study natural variation in adaptation to freeze tolerance. Menon et al. combined physiological assays of freeze tolerance using electrolyte leakage tests with gene expression and population genetic analyses on candidate genes in the poplar C-repeat binding factor (CBF) gene family. Their ﬁndings indicated that northern genotypes were physiologically more cold tolerant than southern genotypes, and that CBF gene expression was strongly induced by cold in balsam poplar. One gene (CBF2 ) also showed population genetic evidence of selection in the form of an excess of derived alleles, elevated FST , and clinal allele frequency variation. While transcriptional regulation and sequence polymorphisms in

CBF genes were clearly associated with cold adaptation, CBF candidate genes only explain a small fraction of cold-induced changes in gene expression, suggesting the involvement of additional pathways.

Outside of Populus, investigations of the genomics of local adaptation in Quercus and other Fagaceae were underway by the mid-2000’s. Foundational work studying QTLs and gene expression of various tissues identiﬁed a number of candidate genes involved in landscape variation in bud phenology and drought response (Porth, Koch, et al., 2005; Casasoli et al., 2006; Derory, Léger, et al., 2006; Derory, Scotti- Saintagne, et al., 2010; Soler et al., 2008). These studies spurred a growing body

13 of work in Quercus, and provided an important resource for comparison to Populus and other taxa. In the European oak (Q. petrea), FJ Alberto et al. (2013) integrated

FST -outlier, GEA, and genotype-phenotype association tests based on SNPs from 106 phenology-related candidate genes to test for molecular signatures of local adaptation along altitudinal and latitudinal gradients. While 34 genes were significant in at least one of the tests, only 4 were identified in more than one: endochitinase class I, seed maturation protein, ribosomal protein L18a, and plastocyanin A. Some candidate genes were significant in both altitudinal and latitudinal gradients, suggesting convergent selection at dierent spatial scales (ribosomal protein L18a, GI, Photosystem II polypeptide, 5’-adenylylsulfate reductase, ribosomal protein S11). VL Sork, Squire, et al. (2016) also started with candidate genes from other studies and combined multiple tests for signatures of local adaptation in Q. lobata in Califor- nia, including FST -outlier and univariate and multivariate environmental association tests. Five of seven genes with significant SNPs were significant in more than one test: auxin-repressed protein and –-amylase/subtilisin inhibition, which are involved in bud phenology; and heat shock protein 17.4 and elongation factor 1-–, which are involved in temperature stress response. The FST values of outliers in Q. lobata are substantially higher than those in Q. petraea, suggesting that dierences in population history and climate heterogeneity may lead to very dierent degrees and patterns of local adaptation. Furthermore, several candidate genes for cold hardiness in Pseu- dotsuga menziesii (Krutovsky and Neale, 2005) and bud phenology in European oaks (Casasoli et al., 2006; Derory, Scotti-Saintagne, et al., 2010) also show evidence of local adaptation in the Californian oak, Q. lobata (VL Sork, Squire, et al., 2016). For the most part, however, candidate genes identified across a limited number of stud-

14 ies were dierent, leaving it unclear how common convergent adaptation is among species and between angiosperm clades (Martins et al., 2018; FJ Alberto et al., 2013; VL Sork, Squire, et al., 2016; Gugger, Cokus, et al., 2016).

The recent publication of the Eucalyptus grandis genome (Myburg et al., 2014) promises to be a useful tool to study adaptive evolution in the species rich and ecologically diverse Eucalyptus. For example, E. pauciﬂora occurs from sea-level to tree-line and seems like a particularly well-suited species for studying adaptation to temperature, cold hardiness, and drought stress. Adaptation studies in Eucalyptus are in the nascent stages (Dillon et al., 2015), but some traits, e.g. adaptation to aridity (Steane et al., 2014), are better understood. A recent candidate gene study of local adaptation in E. camaldulensis identiﬁed associations between climate (especially temperature and evaporative demand) and two genes – ERECTA (an LRR receptor- like kinase) and PIP2 (an aquaporin), both of which also showed excess population dierentiation in FST outlier tests (Dillon et al., 2015). Additional work in E. tricarpa has uncovered extensive evidence of phenotypic plasticity and local adaptation for ecophysiological traits related to drought avoidance, although no associations with molecular data were sought (McLean et al., 2014). We eagerly await further results from landscape genomic studies in Eucalyptus that expand on our current knowledge of candidate genes involved in local adaptation to environmental gradients, especially drought and other forms of osmotic stress, and facilitate comparison with the existing body of work in northern hemisphere forest trees.

15 Genome-wide studies of local adaptation

While candidate gene studies show promise for understanding local adaptation on the landscape, they are still biased by the choice of genes to investigate. The development of new SNP arrays (Geraldes, Difazio, et al., 2013) allowed for genotyping tens of thousands of loci across the genome, while whole genome, exome, or transcriptome re-sequencing studies enabled testing millions of SNPs and identiﬁcation of locally adapted genomic regions without any a priori speciﬁcation of the number or types of genes involved (Evans, Allan, et al., 2015; Holliday et al., 2016). To date, genome- wide association studies (GWAS) have received the most attention in P. trichocarpa (McKown, Guy, Klápöt, et al., 2014; McKown, Klápöt, et al., 2014; McKown, Guy, Quamme, et al., 2014; Zhou et al., 2014; Porth, Klapöte, et al., 2013; Evans, GT Slavov, et al., 2014; Holliday et al., 2016). Loci associated with adaptation to climate (Evans, GT Slavov, et al., 2014), aridity (Steane et al., 2014), adaptation across latitude and elevational gradients (Holliday et al., 2016), and associations with adaptive ecophysiological traits (McKown, Guy, Klápöt, et al., 2014) have been discovered. It is interesting to consider the repeatability of the genomic regions implicated by local adaptation scans in P. trichocarpa, given that they have involved dierent research labs, independent germplasm collections across a similar landscape gradient, and dierent statistical approaches to testing for adaptation. We compared local adaptation outliers from four studies in P. trichocarpa based on a range-wide sampling: two studies (McKown, Klápöt, et al., 2014; Geraldes, Farzaneh, et al., 2014) generated genomic data from genes discovered in the Geraldes, Difazio, et al. (2013) SNP array, while Holliday et al. (2016) sequenced whole exomes and Evans,

16 GT Slavov, et al. (2014) sequenced whole genomes. Gene annotations from Geraldes,

Farzaneh, et al. (2014) were updated from the P. trichocarpa genome version 2.2 to 3.0. Our comparisons were based on unique P. trichocarpa gene IDs identified as locally adapted in each study. These include 579 genes from Geraldes et al. (2014), 276 from McKown, Klápöt, et al. (2014c), 1160 from Holliday et al. (2016), and 452 from Evans et al. (2015) (Fig. 1.2). No gene was identified as a selection outlier by all four studies, however six genes were identified as outliers in common across combinations of three studies (Table 1.1). Two of these shared outliers were transcription factors, while a third was a methyltransferase, possibly indicating an important role for transcriptional and epigenetic regulation of gene expression in local adaptation. One gene, Potri.010G165700, a MYB-like DNA binding protein, showed upregulated expression in dormant and pre-chilling vegetative and floral buds (www.popgenie.org), a profile supportive of this gene’s association with bud phenology in GWAS (McKown et al. 2014c). New candidates for local adaptation have also been found using whole-transcriptome sequences (ignoring expression) from natural populations to associate coding SNPs with environmental variation on the landscape. Using this approach in 22 Q. lobata individuals produced 220,427 SNPs for analysis (Gugger, Cokus, et al., 2016). Of these, 79 SNPs from 50 genes were significantly associated with climate gradients, especially minimum temperature and precipitation variables, and most of these genes were dierentially expressed after drought stress (Gugger, Peñaloza-Ramírez, et al., 2017), suggesting they are likely involved in local adaptation. A few SNPs are in genes with known roles in response to stimulus or stress (e.g., SAUR-like auxin-responsive protein family), cold shock protein binding (zinc knuckle family protein), light re-

17 sponse or photosynthesis (e.g., cryptochrome 1), and trichome development (Myosin family protein with Dil domain) in Arabidopsis (Swarbreck et al., 2007). Furthermore, the 50 candidate genes have elevated levels of genetic diversity, consistent with balancing selection or other mechanisms favoring dierent alleles in dierent contexts.

In a similar population RNASeq study in Populus balsamifera, transcriptome-wide gene expression was compared between northern and southern genotypes grown under common garden conditions (Wang et al., 2014). A total of 4,018 of 32,466 genes showed dierential expression between northern and southern genotypes. Genes that were more highly expressed in southern genotypes included homologs of the Ara- bidopsis ﬂowering time pathway (EARLY FLOWERING 3, FLOWERING LOCUS C, PHYA and CAULIFLOWER). However, unlike in the Quercus study of Gugger, S Fitz-Gibbon, et al. (2016b), the dierentially expressed genes did not harbor SNPs that showed signatures of local selection (based on F ST or GEA outlier analyses), suggesting that expression divergence is largely uncoupled from nucleotide divergence in

P. balsamifera. While the underlying goal of these studies is to identify and understand patterns of adaptive genomic evolution, they also simultaneously address fundamental questions of how genomic diversity is shaped in geographic space. For example, Geraldes et al. (2014), found that diversifying selection along a latitudinal gradient principally was enriched for phenology and photoperiod related genes. At the same time, they addressed a classic landscape genetic question about geographic isolation along riparian zones by demonstrating that populations in connected drainages are more genetically similar than between drainages. Similarly, Holliday et al. (2016) identiﬁed genes associated with climatic dierences in growing season length by testing whether

18 adaptation involves selection targeting the same genomic regions between latitudinal and altitudinal gradients in P. trichocarpa. Their finding of significant overlap in the outlier regions associated with latitude and altitude provided rare support for the hypothesis of parallel adaptation along similar environmental gradients in dierent geographic regions of the species range. Merging the unique biology of angiosperm trees with the landscape genetics/genomics program has generated a deep understanding for the connection between landscapes, biodiversity, and the evolutionary processes of gene flow, genetic drift, and natural selection. The initial goals of the field, as outlined by Manel, Schwartz, et al. (2003), continue to be relevant and guide research today. Despite analytical hurdles that still exist, applications of new sequencing technologies combined with more refined analysis methods are generating exciting discoveries.

1.1.2 Challenges and Solutions for Landscape

Studies in Trees

Many landscape genomic studies have borrowed sampling schemes from phylogeography, the study of population lineages across large (e.g. continental) landscapes (Fig. 4.1a). This approach has yielded much information about how genetic and phenotypic variation aligns along environmental gradients (e.g. photoperiod, temperature, and precipitation) that vary across large geographic regions. However, range-wide sampling can produce methodological challenges, false positive associations due to uncontrolled population structure, and biases in our expectations of which genomic regions are locally adapted. In this section, we review some of the challenges currently

19 facing landscape genomics and highlight potential solutions.

Collinear environmental, phenotypic, and genotypic axes

The biological characteristics of trees that make them good candidates for landscape and ecological genomic studies also generate challenges for studies of natural variation that require clever study designs and/or novel methodologies to solve. The collinear variation of environmental, phenotypic, and genomic variation across landscapes is a major challenge to overcome (Fig. 4.1a, b). For many temperate and boreal species, the structure of neutral loci following population expansion from glacial refugia mim- ics the signal of adaptive selection searched for by association methods (Lotterhos and Whitlock, 2014). Disentangling false positives arising from demographic history against true positives arising from selection creates a problem for treating landscape samples as ‘natural experiments’. Algorithms accounting for demography in a sample of loci have been developed to reduce the false-discovery rate (FDR) due to population structure, while maintaining power to avoid false negatives (WY Yang et al., 2012; Günther and Coop, 2013; Fri- chot et al., 2013; Berg and Coop, 2014). However, deciding where to place a cut-o value for signiﬁcance tests remains dicult, given the very large number of tests in most population genomic datasets. Recently, it was recommended that researchers create empirical p-values from test statistic distributions derived from putatively neutral loci (Lotterhos and Whitlock, 2014). While it is challenging to know what kind of locus is truly neutral (e.g. the promoter for TB1, a gene in the QTL that reorga- nizes teosinte-like to maize-like branching, is 41 kb upstream of the gene; Clark et al.,

2006), selecting intergenic loci and conditioning p-values of candidate loci from them

20 should decrease the FDR markedly (Lotterhos and Whitlock, 2014). Additionally, sampling individuals from regions of the landscape that have experienced the same demographic history and have low FST can greatly minimize false positives generated by demography, particularly if covariation of the ecological and environmental variables of interest is present. Thus, future studies on the genomics of local adaptation will beneﬁt from careful consideration of sampling design and generation of appropriate null expectations, both in terms of individuals on the landscape and SNPs across the genome.

Validating adaptive associations discovered in natural populations

Local adaptation studies are generating many associations increasing our knowledge of gene functions (McKown, Klápöt, et al., 2014); however, attempts to validate these loci using independent evidence is still limited (∆aliÊet al., 2016). For the most part, the candidate genes identiﬁed across even the most robust studies are dierent.

Our comparison of four local adaptation studies in P. trichocarpa (McKown, Klápöt, et al., 2014; Evans, GT Slavov, et al., 2014; Geraldes, Farzaneh, et al., 2014; Holliday et al., 2016) using broadly overlapping genomic data sets ﬁnd few loci in common (Fig. 1.2). What portion of the many associations not shared between studies result from false positives, dierences in the biology of the samples, or methodological approaches employed? This brings into focus the growing need for functional validation of adaptation outliers. Functional validation of genes identiﬁed from GWAS can be dicult, however, when successful, the results are compelling (e.g. Meijón et al., 2014). Genetic approaches to gene validation in trees is possible (Kim et al., 2011), and the advent

21 of CRISPR/Cas9 make editing putatively adaptive and non-adaptive alleles of candidate genes even more feasible (Fan et al., 2015; Bono et al., 2015). Thus, we are optimistic that validation of the most promising candidates using transgenic methods will be an increasing trend in the future. Another validation approach for local adaptation studies is to replicate the association, for example by testing the phenotypic eect of the candidate locus segregating within a pedigreed mapping family (Stinchcombe and Hoekstra, 2008). Trees typically have long generation times and waiting upwards of 4-8 years (for example in

Populus, DiFazio, GT Slavov, et al., 2011) for selected lines to reach reproductive maturity is not uncommon, and producing advanced generation mapping families is dicult. Nevertheless, concerted eorts have produced a handful of advanced generation pedigrees for linkage trait mapping in angiosperm trees in Populus (Taylor, 2002) and several members of the Fagaceae, notably Quercus, Fagus, and Castanea (Kremer et al., 2012). Long-term mapping populations comprised of hundreds of progeny are now available for Q. alba at the University of Tennessee, Knoxville and University of Missouri Center for Agroforesty (http://hardwoodgenomics.org/), as well as for Q. robur in Europe (Bodénès et al., 2012; Plomion et al., 2016). However, researchers remain limited when established mapping families do not segregate for the alleles or traits of interest. Nature, however, has provided additional solutions to some of these problems, particularly in taxa that extensively hybridize with congeners, allowing for strategic admixture mapping from natural hybrid zones.

In these zones, naturally occurring F1s can be selected to produce F2 or backcross recombinant generations. In Populus, a well known tri-hybrid zone exists in southern Alberta between P. balsamifera, P. deltoides, and P. angustifolia (Floate, 2004).

22 Aided by diagnostic SNP genotyping arrays that identify Populus species and their hybrids (Isabel et al., 2013), hybrid F1s as well as every combination of backcrossed

F2s can be found there (Floate et al., 2016). Hybrid zones between P. balsamifera and P. trichocarpa are also well known (Geraldes, Farzaneh, et al., 2014), and in Wyoming and Colorado another tri-hybrid zone has been observed between P. balsamifera, P. trichocarpa, and P. angustifolia (Chhatre et al., 2018). Using the diverse array of interspeciﬁc hybrids as parents in targeted crosses has the potential to accelerate the time to production of advanced generation mapping families.

Investment in phenotyping

The exponential decline of genotyping costs (Manel, Perrier, et al., 2016) has shifted the data collection bottleneck in ecological and landscape genomic studies towards phenotyping. Phenotyping traits representative of the diversity of developmental stages and tissue types can be dicult for trees, given their large size, delayed reproductive maturity, massive root systems, and the high cost of establishing and maintaining long-term common gardens. Increased attention is being devoted to developing standardized methods for high-throughput phenotyping (Brown et al., 2014), and automated leaf morphology software developed for Populus (Bylesjö et al., 2008) has been successfully applied to other species (IvetiÊet al., 2014). Similarly, researchers are beginning to investigate the complexity and adaptive contribution of below-ground tissues (Madritch et al., 2014) and their symbiotic interactions (Hac- quard and Schadt, 2015). However, renewed attention needs to be paid to funding and maintaining common gardens, sharing and integrating data across gardens for purposes of meta-analysis, and innovating new methods and approaches to capturing

23 phenotypes from common gardens and natural landscapes. Common gardens play a critical role in landscape genomic studies and connect genotype to phenotype to environment (Porth, Klapöte, et al., 2013; Evans, GT Slavov, et al., 2014). Gardens with diverse landscape-scale sampling have been established in Populus, including extensive collections of P. trichocarpa by dierent research groups (448 individuals from McKown et al. 2014c, >1,000 individuals from Evans et al. 2015, and 391 individuals from Holliday et al. 2016), as well as the Ag-

CanBap collection of P. balsamifera (525 individuals; Soolanayakanahally, Guy, Silim, Drewes, et al., 2009; Soolanayakanahally, Guy, Silim, and Song, 2013), the Swedish

Aspen collection (SwAsp) of P. tremula (116 individuals; Luquez et al., 2008), collections of P. fremontii, P. angustifolia, and their hybrids (hundreds of individuals; Holeski et al., 2013), and the Chinese P. tomentosa collection (460 individuals; Du et al., 2014). In oaks, a comprehensive provenance trial for Q. lobata was established in 2013 at two sites from seeds sourced from 95 locations throughout its distribution (Delﬁno-Mix et al., 2015). Funding cycles, however, are too short to support the long-term maintenance needed for tree studies, and diverse collections of provenance trials established in the past century have been abandoned and overlooked. Few trials are still maintained, but interest is rebounding in taking advantage of these historical resources. For example, a mature provenance trial of approximately 300 trees from

30 populations sampled across the range of the tuliptree (Liriodendron tulipifera)was planted in the mid-1980’s in Chapel Hill, NC, but was orphaned for much of its existence (Fetter and A Weakley, 2019). Older, but smaller, trials are also available for

Q. alba , including one comprised of 23-year-old trees from 17 populations planted in three locations in Indiana (Huang et al., 2015).

24 While the challenge of establishing and maintaining common gardens will likely remain in landscape genomics, we need to start thinking more strategically about how to use and combine these resources, and where applicable, study tree phenotypes in natura. Aggregating trait data from many common gardens would prove useful for meta-analyses of adaptive phenotypic trends across spatial scales, times, and even closely related species. Comparative work of this type has begun in Pop- ulus (Soolanayakanahally, Guy, Street, et al., 2015), but more awaits. Finding new opportunities to collaborate and integrate data across research labs hosting common gardens is needed, and tree biologists should target funding mechanisms to build research coordination networks around common garden resources. Landscape genomics would also benefit from testing the transferability of genotype- phenotype associations from common gardens to the field. The challenge of landscape phenomics will define the next generation of studies that seek to close the genotype- phenotype-environment triangle of adaptive association. The hurdles, while logisti- cally challenging, are largely technological, and progress in this area is essential to make meaningful applications of tree management and conservation of genetic resources under environmental change (Pettorelli et al., 2014). Exciting advances in remote sensing are now being applied to phenotyping forest trees at scales of clonal stands to entire landscapes (Homolova et al., 2013). For example, remote sensing of seasonal phenology is being deployed at fine (Toomey et al., 2015) to continental (Graham et al., 2010) scales, and satellite-based phenology products such as MODIS provide coarse landscape-scale phenotyping of phenology that correlates with ground- based measurements in Populus (Elmore et al., 2016). This opens doors to connecting landscape variation in forest phenology with adaptive gradients in allele frequencies

25 in genomic regions associated with phenology under controlled experimental conditions. At finer scales of individual trees, measurement of several key phenotypes (e.g. spring bud flush and plant water balance) may be possible through use of field deployed sensors that detect dierences in tree mass using acceleration or reflectance using light-emitting detectors (e.g., Kleinknecht et al., 2015). While still in need of further development, landscape-scale phenotyping holds the potential to connect adaptive genomic variation to forest productivity in natural stands, and is a challenge that future studies will surely begin to address.

1.1.3 Future Research Directions for Forest Tree

Landscape Genomics

Reﬁnements in genomic sequencing, association methods accounting for demography, and inclusion of broader evolutionary questions has driven the steady increase of the impact of landscape genomic studies using angiosperm trees. We make the following recommendations and predictions for the future of the ﬁeld:

Growing genomic resources outside of Populus

Genomic resources in Populus are now well-developed and easily accessible (e.g. www.popgenie.org), but are still lacking for many other angiosperm trees. Increasing genomic resources in other taxa will allow for more comparative studies that will increase the impact of what is known from Populus (see Sollars and Buggs, this volume). Genomic resources in Quercus are steadily increasing and are most developed for the white oaks (Quercus section Quercus). Recently, the ﬁrst draft oak genome,

26 based on a combination of Roche 454, Illumina, and Sanger sequences, was announced for the ecologically and economically important European white oak, Quercus robur (Plomion et al., 2016). The genome contains 18 k scaolds (>2 kb) totaling about 1.3 Gb, which is about 1.7 times the haploid genome size (Kremer et al., 2012). To enable research in ecological genomics, there are several important improvements underway. These include the development of a genome sequence with collapsed haplotypes to facilitate SNP calling, further assembly that utilizes linkage maps and additional sequencing to determine gene order, and structural and functional annotation (Plomion et al., 2016) that will leverage abundant expressed sequence tags (EST) and RNA-Seq libraries from a variety of tissues (Ueno et al., 2010; Tarkka et al., 2013; Lesur et al., 2015).

A draft oak genome is now available for Q. lobata (VL Sork, ST Fitz-Gibbon, et al., 2016), a threatened western North American species with a population history that stands in contrast to many studied trees (Grivet et al., 2006); entirely on Illumina sequencing, two complementary assemblies were produced. The ﬁrst (v0.5) aggressively collapses haplotypes to optimize its use for SNP calling (Gugger, S Fitz- Gibbon, et al., 2016). The second (v1.0) retains haplotypes and is composed of 18.5 k scaolds (> 2 kb) with total length of 1.15 Gb (VL Sork, ST Fitz-Gibbon, et al.,

2016). This latter assembly is similar in both size and sequence to that of Q. robur , with 93% sequence identity and 97% of scaolds reciprocally aligning. Functional and structural annotations are derived from an annotated reference transcriptome assembly for Q. lobata (Cokus et al., 2015) in addition tode novo annotations. Other resources for eastern North American trees are available from the Hardwood Genomics Project (http://www.hardwoodgenomics.org/) and from the Fagaceae Ge-

27 nomics Web (http://www.fagaceae.org) for many important tree species, including oaks, chestnuts, sweet gum, beech, and tuliptrees, among others. These include EST libraries and transcriptome assemblies for Q. alba and Q. rubra that were generated from a variety of tissues and sequencing platforms, as well as a low-coverage genome assembly for Q. alba that was constructed for identifying SSRs. Early analyses suggest high similarity with other oaks and broad synteny within Fagaceae (Kremer et al., 2012; Bodénès et al., 2012; Cokus et al., 2015). Along with the development of sequence resources, there has been an explosion in the number of SNPs and SSRs available. Over one million SNPs were discovered in

California oaks, especially Q. lobata (Cokus et al., 2015; Gugger, Cokus, et al., 2016). Tens of thousands of SNPs are known in Q. robur (Ueno et al., 2010), and recently, a cross-species genotyping array with over 7k SNPs was developed for European oaks

(Lepoittevin et al., 2015). In addition, thousands of SSRs have been identiﬁed in Q. alba and Q. robur (Durand et al., 2010; Bodénès et al., 2012).

Incorporating gene expression studies

Advances in transcriptomics (JH Lee et al., 2015) and isoform detection (Bernard et al., 2015) are making it possible to ask new questions about the genomic mechanisms of local adaptation and the evolution of phenotypes. Integrating gene expression into selection scans can provide complementary information to SNP genotypes when con- ducting selection scans and GWAS. Experimental designs that analyze dierential expression dependent on both treatment and source population can produce excellent candidates for inferring local adaptation. In a small greenhouse experiment that exposed Q. lobata seedlings from three source populations to a drought treatment,

28 whole-transcriptome gene expression changes identified 56 genes that had significant interaction terms (Gugger, Peñaloza-Ramírez, et al., 2017). These genes are primarily involved in metabolic processes, stress response, transport/transfer of molecules, signaling, and transcription regulation. Given the prevalence of molecules involved in metabolism, it was hypothesized that local adaptation to drought in oaks may involve dierent abilities to metabolize or dierent ways of altering metabolic activities in the face of drought. Dierentially expressed genes or isoforms may experience dierent forms of selection (Hartfield, 2016). Some of these expression dierences may be linked to sex function in dioecious species or monoecious species with imperfect flowers (Geraldes, Hefer, et al., 2015). Analysis of dierential expression during the development of imperfect flowers in Betula revealed a transcription factor, BPMADS2, involved in specifying petal and stamen identity and is expressed only in developing male flowers (Järvinen et al., 2003). Local adaptation in the expression of this transcription factor may allow for precise timing of male flower development.

Testing the role of epigenetics in local adaptation

NGS is enabling research on the potential role of epigenetic mechanisms of local adaptation in trees. Cytosine methylation varies among individuals and populations (Schmitz et al., 2013). It can arise through random mutation (Becker, Hagmann, et al., 2011), be induced by the environment (Verhoeven et al., 2010), can be stably inherited (Becker and Weigel, 2012), and may be an additional source of variability for local adaptation. Early evidence in oaks shows that single methylation variants (SMVs) in CpG contexts are exceptionally dierentiated among populations (Platt,

29 Gugger, et al., 2015). In addition, CpG SMVs have strong associations with climate and speciﬁc variants show evidence of local adaptation to temperature stress (Gugger, Cokus, et al., 2016). Therefore, it is hypothesized that either CpG SMVs are markers for loci involved in local adaptation or they are under divergent natural selection themselves. Additional research is needed to better understand the mutation process, assess heritability, and link methylation to phenotypes.

Genomic predictions of climate change and local adaptation

Understanding the genomics of local adaptation to current environmental conditions allows us to ask how populations might respond when climates change (V Sork et al., 2013; Wang et al., 2014; Fitzpatrick and Keller, 2015). For example, research in

Arabidopsis thaliana suggests that the ﬁtness of populations under novel conditions can be predicted from SNPs in genes involved in local adaptation (Hancock et al., 2011). 2011). Prediction of ﬁtness from SNPs might be powerfully applied to tree genomics in studies that pair genotypes with growth traits from common garden trials. Furthermore, the future genetic composition of populations under climate change can be modeled, and the degree of mismatch between current and projected genomic composition can be considered in management planning (Fitzpatrick and Keller, 2015). Identifying adaptive genotypes under projected climate change may be critical for conservation of biodiversity and economically important genotypes, core arguments in favor of assisted migration (Aitken and Bemmels, 2016).

30 Assessing the prevalence of adaptive introgression

Oaks, poplars, and many other angiosperm trees are notorious hybridizers with large eective population sizes readily capable of generating adaptive molecular variation. An important question for tree genomics is how species are able to maintain their ecological, morphological, and genomic identities in the face of widespread hybridization, and whether hybridization contributes actively to adaptation in hybrid zones and outside of them. Genomic tools oer unprecedented opportunities to investigate the genes relevant to speciation (Wu 2001; Bawa and Holliday, this volume), whether hybridization facilitates the transfer of beneﬁcial alleles across species boundaries (Suarez-Gonzalez et al., 2016), the ecological factors that promote or restrict hybridization (Ortego et al., 2014), and how important these processes are for evolution in trees (Gailing 2014, Gailing and Curtu 2014). Evidence from European poplars suggests that advanced generation hybrids accumulate genetic incompatibilities that not only decrease their ﬁtness, but are frequently lethal (Christe et al., 2016). Nevertheless, unidirectional or biased introgression of genes occurs frequently in angiosperm trees (Thompson et al., 2010; Geraldes, Farzaneh, et al., 2014) and fragments of nDNA can introgress across hybrid zones

(Martinsen et al., 2001). In P. tremula , adaptive introgression from Russian to Swedish populations may have contributed to adaptive shifts in the phenology of Swedish trees (De Carvalho et al., 2010). Similarly, the high rate of hybridization in oaks has long been a topic of interest (CH Muller, 1952), and has inﬂuenced the development of the ecological species concept (Van Valen, 1976). One promising area for future study is to investigate the role of introgression in responses to pathogens or other sources of plant stress (e.g., cold, drought, soil-metal

31 tolerance). What role does adaptive introgression play when interspeciﬁc populations meet in an environment where hybridization increases pathogen or herbivore pressures (Holeski et al., 2013; Busby et al., 2013)? Do genetic incompatibilities accumulate in hybridizing populations that limit the ability for adaptive immune responses to track the movement of pathogens/herbivores as they ’step’ across species boundaries

(Floate et al., 2016)? Some evidence from Populus protease inhibitor genes hints that introgression may accompany adaptive responses to herbivore enemies (Neiman et al., 2009). When multiple hybrid zones exist between the same set of species, does unidirectional gene ﬂow create a ’sink species’, biasing the pattern of introgression, or does local selection play a dominant role regulating which genomic regions preferentially introgressed? Geraldes, Farzaneh, et al. (2014) hypothesize that hybridizing populations of P. trichocarpa and P. balsamifera in Alaska may be adaptively introgressing cold tolerance or drought tolerance genes that would experience negative selection in the southern hybrid zone. Trees seem to be ideal systems studying the interaction between hybridization and natural selection.

Seeking out understudied systems and questions

The studies we have discussed oer initial insights into how tree genomes evolve in response to ecological selection pressures, but much work remains to thoroughly understand the molecular basis of local adaptation. Much of the research to date has emphasized local adaptation with regard to bud phenology, growing season length, and other sources of abiotic stress. Biotic interactions and non-climate-related environmental adaptation have received far less attention in landscape studies, particularly in the area of pathogen responses, which is an area of critical importance for

32 human economies and culture (e.g. coee trees and leaf rust, Avelino et al., 2012; the banana and Panama disease, Ordonez et al., 2015). Thus, additional research is needed to understand local adaptation in a broader array of contexts, especially those focused on co-evolution and reciprocal selection driven by species interactions (Whitham et al., 2008). Trees, as foundational species that structure biodiversity, are ideal subjects for community-level genomic studies. We must also expand our geographic and taxonomic focus to include understudied regions, environments, and taxa. A quick survey of the angiosperm (and non- angiosperm) tree landscape genetics literature reveals a strong geographical bias for boreal forest, and, increasingly, temperate forest trees. Although trees from these regions hold much of the Earth’s biomass, a broader taxonomic and geographic sampling of forest trees is needed to understand the diversity of adaptation in natural landscapes. Tropical forests harbor incredible ecological diversity (Gentry, 1988) and the evolutionary ﬁndings from boreal and temperate trees regarding the selective environments and the genomic pathways of adaptation in trees may be the exception, not the rule. Negative density dependent selection, principally from fungal pathogens (Augspurger, 1984), acts to keep tropical tree diversity high and abundance low (Bagchi et al., 2014). What eect does negative density dependent selection from pests and pathogens have on patterns of adaptive genomic diversity in tropical trees? How is diversifying selection for genes involved in an immune response achieved in the face of limited gene ﬂow and low abundance (Hamilton, 1999)? Are large eective population sizes sucient to maintain the supply of standing variation and new mutations required for trees to keep pace in the arms race against pathogens and herbivores? Are the spatial scales of local adaptation to pathogens small, moderate,

33 or large in tropical forests?

Revisiting the scale and scope of landscape genomic studies

Understanding the genetic determinants of locally adaptive clines is a long-standing goal of landscape genetics (Manel, Schwartz, et al., 2003). The large-scale, often range-wide sampling schemes used to investigate clines have become standard in the tree genomics literature. Multiple studies have discovered adaptive evolution of genes associated with phenology clines (Keller, N Levsen, Olson, et al., 2012; McKown, Klápöt, et al., 2014; Evans, GT Slavov, et al., 2014; Geraldes, Farzaneh, et al., 2014), and indeed, given the high broad-sense heritability of phenology traits (McK- own, Guy, Klápöt, et al., 2014) and their direct connection to ﬁtness, one expects strong local selection. Additional ecological dimensions to local adaptation should receive greater focus, including dierent soil chemistries and biotic interactions that vary across landscapes. A modiﬁed scope and sampling design is needed to detect local adaptation that does not vary clinally. Designing studies to more precisely isolate dierent environmental gradients can help disassociate correlated predictors that are common in range-wide studies. For example, temperature and photoperiod both typically covary with latitude, and a researcher interested in temperature adaptation may have trouble interpreting results from a sampling design oriented along a north-south gradient. However, a sample from dierent elevations along a gradient, while keeping latitude constant, will remove eects of photoperiod. Similar designs have been employed recently to detect parallel adaptation to climate across altitudinal transects in P. trichocarpa (Holliday et al., 2016). Smaller scale, targeted landscape genomic studies have much to oer and can unravel new dimensions of local adaptation.

34 1.2 Conclusions

The challenge of understanding genomic variation responsible for adaptive phenotypes on the landscape has been fully engaged for more than a decade by scientists working with angiosperm forest trees. Sequencing and methodological advances have enabled a deeper understanding of neutral and adaptive variation, but the field stands short of proving causative relationships between genotype, phenotype, and the environment. The breadth of evolutionary questions continues to grow, as does the taxonomic sampling, and the field is expanding the availability of genomic resources for a broader array of economically and ecologically important forest trees. Identifying unique geographic hotspots of genomic diversity on landscapes, even before we fully understand their adaptive significance, is a valuable contribution to conservation of tree germplasm during climate change. However, careful consideration of research questions and the sampling design most likely to yield interpretable results is critical and will allow for more diverse studies across the many ecological dimensions of local adaptation. Angiosperm trees will remain important models for studies of adaptive evolution in the coming years, and the combination of landscape genomics with validation methods have the potential to make significant contributions to conservation, sustainable development, and our understanding of the natural world.

1.3 Acknowledgements

Support for this work was provided by National Science Foundation (NSF) awards IOS-1461868 to Keller and IOS-1444611 to Gugger. In addition, we appreciate the

35 thoughtful comments of the editors, Andrew Groover and Quentin Cronk.

1.3.1 Author Contributions

KCF and SRK conceived the scope of the work. KCF performed the statistical analyses, wrote, and edited the manuscript. SRK provided extensive edits of the manuscript. PFG wrote the text speciﬁc to Quercus.

36 1.4 Tables

Table 1.1: Annotations for six genes implicated in local adaptation from range-wide samples of Populus trichocarpa by four separate studies. (M) McKown et al. (2014c), (G) Geraldes et al. (2014), (H) Holliday et al. (2016), (E) Evans et al. (2014). Annotations retrieved from www.popgenie.org.

Gene model Arabidopsis homolog Gene function Overlap Potri.010G212900 AT5G13560 NA M, H, E Potri.005G073000 AT5G65270.1 GTP binding G, M, E Potri.006G257000 NA membrane protein G, M, E Potri.010G165700 AT3G01140 transcription factor G, M, E Potri.006G225600 AT2G26200 methyltransferase G, M, H Potri.009G017400 AT2G35940.1 transcription factor HOX domain proteins G, M, H

37 1.5 Figures

38 Figure 1.1: Populations for landscape genomic studies are frequently sampled out of a context of spatially varying genomic structure and environmental change. The genetic structure of P. balsamifera is represented from a sample from 412 reference SNPs genotyped by (Keller et al. 2010) with three demes observed: low diversity central and western demes, and a higher diversity eastern deme (a). Environmental gradients abound in nature, and greenness onset is a classic example of a biological phenomenon, controlled by both photoperiod and temperature, that changes across large geographic areas (b). Eliminating false positives in selection scans is always challenging; however, fewer false discoveries are observed from the tail of empirical p-value distributions derived from putatively neutral reference SNPs (c). This approach is demonstrated with the results of a BayeScan analysis (Foll and Gaggiotti, 2008) of 412 reference SNPs and 335 candidate SNPs from 27 ﬂowering-time genes that were sampled from the localities in (a). Setting a threshold conditioned with empirical p- values identiﬁes local selection in SNPs from GIGANTEA 5 (GI5), EARLY BOLTING IN SHORT DAYS (EBS), PHYTOCHROME-INTERACTING TRANSCRIPTION FACTOR 3 (PIF3), andCASEIN KINASE II BETA CHAIN 3 (CKB3.4). Allele frequency variation in four SNPs from GI5 are associated with latitude (d). GI5 is involved in light signaling to the circadian clock and local adaptation of these alleles may allow precise timing of phenology. Figures reproduced, with permission, from M. Fitzpatrick, (a); A. Elmore (b); S. R. Keller, (c, d).

39 McKown Holliday

Geraldes 182 1116 Evans 4 80 19 465 414 2 1

18 4

0 3 11

Figure 1.2: Overlap of candidate genes for local adaptation inPopulus trichocarpa identiﬁed in selection scans by Geraldes et al. (2014), McKown et al. (2014)c, Holliday et al. (2016), and Evans et al. (2014). No overlap was detected by all studies, but six genes overlapped among three studies (underlined). Version 2.2 Gene annotations from Geraldes et al. (2014) were updated to 3.0 for comparisons.

40 References

Aitken, SN and JB Bemmels (2016). “Time to get moving: assisted gene ﬂow of forest

trees”. In: Evolutionary Applications 9.1, pp. 271–290. Alberto, FJ, J Derory, C Boury, JM Frigerio, NE Zimmermann, and A Kremer (2013). “Imprints of natural selection along environmental gradients in phenology-related

genes of Quercus petraea”. In: Genetics 195.2, pp. 495–512. Ally, D, K Ritland, and S Otto (2008). “Can clone size serve as a proxy for clone

age? An exploration using microsatellite divergence in Populus tremuloides”. In: Molecular Ecology 17.22, pp. 4897–4911. Anderson, MK (2013). Tending the wild: Native American knowledge and the management of California’s natural resources. University of California Press. Arens, P, H Coops, J Jansen, and B Vosman (1998). “Molecular genetic analysis of

black poplar (Populus nigra L.) along Dutch rivers”. In: Molecular Ecology 7.1, pp. 11–18. Augspurger, CK (1984). “Seedling survival of tropical tree species: interactions of

dispersal distance, light-gaps, and pathogens”. In: Ecology 65.6, pp. 1705–1712. Avelino, J, A Romero-Gurdián, HF Cruz-Cuellar, and FA Declerck (2012). “Landscape context and scale dierentially impact coee leaf rust, coee berry borer, and coee

root-knot nematodes”. In: Ecological applications 22.2, pp. 584–596. Bagchi, R, RE Gallery, S Gripenberg, SJ Gurr, L Narayan, CE Addis, RP Freckleton, and OT Lewis (2014). “Pathogens and insect herbivores drive rainforest plant

diversity and composition”. In: Nature 506.7486, p. 85.

41 Becker, C, J Hagmann, J Müller, D Koenig, O Stegle, K Borgwardt, and D Weigel

(2011). “Spontaneous epigenetic variation in the Arabidopsis thaliana methylome”. In: Nature 480.7376, p. 245. Becker, C and D Weigel (2012). “Epigenetic variation: origin and transgenerational

inheritance”. In: Current opinion in plant biology 15.5, pp. 562–567. Berg, JJ and G Coop (2014). “A population genetic signal of polygenic adaptation”.

In: PLoS genetics 10.8, e1004412. Bernard, E, L Jacob, J Mairal, E Viara, and JP Vert (2015). “A convex formulation for joint RNA isoform detection and quantiﬁcation from multiple RNA-seq samples”.

In: BMC bioinformatics 16.1, p. 262. Block, WM, ML Morrison, and J Verner (1990). “Wildlife and oak-woodland inter-

dependency”. In: Fremontia 18.3, pp. 72–76. Bodénès, C, E Chancerel, O Gailing, GG Vendramin, F Bagnoli, J Durand, PG Goicoechea, C Soliani, F Villani, C Mattioni, et al. (2012). “Comparative mapping

in the Fagaceae and beyond with EST-SSRs”. In: BMC plant biology 12.1, p. 153. Böhlenius, H, T Huang, L Charbonnel-Campaa, AM Brunner, S Jansson, SH Strauss, and O Nilsson (2006). “CO/FT regulatory module controls timing of ﬂowering and

seasonal growth cessation in trees”. In: Science 312.5776, pp. 1040–1043. Bono, JM, EC Olesnicky, and LM Matzkin (2015). “Connecting genotypes, phenotypes and ﬁtness: harnessing the power of CRISPR/Cas9 genome editing”. In:

Molecular ecology 24.15, pp. 3810–3822. Bragg, JG, MA Supple, RL Andrew, and JO Borevitz (2015). “Genomic variation

across landscapes: insights and applications”. In: New Phytologist 207.4, pp. 953– 967.

42 Breen, AL, DF Murray, and MS Olson (2012). “Genetic consequences of glacial sur-

vival: the late Quaternary history of balsam poplar (Populus balsamifera L.) in North America”. In: Journal of Biogeography 39.5, pp. 918–928. Brown, TB, R Cheng, XR Sirault, T Rungrat, KD Murray, M Trtilek, RT Furbank, M Badger, BJ Pogson, and JO Borevitz (2014). “TraitCapture: genomic and envi-

ronment modelling of plant phenomic data”. In: Current opinion in plant biology 18, pp. 73–79.

Burns, RM and BH Honkala (1990). Silvics of north America. Vol. 2. United States Department of Agriculture Washington, DC. Busby, PE, N Zimmerman, DJ Weston, SS Jawdy, J Houbraken, and G Newcombe

(2013). “Leaf endophytes and Populus genotype aect severity of damage from the necrotrophic leaf pathogen, Drepanopeziza populi”. In: Ecosphere 4.10, pp. 1–12. Bylesjö, M, V Segura, RY Soolanayakanahally, AM Rae, J Trygg, P Gustafsson, S Jansson, and NR Street (2008). “LAMINA: a tool for rapid quantiﬁcation of leaf

size and shape parameters”. In: BMC Plant Biology 8.1, p. 82. ∆aliÊ, I, F Bussotti, PJ Martínez-García, and DB Neale (2016). “Recent landscape ge-

nomics studies in forest trees—what can we believe?” In: Tree genetics & genomes 12.1, p. 3. Callahan, CM, CA Rowe, RJ Ryel, JD Shaw, MD Madritch, and KE Mock (2013). “Continental-scale assessment of genetic diversity and population structure in

quaking aspen (Populus tremuloides)”. In: Journal of biogeography 40.9, pp. 1780– 1791. Casasoli, M, J Derory, C Morera-Dutrey, O Brendel, I Porth, JM Guehl, F Villani, and A Kremer (2006). “Comparison of quantitative trait loci for adaptive traits

43 between oak and chestnut based on an expressed sequence tag consensus map”.

In: Genetics 172.1, pp. 533–546. Chhatre, VE, LM Evans, SP DiFazio, and SR Keller (2018). “Adaptive introgression and maintenance of a trispecies hybrid complex in range-edge populations of

Populus”. In: Molecular ecology 27.23, pp. 4820–4838. Christe, C, KN Stölting, L Bresadola, B Fussi, B Heinze, D Wegmann, and C Lexer (2016). “Selection against recombinant hybrids maintains reproductive isolation

in hybridizing Populus species despite F1 fertility and recurrent gene ﬂow”. In: Molecular ecology 25.11, pp. 2482–2498. Clark, RM, TN Wagler, P Quijada, and J Doebley (2006). “A distant upstream enhancer at the maize domestication gene tb1 has pleiotropic eects on plant

and inﬂorescent architecture”. In: Nature genetics 38.5, p. 594. Cokus, SJ, PF Gugger, and VL Sork (2015). “Evolutionary insights from de novo

transcriptome assembly and SNP discovery in California white oaks”. In: BMC genomics 16.1, p. 552. Dahlgren, RA, MJ SINGER, and X Huang (1997). “Oak tree and grazing impacts on

soil properties and nutrients in a California oak woodland”. In: Biogeochemistry 39.1, pp. 45–64. De Carvalho, D, PK Ingvarsson, J Joseph, L Suter, C Sedivy, D MACAYA-SANZ, J Cottrell, B Heinze, I Schanzer, and C Lexer (2010). “Admixture facilitates adap-

tation from standing variation in the European aspen (Populus tremula L.), a widespread forest tree”. In: Molecular Ecology 19.8, pp. 1638–1650. Delﬁno-Mix, A, JW Wright, PF Gugger, C Liang, and VL Sork (2015). “Establishing

a range-wide provenance test in valley oak (Quercus lobata Née) at two Califor-

44 nia sites”. In: Gen. Tech. Rep. PSW-GTR-251. Berkeley, CA: US Department of Agriculture, Forest Service, Paciﬁc Southwest Research Station: 413-424 251, pp. 413–424. Derory, J, C Scotti-Saintagne, E Bertocchi, L Le Dantec, N Graignic, A Jaures, M Casasoli, E Chancerel, C Bodénès, F Alberto, et al. (2010). “Contrasting relationships between the diversity of candidate genes and variation of bud burst in

natural and segregating populations of European oaks”. In: Heredity 104.5, p. 438. Derory, J, P Léger, V Garcia, J Schaeer, MT Hauser, F Salin, C Luschnig, C Plomion, J Glössl, and A Kremer (2006). “Transcriptome analysis of bud burst in sessile

oak (Quercus petraea)”. In: New Phytologist 170.4, pp. 723–738. Di Lonardo, S, M Capuana, M Arnetoli, R Gabbrielli, and C Gonnelli (2011). “Ex-

ploring the metal phytoremediation potential of three Populus alba L. clones using an in vitro screening”. In: Environmental Science and Pollution Research 18.1, pp. 82–90. DiFazio, SP, S Leonardi, GT Slavov, SL Garman, WT Adams, and SH Strauss (2012). “Gene ﬂow and simulation of transgene dispersal from hybrid poplar plantations”.

In: New Phytologist 193.4, pp. 903–915. DiFazio, SP, GT Slavov, and CP Joshi (2011). “Populus: a premier pioneer system for plant genomics”. In: Enﬁeld, NH: Science Publishers, pp. 1–28. Dillon, S, R McEvoy, DS Baldwin, S Southerton, C Campbell, Y Parsons, and GN

Rees (2015). “Genetic diversity of Eucalyptus camaldulensis Dehnh. following population decline in response to drought and altered hydrological regime”. In: Austral Ecology 40.5, pp. 558–572.

45 Dosskey, M, R Schultz, and T Isenhart (1997). “Riparian Buers for Agricultural

Land in Agroforestry Notes”. In: National Agroforestry Center, USDA Forest Service, Rocky Mountain Station USDA Natural Resources Conservation Service, East Campus-UNL, Lincoln, Nebraska. Du, Q, B Xu, C Gong, X Yang, W Pan, J Tian, B Li, and D Zhang (2014). “Variation

in growth, leaf, and wood property traits of Chinese white poplar (Populus tomentosa), a major industrial tree species in Northern China”. In: Canadian journal of forest research 44.4, pp. 326–339. Durand, J, C Bodénès, E Chancerel, JM Frigerio, G Vendramin, F Sebastiani, A Buonamici, O Gailing, HP Koelewijn, F Villani, et al. (2010). “A fast and cost- eective approach to develop and map EST-SSR markers: oak as a case study”.

In: BMC genomics 11.1, p. 570. Elmore, A, C Stylinski, and K Pradhan (2016). “Synergistic use of citizen science and remote sensing for continental-scale measurements of forest tree phenology”. In:

Remote Sensing 8.6, p. 502. Espinoza, O, U Buehlmann, M Bumgardner, and B Smith (2011). “Assessing changes in the US hardwood sawmill industry with a focus on markets and distribution”.

In: BioResources 6.3, pp. 2676–2689. Evans, LM, GJ Allan, SP DiFazio, GT Slavov, JA Wilder, KD Floate, SB Rood, and TG Whitham (2015). “Geographical barriers and climate inﬂuence demographic

history in narrowleaf cottonwoods”. In: Heredity 114.4, p. 387. Evans, LM, GT Slavov, E Rodgers-Melnick, J Martin, P Ranjan, W Muchero, AM Brunner, W Schackwitz, L Gunter, JG Chen, et al. (2014). “Population genomics

46 of Populus trichocarpa identiﬁes signatures of selection and adaptive trait associations”. In: Nature genetics 46.10, p. 1089. Fan, D, T Liu, C Li, B Jiao, S Li, Y Hou, and K Luo (2015). “Ecient CRISPR/Cas9-

mediated targeted mutagenesis in Populus in the ﬁrst generation”. In: Scientiﬁc reports 5, p. 12217. Fang, GC, BP Blackmon, ME Staton, CD Nelson, TL Kubisiak, BA Olukolu, D Henry, T Zhebentyayeva, CA Saski, CH Cheng, et al. (2013). “A physical map of

the Chinese chestnut (Castanea mollissima) genome and its integration with the genetic map”. In: Tree genetics & genomes 9.2, pp. 525–537. Fetter, KC and A Weakley (2019). “Reduced Gene Flow from Mainland Populations

of Liriodendron tulipifera into the Florida Peninsula Promotes Diversiﬁcation”. In: International Journal of Plant Sciences 180.3, pp. 253–269. Feurdean, A, SA Bhagwat, KJ Willis, HJB Birks, H Lischke, and T Hickler (2013). “Tree migration-rates: narrowing the gap between inferred post-glacial rates and

projected rates”. In: PLoS One 8.8, e71797. Fitzpatrick, MC and SR Keller (2015). “Ecological genomics meets community-level modelling of biodiversity: Mapping the genomic landscape of current and future

environmental adaptation”. In: Ecology letters 18.1, pp. 1–16. Floate, KD (2004). “Extent and patterns of hybridization among the three species

of Populus that constitute the riparian forest of southern Alberta, Canada”. In: Canadian Journal of Botany 82.2, pp. 253–264. Floate, KD, J Godbout, MK Lau, N Isabel, and TG Whitham (2016). “Plant–

herbivore interactions in a trispeciﬁc hybrid swarm of Populus: assessing support

47 for hypotheses of hybrid bridges, evolutionary novelty and genetic similarity”. In:

New Phytologist 209.2, pp. 832–844. Foll, M and O Gaggiotti (2008). “A genome-scan method to identify selected loci appropriate for both dominant and codominant markers: a Bayesian perspective”.

In: Genetics 180.2, pp. 977–993. Foster, AJ, G Pelletier, P Tanguay, and A Séguin (2015). “Transcriptome Analysis

of Poplar during Leaf Spot Infection with Sphaerulina spp.” In: PLoS One 10.9, e0138162. Fralish, JS (2004). “The keystone role of oak and hickory in the central hardwood

forest”. In: Gen. Tech. Rep. SRS-73. Asheville, NC: US Department of Agriculture, Forest Service, Southern Research Station. pp. 78-87. Frichot, E, SD Schoville, G Bouchard, and O François (2013). “Testing for associations between loci and environmental gradients using latent factor mixed models”. In:

Molecular biology and evolution 30.7, pp. 1687–1699. Gentry, AH (1988). “Tree species richness of upper Amazonian forests”. In: Proceed- ings of the National Academy of Sciences 85.1, pp. 156–159. Geraldes, A, S Difazio, G Slavov, P Ranjan, W Muchero, J Hannemann, L Gunter, A Wymore, C Grassa, N Farzaneh, et al. (2013). “A 34K SNP genotyping array for

Populus trichocarpa: design, application to the study of natural populations and transferability to other Populus species”. In: Molecular Ecology Resources 13.2, pp. 306–323. Geraldes, A, N Farzaneh, CJ Grassa, AD McKown, RD Guy, SD Mansﬁeld, CJ Dou-

glas, and QC Cronk (2014). “Landscape genomics of Populus trichocarpa: the role

48 of hybridization, limited gene ﬂow, and natural selection in shaping patterns of

population structure”. In: Evolution 68.11, pp. 3260–3280. Geraldes, A, CA Hefer, A Capron, N Kolosova, F Martinez-Nuñez, RY Soolanayakana- hally, B Stanton, RD Guy, SD Mansﬁeld, CJ Douglas, et al. (2015). “Recent Y

chromosome divergence despite ancient origin of dioecy in poplars (Populus)”. In: Molecular ecology 24.13, pp. 3243–3256. Gonzalez-Martinez, SC, KV Krutovsky, and DB Neale (2006). “Forest-tree population

genomics and adaptive evolution”. In: New Phytologist 170.2, pp. 227–238. Graham, EA, EC Riordan, EM Yuen, D Estrin, and PW Rundel (2010). “Public internet-connected cameras used as a cross-continental ground-based plant phe-

nology monitoring system”. In: Global Change Biology 16.11, pp. 3014–3023. Grivet, D, MF Deguilloux, RJ Petit, and VL Sork (2006). “Contrasting patterns of

historical colonization in white oaks (Quercus spp.) in California and Europe”. In: Molecular Ecology 15.13, pp. 4085–4093. Gugger, PF, SJ Cokus, and VL Sork (2016). “Association of transcriptome-wide se-

quence variation with climate gradients in valley oak (Quercus lobata)”. In: Tree Genetics & Genomes 12.2, p. 15. Gugger, PF, S Fitz-Gibbon, M Pellegrini, and VL Sork (2016). “Species-wide patterns

of DNA methylation variation in Quercus lobata and their association with climate gradients”. In: Molecular Ecology 25.8, pp. 1665–1680. Gugger, PF, JM Peñaloza-Ramírez, JW Wright, and VL Sork (2017). “Whole-transcriptome

response to water stress in a California endemic oak, Quercus lobata”. In: Tree physiology 37.5, pp. 632–644.

49 Günther, T and G Coop (2013). “Robust identiﬁcation of local adaptation from allele

frequencies”. In: Genetics 195.1, pp. 205–220. Hacquard, S and CW Schadt (2015). “Towards a holistic understanding of the ben-

eﬁcial interactions across the Populus microbiome”. In: New Phytologist 205.4, pp. 1424–1430.

Haldane, J (1948). “The theory of a cline”. In: Journal of genetics 48.3, pp. 277–284. Hall, D, V Luquez, VM Garcia, KR St Onge, S Jansson, and PK Ingvarsson (2007). “Adaptive population dierentiation in phenology across a latitudinal gradient in

European aspen (Populus tremula, L.): a comparison of neutral markers, candidate genes and phenotypic traits”. In: Evolution 61.12, pp. 2849–2860. Hall, D, XF MA, and PK Ingvarsson (2011). “Adaptive evolution of the Populus tremula photoperiod pathway”. In: Molecular ecology 20.7, pp. 1463–1474. Hamilton, MB (1999). “Tropical tree gene ﬂow and seed dispersal”. In: Nature 401.6749, p. 129. Hancock, AM, B Brachi, N Faure, MW Horton, LB Jarymowycz, FG Sperone, C Toomajian, F Roux, and J Bergelson (2011). “Adaptation to climate across the

Arabidopsis thaliana genome”. In: Science 334.6052, pp. 83–86. Hartﬁeld, M (2016). “Evolutionary genetic consequences of facultative sex and out-

crossing”. In: Journal of evolutionary biology 29.1, pp. 5–22. Herman, DJ, LJ Halverson, and MK Firestone (2003). “Nitrogen dynamics in an

annual grassland: oak canopy, climate, and microbial population eects”. In: Eco- logical Applications 13.3, pp. 593–604. Hewitt, GM (1999). “Post-glacial re-colonization of European biota”. In: Biological journal of the Linnean Society 68.1-2, pp. 87–112.

50 Holeski, LM, MS Zinkgraf, JJ Couture, TG Whitham, and RL Lindroth (2013). “Transgenerational eects of herbivory in a group of long-lived tree species: maternal damage reduces ospring allocation to resistance traits, but not growth”.

In: Journal of Ecology 101.4, pp. 1062–1073. Holliday, JA, L Zhou, R Bawa, M Zhang, and RW Oubida (2016). “Evidence for extensive parallelism but divergent genomic architecture of adaptation along al-

titudinal and latitudinal gradients in Populus trichocarpa”. In: New Phytologist 209.3, pp. 1240–1251. Homolova, L, Z Malenovsk˝, JG Clevers, G García-Santos, and ME Schaepman (2013). “Review of optical-based remote sensing for plant trait mapping”. In:

Ecological Complexity 15, pp. 1–16. Huang, YN, H Zhang, S Rogers, M Coggeshall, and K Woeste (2015). “White oak growth after 23 years in a three-site provenance/progeny trial on a latitudinal

gradient in Indiana”. In: Forest Science. Imbert, E and F Lefévre (2003). “Dispersal and gene ﬂow of Populus nigra (Salicaceae) along a dynamic river system”. In: Journal of Ecology 91.3, pp. 447–456. Ingvarsson, PK (2008). “Multilocus patterns of nucleotide polymorphism and the

demographic history of Populus tremula”. In: Genetics 180.1, pp. 329–340. Ingvarsson, PK, MV García, D Hall, V Luquez, and S Jansson (2006). “Clinal variation in phyB2, a candidate gene for day-length-induced growth cessation and

bud set, across a latitudinal gradient in European aspen (Populus tremula)”. In: Genetics 172.3, pp. 1845–1853. Ingvarsson, PK, MV Garcia, V Luquez, D Hall, and S Jansson (2008). “Nucleotide polymorphism and phenotypic associations within and around the phytochrome

51 B2 locus in European aspen (Populus tremula, Salicaceae)”. In: Genetics 178.4, pp. 2217–2226. Isabel, N, M Lamothe, and SL Thompson (2013). “A second-generation diagnostic single nucleotide polymorphism (SNP)-based assay, optimized to distinguish among

eight poplar (Populus L.) species and their early hybrids”. In: Tree genetics & genomes 9.2, pp. 621–626. IvetiÊ, V, S StjepanoviÊ, J DevetakoviÊ, D StankoviÊ, and M äkoriÊ(2014). “Re- lationships between leaf traits and morphological attributes in one-year bareroot

Fraxinus angustifolia Vahl. seedlings”. In: Annals of Forest Research 57.2, pp. 197– 203. Järvinen, P, J Lemmetyinen, O Savolainen, and T Sopanen (2003). “DNA sequence

variation in BpMADS2 gene in two populations of Betula pendula”. In: Molecular ecology 12.2, pp. 369–384. Keller, SR, N Levsen, PK Ingvarsson, MS Olson, and P Tin (2011). “Local selection

across a latitudinal gradient shapes nucleotide diversity in balsam poplar, Populus balsamifera L”. In: Genetics 188.4, pp. 941–952. Keller, SR, N Levsen, MS Olson, and P Tin (2012). “Local adaptation in the

ﬂowering-time gene network of balsam poplar, Populus balsamifera L.” In: Molec- ular biology and evolution 29.10, pp. 3143–3152. Keller, SR, MS Olson, S Silim, W Schroeder, and P Tin (2010). “Genomic diversity, population structure, and migration following rapid range expansion in the Balsam

Poplar, Populus balsamifera”. In: Molecular Ecology 19.6, pp. 1212–1226.

52 Kempes, CP, GB West, K Crowell, and M Girvan (2011). “Predicting maximum tree

heights and other traits from allometric scaling and resource limitations”. In: PLoS One 6.6, e20551. Kim, YH, MD Kim, YI Choi, SC Park, DJ Yun, EW Noh, HS Lee, and SS Kwak

(2011). “Transgenic poplar expressing Arabidopsis NDPK2 enhances growth as well as oxidative stress tolerance”. In: Plant biotechnology journal 9.3, pp. 334– 347. Kleinknecht, GJ, HE Lintz, A Kruger, JJ Niemeier, MJ Salino-Hugg, CK Thomas, CJ Still, and Y Kim (2015). “Introducing a sensor to measure budburst and its

environmental drivers”. In: Frontiers in Plant Science 6, p. 123. Kremer, A, AG Abbott, JE Carlson, PS Manos, C Plomion, P Sisco, ME Staton, S

Ueno, and GG Vendramin (2012). “Genomics of Fagaceae”. In: Tree Genetics & Genomes 8.3, pp. 583–610. Kroeger, T, F Casey, P Alvarez, M Cheatum, and L Tavassoi (2010). An Economic Analysis of the Beneﬁts of Habitat Conservation on California Rangelands. Wash- ington, D.C.: Conservation Economics Program, Defenders of Wildlife. Krutovsky, KV and DB Neale (2005). “Nucleotide diversity and linkage disequilibrium in cold-hardiness-and wood quality-related candidate genes in Douglas ﬁr”. In:

Genetics 171.4, pp. 2029–2041. Langlet, O (1971). “Two hundred years genecology”. In: Taxon, pp. 653–721. Le Corre, V, N Machon, RJ Petit, and A Kremer (1997). “Colonization with long- distance seed dispersal and genetic structure of maternally inherited genes in forest

trees: a simulation study”. In: Genetics Research 69.2, pp. 117–125.

53 Lee, JH, ER Daugharthy, J Scheiman, R Kalhor, TC Ferrante, R Terry, BM Tur- czyk, JL Yang, HS Lee, J Aach, et al. (2015). “Fluorescent in situ sequencing (FISSEQ) of RNA for gene expression proﬁling in intact cells and tissues”. In:

Nature protocols 10.3, p. 442. Lepoittevin, C, C Bodénés, E Chancerel, L Villate, T Lang, I Lesur, C Boury, F Ehrenmann, D Zelenica, A Boland, et al. (2015). “Single-nucleotide polymorphism discovery and validation in high-density SNP array for genetic analysis in Euro-

pean white oaks”. In: Molecular ecology resources 15.6, pp. 1446–1459. Lesur, I, G Le Provost, P Bento, C Da Silva, JC Leplé, F Murat, S Ueno, J Bartholomé, C Lalanne, F Ehrenmann, et al. (2015). “The oak gene expression atlas: insights into Fagaceae genome evolution and the discovery of genes regulated during bud

dormancy release”. In: BMC genomics 16.1, p. 112. Levsen, ND, P Tin, and MS Olson (2012). “Pleistocene speciation in the genus

Populus (Salicaceae)”. In: Systematic biology 61.3, p. 401. Liang, H, S Ayyampalayam, N Wickett, A Barakat, Y Xu, L Landherr, PE Ralph, Y Jiao, T Xu, SE Schlarbaum, et al. (2011). “Generation of a large-scale genomic

resource for functional and comparative genomics in Liriodendron tulipifera L.” In: Tree Genetics & Genomes 7.5, pp. 941–954. Linhart, YB and MC Grant (1996). “Evolutionary signiﬁcance of local genetic dier-

entiation in plants”. In: Annual review of ecology and systematics 27.1, pp. 237– 277. Lotterhos, KE and MC Whitlock (2014). “Evaluation of demographic history and

neutral parameterization on the performance of FST outlier tests”. In: Molecular ecology 23.9, pp. 2178–2192.

54 Luppold, W and M Bumgardner (2008). “Regional analysis of hardwood lumber pro-

duction: 1963–2005”. In: Northern journal of applied forestry 25.3, pp. 146–150. Luquez, V, D Hall, BR Albrectsen, J Karlsson, P Ingvarsson, and S Jansson (2008).

“Natural phenological variation in aspen (Populus tremula): the SwAsp collection”. In: Tree Genetics & Genomes 4.2, pp. 279–292. Ma, XF, D Hall, KRS Onge, S Jansson, and PK Ingvarsson (2010). “Genetic dierentiation, clinal variation and phenotypic associations with growth cessation across

the Populus tremula photoperiodic pathway”. In: Genetics 186.3, pp. 1033–1044. Madritch, MD, CC Kingdon, A Singh, KE Mock, RL Lindroth, and PA Townsend (2014). “Imaging spectroscopy links aspen genotype with below-ground processes

at landscape scales”. In: Philosophical Transactions of the Royal Society B: Bio- logical Sciences 369.1643, p. 20130194. Manel, S, C Perrier, M Pratlong, L Abi-Rached, J Paganini, P Pontarotti, and D Aurelle (2016). “Genomic resources and their inﬂuence on the detection of the

signal of positive selection in genome scans”. In: Molecular ecology 25.1, pp. 170– 184. Manel, S, MK Schwartz, G Luikart, and P Taberlet (2003). “Landscape genetics:

combining landscape ecology and population genetics”. In: Trends in ecology & evolution 18.4, pp. 189–197. Martins, K, PF Gugger, J Llanderal-Mendoza, A González-Rodríguez, ST Fitz-Gibbon, JL Zhao, H Rodríguez-Correa, K Oyama, and VL Sork (2018). “Landscape genomics provides evidence of climate-associated genetic variation in Mexican pop-

ulations of Quercus rugosa”. In: Evolutionary Applications 11.10, pp. 1842–1858.

55 Martinsen, GD, TG Whitham, RJ Turek, and P Keim (2001). “Hybrid populations

selectively ﬁlter gene introgression between species”. In: Evolution 55.7, pp. 1325– 1335. McCauley, DE (1995). “The use of chloroplast DNA polymorphism in studies of gene

ﬂow in plants”. In: Trends in ecology & evolution 10.5, pp. 198–202. McKown, AD, J Klápöt, RD Guy, A Geraldes, I Porth, J Hannemann, M Friedmann, W Muchero, GA Tuskan, J Ehlting, QCB Cronk, YA El-Kassaby, SD Mansﬁeld, and CJ Douglas (2014). “Genome-wide association implicates numerous genes un-

derlying ecological trait variation in natural populations of Populus trichocarpa”. In: New Phytologist 203.2, pp. 535–553. McKown, AD, RD Guy, J Klápöt, A Geraldes, M Friedmann, QC Cronk, YA El- Kassaby, SD Mansﬁeld, and CJ Douglas (2014). “Geographical and environmental

gradients shape phenotypic trait variation and genetic structure in Populus trichocarpa”. In: New Phytologist 201.4, pp. 1263–1276. McKown, AD, RD Guy, L Quamme, J Klápöt, J La Mantia, C Constabel, YA El-Kassaby, RC Hamelin, M Zifkin, and M Azam (2014). “Association genet-

ics, geography and ecophysiology link stomatal patterning in Populus trichocarpa with carbon gain and disease resistance trade-os”. In: Molecular Ecology 23.23, pp. 5771–5790. McLachlan, JS, JS Clark, and PS Manos (2005). “Molecular indicators of tree migra-

tion capacity under rapid climate change”. In: Ecology 86.8, pp. 2088–2098. McLean, EH, SM Prober, WD Stock, DA Steane, BM Potts, RE Vaillancourt, and M Byrne (2014). “Plasticity of functional traits varies clinally along a rainfall

56 gradient in Eucalyptus tricarpa”. In: Plant, Cell & Environment 37.6, pp. 1440– 1451. McShea, WJ, WM Healy, P Devers, T Fearer, FH Koch, D Stauer, and J Waldon (2007). “Forestry matters: decline of oaks will impact wildlife in hardwood forests”.

In: The Journal of Wildlife Management 71.5, pp. 1717–1728. Meijón, M, SB Satbhai, T Tsuchimatsu, and W Busch (2014). “Genome-wide association study using cellular traits identiﬁes a new regulator of root development in

Arabidopsis”. In: Nature genetics 46.1, p. 77. Meirmans, PG, M Lamothe, MC Gros-Louis, D Khasa, P Périnet, J Bousquet, and N Isabel (2010). “Complex patterns of hybridization between exotic and native

North American poplar species”. In: American Journal of Botany 97.10, pp. 1688– 1697. Menon, M, WJ Barnes, and MS Olson (2015). “Population genetics of freeze tolerance

among natural populations of Populus balsamifera across the growing season”. In: New Phytologist 207.3, pp. 710–722. Mishra, P and KC Panigrahi (2015). “GIGANTEA–an emerging story”. In: Frontiers in plant science 6, p. 8. Mock, KE, CM Callahan, MN Islam-Faridi, JD Shaw, HS Rai, SC Sanderson, CA Rowe, RJ Ryel, MD Madritch, RS Gardner, and PG Wolf (2012). “Widespread

triploidy in western North American aspen (Populus tremuloides)”. In: PLoS One 7.10, e48406. Mock, KE, C Rowe, MB Hooten, J Dewoody, and V Hipkins (2008). “Clonal dynamics

in western North American aspen (Populus tremuloides)”. In: Molecular Ecology 17.22, pp. 4827–4844.

57 Moran, EV, J Willis, and JS Clark (2012). “Genetic evidence for hybridization in red

oaks (Quercus sect. Lobatae, Fagaceae)”. In: American Journal of Botany 99.1, pp. 92–100.

Muller, CH (1952). “Ecological control of hybridization in Quercus: a factor in the mechanism of evolution”. In: Evolution, pp. 147–161. Myburg, AA, D Grattapaglia, GA Tuskan, U Hellsten, RD Hayes, J Grimwood, J

Jenkins, E Lindquist, H Tice, D Bauer, et al. (2014). “The genome of Eucalyptus grandis”. In: Nature 510.7505, p. 356. Myneni, R, J Dong, C Tucker, R Kaufmann, P Kauppi, J Liski, L Zhou, V Alexeyev, and M Hughes (2001). “A large carbon sink in the woody biomass of Northern

forests”. In: Proceedings of the National Academy of Sciences 98.26, pp. 14784– 14789. Nathan, R, GG Katul, HS Horn, SM Thomas, R Oren, R Avissar, SW Pacala, and SA Levin (2002). “Mechanisms of long-distance dispersal of seeds by wind”. In:

Nature 418.6896, p. 409. Neale, DB and A Kremer (2011). “Forest tree genomics: growing resources and ap-

plications”. In: Nature Reviews Genetics 12.2, p. 111. Neiman, M, MS Olson, and P Tin (2009). “Selective histories of poplar protease in- hibitors: elevated polymorphism, purifying selection, and positive selection driving

divergence of recent duplicates”. In: New Phytologist 183.3, pp. 740–750. Oliverio, KA, M Crepy, EL Martin-Tryon, R Milich, SL Harmer, J Putterill, MJ Yanovsky, and JJ Casal (2007). “GIGANTEA regulates phytochrome A-mediated

photomorphogenesis independently of its role in the circadian clock”. In: Plant Physiology 144.1, pp. 495–502.

58 Olson, MS, N Levsen, RY Soolanayakanahally, RD Guy, WR Schroeder, SR Keller,

and P Tin (2013). “The adaptive potential of Populus balsamifera L. to phenology requirements in a warmer global climate”. In: Molecular Ecology 22.5, pp. 1214–1230. Ordonez, N, MF Seidl, C Waalwijk, A Drenth, A Kilian, BP Thomma, RC Ploetz, and GH Kema (2015). “Worse comes to worst: bananas and Panama disease—when

plant and pathogen clones meet”. In: PLoS pathogens 11.11, e1005197. Ortego, J, PF Gugger, EC Riordan, and VL Sork (2014). “Inﬂuence of climatic niche suitability and geographical overlap on hybridization patterns among southern

Californian oaks”. In: Journal of Biogeography 41.10, pp. 1895–1908. Pavlik, BM, SG Johnson, and P M. (1991). Oaks of California. Los Olivos, CA: Cachuma Press. Petit, RJ, S Brewer, S Bordács, K Burg, R Cheddadi, E Coart, J Cottrell, UM Csaikl, B van Dam, JD Deans, et al. (2002). “Identiﬁcation of refugia and post-glacial colonization routes of European white oaks based on chloroplast DNA and fossil

pollen evidence”. In: Forest ecology and management 156.1-3, pp. 49–74. Petit, RJ, J Carlson, AL Curtu, ML Loustau, C Plomion, A González-Rodríguez, V Sork, and A Ducousso (2013). “Fagaceae trees as models to integrate ecology,

evolution and genomics”. In: New Phytologist 197.2, pp. 369–371. Petit, RJ and A Hampe (2006). “Some evolutionary consequences of being a tree”.

In: Annu. Rev. Ecol. Evol. Syst. 37, pp. 187–214. Petit, RJ, FS Hu, and CW Dick (2008). “Forests of the past: a window to future

changes”. In: Science 320.5882, pp. 1450–1452.

59 Petit, RJ, E Pineau, B Demesure, R Bacilieri, A Ducousso, and A Kremer (1997).

“Chloroplast DNA footprints of postglacial recolonization by oaks”. In: Proceed- ings of the National Academy of Sciences 94.18, pp. 9996–10001. Pettorelli, N, K Saﬁ, and W Turner (2014). Satellite remote sensing, biodiversity research and conservation of the future. Platt, A, PF Gugger, M Pellegrini, and VL Sork (2015). “Genome-wide signature of local adaptation linked to variable CpG methylation in oak populations”. In:

Molecular Ecology 24.15, pp. 3823–3830. Platt, A, BJ Vilhjálmsson, and M Nordborg (2010). “Conditions under which genome-

wide association studies will be positively misleading”. In: Genetics 186.3, pp. 1045– 1052. Plomion, C, JM Aury, J Amselem, T Alaeitabar, V Barbe, C Belser, H Bergès, C Bodénès, N Boudet, C Boury, A Canaguier, A Couloux, C Da Silva, S Duplessis, F Ehrenmann, B Estrada-Mairey, S Fouteau, N Francillonne, C Gaspin, C Guichard, C Klopp, K Labadie, C Lalanne, I Le Clainche, JC Leplé, G Le Provost, T Leroy, I Lesur, F Martin, J Mercier, C Michotey, F Murat, F Salin, D Steinbach, P Faivre- Rampant, P Wincker, J Salse, H Quesneville, and A Kremer (2016). “Decoding the oak genome: public release of sequence data, assembly, annotation and publication

strategies”. In: Molecular ecology resources 16.1, pp. 254–265. Porth, I, J Klápöt, AD McKown, J La Mantia, RD Guy, PK Ingvarsson, R Hamelin, SD Mansﬁeld, J Ehlting, CJ Douglas, et al. (2015). “Evolutionary quantitative

genomics of Populus trichocarpa”. In: PloS one 10.11, e0142864. Porth, I, J Klapöte, O Skyba, J Hannemann, AD McKown, RD Guy, SP DiFazio, W Muchero, P Ranjan, GA Tuskan, MC Friedmann, J Ehlting, QCB Cronk, YA

60 El-Kassaby, CJ Douglas, and SD Mansﬁeld (2013). “Genome-wide association

mapping for wood characteristics in Populus identifies an array of candidate single nucleotide polymorphisms”. In: New Phytologist 200.3, pp. 710–726. Porth, I, M Koch, M Berenyi, A Burg, and K Burg (2005). “Identification of adaptation- specific dierences in mRNA expression of sessile and pedunculate oak based on

osmotic-stress-induced genes”. In: Tree physiology 25.10, pp. 1317–1329. Rockman, MV (2012). “The QTN program and the alleles that matter for evolution:

all that’s gold does not glitter”. In: Evolution: International Journal of Organic Evolution 66.1, pp. 1–17. Savolainen, O, T Pyhäjärvi, and T Knürr (2007). “Gene ﬂow and local adaptation in

trees”. In: Annu. Rev. Ecol. Evol. Syst. 38, pp. 595–619. Schmitz, RJ, MD Schultz, MA Urich, JR Nery, M Pelizzola, O Libiger, A Alix, RB McCosh, H Chen, NJ Schork, and J Ecker (2013). “Patterns of population epige-

nomic diversity”. In: Nature 495.7440, p. 193. Sheehan, MJ, LM Kennedy, DE Costich, and TP Brutnell (2007). “Subfunctional- ization of PhyB1 and PhyB2 in the control of seedling and mature plant traits in

maize”. In: The Plant Journal 49.2, pp. 338–353. Slavov, GT, SP DiFazio, J Martin, W Schackwitz, W Muchero, E Rodgers-Melnick, MF Lipphardt, CP Pennacchio, U Hellsten, LA Pennacchio, LE Gunter, P Ran- jan, K Vining, KR Pomraning, LJ Wilhelm, M Pellegrini, TC Mockler, M Freitag, A Geraldes, YA El-Kassaby, SD Mansﬁeld, QCB Cronk, CJ Douglas, SH Strauss, D Rokhsar, and GA Tuskan (2012). “Genome resequencing reveals multiscale ge-

ographic structure and extensive linkage disequilibrium in the forest tree Populus trichocarpa”. In: New Phytologist 196.3, pp. 713–725.

61 Soler, M, O Serra, M Molinas, E García-Berthou, A Caritat, and M Figueras (2008). “Seasonal variation in transcript abundance in cork tissue analyzed by real time

RT-PCR”. In: Tree Physiology 28.5, pp. 743–751. Soolanayakanahally, RY, RD Guy, SN Silim, EC Drewes, and WR Schroeder (2009). “Enhanced assimilation rate and water use eciency with latitude through in-

creased photosynthetic capacity and internal conductance in balsam poplar (Pop- ulus balsamifera L.)” In: Plant, Cell & Environment 32.12, pp. 1821–1832. Soolanayakanahally, RY, RD Guy, SN Silim, and M Song (2013). “Timing of pho-

toperiodic competency causes phenological mismatch in balsam poplar (Populus balsamifera L.)” In: Plant, cell & environment 36.1, pp. 116–127. Soolanayakanahally, RY, RD Guy, NR Street, KM Robinson, SN Silim, BR Albrect-

sen, and S Jansson (2015). “Comparative physiology of allopatric Populus species: geographic clines in photosynthesis, height growth, and carbon isotope discrimi-

nation in common gardens”. In: Frontiers in plant science 6, p. 528. Sork, VL, FW Davis, PE Smouse, VJ Apsit, RJ Dyer, J Fernandez-M, and B Kuhn

(2002). “Pollen movement in declining populations of California Valley oak, Quer- cus lobata: where have all the fathers gone?” In: Molecular Ecology 11.9, pp. 1657– 1668. Sork, VL, ST Fitz-Gibbon, D Puiu, M Crepeau, PF Gugger, R Sherman, K Stevens, CH Langley, M Pellegrini, and SL Salzberg (2016). “First draft assembly and anno-

tation of the genome of a California endemic oak Quercus lobata Née (Fagaceae)”. In: G3: Genes, Genomes, Genetics 6.11, pp. 3485–3495. Sork, VL and PE Smouse (2006). “Genetic analysis of landscape connectivity in tree

populations”. In: Landscape ecology 21.6, pp. 821–836.

62 Sork, VL, PE Smouse, VJ Apsit, RJ Dyer, and RD Westfall (2005). “A two-generation

analysis of pollen pool genetic structure in ﬂowering dogwood, Cornus ﬂorida (Cornaceae), in the Missouri Ozarks”. In: American journal of botany 92.2, pp. 262– 271. Sork, VL, K Squire, PF Gugger, SE Steele, ED Levy, and AJ Eckert (2016). “Land- scape genomic analysis of candidate genes for climate adaptation in a California

endemic oak, Quercus lobata”. In: American Journal of Botany 103.1, pp. 33–46. Sork, V, S Aitken, R Dyer, A Eckert, P Legendre, and D Neale (2013). “Putting the landscape into the genomics of trees: approaches for understanding local adapta-

tion and population responses to changing climate”. In: Tree Genetics & Genomes 9.4, pp. 901–911. Sorrie, BA and AS Weakley (2001). “Coastal plain vascular plant endemics: phyto-

geographic patterns”. In: Castanea, pp. 50–82. Standiford, RB and RE Howitt (1993). “Multiple use management of California’s

hardwood rangelands.” In: Rangeland Ecology & Management/Journal of Range Management Archives 46.2, pp. 176–182. Steane, DA, BM Potts, E McLean, SM Prober, WD Stock, RE Vaillancourt, and M Byrne (2014). “Genome-wide scans detect adaptation to aridity in a widespread

forest tree species”. In: Molecular Ecology 23.10, pp. 2500–2513. Stinchcombe, JR and HE Hoekstra (2008). “Combining population genomics and quantitative genetics: ﬁnding the genes underlying ecologically important traits”.

In: Heredity 100.2, p. 158.

63 Strei, R (1999). “Pollen dispersal inferred from paternity analysis in a mixed oak

stand of Quercus robur L. and Q. petraea (Matt.) Liebl”. In: Molecular Ecology 8, pp. 831–841. Suarez-Gonzalez, A, CA Hefer, C Christe, O Corea, C Lexer, QC Cronk, and CJ Douglas (2016). “Genomic and functional approaches reveal a case of adaptive

introgression from Populus balsamifera (balsam poplar) in P. trichocarpa (black cottonwood)”. In: Molecular ecology 25.11, pp. 2427–2442. Swarbreck, D, C Wilks, P Lamesch, TZ Berardini, M Garcia-Hernandez, H Foerster, D Li, T Meyer, R Muller, L Ploetz, et al. (2007). “The Arabidopsis Informa-

tion Resource (TAIR): gene structure and function annotation”. In: Nucleic acids research 36.suppl_1, pp. D1009–D1014. Tarkka, MT, S Herrmann, T Wubet, L Feldhahn, S Recht, F Kurth, S Mailänder, M Bönn, M Neef, O Angay, M Bacht, M Graf, H Maboreke, F Fleischmann, TEE Grams, L Ruess, M Schädler, R Brandl, S Scheu, SD Schrey, I Grosse, and F Buscot (2013). “OakContig DF 159.1, a reference library for studying dierential

gene expression in Quercus robur during controlled biotic interactions: use for quantitative transcriptomic proﬁling of oak roots in ectomycorrhizal symbiosis”.

In: New Phytologist 199.2, pp. 529–540. Taylor, G (2002). “Populus: Arabidopsis for forestry. Do we need a model tree?” In: Annals of Botany 90.6, pp. 681–689. Thompson, SL, M Lamothe, PG Meirmans, P Perinet, and N Isabel (2010). “Repeated

unidirectional introgression towards Populus balsamifera in contact zones of exotic and native poplars”. In: Molecular Ecology 19.1, pp. 132–145.

64 Tin, P and J Ross-Ibarra (2014). “Advances and limits of using population genetics

to understand local adaptation”. In: Trends in ecology & evolution 29.12, pp. 673– 680. Toomey, M, MA Friedl, S Frolking, K Hufkens, S Klosterman, O Sonnentag, DD Baldocchi, CJ Bernacchi, SC Biraud, G Bohrer, et al. (2015). “Greenness indices from digital cameras predict the timing and seasonal dynamics of canopy-scale

photosynthesis”. In: Ecological Applications 25.1, pp. 99–115. Tuskan, GA et al. (2006). “The genome of black cottonwood, Populus trichocarpa (Torr. & Gray)”. In: science 313.5793, pp. 1596–1604. Ueno, S, G Le Provost, V Léger, C Klopp, C Noirot, JM Frigerio, F Salin, J Salse, M Abrouk, F Murat, O Brendel, J Derory, P Abadie, P Léger, C Cabane, A Barré, A de Daruvar, A Couloux, P Wincker, MP Reviron, A Kremer, and C Plomion (2010). “Bioinformatic analysis of ESTs collected by Sanger and pyrosequencing

methods for a keystone forest tree species: oak”. In: BMC genomics 11.1, p. 650. Van Valen, L (1976). “Ecological species, multispecies, and oaks”. In: Taxon, pp. 233– 239. Verhoeven, KJ, JJ Jansen, PJ van Dijk, and A Biere (2010). “Stress-induced DNA

methylation changes and their heritability in asexual dandelions”. In: New Phy- tologist 185.4, pp. 1108–1118. Wang, L, P Tin, and MS Olson (2014). “Timing for success: expression pheno-

type and local adaptation related to latitude in the boreal forest tree, Populus balsamifera”. In: Tree genetics & genomes 10.4, pp. 911–922.

65 Whitham, TG, SP DiFazio, JA Schweitzer, SM Shuster, GJ Allan, JK Bailey, and SA Woolbright (2008). “Extending genomics to natural communities and ecosystems”.

In: Science 320.5875, pp. 492–495. Yang, WY, J Novembre, E Eskin, and E Halperin (2012). “A model-based approach

for analysis of spatial structure in genetic data”. In: Nature genetics 44.6, p. 725. Zhang, P, F Wu, X Kang, C Zhao, and Y Li (2015). “Genotypic variations of biomass

feedstock properties for energy in triploid hybrid clones of Populus tomentosa”. In: BioEnergy Research 8.4, pp. 1705–1713. Zhou, L, R Bawa, and J Holliday (2014). “Exome resequencing reveals signatures of demographic and adaptive processes across the genome and range of black

cottonwood (Populus trichocarpa)”. In: Molecular ecology 23.10, pp. 2486–2499.

66 Chapter 2

StomataCounter: a neural network for automatic stomata identification and counting.

Karl C. Fettera,b,1,2, Sven Eberhardtc,1, Rich S. Barclayb, Scott Wingb, and Stephen R. Kellera aDepartment of Plant Biology, University of Vermont bDepartment of Paleobiology, Smithsonian Institution, National Museum of Natural History cAmazon.com, Inc. 1S.E. and K.C.F. contributed equally to this work 2Corresponding author: Karl C. Fetter (e-mail: [email protected], tel: +1 802 656 2930)

67 Abstract

Stomata regulate important physiological processes in plants and are often phenotyped by researchers in diverse fields of plant biology. Currently, there are no user friendly, fully-automated methods to perform the task of identifying and counting stomata, and stomata density is generally estimated by manually counting stomata. We introduce StomataCounter, an automated stomata counting system using a deep convolutional neural network to identify stomata in a variety of dierent microscopic images. We use a human-in-the-loop approach to train and refine a neural network on a taxonomically diverse collection of microscopic images. Our network achieves 98.1% identification accuracy on Ginkgo SEM micrographs, and 94.2% transfer accuracy when tested on untrained species. To facilitate adoption of the method, we provide the method in a publicly available website at http://www.stomata.uvm.edu. 2.1 Introduction

Stomata are important microscopic organs which mediate important biological and ecological processes and function to exchange gas with the environment. A stoma is composed of a pair of guard cells forming an aperture that, in some species, are flanked by a pair of subsidiary cells (Bergmann and Sack, 2007). Regulation of the aperture pore size, and hence, the opening and closing of the pore, is achieved through changing turgor pressure in the guard cells (Shimazaki et al., 2007). Stomata evolved to permit exchange between internal and external sources of gases, most notably CO2 and water vapor, through the impervious layer of the cuticle (Kim et al., 2010). Because of their importance in regulating plant productivity and response to the environment, stomata have been one of the key functional traits of interest to researchers working across scales in plant biology. At the molecular level, regulation of stomata has been the subject of numerous genetic studies (see Shimazaki et al., 2007 and (Kim et al., 2010) for detailed reviews) and crop improvement programs have modified stomata phenotypes to increases yield (Fischer et al., 1998). Stomata also mediate trade-os between carbon gain and pathogen exposure that are of interest to plant ecophysiologists and pathologists. For example, foliar pathogens frequently exploit the aperture pore as a site of entry. In Populus, some species, and even populations within species, evolve growth strategies that maximize carbon fixation through increased stomatal density and aperture pore size on the adaxial leaf surface. This adaptation results in a cost of increased infection by fungal pathogens which have more sites of entry to the leaf (McKown et al., 2014). Stomata, as sites of water vapor exchange, are also implicated in driving environmental change across biomes

69 (Hetherington and Woodward, 2003) and variation of stomatal density and aperture pore length are linked to changes of ecosystem productivity (Wang et al., 2015). Stomatal traits are of particular interest to paleoecologists and paleoclimatologists due to the relationship between stomatal traits and gas exchange. Measurements of stomatal density from fossil plants have been proposed as an indicator of paleoatmospheric CO2 concentration (Royer, 2001), and measuring stomatal traits to predict paleoclimates has become widely adopted (JC McElwain and Steinthorsdottir, 2017). It is clear from their physiological importance that researchers across a wide variety of disciplines in plant biology will phenotype stomatal traits for decades to come. A typical stomatal phenotyping pipeline consists of collecting plant tissue, creating a mounted tissue for imaging, imaging of the specimen, and manual phenotyping of a trait of interest. This last step can be the most laborious, costly, and time-consuming task, reducing the eciency of the data acquisition and analysis pipeline. This is especially important in large-scale plant breeding and genome-wide association studies, where phenotyping has been recognized as the new data collection bottleneck, in comparison to the relative ease of generating large genome sequence datasets (Hudson, 2008). Here, we seek to minimize the burden of high-throughput phenotyping of stomatal traits by introducing an automated method to identify and count stomata from micrographs. Although automated phenotyping methods using computer vision have been developed (Higaki et al., 2014; Laga et al., 2014; Duarte et al., 2017; Jayakody et al., 2017), these highly specialized approaches require feature engineering speciﬁc to a collection of images. These methods do not transfer well to images created with novel imaging and processing protocols. Additionally, tuning hand-crafted methods

70 to work on a general set of conditions is cumbersome and often impossible to achieve. Recent applications of deep learning techniques to stomata counting have demonstrated success (Bhugra et al., 2018; Aono et al., 2019), but are not easily accessible to the public and use training data sets sampled from few taxa. Deep convolutional neural networks (DCNN) circumvent specialized approaches by training the feature detector along with the classifier (LeCun et al., 2015). The method has been widely successful on a range of computer vision problems in biology such as medical imaging (see Shen et al. (2017) for a review) or macroscopic plant phenotyping (Ubbens and Stavness, 2017). The main caveat of deep learning methods is that a large number of parameters have to be trained in the feature detector. Im- provements in network structure (He et al., 2016) and training procedure (Simonyan and Zisserman, 2014) have helped training of large networks which incorporate gradient descent learning methods and prove to be surprisingly resilient against overfitting (Poggio et al., 2017). Nevertheless, a large number of correctly annotated training images is still required to allow the optimizer to converge to a correct feature repre- sentation. Large labeled training sets like the ImageNet database exist (Deng et al., 2009), but for a highly specialized problems, such as stomata identification, publicly available datasets are not available at the scale required to train a typical DCNN. We solve these problems by creating a large and taxonomically diverse training dataset of plant cuticle micrographs and by creating a network with a human-in- the-loop approach. Our development of this method is provided to the public as a web-based tool called StomataCounter which allows plant biologists to rapidly upload plant epidermal image datasets to pre-trained networks and then annotate stomata on cuticle images when desired. We apply this tool to a the training dataset and

71 achieve robust identiﬁcation and counts of stomata on a variety of angiosperm and pteridosperm taxa.

2.2 Materials and Methods

2.2.1 Biological Material

Micrographs of plant cuticles were collected from four sources: the cuticle database

(https://cuticledb.eesi.psu.edu/, Barclay, J McElwain, et al., 2007); a Ginkgo common garden experiment (Barclay and Wing, 2016); a new intraspeciﬁc collection of balsam poplar (Populus balsamifera); and a new collection from living material at the Smithsonian, National Museum of Natural History (USNM) and the United States Botanic Garden (USBG) (Table 2.1). Specimens in the cuticle database collection were previously prepared by clearing and staining leaf tissue and then imaged. The entire collection of the cuticle database was downloaded on November 16, 2017. Downloaded images contained both the abaxial and adaxial cuticles in a single image, and were automatically separated with a custom bash script. Abaxial cuticle micrographs were discarded if no stomata were visible or if the image quality was so poor that no stomata could be visually identiﬁed by a human. Gingko micrographs were prepared by chemically separating the upper and lower cuticles and imaging with an environmental SEM Barclay and Wing, 2016. To create the poplar dataset, dormant cuttings of P. balsamifera genotypes were collected across the eastern and central portions of its range in the United States and Canada by S.R. Keller et al. and planted in common gardens in Vermont and Saskatchewan in 2014. Fresh leaves were sampled

72 from June to July 2015 and immediately placed in plastic bags and then a cooler for temporary storage, up to three hours. In a laboratory, nail polish (Sally Hansen, big kwik dry top coat #42494) was applied to a 1 cm2 region of the adaxial and abaxial leaf surfaces, away from the mid-vein, and allowed to dry for approximately twenty minutes. The dried cast of the epidermal surface was lifted with clear tape and mounted onto a glass slide. The collection at the USNM and USBG was made similarly from opportunistically sampled tissues of the gardens around the National Museum of Natural History building. USBG collections were focused on the Hawai- ian, southern exposure, and medicinal plant, and tropical collections. The taxonomic identity of each specimen was recorded according to the existing label next to each plant. The USNM/USBG collection was imaged with an Olympus BX-63 microscope using dierential interference contrast (DIC). Some specimens had substantial relief and z-stacked images were created using cellSens image stacking software (Olympus Corporation). Slides were imaged in three non-adjacent areas per slide. The poplar collection was imaged with an Olympus BX-60 using DIC and each slide was imaged in two areas. Mounted material for the USNM/USBG and Ginkgo collections are deposited in the Smithsonian Institution, National Museum of Natural History in the Department of Paleobiology. Material for the poplar collection is deposited in the Keller laboratory in the Department of Plant Biology at the University of Vermont, USA. The training dataset totaled 4618 images (Table 2.1).

2.2.2 Deep Convolutional Neural Network

We used a deep convolutional neural network (DCNN) to generate a stomata likelihood map for each input image, followed by thresholding and peak detection to

73 localize and count stomata (Fig. 1). Because dense per-pixel annotations of stoma versus non-stoma are dicult to acquire in large quantity, we trained a simple image classification DCNN based on the AlexNet structure instead (Krizhevsky et al., 2012), and copied the weights into a fully convolutional network to allow per-location assessment of stomata likelihood. Although this method does not provide dense per- pixel annotations, the resulting resolution proved to be high enough to dierentiate and count individual stomata. We used pre-trained weights for the lowest five convolutional layers from conv1 to conv5. The weights were taken from the ILSVRC image classification tasks (Rus- sakovsky et al., 2015) made available in the caenet distribution (Jia et al., 2014). All other layers were initialized using Gaussian initialization with scale 0.01. Training was performed using a standard stochastic gradient descent solver as in Krizhevsky et al. (2012), with learning rate 0.001 for pre-trained layers and 0.01 for randomly initialized layers with a momentum of 0.9. Because the orientation of any individual stomata does not hold information for identification, we augmented data by rotating all training images into eight dierent orientations, applied random flipping and randomly positioned crop regions of the input size within the extracted 256x256 image patch. For distractors, we sampled patches from random image regions on human-annotated images that were at least 256 pixels distant from any labeled stoma. The trained network weights were transferred into a fully convolutional network (Long et al., 2015), which replaces the final fully connected layers by convolutions. To increase the resolution of the detector slightly, we reduced the stride of layer pool5 from two to one, and added a dilation (Yu and Koltun, 2015) of two to layer fc6conv to compensate. Due to margins (96 pixels) and stride (32 over all layers), application

74 of the fully convolutional network to an image of size s yields an output probability map p of size s (96 2)/32 along each dimension. ≠ ú To avoid detecting low-probability stomata within the noise, the probability map p was thresholded and all values below the threshold p-thresh=0.98 were set to zero. Local peak detection was run on a 3x3 pixel neighborhood on the thresholded map, and each peak, excluding those located on a 96-pixel width border, was labeled as a stoma center. This intentionally excludes stomata for which the detection peak is found near the border within the model margin to match the instructions given to human annotators. Resulting stomata positions were projected back onto the original image (see Fig. 2.2).

We built a user-friendly web service, StomataCounter, freely available at http:

//stomata.uvm.edu/, to allow the scientific community easy access to the method. We are using a flask/jquery/bootstrap stack. Source codes for network training, as well as the webserver are available at http://stomata.uvm.edu/source.To use StomataCounter, users upload single .jpg images or a zip files of of their .jpg images containing leaf cuticles prepared. Z-stacked image sets should be combined into single image before uploading. A new dataset is then created where the output of the automatic counts, image quality scores and image metadata are recorded and can easily be exported for further analysis. In addition to automatic processing, the user can manually annotate stomata and determine the empirical error rate of the automatic counts through a straightforward and intuitive web interface. The annotations can then be reincorporated into the training dataset to improve future performance of StomataCounter by contacting the authors and requesting retraining of the DCNN.

75 2.2.3 Statistical Analyses

We tested the performance of the DCNN with a partitioned set of images from each dataset source. Whenever possible, images from a given species were used in either the training or test set, but not both, and only 7 out of 1467 species are included in both. In total, 1941 images were used to test the performance of the network (Table 2.1). After running the test set through the network, stomata were manually annotated. If the center of a stoma intersected the bounding box around the perimeter of the image, it was not counted. To evaluate the DCNN, we ﬁrst determined if it could identify stomata when they are known to be present and fail to identify them when they are absent. To execute this test, a set of 25 randomly selected abaxial plant cuticle micrographs containing stomata were chosen from each of the four datasets for a total of 100 images. To create a set of test images known to lack stomata, 100 adaxial cuticle micrographs were randomly sampled from the cuticle database. Visual inspection conﬁrmed that none of the adaxial images contained stomata. Micrographs of thoracic aorta from an experimental rat model of preeclampsia (Johnson and Cipolla, 2017), and breast cancer tissue micrographs (Gelasca et al., 2008) were used as negative controls from non-plant material. As a second test, we deteremined how well the method could identify stomata from stomatal patterning mutants in Arabidopsis that violate the single cell spacing rule. We sampled images from Papanatsiou et al. (2016) including

88 tmm1 and 90 wild type micrographs, and used a paired t-test to determine if counts from manual or automatic were statistically dierent. We then determined if the precision (see below) was dierent between mutant and wild type accessions with

76 ANOVA. Identification accuracy is tested by applying the DCNN to a small image patch either centered on a stoma (target), or taken from at least 256 pixels distance to any labeled stomata (distractor). This yields true positive (NTP), true negative (NTN), false positive (NFP) and false negative (NFN) samples. We define the classification accuracy A as:

N + N A = TP TN NTP + NTN + NFP + NTP

We deﬁne classiﬁcation precision P as:

N P = TP NTPNFP

We assume the human counts contain only true positives and the automatic count contain true positives and false positives. As such, we deﬁne precision as:

Human Count Precision = log AAutomatic CountB

This definition of precision identifies over-counting errors as negative values and under-counting as positive values. This measure of precision is undefined if either manual or automatic count is zero, and 30 of the 1772 observations were discarded. These samples were either out of focus, lacked stomata entirely, or too grainy for human detection of stomata. Classification recall R was defined as:

N R = TP NTP + NFN

77 Classification accuracy, precision, and recall, were calculated from groups of images constructed to span the diversity of imaging capture methods (i.e. brightfield, DIC, and SEM) and magnification (i.e. 200x and 400x) to determine how well the training set from one group transfers to identifying stomata in another group. Groups used for transfer assessment were the cuticle database, the Ginkgo collection, micrographs imaged at 200x, at 400x, and the combined set of images. We use linear regression to understand the relationship between human and automatic stomata counts. For the purpose of calculating error statistics, we consider deviation from the human count attributable to error in the method. Images were partitioned for linear models by collection source, higher taxonomic group (i.e. Eu- dicot, Monocot, Magnoliid, Gymnosperm, Fern, or Lycophyte), and magnification. To understand how dierent sources of variance contribute to precision variance, we collected data on the taxonomic family, magnification, imaging method, and three measures of image quality. The taxonomic family of each image was determined using the open tree of life taxonomy accessed with the rotl package (Michonneau et al., 2016). Two image quality measures used were based on the properties of an image’s power-frequency distribution tail and described the standard deviation (fSTD) and mean (fMean) of the tail. Low values of fSTD and fMean indicate a blurry image, while high values indicate non-blurry images. The third image quality measure, tEn- tropy, is a measure an image’s information content. High entropy values indicate high contrast/noisy images while low values indicate lower contrast. These image quality measures were created with PyImq (Koho et al., 2016) and standardized between zero and one. Random eects linear models were created with the R package lme4 (Bates et al., 2014) by fitting log precision to taxonomic family, magnification, and

78 imaging method as factors. Linear models were ﬁt for the scaled error and image quality scores. We used the root mean square error (RMSE) of the model residuals to understand how the factors and quality scores described the variance of log precision. Higher values of RMSE indicate larger residuals. Statistical analyses were conducted in python and R (R Core Team, 2013). Stomata counts, image quality scores, and taxonomy of training and test set micrographs are provided in supporting information table S1. Images used for the training and test sets are available for download from dryad (https://doi.org/10.5061/dryad.kh2gv5f).

2.3 Results

2.3.1 Stomata detection

StomataCounter was able to accurately identify and count stomata when they were present in an image. False positives were detected in the adaxial cuticle, aorta, and breast cancer cell image sets at low frequency (Fig. S2.1). The mean number of stomata detected in the adaxial, aorta, and breast cancer image sets was 1.5, 1.4, and 2.4, respectively, while the mean value of the abaxial set containing stomata was 24.1.

Stomata measured from micrographs of mutant and wild type Arabidopsis by human and automatic counting showed few dierences between counting method (Fig. S2.2). A paired T-test of human and automatic counts failed to reject the null hypothesis of no dierence in means (t = -0.91, df = 175, p-value = 0.37), but precision was dierent between wild type and mutant genotypes (ANOVA: F-value =

79 07 2 28.82, p-value = 2.51e≠ ). The variance of precision in Col-0 accessions (‡ = 0.189) was higher than the mutant accessions (‡2 = 0.153), possibly due to image capture settings on the microscope. Correspondence between automated and human stomata counting varied among the respective sample sets. There was close agreement among all datasets to the human count, with the exception of some of the samples imaged at 200x magnification (Fig. 2.3a-c). In these samples, the network tended to under count relative to human observers. Despite the variation among datasets and the large error present in the 200x dataset, the slopes of all models were close to 1 (Table 2.2). The 400x, Ginkgo, and cuticle database sets all performed well at lower stomata counts, as indicated by their proximity to the expected one-to-one line and decreased in precision as counts increased. Precision has a non-linear relationship with image quality, and changes in variance of precision are correlated with variation of image quality. With a ratio of 17:1 images in the training vs. test sets, the Poplar dataset had the best precision, followed by the Ginkgo, Cuticle database, and USNM/USBG datasets (Fig. 2.3d). Among the dierent imaging methods, SEM had the best precision, followed by DIC, and finally and brightfield microscopy (Fig. 2.3e). The variance of precision is higher for images with 200x magnification (Fig. 2.3f). RMSE values were lowest for taxonomic family and the family:magnification interaction, suggesting these factors contributed less to deviations between human and automated stomata counts than image quality or imaging method (Table 2.3).

80 2.3.2 Classification accuracy

The peak accuracy (94.2%) on the combined test sets is achieved when all training sets are combined (Fig. 2.6). The combined dataset performs best on all test subsets of the data; i. e. adding additional training data - even from dierent sets - is always beneﬁcial for the generalization of the network. Accuracy from train to test within a single species is higher (e. g. Ginkgo training for Ginkgo test at 97.4%) than transfer within datasets with a large number species across families (400x training to 400x test: 85.5%). The network does not generalize well between vastly dierent scales, i.e. the 200x- dataset, which contains images downscaled to half the image width and height. In this case, only training within the same scale achieves high accuracy (97.3%), while adding additional samples from the larger scale reduces the performance (to 90.7%). Precision values are generally higher than recall (0.99 precision on the combined training and test sets; 0.93 recall, Fig. S2.3), which shows that we mostly miss stomata rather than misidentifying non-existing stomata. Increased training size is correlated with increased accuracy (Fig. 2.4), and providing a large number of annotated images is beneﬁcial, as it lifts training accuracy from 72.8% with a training set of ten images to 94.2% with the complete set of training images.

81 2.4 Discussion

Stomata are an important functional trait to many fields within plant biology, yet manual phenotyping of stomata counts is a laborious method that has few controls on human error and reproducibility. We created a fully automatic method for counting stomata that is both highly sensitive and reproducible, allows the user to quantity error in their counts, but is also entirely free of parameter optimization from the user. Furthermore, the DCNN can be iteratively retrained with new images to improve performance and adjust to the needs of the community. This is a particular advantage of this method for adjusting to new taxonomic sample sets. However, new users are not required to upload new image types or images from new species for the method to work on their material. The pre-loaded neural network was specifically trained on diverse set of image types and from many species (N = 739). As the complexity of processing pipelines in biological studies increases, reproducibility of studies increasingly becomes a concern (Vieland, 2001). Apart from the reduced workload, automated image processing provides better reproducibility than manual stomata annotations. For instance, if multiple experts count stomata, they may not agree, causing artificial dierences between compared populations. This includes how stomata at the edge of an image are counted, and what to do with dicult to identify edge cases. Automatic counters will have an objective measure, and introduce no systematic bias between compared sets as long as the same model is used. Additionally, our human counter missed stomata that the machine detected (Fig. 2.5). Our method is not the first to identify and count stomata. However, previous

82 methods have not been widely adopted by the community and a survey of recent literature indicates manual counting is the predominant method (Liu et al., 2018; Morales-Navarro et al., 2018; Sumathi et al., 2018; Peel et al., 2017; Takahashi et al., 2015). Previous methods relied on substantial image pre-processing to generate images for thresholding to isolate stomata for counting (Oliviera et al., 2014; Duarte et al., 2017). Thresholding can perform well in a homogenous collection of images, but quickly fails when images collected by dierent preparation and microscopy methods are provided to the thresholding method (K. Fetter, pers. obs.). Some methods also require the user to manually segment stomata and subsequently process those images to generate sample views to supply to template matching methods (Laga et al., 2014). Object-oriented methods (Jian et al., 2011) also require input from the user to deﬁne model parameters. These methods invariably requires the user to participate in the counting process to tune parameters and monitor the image processing, and are not fully autonomous. More recently, cascade classiﬁer methods have been developed which perform well on small collections of test sets (Vialet-Chabrand and Brendel, 2014; Jayakody et al., 2017). Additionally, most methods rely on a very small set of images (50 to 500) typically sampled from just a few species or cultivars to create the training and/or test set (e.g., Bhugra et al., 2018). Apart from generalization concerns, several published methods require the user to have some experience coding in python or C++, a requirement likely to reduce the potential pool of end users. Our method resolves these issues by being publicly available, fully autonomous of the user, who is only required to upload jpeg formatted images, is free of any requirement for the user to code, and is trained on a relatively large and taxonomically diverse set of cuticle images. Users may wish to set up

83 StomataCounter locally on their own servers, and we have integrated the python scripts into an oine program users can run at the command line. The command line version is available in the github repository: https://github.com/SvenTwo/epidermal. Users can generate model weights from a custom set of training images and supply it to the image processing script for stomatal identification, or use the predefined weights generated from this work. We have demonstrated that this method is capable of accurately identifying stomata when they are present, but false positives may still be generated by shapes in images that approximate the size and shape of stomata guard cells. Conversely, false negatives are generated when a stomata is hidden by a feature of the cuticle or if poor sample preparation/imaging introduces blur. This issue is likely avoidable through increased sample size of the training image set and good sample preparation and microscopy techniques by the end user. The method is also able to detect stomata from mutant Arabidopsis genotypes that violate the single spacing rule. The importance of having a well-matched training and testing image set was apparent at 200x, where there was a subset of observations with low transfer accuracy (Fig. 2.6), and StomataCounter consistently under counted relative to human observation (Fig. 2.3). We argue that transferring architecture between scales is not advisable and images should be created by the user to match the predominant size (2048x2048) and magnification (400x) of images in the training network. Our training set of images spanned 82 dierent families and was over-represented by angiosperms. Stomata in gymnosperms are typically sunken into pores that make it dicult to obtain good nail polish casts. Models tested to explain variation in scaled error revealed that taxonomic family and its interaction with magnification were the factors

84 that had the best explanatory power for scaled error. Future collections of gymnosperm cuticles could be uploaded to the DCNN to retrain it in order to improve the performance of the method for gymnosperms. More generally however, this highlights how users will need to be thoughtful about matching of training and test samples for taxa that may deviate in stomata morphology from the existing reference database. We therefore recommend that users working with new or morphologically divergent taxa first run several pilot tests with dierent magnification and sample preparation techniques to find optimal choices that minimize error for their particular study system. SEM micrographs had the least amount of error, followed by DIC, and finally brightfield (Fig. 2.3g-i). Lastly, image quality was strongly related to log precision; predictably, images that are too noisy (i.e. high entropy) and out-of-focus (low fMean or fSTD) will generate higher error. Obtain- ing high quality, in-focus images should be a priority during data acquisition. We provide these guidelines for using the method, and recommend that users read these guidelines before collecting a large quantity of images:

• Collect sample images using dierent microscopy methods from the same tissues. We recommend an initial collection of 25 to 100 images prior to initiating a new large-scale study.

• Run images through StomataCounter.

• Establish a ’true’ stomata count using the annotation feature.

• Regress image quality scores (automatically provided in output csv ﬁle) against log precision.

85 • Regress human versus automatic counts and assess error.

• Choose the microscopy method that minimizes error and image the remaining samples.

Dierent microscopy methods can include using DIC or phase contrast ﬁlters, adjusting the aperture to increase contrast, or staining tissue and imaging with under ﬂuorescent light (see Eisele et al., 2016 for more suggestions). If a large collection of images is already available and re-imaging is not feasible, we recommend the users take the following actions:

• Randomly sample 100 images.

• Upload the images to StomataCounter.

• Annotate images to establish the ’true’ count.

• Explore image quality scores with against the log(precision) to determine a justiﬁable cut-ovalue for ﬁltering images.

• Discard images below the image quality cut-ovalue.

New users can also contact the authors through the StomataCounter web interface and we may retrain the model to include the 100 annotated images. Fast and accurate counting of stomata increases productivity of workers and decreases the time from collecting a tissue to analyzing the data. Until now, assessing

86 measurement error required phenotyping a reduced set of images multiple times by, potentially, multiple counters. With StomataCounter, users can instantly phenotype their images and annotate them to create empirical error rates. The open source code and ﬂexibility of using new and customized training sets will make StomataCounter and important resource for the plant biology community.

2.5 Acknowledgements

We thank Kyle Wallick at the United States Botanic Garden for facilitating access to the living collections of the Garden. Scott Whitaker aided in imaging cuticle specimens with SEM and DIC microscopy in the Laboratories of Analytical Biology at the Smithsonian Institution, National Museum of Natural History. Rat aorta images were provided by Dr. A. Chapman and Dr. M. Cipolla at the University of Vermont. Arabidopsis stomatal patterning mutants were provided by Dr. Maria Papanatsiou and Dr. Michael Blatt. Dr. Terry Delaney at the University of Vermont kindly allowed KCF to use his microscope for imaging. Funding to create Stomata Counter was provided by an NSF grant to SRK (IOS-1461868) and a Smithsonian Institution Fellowship to KCF.

2.6 Data availability

Training and test set micrographs, model weights, and caenet model deﬁnition protocol are available as downloads from Dryad (https://doi.org/10.5061/dryad. kh2gv5f).

87 2.7 Author contributions

The research was conceived and performed by KCF and SE. The website and python scripts were written by SE. Data and cuticle images were collected and analyzed by KCF. Ginkgo Images were submitted by RSB and SW. All authors interpreted the results. The manuscript was written by KCF, SE, and SRK. All authors edited and approved the manuscript.

88 2.8 Tables

Table 2.1: Description of the datasets used for training and testing the network. Abbrevia- tions: DIC, Dierential interference contrast; SEM, scanning electron microscopy.

Training Test Preparation Imaging Z-stacks Dataset N N N N Magnification Citation images species images species method method applied Poplar 3123 1 175 1 Nail polish DIC 400 no This paper Ginkgo 408 1 200 1 Lamina peel SEM 200 no Barclay and Wing, 2016 Cuticle DB 678 613 696 599 Clear & stain Brightfield 400 no Barclay, J McElwain, et al., 2007 USNM/USBG 409 124 696 132 Nail polish DIC, SEM1 1002, 200, 400 yes This paper Aorta - - 116 1 Elastic-Van Gieson Brightfield 40 no Johnson and Cipolla, 2017 Hematoxylin Breast Cancer - - 58 1 Brightfield no Gelasca et al., 2008 & eosin stain Total 4618 739 1941 735 Note: 1) Training SEM N = 15; 2) training set, N = 4.

89 Table 2.2: Summary of linear model fit parameters in Figure 2.3 for dierent test datasets. Dataset definitions given in text. Abbreviations: y-intercept, a; standard error, SE. 200x † images removed from this USNM/USBG set. Significance indicated by: ú: P<0.05; úú: P<0.01; úúú: P<0.001. (A) Dataset Cuticle DB Poplar Ginkgo USNM/USBG USNM/USBG † a 3.315úúú 0.823 -0.364 10.068úúú -0.695 SEa (0.635) (0.456) (0.677) (2.041) (0.5) s 0.853úúú 0.978úúú 0.832úúú 0.997úúú 1.167úúú SEs (0.029) (0.014) (0.028) (0.089) (0.026) r2 0.578 0.964 0.812 0.157 0.835

(B) Taxonomic group Eudicot Monocot Magnoliid Gymnosperm Fern Lycophyte a 10.405úúú 2.360úúú 4.360ú 0.004 0.624 0.771 SEa (1.643) (0.598) (1.844) (0.735) (0.474) (1.414) s 0.808úúú 0.808úúú 0.912úúú 0.830úúú 0.799úúú 0.751úú SEs (0.063) (0.056) (0.086) (0.031) (0.039) (0.169) r2 0.141 0.617 0.272 0.778 0.938 0.739

(C) Magniﬁcation 200x 400x a 18.346*** 1.446*** SEa (3.192) (0.381) s 0.654*** 0.970*** SEs (0.123) (0.017) r2 0.055 0.740

90 Table 2.3: Root mean square error (RMSE) of the model residuals. Lower RMSE values suggest a better ﬁt of the model. The response in each model was precision. Abbreviations: fMean, mean power-frequency tail distribution; fSTD, standard deviation of power-frequency tail distribution.

Model RMSE

family 0.463 ≥ magniﬁcation 0.523 ≥ family:magniﬁcation 0.399 ≥ imaging method 0.519 ≥ fMean 0.523 ≥ fSTD 0.523 ≥ tEntropy 0.521 ≥ tEntropy:imaging method 0.515 ≥

91 2.9 Figures

Figure 2.1: Architecture of the deep convolutional neural network (DCNN) and classification tasks. Left: Training and testing procedure. First column: Target patches were extracted and centered around human-labeled stomata center positions; distractor patches were extracted in all other regions. A binary image classification network was trained. Second column: The image classification network was applied fully convolutional to the test image to produce a prediction heatmap. On the thresholded heatmap, peaks were detected and counted.

92 Figure 2.2: Sample sizes of images for the top 20 families represented in the training (yellow) and test (blue) datasets (a). Examples of pre- and post-analysis images. A probability heatmap map is overlain onto the input image in the red channel. Detected stomata marked with circles with peak values given in green (b). Species in image pairs: Adiantum peru- vianum, Begonia loranthoides (top row); Chaemaerauthum gadacardii, Echeandia texensis (second row); Ginkgo biloba, Populus balsamifera (third row); Trillium luteum, Pilea libanensis (bottom row). Scale bars in row one and two equal 20 µm, scale bars in row three equal 25 µm, and scale bars in row four equal 50 µm.

93 Figure 2.3: Results of human and automatic counts and precision for each dataset organized by collection source (a,d), taxonomic group (b), and magnification (c), and imaging method (e). Precision and image quality scores have non-linear relationships, and changes in variance correlate to image quality (g-i). Positive values of precision indicate undercounting of stomata relative to human counts, while negative values indicate overcounting. The dotted black line is the 1:1 line. See Table 2.2 for model summaries. Abbreviations: tEntropy, image entropy ;fMean, mean power-frequency tail distribution; fSTD, standard deviation of power-frequency tail distribution. 94 Figure 2.4: Accuracy by training image count and the eect of increasing the training set size on classification accuracies. Since classification is a binary task, chance level is at 50%. Training image count “all” includes all 4,618 annotated training images.

False Negatives False Positives Human-annotated stomata Human-annotated as non-stomata Machine-annotated as non-stomata Machine-annotated as stomata

Figure 2.5: Samples that were mislabeled with high conﬁdence by either the machine or human. False negatives (left panel) from stomata that were detected by the human, but not by the machine. Image features typically generating false negatives are blurry images, artifacts obstruction the stomata, low contrast between epidermal and guard cells, and very small scale stomata. False positives (right panel) are image features that are labeled by the machine as stomata, but not by the human counter, or are missed by the human counter (e.g. top and bottom right images in the panel). Errors typically generating false positives are image artifacts that superﬁcially resemble stomata, particularly shapes mimicking interior guard cell structure.

95 Figure 2.6: Accuracies for models trained on dierent training datasets (vertical), tested on dierent test datasets. Combined is a union of all training and test sets. For precision and recall values, see Supporting Information Fig. S2.3.

96 Supporting Information

*** b

a 20

Automatic stomata count Automatic a a

0 abaxial_cuticle adaxial_cuticle aorta breast_cancer

Figure S2.1: Stomata identiﬁcation test results

a b ● c

9 9 ● ● ●

● 1.0 ● ● ● ● 6 *** a a*** 6 ● ● ● ● ● ● ● ● ● Freq

*** ● ● ● ● ● ● ● Human Count Stomata count b 0.5 *** 3 b 3 ● ● ● ● ●

● ● ● ● ●

0 0 0.0

mutant wt 0.0 2.5 5.0 7.5 −1 0 1 Automatic Count Precision

Counting method

Human Automatic ● mutant ● wt mutant wt

Figure S2.2: Stomata identiﬁcation in Arabidopsis cell-patterning mutants.

97 Figure S2.3: Precision and recall of transferred training and testing datasets.

98 References

Aono, A, J Nagai, G Dickel, R Marinho, P Oliveira, and F Faria (2019). “A stomata classiﬁcation and detection system in microscope images of maize cultivars”. In:

bioRxiv, p. 538165. Barclay, R, J McElwain, D Dilcher, and B Sageman (2007). “The cuticle database: developing an interactive tool for taxonomic and paleoenvironmental study of the

fossil cuticle record”. In: Courier-Forschungsinstitut Senckenberg 258, p. 39.

Barclay, R and S Wing (2016). “Improving the Ginkgo CO2 barometer: implications for the early Cenozoic atmosphere”. In: Earth and Planetary Science Letters 439, pp. 158–171. Bates, DM, M Maechler, BM Bolker, and S Walker (2014). “Fitting Linear Mixed-

Eects Models Using lme4”. In: Journal of Statistical Software arXiv, p. 1406.5823. Bergmann, DC and FD Sack (2007). “Stomatal development”. In: Annual Review of Plant Biology 58, pp. 163–181. Bhugra, S, D Mishra, A Anupama, S Chaudhury, B Lall, A Chugh, and V Chinnusamy (2018). “Deep convolutional neural networks based framework for estimation of

stomata density and structure from microscopic images”. In: European Conference on Computer Vision. Springer, pp. 412–423. Deng, J, W Dong, R Socher, LJ Li, K Li, and L Fei-Fei (2009). “Imagenet: A large-

scale hierarchical image database”. In: 2009 IEEE conference on computer vision and pattern recognition. Ieee, pp. 248–255. Duarte, KT, MAG de Carvalho, and PS Martins (2017). “Segmenting high-quality digital images of stomata using the wavelet spot detection and the watershed

99 transform.” In: VISIGRAPP (4: VISAPP). Setubal, Portugal: Science and Tech- nology Publications, Lda, pp. 540–547. Eisele, JF, F Fäßler, PF Bürgel, and C Chaban (Oct. 2016). “A rapid and simple

method for microscopy-based stomata analyses”. In: PLOS ONE 11.10, pp. 1–13. Fischer, R, D Rees, K Sayre, Z Lu, A Condon, and L Saavedra (1998). “Wheat yield progress associated with higher stomatal conductance and photosynthetic rate,

and cooler canopies”. In: Crop science 38.6, pp. 1467–1475. Gelasca, ED, J Byun, B Obara, and BS Manjunath (2008). “Evaluation and bench-

mark for biological image segmentation”. In: Proceedings - International Confer- ence on Image Processing, ICIP, pp. 1816–1819. He, K, X Zhang, S Ren, and J Sun (2016). “Deep residual learning for image recog-

nition”. In: Proceedings of the IEEE conference on computer vision and pattern recognition. Los Alamitos, California: IEEE Computer Society, pp. 770–778. Hetherington, AM and FI Woodward (2003). “The role of stomata in sensing and

driving environmental change”. In: Nature 424.6951, p. 901. Higaki, T, N Kutsuna, and S Hasezawa (2014). “CARTA-based semi-automatic de-

tection of stomatal regions on an Arabidopsis cotyledon surface”. In: Plant Mor- phology 26.1, pp. 9–12. Hudson, ME (2008). “Sequencing breakthroughs for genomic ecology and evolutionary

biology”. In: Molecular ecology resources 8.1, pp. 3–17. Jayakody, H, S Liu, M Whitty, and P Petrie (2017). “Microscope image based fully automated stomata detection and pore measurement method for grapevines”. In:

Plant methods 13.1, p. 94.

100 Jia, Y, E Shelhamer, J Donahue, S Karayev, J Long, R Girshick, S Guadarrama, and T Darrell (2014). “Cae: convolutional architecture for fast feature embedding”.

In: arXiv preprint arXiv:1408.5093. Jian, S, C Zhao, and Y Zhao (2011). “Based on remote sensing processing technol-

ogy estimating leaves stomatal density of Populus euphratica”. In: 2011 IEEE International Geoscience and Remote Sensing Symposium. IEEE, pp. 547–550. Johnson, AC and MJ Cipolla (2017). “Altered hippocampal arteriole structure and function in a rat model of preeclampsia: potential role in impaired seizure-induced

hyperemia”. In: Journal of Cerebral Blood Flow & Metabolism 37.8, pp. 2857–2869. Kim, TH, M Böhmer, H Hu, N Nishimura, and JI Schroeder (2010). “Guard cell

signal transduction network: advances in understanding abscisic acid, CO2, and Ca2+ signaling”. In: Annual review of plant biology 61, pp. 561–591. Koho, S, E Fazeli, JE Eriksson, and PE Hänninen (2016). “Image quality ranking

method for microscopy”. In: Scientiﬁc reports 6, p. 28962. Krizhevsky, A, I Sutskever, and GE Hinton (2012). “Imagenet classiﬁcation with deep

convolutional neural networks”. In: Advances in neural information processing systems, pp. 1097–1105. Laga, H, F Shahinnia, and D Fleury (2014). “Image-based plant stornata phenotyp-

ing”. In: 2014 13th International Conference on Control Automation Robotics and Vision, ICARCV 2014, pp. 217–222. LeCun, Y, Y Bengio, and G Hinton (2015). “Deep learning”. In: Nature 521.7553, p. 436.

101 Liu, C, N He, J Zhang, Y Li, Q Wang, L Sack, and G Yu (2018). “Variation of stomatal traits from cold temperate to tropical forests and association with water-

use eciency”. In: Functional Ecology 32.1, pp. 20–28. Long, J, E Shelhamer, and T Darrell (2015). “Fully convolutional networks for seman-

tic segmentation”. In: Proceedings of the IEEE conference on computer vision and pattern recognition. Los Alamitos, California: IEEE Computer Society, pp. 3431– 3440. McElwain, JC and M Steinthorsdottir (2017). “Paleoecology, ploidy, paleoatmospheric composition, and developmental biology: a review of the multiple uses of fossil

stomata”. In: Plant Physiology 174.2, pp. 650–664. McKown, AD, RD Guy, L Quamme, J Klápöt, J La Mantia, C Constabel, YA El-Kassaby, RC Hamelin, M Zifkin, and M Azam (2014). “Association genet-

ics, geography and ecophysiology link stomatal patterning in Populus trichocarpa with carbon gain and disease resistance trade-os”. In: Molecular Ecology 23.23, pp. 5771–5790. Michonneau, F, JW Brown, and DJ Winter (2016). “rotl: an R package to interact

with the Open Tree of Life data”. In: Methods in Ecology and Evolution 7.12, pp. 1476–1481. Morales-Navarro, S, R Pérez-Díaz, A Ortega, A de Marcos, M Mena, C Fenoll, E González-Villanueva, and S Ruiz-Lara (2018). “Overexpression of a SDD1-Like gene from wild tomato decreases stomatal density and enhances dehydration

avoidance in Arabidopsis and cultivated tomato”. In: Frontiers in plant science 9.

102 Oliviera, MWdS, NR da Silva, D Casanova, LFS Pinheiro, RM Kolb, and OM Bruno

(2014). “Automatic counting of stomata in epidermis microscopic images”. In: X Workshop de Visao Computacional 3, pp. 253–257. Papanatsiou, M, A Amtmann, and MR Blatt (2016). “Stomatal spacing safeguards stomatal dynamics by facilitating guard cell ion transport independent of the

epidermal solute reservoir”. In: Plant physiology 172.1, pp. 254–263. Peel, JR, MC Mandujano Sánchez, J López Portillo, and J Golubov (2017). “Stomatal

density, leaf area and plant size variation of Rhizophora mangle (Malpighiales: Rhizophoraceae) along a salinity gradient in the Mexican Caribbean”. In: Revista de Biología Tropical 65.2, pp. 701–712. Poggio, T, K Kawaguchi, Q Liao, B Miranda, L Rosasco, X Boix, J Hidary, and H Mhaskar (2017). “Theory of deep learning III: explaining the non-overﬁtting

puzzle”. In: arXiv preprint arXiv:1801.00173. RCoreTeam(2013).R: A Language and Environment for Statistical Computing.R Foundation for Statistical Computing. Vienna, Austria. Royer, DL (2001). “Stomatal density and stomatal index as indicators of paleoatmo-

spheric CO2 concentration”. In: Review of Palaeobotany and Palynology 114.1-2, pp. 1–28. Russakovsky, O, J Deng, H Su, J Krause, S Satheesh, S Ma, Z Huang, A Karpathy, A Khosla, M Bernstein, et al. (2015). “Imagenet large scale visual recognition

challenge”. In: International journal of computer vision 115.3, pp. 211–252. Shen, D, G Wu, and HI Suk (2017). “Deep learning in medical image analysis”. In:

Annual review of biomedical engineering 19, pp. 221–248.

103 Shimazaki, Ki, M Doi, SM Assmann, and T Kinoshita (2007). “Light regulation of

stomatal movement”. In: Annual Review of Plant Biology 58, pp. 219–247. Simonyan, K and A Zisserman (2014). “Very deep convolutional networks for large-

scale image recognition”. In: arXiv preprint arXiv:1409.1556. Sumathi, M, V Bachpai, B Deeparaj, A Mayavel, MG Dasgupta, B Nagarajan, D Rajasugunasekar, V Sivakumar, and R Yasodha (2018). “Quantitative trait loci

mapping for stomatal traits in interspeciﬁc hybrids of Eucalyptus”. In: Journal of genetics, pp. 1–7. Takahashi, S, K Monda, J Negi, F Konishi, S Ishikawa, M Hashimoto-Sugimoto, N Goto, and K Iba (2015). “Natural variation in stomatal responses to environmental

changes among Arabidopsis thaliana ecotypes”. In: PloS one 10.2, e0117449. Ubbens, JR and I Stavness (2017). “Deep plant phenomics: a deep learning platform

for complex plant phenotyping tasks”. In: Frontiers in plant science 8, p. 1190. Vialet-Chabrand, S and O Brendel (2014). “Automatic measurement of stomatal den-

sity from microphotographs”. In: Trees 28.6, pp. 1859–1865. Vieland, VJ (2001). “The replication requirement”. In: Nature Genetics 29.3, p. 244. Wang, R, G Yu, N He, Q Wang, N Zhao, and Z Xu (2015). “Latitudinal variation of leaf stomatal traits from species to community level in forests: linkage with

ecosystem productivity”. In: Nature Scientiﬁc Reports 5.14454, pp. 1–11. Yu, F and V Koltun (2015). “Multi-scale context aggregation by dilated convolutions”.

In: arXiv preprint arXiv:1511.07122.

104 Chapter 3

Trade-offs and selection conflicts in hybrid poplars indicate the stomatal ratio as an important trait regulating disease resistance.

Karl C. Fettera,1,DavidNelsonb, Stephen R. Kellera,1 aDepartment of Plant Biology, University of Vermont bUniversity of Maryland Center for Environmental Science 1Corresponding authors: Stephen R. Keller (e-mail: [email protected], tel: +1 802 656 2930) 1Karl C. Fetter (e-mail: [email protected])

105 Abstract

Natural selection can remove maladaptive genotypic variance from populations, leaving reduced phenotypic variation as a signal of its action. Hybrid populations oer a unique opportunity to study phenotypic variance before selection purifies it, as these populations can have increase genotypic and phenotypic variance than can reveal trade-os and selection conflicts not visible, or visible to a lesser extent, in unadmixed populations. Here, we study the interactions between a fungal leaf rust disease (Melampsora medusae) and stomata and ecophysiology traits in a set of hybrid and unadmixed individuals formed by natural matings between Populus balsamifera, P. trichocarpa, P. angustifolia, and P. deltoides. Phenotypes were measured on cloned genotypes grown in a common garden and genotyped at 227K SNPs with GBS. Our analyses indicate hybridization decreases disease resistance and increases the variance of stomatal ratio (SR), or the ratio of upper leaf surface stomata density to total stomata density. Heritability of SR was high in admixed populations (H2 =0.72) and covaries strongly with the proportion of P. balsamifera ancestry in a genome; thus, selection could eectively reduce disease by selecting for low values of stomatal ratio. However, selection conflicts present in some admixed populations may prevent adaptation to pathogens as a result of these populations being unable to occupy adaptive trait space along the growth-defense spectrum. These results suggest an important role for SR and stomatal patterning traits as a target of pathogen induced selection to achieve local fitness optima. Additionally, we demonstrate that hybridization is capable of generating, or magnifying maladaptive intermediate phenotypes to reveal trade-os and selection conflicts. 3.1 Introduction

Trade-os arise when selection on fitness components is constrained by negative correlations between pairs of traits (Stearns, 1989; Schluter et al., 1991). In plants, an important life history trade-ois between size and the age of first reproduction, with delayed reproduction correlating with longer fertility and higher quality ospring that have a greater chance of survival (Stearns, 1989). However, plants must cope with numerous natural enemies (e.g. herbivores and pathogens), and delaying reproduction can come at a high cost if diverting limited resources to defense away from reproduction reduces fitness (Obeso, 2002). Depending on pathogen prevalence, plant life histories generally evolve towards increased investment in defensive enzymes and compounds paired with slower growth, or decreased investment in defense paired with faster growth. The fast-slow trade-ohas been consistently identified in annuals (Tian et al., 2003), perennials (Messina et al., 2002), and long-lived trees (McKown, Guy, et al., 2014; McKown, Klápöt, et al., 2019), although this hypothesis is not without its critics (Kliebenstein, 2016). At the molecular level, the plant growth-defense trade-ois regulated by cellular signaling and hormonal regulation to direct metabolic activity away from growth and towards defense, or vice versa (Tian et al., 2003; Wang, 2012; Chandran et al., 2014). Host plant species may evolve a variety of costly defenses to combat disease, including the evolution of structural phenotypes to reduce exposure to pathogens (Gonzales-Vigil et al., 2017), immune systems to detect pathogens (Dangl and Jones, 2001) in order to initiate appropriate responses (Cai et al., 2018), and resistance via constitutive or inducible synthesis of defensive compounds (Bailey et al., 2005; Ullah

107 et al., 2018). Hosts may also avoid disease by migrating out of environments where a pathogen spatially or temporally occurs (Altizer et al., 2011), by colonizing new environments where the pathogen is absent (Bruns et al., 2018), or by evolving life history strategies that avoid or compensate for pathogen exposure (Stearns, 1989). The phenotypic expression of growth-defense trade-os may be altered by hybridization between parental species that have evolved dierent strategies of growth or defense. Hybridization is an important mechanism for evolutionary change, and has been implicated in the maintenance of species boundaries due to negative selection on advanced generation hybrids (Christe et al., 2016), introgression of beneﬁcial alleles across species barriers (Chhatre et al., 2018), and, among other evolutionary phenomena, speciation (Mallet, 2007). It is well-known that hybridization changes the mean and variance of traits in hybrid populations with regards to either parental species; indeed, crop breeders intentionally create hybrids to take advantage of hybrid vigor and transgressive segregation in F1 hybrids to increase yields and select for novel phenotypic combinations (Goulet et al., 2017). Transgressive segregation in hybrids is a source of new phenotypes for selection to act upon in natural populations (Goulet et al., 2017), and the balance between competing ﬁtness components underlying trade-os can change in hybrid populations (Ellstrand and Schierenbeck, 2000).

For example, hybrid Salix eriocephala x sericea have an increase in disease severity to willow leaf rust (Melampsora sp.) by a factor of 3.5 in comparison to unadmixed parents (Roche and Fritz, 1998). Thus, it is apparent that hybridization can have beneficial and deleterious eects on fitness, but the role of hybridization in revealing fitness trade-os that are not evident, or cryptic, within the parental species has been little explored.

108 One scenario where a trade-obetween plant growth and defense may arise is the relationship between stomatal traits and infection by fungal pathogens. Stomata are microscopic valves on the surface of the leaf that regulate gas exchange during photosynthesis. Some foliar pathogens enter their hosts via stoma by sensing the topography of the leaf surface for guard cells, and forming an appressorium over a stoma from which a penetration peg is used to invade the mesophyll tissue (Allen et al., 1991). The plant immune system has evolved mechanisms for detecting pathogens which can activate signal transduction pathways that lead stomatal closure (Melotto et al., 2006). However, closing stomata comes at the expense of limiting gas exchange, which may have negative eects on carbon assimilation depending on the water-use eciency of photosynthesis. Thus, the conditions for the evolution of a growth-defense trade-omediated by selection on stomata exists. Hybridization may be expected to shift genotypes along the trade-ocontinuum, and when competing strategies inherited from either parent meet in a hybrid, we may expect a mismatch between the optimal survival and growth strategy to result in increased disease. In this study, we test for the eects of hybridization on disease severity and stomatal related growth traits, and investigate if hybridization reveals trade-os and selection conflicts that may be masked in unadmixed populations. We use a sample of naturally formed hybrid genotypes between North American poplars infected by a common leaf rust of poplars, Melampsora medusae. We specifically ask the following questions: 1) how does hybridization change trait distributions, and does it alter heritable variation; 2) which traits correlate most strongly with disease severity; 3) are there selection trade-os between growth, disease severity, and stomatal traits; and 4) does hybridization reveal selection conflicts between traits underlying growth and

109 defense?

3.2 Materials & Methods

3.2.1 Study system

Poplars (Populus) are a genus of predominantly holarctic tree species (DiFazio et al., 2011). Extensive hybridization between species within a section of the genus, as well as between some sections, make the taxonomy of the genus dicult (Ecken- walder, 1996), with some authors identifying 29 (Eckenwalder, 1996) to as many as 60 species (Raven and Wu, 1999). Hybrids can be formed from species pairs within and between Populus sections Tacamahaca and Aigeiros, and extensive hybrid zones spontaneously form in where the ranges of two reproductively compatible species meet (Suarez-Gonzalez, Lexer, et al., 2018). Western North America contains several well documented hybrid zones (Martinsen et al., 2001; LeBoldus et al., 2013; Chhatre et al., 2018; Suarez-Gonzalez, Hefer, Lexer, Douglas, et al., 2018), including a tri-hybdrid zone in Alberta, Canada (Floate et al., 2016). In this study, we work with hybrids formed from crosses between the section Tacamahaca species balsam poplar (P. balsamifera), and either a black balsam poplar (P. trichocapra), a narrowleaf cottonwood (P. angustifolia), or an eastern cottonwood (P. deltoides) from the Populus section Aigeiros. We lack information on the which species served as the maternal vs. paternal parent for each hybrid, so we adopt the convention of listing the P. balsamifera parent ﬁrst. The disease we study is from the fungal leaf pathogen Melampsora medusae,a

110 macrocyclic basidiomycete whose aecial host is a larch (Larix), and telial host is a poplar (Feau et al., 2007). Uredospores (N + N) of M. medusae can clonally reproduce on poplar leaves within a single season, and the closely related M. larici-populina is an agricultural pest that can reduce yields of hybrid poplars grown for agroforestry (Pinon and Frey, 2005).

3.2.2 Materials

During winter 2013, dormant stem cuttings were collected from 534 trees from 59 populations spanning 9 Canadian provinces and 7 US States (longitudes -55 to -128 °W and latitudes 39 to 60 °N; Fig. 3.1, Table S3.1). The main focus of the 2013 collection was to sample P. balsamifera cuttings, but as the trees were dormant and some of the diagnostic traits were not immediately visible, a number of putative hybrids were also collected. Additionaly, 32 presumably unadmixed eastern cottonwood (P. deltoides) genotypes were collected in central Vermont, USA to serve as a reference population for identifying admixed P. deltoides hybrids with population genetic methods. For the 2013 collection, cuttings were grown for one year in a greenhouse, and then planted in the summer of 2014 in a common garden near Burlington, VT (44.444422 °N, -73.190164 °W). Replicates were planted in a randomized design with 2x2 meter spacing and 1,000 ramets were planted. The minimum, median, and maximum number of ramets per genet was 1, 2, and 6, respectively. Plants were not fertilized, but were irrigated as-needed during the 2014 growing season to ensure establishment, and then received no supplemental water.

111 3.2.3 Molecular analyses

Fresh foliage from greenhouse grown plants was used for extracting whole genomic DNA using DNeasy 96 Plant Mini Kits (Qiagen, Valencia, CA, USA). DNA was quan- tified using a fluorometric assay (Qubit BR, Invitrogen) and confirmed for high molecular weight using 1% agarose gel electrophoresis. We used genotyping-by-sequencing (GBS) (Elshire et al., 2011) to obtain genome-wide polymorphism data for all 534 trees. Genomic sequencing libraries were prepared from 100 ng of genomic DNA per sample digested with EcoT221 followed by ligation of barcoded adapters of varying length from 4-8 bp, following Elshire et al. (2011). Equimolar concentrations of barcoded fragments were pooled and purified with QIAquick PCR purification kit. Purified products were amplified with 18 PCR cycles to append Illumina sequencing primers, cleaned again using a PCR purification kit. The resulting library was screened for fragment size distribution using a Bioanalyzer. Libraries were sequenced at 48 plex (i.e., each library sequenced twice) using an Illumina HiSeq 2500 to generate 100 bp single end reads. Cornell University Institute of Genomic Diversity (Ithaca,

NY) performed the library construction and sequencing steps. For the P. deltoides reference population, library preparation was performed in the Keller lab using the same protocol. DNA sequencing was performed on an Illumina HiSeq 2500 at the Vermont Genetics Network core facility. Raw sequences reads are deposited in NCBI SRA under accession number SRP070954. We employed the Tassel GBS Pipeline (Glaubitz et al., 2014) to process raw sequence reads and call variants and genotypes. In order to pass the quality control, sequence reads had to have perfect barcode matches, the presence of a restriction

112 site overhang and no undecipherable nucleotides. Filtered reads were trimmed to

64 bp and aligned to the P. trichocarpa reference assembly version 3.0 (Tuskan et al., 2006) using the Burrows-Wheeler Aligner (BWA) (Li and Durbin, 2009). Single nucleotide polymorphisms (SNPs) were determined based on aligned positions to the reference, and genotypes called with maximum likelihood in Tassel (Glaubitz et al., 2014). SNP genotype and sequence quality scores were stored in Variant Call Format v4.1 (VCF) files, which were further processed with VCFTools 0.1.11 (Danecek et al., 2011). SNPs with a minor allele frequency < 0.001 were removed, and only biallelic sites were retained. Sites with with a mean depth < 5, genotype quality > 90, and indels were removed. Missing data was imputed with Beagle v5.0 (Browning et al., 2018), and sites with post-imputation genotype probability < 90 and sites with any missingness were removed. After filtering, the final dataset contained 227,607 SNPs for downstream analyses. The filial generation was estimated using NewHybrids (Anderson and Thompson, 2002) separately for each set of hybrids, (i.e. BxT, BxA, and BxD). Filial generation of P. balsamifera x angustifolia genotypes were previously estimated by Chhatre et al. (2018) and used here. NewHybrids requires reference populations for each species, and

P. trichocarpa reference genotypes were downloaded from previously published work, and we selected 25 genotypes from Pierce Co, Washington in populations known to lack admixture with P. balsamifera (Evans et al., 2014). P. deltoides reference genotypes were collected from the Winooski and Mad River watersheds in central

Vermont. The reference population for P. balsamifera was selected from individuals in the SLC, LON, and DPR populations that are known to lack admixture with other Populus species (see below). To select loci for distinguishing ﬁlial generations,

113 we determined the locus-wise FST dierence between reference populations. For the

BxT analysis, 355 loci with an FST dierence greater than 0.8 were randomly selected. In the BxD analysis, 385 loci that segregated completely between parental species (i.e.

FST dierence = 1) were randomly sampled. NewHybrids was run using Jerey’s prior for ﬁ and ◊ for 200,000 sweeps with 100,000 discarded as burn-in. The cross type (BxB, BxT, BxA, or BxD) was determined from the NewHybrids result and the expected proportion of the genome from each parental species was calculated as:

(2n 1) Expected ancestry proportion = ≠ 2n

where n = the number crossing events.

3.2.4 Trait measurements

All traits were measured from common garden grown trees in 2015, and disease severity was measured again in 2016 (Table 3.1). The severity of naturally innoculated leaf rust disease was phenotyped in both years using an ordinal scale from zero to four created by LaMantia et al. (2013) where: 0 = no uredinia visible, 1 = less than five uredinia per leaf on less than five leaves, 2 = less than five uredinia per leaf on more than five leaves, 3 = more than five uredinia per leaf on more than five leaves, and 4 = more than five uredinia on all leaves. In 2016, we also scored disease using the method of Dowkiw and Bastien (2004), where the single most infected leaf on a tree was classified as: 1 = no uredinia, 2 = 1 to 10 uredinia, 3 = 11 uredinia to 25 % of the leaf area, 4 = 25 to 50 % of the leaf area, 5 = 50 to 75 % of the leaf area, and 6 75 % of the leaf area covered in uredinia. We refer to the ordinal scale Ø

114 of La Mantia et al. (2013) as ordinal scale 1 and the ordinal scale of Dowkiw and Bastien (2004) as ordinal scale 2. A mature larch tree (approx. 20m tall) was located approximately 100m from the garden site, and we assume a uniform distribution of aeciospore inoculum into the garden. The pathogen was visually confirmed to be M. medusae using microscopy to view the ellipsoid to obovoid shape of uredospores, the size range (min = 19.6 µ m, mean = 28.3 µ m, max = 34.45 µ m, N = 17, Fig. 3.1f), and the presence of a smooth equatorial region on the spore flanked by polar regions with papillae (Van Kraayenoord et al., 1974) (Fig. 3.1f). To collect isotopic, elemental, and specific leaf area (SLA) data, three hole punches ( = 3 mm) were sampled in June 2015 from a central portion of each leaf adjacent to, but avoiding the central leaf vein. Hole punches were dried at 65 °C to constant mass. Approximately 2 mg of foliar tissue from each sample was weighed into a tin capsule and analyzed for %C, %N, ”13C, and ”15N using a Carlo Erba NC2500 elemental an- alyzer (CE Instruments, Milano, Italy) interfaced with a ThermoFinnigan Delta V+ isotope ratio mass spectrometer (Bremen, Germany) at the Central Appalachians Stable Isotope Facility (CASIF) at the Appalachian Laboratory (Frostburg, Mary- land, USA). %C and %N were calculated using a size series of atropine. The ”13C and ”15N data were normalized to the VPDB and AIR scales, respectively, using a two-point normalization curve with laboratory standards calibrated against USGS40 and USGS41. The long-term precision of an internal leaf standard analyzed along- side samples was 0.28%for ”13C and 0.24%for ”15N. Isotopic results are reported in units per mil (%) and are expressed as:

”13C =[(13C/12C )/(13C/12C ) 1] 103 sample V-PDB ≠ ú

115 where ”13C is the isotopic composition of the sample in %, and V-PDB is the Vienna-Pee Dee Belemnite international standard. Carbon isotope discrimination against 13C(13C) was calculated according to Farquhar et al. (1982) as:

13C =(”13C ”13C )/(1 + ”13C /103) a ≠ i i

13 where the ” Ca value (-8.456 %) was the mean value in 2015 measured at NOAA Mauna Loa observatory (White et al., 2011). Higher 13C values indicate increased intracellular (Ci) relative to atmospheric (Ca)CO2 concentrations as a result of greater stomatal conductance and/or lower photosynthetic assimilation rates in C3 plants. 13C is a useful metric of intrinsic water use eciency (ratio of photosynthesis to stomatal conductance) since both are inﬂuenced by Ci/Ca. To calculate SLA, or the ratio of fresh punch area to dry leaf mass, three oven dried hole punches per leaf were massed collectively. Relative growth rate (G) was measured as the height increment gain (cm) between the apical bud in 2015 and the previous year’s bud scar on the most dominant stem. The chlorophyll content index (CCI) was measured with a Konica Minolta SPAD 502 (Konica Minolta Sensing Americas, Inc) and the average of three measurements from the central portion of a leaf was recorded. Stomata traits were measured from micrographs of nail polish casts sampled in 2014 from the abaxial (lower) and adaxial (upper) leaf surfaces. Leaves were collected from the ﬁeld and placed in a cooler until processed in the lab. Nail polish peels were made and mounted on slides without a cover slip. Two non-overlapping areas without large veins from each peel were imaged (N = 1894) with an Olympus BX- 60 microscope using dierential interference contrast. Stomata density (SD) was estimated using the machine learning protocol of StomataCounter (Fetter et al., 2019).

116 To verify the automatic counts, stomata were manually annotated on each image using the image annotation tool. The correlation between automatic and manual count was r = 0.99, and we used the automatic count data to calculate stomata density. To measure stomata pore length (PL), each image was segmented with four equally spaced horizontal lines in ImageJ (Schneider et al., 2012) and the length of a single pore was measured in each segmented region for a total of ﬁve observations per image. A measure of pore surface area, porosity, was calculated using the following formula: porosity = density * (ﬁ * (pore length / 2)2). The stomata density ratio (SR), stomata pore length ratio (LR), and porosity ratio (PR) were calculated by dividing the value of the adaxial trait from the summed value of the abaxial and adaxial traits. For example, SR = AD_SD/(AD_SD + AB_SD).

3.2.5 Quantitative genetic analyses

To estimate clonal values of genets from the phenotyped ramets, mixed-eects models were fit with lme4 (Bates et al., 2015) and the response, here a trait, was modelled as a function of the fixed eects of garden row and column position, and random eects of genet. A normal distribution was assumed for all traits except: disease ordinal 1 (2015), disease ordinal 1 and 2 (2016), SR, LR, and PR. To model non- normal responses of disease ordinal replicate values, four error distributions were fit to the data and compared with AIC using glmmADMB (Fournier et al., 2012; Skaug et al., 2016) and lme4, which include: (1) a standard mixed-eects model; (2) a zero-inflated negative binomial model with variance = µ(1 + µ/k); (3) zero-inflated negative binomial model with variance = „µ; and (4) a zero-inflated Poisson model. A zero-inflated Poisson model was used for the error distribution for disease ordinal

117 1 (2015) and (2016). A zero-inflated negative binomial error model with variance = µ(1 + µ/k) was used for disease ordinal 2 (2016). For the SR, LR, and PR traits, the response was log-transformed and clonal values created with a standard mixed eects model using a normal error distribution. To investigate how hybridization alters trait means and heritable variation, we used partial least squares regression (PLS) and estimated broad-sense heritability. PLS was performed in R (R Core Team, 2018) with the package mixOmics (Rohart et al., 2017) using canonical correlation where the X matrix was a column vector of the proportion of ancestry attributable to each species (determined by NewHybrids), and the Y matrix was a column vector of traits. To understand how backcrossing into P. balsamifera alters trait values, we plotted the filial generation estimated by NewHybrids against the trait as boxplots and performed Tukey’s post-hoc tests to determine significance at –<0.05. Broad-sense heritability (H2) was estimated with hybrids present and absent with MCMCglmm (Hadfield, 2010) by estimating the variance-covariance matrix of the fixed and random eects and the intercept. Models were run for 50,000 to 500,000 iterations until eective sample sizes of posterior parameters were above 1000. Normal distributions were assumed for continuous traits. Broad-sense heritability was estimated using the following equation:

‡2 genet H2 = ‡2 genet + ‡2 error

where ‡2 represent error attributable to position of the ramet in the garden and measurement error. The posterior mode and highest posterior density interval of the heritability estimate were reported. To understand which traits correlated most strongly with the three disease phe-

118 notypes, we fit a MANOVA in R using the three disease phenotypes as a multivariate response and the remaining stomatal and ecophysiology traits as predictors. To test for selection trade-os between growth, disease resistance, and stomatal ratio, we fit linear models in a path analysis framework. Disease resistance was calculated as -1 * D2”. Traits were centered and standardized before regression models were fit. Model specifications are provided in Table 3.4. Each model was fit with three sets of data: hybrids absent, unadmixed P. balsamifera and Tacamahaca hybrids, and unadmixed P. balsamifera and Aigeiros hybrids. Evidence for trade-os between traits was considered present if the product of the slopes was negative (Schluter et al., 1991); similarly, selection conflicts were inferred if the product of the slopes from a trade-o were negative (Schluter et al., 1991). Significance testing of the observed regression coecients was performed by generating a permuted distribution of regression coecient products, and comparing the observed product to the distribution of permuted products. To generate permuted regression coecient product distributions, the underlying trait data was permuted 1000 times and regression models were fit. The regression coecient products were recorded, and a two-tailed probability test was used to generate a p-value for each observed regression coecient product. Addi- tionally, we performed a PCA of the three disease phenotypes with relative growth rate to decompose the relationship into orthogonal vectors. PC1 and PC2 was re- gressed against stomatal ratio to determine if the slopes from each set of hybrids were conflicting in sign, and hence, suggestive of a selection conflict on stomatal ratio in relation to growth and disease.

119 3.3 Results

3.3.1 Genetic ancestry and trait variation

Extensive backcrossing and hybridization was revealed from the NewHybrids analyses (Fig 3.1c, Table 3.2). Our sample collection protocol was designed to target non- hybrid genotypes, thus the distribution of hybrids in our sample is less than what is expected on the landscape. Nevertheless, we observe hybrids at the F1 to advanced generations backcrossing into P. balsamifera from P. angustifolia. Hybrids formed with P. trichocarpa were observed from the P1.F1 generation and beyond, while P. deltoides hybrids were only observed at the F1 generation. Disease severity was measured across two years with two dierent ordinal scales. Distributions of all three disease severity phenotypes revealed a zero-inﬂated distribution, along with highly variable patterns of disease between BxB and hybrid genotypes (Fig. 3.1d). Zero-inﬂated distributions were observed for the underlying disease traits, and traits which incorporated adaxial stomata (Fig. S3.1). Section

Tacamahaca hybrids exhibited decreased growth in comparisson to unadmixed P. balsamifera or the hybrids with a Aigeiros parent. The stomatal ratio was zero-inﬂated and the tail of had a median of 0.18 with a min. of 0.05 and a max. of 0.44. Hybrid genotypes, along with 43 unadmixed BxB individuals, had an elevated stomatal ratio. A PCA of the non-disease traits revealed stomata variation drives the principal axis of variation, while variation along PC2 was largely driven by of ecophysiology traits. Mapping hybrid cross type and disease presence or absence into the PC space revealed patterns of disease that correlated with negative values of PC1, an axis largely

120 describing adaxial stomatal patterning (Fig. 3.1e). Segregation of the trait space was loosely clustered by the identity of the non-balsamifera parent (Fig. 3.1e). As expected, hybridization altered trait means and variances (Table 3.1, Fig. S3.2, Fig. S3.3). Disease traits increased with increasing phylogenetic distance of the non- balsamifera parent, and transgressive segregation was observed for the adaxial and abaxial stomata traits as well as for 13C in BxA hybrids (Fig. S3.2). PLS analyses were conducted to investigate the eects of variation of the expected proportion of ancestry from each species (X matrix) on trait variation (Y matrix). The scalar product between pairs of vectors in the X and Y matrices indicate correlation between sets of variables. Using hierarchical clustering of the scalar products, we observed two large blocks: one characterized by high scalar product values between the ancestry proportions to SR, LR, PR, adaxial stomata, and the disease traits; and the second block contained abaxial stomata, and ecophysiology traits with generally lower scalar product values to the ancestry proportions (Fig. 3.2). In addition to shifting trait means, hybridization altered genotypic variance, resulting in increased heritability for the disease severity, SR, LR, PR, and adaxial stomatal patterning traits (Fig 3.2, Table S3.2). The H2 for ecophysiology and abaxial stomatal patterning traits was less inﬂuenced by hybridization, and generally lower.

As a whole, H 2 estimates ranged from 0.08 (CN) to 0.79 (SD_AD) (Fig 3.2). Backcrossing into P. balsamifera decreased disease severity in advanced generation hybrids (Fig. 3.3a). Tukey’s post-hoc tests indicated signiﬁcant dierences in disease severity between F1,F2, and P1.F2 backcrosses to advanced generation backcrosses and unadmixed P. balsamifera. The stomatal ratio remained high among the ﬁlial generations and only decreased in unadmixed P. balsamifera (Fig. 3.3b). When

121 accounting for the eect of SR on PC1 of disease, residuals were significantly dierent for the F1 generation and remained uniform for all other filial generations (Fig. 3.3c). Growth was not significantly dierent among filial generations and was only significantly elevated in unadmixed P. balsamifera (Fig. 3.3d). The eect of hybrid cross type on disease, stomatal patterning, and ecophysiology traits was explored with boxplots (Fig. S3.2). BxB genotypes generally had low disease, stomata patterning ratios, and adaxial stomatal patterning values. Patterns between ecophysiology traits were more nuanced. The eect of backcrossing on the traits was determined (Fig. S3.3), and in generally, stomatal patterning and disease traits demonstrated large variation between filial classes. Transgressive segregation

13 in F2 filial generation hybrids was observed for C, C:N, and for CCI in P1.F2 hybrids. A MANOVA was fit using the three disease traits as the response and the remaining traits as predictors to model which traits best predicted variation of disease severity. stomatal ratio and growth were the most significant predictors (p-value < 0.001), followed by the stomata pore length ratio, adaxial stomata density, porosity ratio, and abaxial porosity (Table 3.3). Plotting the regression coecients of each response revealed that multivariate disease traits did not always respond in the same direction to the predictors (Fig. 3.4). For example, the stomatal ratio had a positive regression coecient to each of the disease responses (— =1.1, 0.5, 0.4), while stomatal pore length ratio had one positive and two negative regression coecients to the disease responses (— =0.7, 1.3, 0.6). ≠ ≠

122 3.3.2 Selection trade-offs and conflicts

To determine if there was evidence for evolutionary trade-os in unadmixed and admixed populations, linear and multiple regression models were ﬁt for two components of ﬁtness, growth and disease resistance, and one metric trait, stomatal ratio. Trade- os were dierentially expressed if hybrids were present or absent from the analysis, and whether they involved Tacamahaca or Aigeiros hybrids (Table 3.4, Table S3.3). Disease resistance had a negative correlation with growth when hybrids were absent

ns (— = 0.18 ), and in the Aigeiros set (— = 0.17ú); however, in the Tacamahaca ≠ ≠ set, disease resistance had a positive relationship with growth (— =0.15ns). The positive slope in the Tacamahaca hybrids was due to the decreased disease resistance in the hybrid genotypes. For the multiple regression of SR on growth while including disease resistance, negative regression coecients were found for all three data sets, but the regression coecient was not significant in the Aigeiros data set. Significant negative regression coecients were found for all three data sets in the linear regression of stomatal ratio on disease resistance, indicating a decrease in the stomatal ratio increased disease resistance. The cumulative eect of both paths of stomatal ratio on growth was negative for the unadmixed P. balsamifera data set (R = 0.24) and SR,G ≠ in the Tacamahaca data set (R = 0.40), but closer to zero in the Aigeiros data SR,G ≠ set (R = 0.07). The smaller value of the path analysis in Aigeiros hybrids may SR,G ≠ have indicated the trade-owas less severe. Selection conflicts exist between pairs of traits if the product of their selection gradients are negative (Schluter et al., 1991). We observe a selection conflict between disease resistance and growth (—1) and SR and growth (—2) in the data set that

123 includes the Tacamahaca and unadmixed P. balsamifera (Table 3.4). This selection conflict may indicate that selection for disease resistance at the expense of decreased growth cannot be achieved through selection for lower values of SR. We can further view how hybridization reveals trade-os in admixed and undad- mixed populations by plotting stomatal ratio as a predictor to principal components reflecting the trade-obetween growth and disease severity (Fig. 3.5). PC1 largely represents disease severity variance, while PC2 largely describes growth variance. Un- admixed P. balsamifera and Tacamahaca hybrids have a positive relationship to PC1, while poplars admixed with P. deltoides have a negative slope. Regressing stomatal ratio on PC2 reveals a slight positive relationship in BxB and BxT genotypes and negative slopes in BxA and BxD hybrids, although these are not statistically significant.

3.4 Discussion

We conducted a common garden study with naturally formed Populus hybrids to observe how hybridization alters trait value distributions, heritable variation, and the potential to reveal trade-os and selection conﬂicts. We focused on growth, disease, and the stomatal ratio, as these traits contribute to ﬁtness and were previously implicated in disease variation within unadmixed P. trichocarpa (McKown, Guy, et al., 2014; McKown, Klápöt, et al., 2019). These results indicate hybridization increases disease severity, decreases growth of hybrids formed from within-Tacamahaca crosses, and increases the stomatal ratio. Our results demonstrate increases in stomatal ratio are consistently correlated across years and measurement scales with an increase in

124 disease severity. The negative correlation between stomatal ratio and disease resistance may be indicative of a locally adaptive trade-oevolved in the non-balsamifera parental species, which is still being expressed in the hybrids we sampled. The high

H 2 of the stomatal ratio and its strong covariation to the proportion of P. balsamifera ancestry suggest selection could eectively reduce disease within a population by selecting for lower values of stomatal ratio, and, in this system, increased growth is a likely result. Although unadmixed P. balsamifera did experience disease, hybridization introduced more variation than was observed in unadmixed populations; demonstrating that hybridization studies can use the increased variance present in admixed populations to reveal trade-os and selection conflicts that generations of natural selection has purged from unadmixed populations. Given sucient selection from pathogens and heritable variation, plant species will evolve an optimum along the growth-defense spectrum that maximizes their fitness based on the likelihood of pathogen exposure, physiological severity of the disease, and the cost of mounting a defense (Stearns, 1989; Schluter et al., 1991). That optimum can be locally adapted, change between populations within a species (e.g. northern versus southern populations of P. trichocarpa, McKown, Guy, et al., 2014), as well as evolve dierent adaptive combinations of traits between species. When hybrids are formed, misaligned growth-defense strategies expressed in a hybrid genotype can negatively impact its fitness through outbreeding depression (Goldberg et al., 2005), perhaps even in the presence of heterosis, as suggested by the increased disease of F1 BxD hybrids we study here. The eect of phylogenetic distance on these traits may be indirectly inferred from the distribution of traits arranged by hybrid cross type. Increasing the phylogentic

125 distance of the non-balsamifera parent tended to increase disease severity and decrease growth. Although, the increased phylogenetic distance between P. balsamifera and P. deltoides likely induced heterosis for growth, but is accompanied by increased disease (Fig. 3.1d) and increased stomatal ratio; thereby revealing a growth-defense trade-o. Interestingly, the BxD hybrids do not have increased 13C, which is an indicator of how closed stomata are, and analysis of the residuals from D SR PC1 ≥ indicate eects in addition to the stomatal ratio contribute to signiﬁcantly elevated disease (Fig. 3.3). These results allow us to speculate that the increased disease in BxD F1 hybrids may relate to a loss of resistance loci as a result of the accumulation of genetic incompatibilities, in addition to variance of stomatal patterning traits. While we do not have trait data for unadmixed P. trichocarpa, P. angustifolia, or P. deltoides, the growth-defense phenotypes in hybrids with ancestry from those species suggests dierent adaptive strategies from what has evolved in P. balsamifera. Common garden studies in P. trichocarpa have demonstrated that genotypes from Northern latitudes have more adaxial stomata, larger stomatal aperture pores, and increased carbon gain, but lower levels of defensive phenolic compounds and increased foliar pathogen disease (McKown, Guy, et al., 2014). This suggests one

ﬁtness strategy, typiﬁed by P. trichocarpa, is to maximize growth while tolerating the negative eects of disease. Further research is needed to compare the growth-defense strategies of multiple unadmixed Populus species in a common environment. In addition to increasing genetic variance, hybridization can disrupt co-adapted immune systems responses in host plants, possibly leading to increased disease susceptibility. Plant populations will typically evolve towards increased disease resistance with lower growth, or increased growth with lower disease resistance (Stearns, 1989;

126 Schluter et al., 1991). In line with those expectation, we observe a negative relationship between disease resistance and growth in unadmixed P. balsamifera and Aigerios

F1 hybrids (Table 3.4). However, in Tacamahaca hybrids, we observe decreased disease resistance correlated with decreased growth. This correlation may be the result of negative physiological eects of increased pathogen presence on the host trees leading to a reduction in growth. Comparison of the slopes of G D and G SR indicate a ≥ ≥ selection conflict is present in Tacamahaca hybrids that have lower growth, decreased disease resistance, and increased stomatal ratio. One possible interpretation of the selection conflict is the positive slope between disease resistance and growth observed in Tacamahaca hybrids indicates that these populations cannot evolve towards one of the two optimal trait spaces along the growth-defense spectrum by decreasing stomatal ratio values, or by tolerating increased disease and instead investing more in growth. The presence of amphistomy (having stomata on both leaf surfaces) as a result of hybridization raises interesting evolutionary consequences. Theoretical models of the benefits and costs of amphistomy indicate the increased eciency of photosynthesis under amphistomy should lead to more species organizing their stomata on both surfaces (Muir, 2015). Yet, hypostomy (stomata only on the abaxial surface) predominates, particularly in trees and shrubs. Models which incorporate costs of amphistomy indicate a narrow range of optima, with few intermediate values of stomatal ratio (Muir, 2015). The results from this study support theoretical conclusions that the costs of amphistomy are suciently high to constrain the available trait space in which plants evolve, and selection from pathogens likely induces a cost of amphistomy that is paid in the currency of fitness. Amphistomy is rare in P. balsamifera, which

127 likely reﬂects the evolution of ﬁtness optima that prioritizes defense over growth, which stands in contrast to northern populations of P. trichocarpa that have possibly evolved increase stomatal ratio to favor growth over defense (McKown, Guy, et al., 2014; McKown, Klápöt, et al., 2019). The high heritability of disease severity, stomatal ratio, and the other adaxial stomata traits indicates selection could eectively purge genetic variance linked to disease susceptibility. Indeed, disease severity and stomata ratio are strongly correlated with the proportion of P. balsamifera ancestry (Fig. 3.2), and to each other (Table 3.3, Fig. 3.4), and have high heritabilities. Thus, we can predict that if pathogens exert a strong selective pressure on these populations, and if disease resistance is linked to low stomatal ratios (which our results indicate), selection will favor low stomatal ratio values. Unadmixed P. balsamifera have low stomatal ratio and high disease resistance, and it is likely that selection for low stomatal ratio values has occurred in this species. The increased disease we observed in hybrid poplars may be a common feature of hybrid zones, and indeed, hybrid zones may even provide refuge for pathogens and pests (Whitham, 1989). For example, hybrids of Salix eriocephala and S. serica have a 3.5 fold increase in disease severity to Melamposora sp. and transgressively segregate for this trait (Roche and Fritz, 1998). Naturally formed F1 P. balsamifera x angustifolia hybrids transgressively segregate for the number of galls per tree and back-crosses into P. balsamifera are even more susceptible to gall forming pests and have higher variance than F1 or unadmixed parental species genotypes (Floate et al.,

2016). Similar trends have been observed in animal hybrid zones, where F1 and F2 largemouth bass (Micropterus salmoides) have lower survival after inoculation with

128 Ranavirus as a result of outbreeding depression (Goldberg et al., 2005). The occurrence of multiple dierent species admixed with P. balsamifera allowed us to indirectly observe the eect of phylogenetic distance between parental species on trade-os between growth and disease resistance, and to speculate on their importance in maintaining species barriers. Disease increased with phylogenetic distance of the non-balsamifera parent (Fig. 3.1, Fig. S3.1), and is ameliorated by back-crossing into P. balsamifera (Fig. 3.3). The evidence we present suggests shifts in stomatal ratio underlie much of the increase in disease; however, other studies point to the accumulation of genetic incompatibilities in hybrids as a cause of disease and disease-like symptoms which contribute to reproductive isolation and the maintenance of species barriers. Simulation studies of genetic incompatibilities have demonstrated that relatively few two-locus incompatibilities are required to maintain reproductive isolation in the presence of selection (Lindtke and Buerkle, 2015). Our finding of significant ancestry proportion eects on disease susceptibility suggests hybridization may also introduce genetic incompatibilities that disrupt the immune system. While any lethal combinations would have been filtered by selection in our sample, incompatibilities at loci in the innate or inducible immune system may persist into later life stages and filial generations, and could yield increased disease. Mapping these loci and tracking incompatibitilites in recombinant genotypes could test the hypothesis generated by the reduction of disease in recombinant genotypes that we observe (Fig. 3.3). Despite the potential loss of fitness in hybrid genotypes, hybrid populations are ubiquitous in nature, and hybridization represents an opportunity for the sharing of adaptive alleles evolved in one species to another (Chhatre et al., 2018; Suarez- Gonzalez, Lexer, et al., 2018). Adaptive introgression of disease resistance loci has

129 been documented in humans (Racimo et al., 2015) and Helianthus annuus (Whitney et al., 2006), among other taxa. Adaptive introgression between the accessions of P. balsamifera and P. angustifolia under investigation here has documented the introgression of genes principally associated with photoperiodic regulation, cold hardiness and terpene synthesis (Chhatre et al., 2018). Although Chhatre et al. (2018) did not

find introgressed loci directly linked to disease, studies between P. balsamifera and P. trichocarpa have identified R-genes and disease-associated loci introgressing in both directions (Suarez-Gonzalez, Hefer, Lexer, Cronk, et al., 2018). By studying traits contributing to fitness, selective regimes can be identified within and between species. Hybridization introduces genetic variance, which can reveal misalignment of modules of linked traits within a hybrid population. Selection can fine tune linked traits, by selecting for increased disease resistance or increased growth, but not both. Here, we have studied hybrid populations formed from four species to observe an expanded trait space. We are able to predict how selection is likely to constrain this trait space to achieve reduced disease; namely, through a decrease in the stomatal ratio. These results demonstrate the important evolutionary and ecological eects of hybridization in plant-pathogen interactions. The dierent trade- os we observed for growth, disease resistance, and stomatal traits, are likely due to the eect of misaligned growth-survival strategies that have evolved in each species, in addition to the eect of heterosis. Future research in this system will focus on using admixture mapping to identify genomic regions which underlie the disease resistance observed in P. balsamifera and the increased variance of disease severity in hybrids. Understanding the core growth-survival strategies that have evolved in each species will allow ecologists to place their findings of disease ecology in hybrid zones into a

130 wider context, and should be undertaken by future researchers.

131 3.5 Acknowledgements

This study was funded by a National Science Foundation Award (# 1461868). Matt Fitzpatrick, Andrew Elmore, Raju Soolanayakanahally, Ronald Zalesny, and Steve Guinn aided in collections of dormant cuttings. Dave Nelson provided isotope data. Charles Goodnight provided extensive feedback on the statistical models used here. Terrence Delaney provided access to his microscopy facilities. Jeanne Harris and Melissa Pespeni provided feedback on manuscript drafts. Sven Eberhardt helped create an early version of StomataCounter used to phenotype stomata density. Madie Hassett collected disease phenotypes in 2015. Many volunteers from the Plant Biology department planted cuttings in 2014 to establish the Spear Street common garden.

3.6 Data Accessibility

Cuticle micrographs are deposited on Dryad (doi:10.5061/dryad.kh2gv5f). Raw sequence reads are available on NCBI SRA (accession # SRP070954). NewHybrids ﬁlail call probabilities are provided in Supplementary Information S4a-c.

3.7 Author Contributions

KCF conceived the study, phenotyped accessions, performed the analyses, and wrote the manuscript. DN provided elemental and stable isotope data. SRK acquired plant material, phenotyped accessions, and provided extensive feedback on earlier drafts, phenotypes, and DNA sequences for study. Both authors read and approved the ﬁnal

132 manuscript.

133 3.8 Tables

Table 3.1: Trait deﬁnitions, abbreviations, and the mean (µ) standard deviation (‡)of ± traits with hybrids present and absent.

Trait Abbr. Min. Max Hybrids present µ ‡ Hybrids absent µ ‡ ± ± Disease severity 2015 scale 1 D1’ 0.30 5.95 1.43 1.44 1.01 0.90 ± ± Disease severity 2016 scale 1 D2’ 0.02 3.83 0.35 0.80 0.10 0.34 ± ± Disease severity 2016 scale 2 D2” 0.04 2.99 0.23 0.43 0.10 0.19 ± ± Relative growth rate (cm) G 15.96 53.94 27.63 6.46 27.78 6.55 ± ± Stomatal ratio SR 0.00 0.44 0.06 0.10 0.03 0.06 ± ± Pore length ratio LR 0.00 0.53 0.16 0.22 0.09 0.18 ± ± Porosity ratio PR 0.00 0.43 0.05 0.08 0.02 0.05 ± ± Abaxial stomata density (mm2) SD_AB 130.22 301.51 219.57 30.55 221.99 27.71 ± ± Adaxial stomata density (mm2) SD_AD 0.00 153.89 16.18 27.43 6.71 15.71 ± ± Abaxial pore length (µm) PL_AB 21.07 35.14 25.65 1.78 25.72 1.57 ± ± Adaxial pore length (µm) PL_AD 0.002 34.75 7.63 10.71 4.45 9.01 ± ± Abaxial porosity PO_AB 7.7e4 1.5e5 1.1e5 1.1e3 1.1e5 1.1e4 ± ± Adaxial porosity PO_AD 0.00 7.3e4 6.2e3 1.1e3 2.7e3 6.5e3 ± ± Total porosity TO_PO 8.2e4 1.7e5 1.2e5 1.3e3 1.2 e5 1.2e4 ± ± Carbon isotope discrimination (%) 13C 20.49 25.23 22.72 0.71 22.62 0.62 ± ± Nitrogen isotope fractionation (%) ”15N 2.25 4.27 3.34 0.35 3.32 0.35 ± ± Carbon:Nitrogen CN 36.13 45.03 39.50 1.30 39.53 1.28 ± ± Carbon (%) C 45.40 48.98 47.62 0.47 47.65 0.45 ± ± Nitrogen (%) N 1.2840 1.2844 1.2840 0.0005 1.2840 0.0005 ± ± Speciﬁc leaf area SLA 11.04 14.11 12.32 0.36 12.31 0.36 ± ± Chlorophyll content index CCI 19.61 38.41 27.40 2.98 27.66 3.09 ± ±

Table 3.2: NewHybrids results. The Ref. non-Pb (0) and Ref. Pb (1) columns refer to the size of the reference populations from which segregating loci were selected. The remaning columns describe the sizes of the ﬁlial generations in unknown samples. Abbreviations: Ref., reference population; Pb., Populus balsamifera.

Parental species Ref. non-Pb. (0) Ref. Pb. (1) Pb. F1 F2 P0-F1 P1-F1 P0-F2 P1-F2 P0-P0F1 P1-P1F1 P0-P0F2 P1-P1F2 P. balsamifera x trichocarpa 46 38 407 - - - 2 - 32 - 15 - 35 P. balsamifera x angustifolia 37 114 216 22 3 - 3 - - 1 3 - 1 P. balsamifera x deltoides 32 38 369 15 ------

134 Table 3.3: Output of MANOVA with three disease severity traits jointly modelled as the response. Signiﬁcance indicated as *** P < 0.001; ** P < 0.01; * P < 0.05; . < 0.1.

Pillai approx F Pr(>F) sig. code SR 0.62632 127.381 <2.2e-16 *** G 0.18223 16.935 5.768e-10 *** LR 0.06101 4.938 0.002427 ** SD_AB 0.03874 3.063 0.028933 * PR 0.03513 2.767 0.042616 * PO_AB 0.03364 2.646 0.049882 * CCI 0.02965 2.323 0.075884 . PL_AD 0.02780 2.173 0.091975 . C 0.01949 1.511 0.212453 SD_AD 0.01781 1.378 0.250319 N 0.01688 1.305 0.273694 PL_AB 0.01343 1.034 0.378196 13C 0.00718 0.550 0.648664 ”15N 0.00695 0.532 0.660778 CN 0.00601 0.460 0.710751 PO_AD 0.00505 0.386 0.763368 SLA 0.00020 0.015 0.997448

135 Table 3.4: Trade-os and selection conflicts inferred from linear and multiple regression models of three traits: D, disease resistance; G, relative growth rate; and SR, stomatal ratio. Negative regression coecients (—n)orpathanalysis(RSR,G) indicate a trade-os is present. The path analysis sums each independent path of the eect of SR on G: path 1, —2;path2, — — . Selection conflicts are inferred when the product of regression coecients are negative 3ú 1 (Schluter et al., 1991). The hybrids absent set includes only unadmixed P. balsamifera, and the Tacamahaca and Aigeiros data sets add hybrid individuals. —1 was taken from model 1. Significance testing for selection conflicts was performed with permutation tests (Fig. S3.4). Significance indicated as *** P < 0.001; ** P < 0.01; * P < 0.05.

Trade-os Data set —1 —2 —3 RSR,G model 1:G µ + — D + ‘ Hybrids absent -0.18 -0.30* -0.30*** -0.24 ≥ 1 model 2:G µ + — D + — SR + ‘ Tacamahaca 0.15 -0.32*** -0.49*** -0.40 ≥ 1 2 model 3:D µ + — SR + ‘ Aigeiros -0.17* -0.21 -0.82*** -0.07 ≥ 3

1 Selection conﬂicts D G Data set —1 —2 —2 —3 ú ú Hybrids absent 0.053ú 0.089úúú Tacamahaca -0.050úúú 0.160úúú 3 2 Aigeiros 0.034úú 0.168úúú SR

136 3.9 Figures

137 Figure 3.1: Summary of sample localities, phenotypes, and expected ancestry proportions for the trees under study. Sample locality map with the geographic ranges of four species of poplar: P. balsamifera in blue, P. trichocarpa in green, P. angustifolia in orange, and P. deltoides in yellow (a). See Table S1 for population coordinates and samples sizes. The phylogenetic relationships of the four species under study are presented as a cladogram (b). The expected proportion of ancestry of hybrid genotypes was calculated from NewHybrid (Anderson and Thompson, 2002) estimates of filial generation (c). Histograms and box plots of three disease traits summarized by PC1 of a principal components analysis (DPC1), relative growth rate (G), and stomatal ratio (SR). Tukey tests indicate significant dierences at –<0.05. PCA of stomatal patterning and ecophysiology trait data. The presence or absence of disease and the hybrid cross type is mapped onto the points, but was not used to decompose variance of the traits in principal component space (e). Disease presence/absence calls in e were made from the eigenvalues for DPC1, where PC1 eigenvalues greater than -0.8 were classified as diseased. Images of the uredospore (f) and disease sign of orange-yellow uredinia (g) of Melampsora medusae on an BxD hybrid.

138 1.00 1.00 Hybrids

● ● ● ● present 0.75 ● ● ● 0.75 ● ● ● absent

2 ● ● ● ● 0.50 ● ● 0.50 H ● ● ● ● ● ● ● ● ● ● ● ● ● 0.25 ● ● ● ● 0.25 ● ● ● ● ● ● ● ● ● ● ● 0.00 0.00 C N C N G 13 15 LR SR PR CN D1' D2' %N %C D2'' 13 CCI SLA δ 15 ∆ PL_AB SR PO_AD PR SD_AD LR PL_AD D1' D2'' D2' SLA CN G %N CCI PL_AB TO_PO ∆ δ %C PO_AB SD_AB PL_AD SD_AB

SD_AD PO_AB

TO_PO PO_AD

TT DD

ScalarColor product key Color key 0.91 −

−0.91 0 0.91

Figure 3.2: Results of partial least squaresPO_AD regression (PLS) of ancestry proportions and traits. B,D,T, and A represent vectors ofHR the expected ancestry proportions for each individual estimated by NewHybrids (Table 3.2SD_AD). The color ramp is the scalar product of variable PL_AD

loadings U’ and V’ from the X and Y matrices,PR respectively, indicating the correlation be- 2 tween pairs of vectors. Broad-sense heritabilityD2'' (H ) estimates for each trait are given from data sets with hybrids present (black) andD1' absent (red). 95% credible intervals indicated as 2 D2' 2 whiskers. See text for PLS and H estimationPL_AB details and Table S3.2 for H estimates.

CCI

RGR

TO_PO

∆13C SD_AB

δ15N SLA

PO_AB

%C T B A D 139 5 0.4 a b b a b 4 c a a ● 0.3 ab abc ab ● c 3 ●

1 c c ● C

P ● 2 0.2 ●

SR ● ● ● D ● ● ● ● c ● ● 1 ● ● ● 0.1 ● ● ● ● ● ● 0 ● 0.0

60 c DPC1 ~ SR d b ab ● 4 ● 50 ● ● b ● ● ab ● a a 2 ● 40 ● ab a ● G a a ● ● ab ab ● 0 30 ab ● Residuals ● ● ● 20 −2 ● F1 F2 F1 F2 P1.F2 P1.F1 P1.F2 Pure_1 Pure_1 P1.P1F1 P1.P1F2 P1.P1F1 P1.P1F2 Filial generation Filial generation

Figure 3.3: Backcrossing into P. balsamifera decreases disease (DPC1)(a)andthestomatal ratio (SR) (b). When accounting for the eect of SR on disease, residual variance decreases from F1 to Fn generations and unadmixed P. balsamifera (c). Growth is uniform across the ﬁlial generations (d). Dierences in Tukey’s post-hoc tests for signiﬁcance (p-value < 0.05) indicated by letter.

140 13 C

15 N

LR D1’ D2’ D2’’

Figure 3.4: MANOVA regression coecient heat map.

141 7.5 0.1

D2..D2. 5.0 0.0 D1. 2.5 PC1 −0.1

PC2 (26.93%) 0.0 G −0.2 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 PC1 (63.22%) SR

PC1_SR P PC2_SR P. 1 BxB 7.352 0 1.601 0.316

PC2 0 BxT 13.814 0 1.108 0.576

BxA 19.679 0.349 −4.474 0.711 −1 BxD −13.397 0.52 −13.983 0.242 −2 0.0 0.1 0.2 0.3 SR

Cross BxA BxB BxD BxT

Figure 3.5: Regression of stomatal ratio onto principal components 1 and 2 of disease severity and growth. Regression coecients and signiﬁcance indicated in table.

142 Supporting Information

143 Table S3.1: Summary of populations, samples sizes and geographic coordinates.

pop_code N lat lon pop_code N lat lon CBI 6 46.10 -61.18 SLC 14 44.79 -75.17 CHL 8 44.58 -76.25 SSR 16 44.46 -109.61 CLK 12 54.23 -110.07 TIM 13 48.36 -81.63 CPL 14 47.52 -83.26 TUR 12 53.20 -108.37 CYH 13 49.64 -109.81 UMI 3 56.34 -76.28 DCK 15 51.60 -101.74 USDA1 3 46.50 -92.98 DMT 1 39.51 -105.19 USDA10 5 46.58 -91.58 DPR 15 46.09 -77.51 USDA11 1 46.33 -89.48 FIS 5 48.20 -58.41 USDA12 9 46.58 -88.03 FNO 14 58.51 -122.40 USDA13 5 46.08 -88.03 GAM 10 49.47 -77.36 USDA14 1 45.58 -88.03 HBY 16 52.90 -102.39 USDA15 4 46.42 -86.87 HST 12 49.76 -84.01 USDA16 6 46.08 -85.85 HWK 10 50.41 -57.00 USDA17 6 46.00 -84.50 JKH 17 43.81 -110.48 USDA18 5 45.42 -84.50 KAP 11 49.38 -83.16 USDA19 3 45.00 -84.30 LON 9 43.18 -81.16 USDA2 7 46.75 -92.98 LPD 11 44.35 -73.73 USDA20 1 44.92 -85.85 LSM 2 39.22 -79.70 USDA21 5 45.42 -89.23 MBK 7 51.12 -55.78 USDA3 1 47.17 -91.67 MMT 15 49.87 -102.59 USDA4 4 47.92 -91.67 MSG 13 44.32 -106.87 USDA5 6 47.42 -92.98 NBY 11 46.33 -79.52 USDA6 2 47.92 -92.98 NEG 15 45.41 -62.14 USDA7 4 48.42 -92.98 OFR 15 53.14 -101.10 USDA8 3 48.33 -94.52 OUT 12 51.17 -106.25 USDA9 4 47.92 -94.52 RAD 13 53.43 -77.67 VER 8 44.38 -72.63 RMP 14 40.36 -105.62 WLK 11 60.05 -128.46 SKN 10 52.34 -106.58 WTR 13 48.59 -85.27

144 Table S3.2: Broad-sense heritability estimates and credible intervals for the hybrids present and absent datasets.

cross X Nind Nrep H2 lwr upr hybrids present G 479 985 0.25 0.17 0.33 hybrids present D1. 282 447 0.59 0.49 0.67 hybrids present D2. 330 545 0.5 0.41 0.59 hybrids present D2.. 330 545 0.46 0.36 0.54 hybrids present dis_pres 330 545 0.74 0.55 0.86 hybrids present D13C 425 805 0.43 0.36 0.51 hybrids present d15N 425 805 0.2 0.11 0.28 hybrids present perc_C 425 805 0.2 0.13 0.31 hybrids present perc_N 425 805 0.17 0.12 0.22 hybrids present sla 425 806 0.1 0.03 0.17 hybrids present cci 428 815 0.39 0.31 0.47 hybrids present CN 426 806 0.08 0.001 0.17 hybrids present ad_pres 417 785 0.92 0.87 0.97 hybrids present ad_ab 416 784 0.66 0.6 0.72 hybrids present ab_sd 417 785 0.36 0.26 0.43 hybrids present ad_sd 417 785 0.79 0.76 0.82 hybrids present ab_pl 417 785 0.47 0.39 0.55 hybrids present ad_pl 417 785 0.76 0.71 0.79 hybrids present ab_po 417 785 0.29 0.21 0.38 hybrids present ad_po 417 785 0.76 0.72 0.79 hybrids present to_po 417 785 0.52 0.47 0.58 hybrids present SR 417 785 0.72 0.65 0.76 hybrids present PR 417 785 0.68 0.62 0.73 hybrids present HR 417 785 0.71 0.65 0.75 hybrids absent G 384 778 0.26 0.19 0.36 hybrids absent D1. 227 356 0.22 0.06 0.36 hybrids absent D2. 276 426 0.72 0.66 0.78 hybrids absent D2.. 276 426 0.22 0.06 0.33 hybrids absent dis_pres 276 426 0.57 0.41 0.79 hybrids absent D13C 339 626 0.3 0.22 0.41 hybrids absent d15N 339 626 0.2 0.1 0.3 hybrids absent sla 339 626 0.1 0.01 0.18 hybrids absent cci 341 632 0.39 0.3 0.49 hybrids absent perc_C 339 626 0.15 0.05 0.26 hybrids absent perc_N 339 626 0.18 0.13 0.24 hybrids absent CN 339 626 0.08 0.01 0.18 hybrids absent ad_pres 332 587 0.36 0.27 0.47 hybrids absent ab_sd 332 587 0.2 0.08 0.32 hybrids absent ad_sd 332 587 0.52 0.43 0.6 hybrids absent ab_pl 332 587 0.41 0.29 0.5 hybrids absent ad_pl 332 587 0.4 0.29 0.49 hybrids absent ab_po 332 587 0.22 0.12 0.33 hybrids absent ad_po 332 587 0.5 0.42 0.6 hybrids absent to_po 332 587 0.24 0.14 0.34 hybrids absent HR 332 587 0.4 0.3 0.5 hybrids absent PR 332 587 0.36 0.26 0.46 hybrids absent SR 332 587 0.37 0.28 0.46

145 Table S3.3: Model output for trade-oregressions. Signiﬁcance indicated as *** P < 0.001; ** P < 0.01; * P < 0.05.

Hybrids absent Variable Statistic Model 1 Model 2 Model 3 (Intercept) Estimate 0.316*** 0.225** 0.168*** Std Err [0.081] [0.093] [0.036] D Estimate -0.179 -0.296* Std Err [0.150] [0.160] SR Estimate -0.295** -0.300*** Std Err [0.138] [0.053] N239232234 R2 0.006 0.025 0.121 adj R2 0.002 0.017 0.117 AIC 698.750 678.857 266.631

Tacamahaca set Variable Statistic Model 1 Model 2 Model 3 (Intercept) Estimate 0.117* 0.114* 0.068* Std Err [0.060] [0.060] [0.036] D Estimate 0.153* -0.090 Std Err [0.080] [0.098] SR Estimate -0.324*** -0.492*** Std Err [0.080] [0.038] N294286288 R2 0.013 0.067 0.367 adj R2 0.009 0.060 0.365 AIC 846.300 812.621 525.122

Aigeiros set Variable Statistic Model 1 Model 2 Model 3 (Intercept) Estimate 0.314*** 0.261*** -0.144*** Std Err [0.066] [0.076] [0.051] D Estimate -0.165** -0.256*** Std Err [0.074] [0.093] SR Estimate -0.205 -0.820*** Std Err [0.126] [0.068] N252245247 R2 0.020 0.030 0.372 adj R2 0.016 0.022 0.369 AIC 737.204 719.189 537.947

146 250 200 75 200 200 200 150 50 150 100 100 100 100 25 50 50

0 0 0 0 0 0.0 0.1 0.2 0.3 0.4 0 2 4 6 0 1 2 3 4 0 1 2 3 0.0 0.2 0.4 D1' D2' D2'' LR PR

60 30 40 200 200 40 30 20 20 100 100 10 20 10

0 0 0 1.2e5 1.4e5 150 200 250 300 100 150 0 0 50 8e4 1e5 20 24 28 32 10 20 30 0 0 2 2 SD_AB (mm ) SD_AD (mm ) PL_AB (µm) PL_AB (µm) PO_AB

60 50 40 50 40 40 200 30 40 30 30 20 100 20 20 20 10 10 10 0

0 1.3e5 1.6e5 0 0 0 2.5 3.0 3.5 4.0 37.5 40.0 42.5 45.0 21 22 23 24 25 3e4 6e4 1e5 0 13 15 PO_AD TO_PO ∆ C (‰) δ N (‰) CN

60 40

40 60 30 40 40 20 20 20 20 0 10 1.28425 1.28450 1.28475 1.28500

0 0 0 46 47 48 49 11 12 13 14 20 25 30 35

%C %N SLA CCI

Figure S3.1: Distribution of clonal values. See Table 3.1 for trait abbreviations.

147 6 4 3 0.5

3 0.4 0.2 4 2 0.3 2 LR PR D1' D2' D2'' 0.2 0.1 2 1 1 0.1

0 0 0.0 0.0 BxB BxT BxA BxD BxB BxT BxA BxD BxB BxT BxA BxD BxB BxT BxA BxD BxB BxT BxA BxD

300 150 30 ) ) 2 2 32 1.4e5 m) 250 100 m) µ µ 20 1.2e5 28 200

50 PO_AB 10 1e5 PL_AB ( 24 PL_AD ( SD_AB (mm SD_AD (mm 150 0 0 8e4 BxB BxT BxA BxD BxB BxT BxA BxD BxB BxT BxA BxD BxB BxT BxA BxD BxB BxT BxA BxD

25 45 4e4 1.6e5 4.0 24 43 3e4 3.5 1.3e5 23 41 C (‰) N (‰) 2e4 CN 13 15 TO_PO PO_AD δ ∆ 22 3.0 1e4 39 1e5 21 0 2.5 37 BxB BxT BxA BxD BxB BxT BxA BxD BxB BxT BxA BxD BxB BxT BxA BxD BxB BxT BxA BxD

49 1.2848 13.5

13.0 35 48 1.2846 12.5 30 %C %N CCI

47 SLA 1.2844 12.0 25 11.5 46 1.2842 11.0 20 BxB BxT BxA BxD BxB BxT BxABxD BxB BxT BxA BxD BxB BxT BxA BxD

Figure S3.2: Variation of traits by cross type. See Table 3.1 for trait abbreviations.

148 4 6 ● ● ● 3 0.5 ● ● ● ● ● ● ● ● ● ● ● ● ● ● 3 0.4 ● ● ● ● ● 0.2 ● ● 4 ● 2 ● ● ● 0.3 ● ● ● ● 2 ● LR

● PR D1' D2' ● ● D2'' 0.2 0.1 ● ● 1 ● ● 2 ● ● ● 1 ● ● ● ● ● ● 0.1 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0 0 0.0 ● 0.0 ●

150 300 ● ● 30 ●

● ● ● ● ) ) ● 1.4e5 ● 2

2 ● 32 ● ● ● m) 250 ● 100 m) ● ● 20 ● µ µ ● ● ● ● 1.2e5 ● 28 200 ● ●

● PO_AB 50 ● ● 10 ● ● 1e5 PL_AB ( ● PL_AD ( SD_AB (mm SD_AD (mm ● 24 ● ● ● ● ● 150 ● 0 0 ● 8e4 ●

● ● ● ● ● 45 25 ● ● 4e4 ● ● ● ● 4.0 ● 1.6e5 ● 43 ● ● ● 24 3e4 ● ● ● ● ● ● 41 ● 3.5 CN ● ● 1.3e5 23 C (‰) 2e4 N (‰) 39 ● 13 ● 15 TO_PO PO_AD ● ● δ ● ∆ 22 3.0 ● 1e4 ● ● 37 ● ● ● 1e5 ● ● 21 ● ●

● F1 F2 ● 2.5 ● 0 P1.F2 Pure_1 P1.P1F1 P1.P1F2

● 1.2848 ● ● 49 13.5 ● ● ● ● ● 13.0 35 48 1.2846 ● 12.5 ● 30 ● %C %N 47 1.2844 CCI SLA 12.0 ● ● 25 ● 46 1.2842 11.5 ● ● ● 11.0 ● 20 ● F1 F2 F1 F2 F1 F2 F1 F2 P1.F2 P1.F2 P1.F2 P1.F2 Pure_1 Pure_1 Pure_1 Pure_1 P1.P1F1 P1.P1F2 P1.P1F1 P1.P1F2 P1.P1F1 P1.P1F2 P1.P1F1 P1.P1F2

Figure S3.3: Variation of traits by ﬁlial generation. See Table 3.1 for trait abbreviations.

149 400 400

300 300

200 200 Freq 100 100

Hybrids absent 0 0 −0.1 0.0 −0.050−0.025 0.000 0.025 0.050 0.075 β1 * β2 β2 * β3

400 a 600 c

a 300 h

a 400

m 200 Freq a

c 200 a 100 T 0 0 −0.04 −0.02 0.00 0.02 0.00 0.04 0.08 0.12 β1 * β2 β2 * β3

300 300

s 200 200 o r i Freq e

g 100 100 i A 0 0 −0.02 0.00 0.02 0.04 −0.04 −0.02 0.00 0.02 β1 * β2 β2 * β3

Figure S3.4: Permutation tests for significance of selection conflict products. The product of the regression coecients for each pair of traits was estimated 1000 times with permuted data (grey distribution). The observed product is considered significant if its value is more extreme than the permuted distribution at –/2=0.025. See Table 3.4 for results.

150 References

Allen, E, B Hazen, H Hoch, Y Kwon, G Leinhos, R Staples, M Stumpf, B Terhune, et al. (1991). “Appressorium formation in response to topographical signals by 27

rust species.” In: Phytopathology 81.3, pp. 323–331. Altizer, S, R Bartel, and BA Han (2011). “Animal migration and infectious disease

risk”. In: science 331.6015, pp. 296–302. Anderson, E and E Thompson (2002). “A model-based method for identifying species

hybrids using multilocus genetic data”. In: Genetics 160.3, pp. 1217–1229. Bailey, JK, R Deckert, JA Schweitzer, BJ Rehill, RL Lindroth, C Gehring, and TG Whitham (2005). “Host plant genetics aect hidden ecological players: links among

Populus, condensed tannins, and fungal endophyte infection”. In: Canadian Jour- nal of Botany 83.4, pp. 356–361. Bates, D, M Mächler, B Bolker, and S Walker (2015). “Fitting Linear Mixed-Eects

Models Using lme4”. In: Journal of Statistical Software 67.1, pp. 1–48. Browning, BL, Y Zhou, and SR Browning (2018). “A one-penny imputed genome from

next-generation reference panels”. In: The American Journal of Human Genetics 103.3, pp. 338–348. Bruns, EL, J Antonovics, and M Hood (2018). “Is there a disease-free halo at species range limits? The co-distribution of anther-smut disease and its host species”. In:

Journal of Ecology. Cai, Q, L Qiao, M Wang, B He, FM Lin, J Palmquist, SD Huang, and H Jin (2018). “Plants send small RNAs in extracellular vesicles to fungal pathogen to silence

virulence genes”. In: Science 360.6393, pp. 1126–1129.

151 Chandran, D, J Rickert, Y Huang, MA Steinwand, SK Marr, and MC Wildermuth (2014). “Atypical E2F transcriptional repressor DEL1 acts at the intersection of

plant growth and immunity by controlling the hormone salicylic acid”. In: Cell host & microbe 15.4, pp. 506–513. Chhatre, VE, LM Evans, SP DiFazio, and SR Keller (2018). “Adaptive introgression and maintenance of a trispecies hybrid complex in range-edge populations of

in hybridizing Populus species despite F1 fertility and recurrent gene ﬂow”. In: Molecular ecology 25.11, pp. 2482–2498. Danecek, P, A Auton, G Abecasis, CA Albers, E Banks, MA DePristo, RE Handsaker, G Lunter, GT Marth, ST Sherry, et al. (2011). “The variant call format and

VCFtools”. In: Bioinformatics 27.15, pp. 2156–2158. Dangl, JL and JD Jones (2001). “Plant pathogens and integrated defence responses

to infection”. In: nature 411.6839, p. 826. DiFazio, SP, GT Slavov, and CP Joshi (2011). “Populus: a premier pioneer system for plant genomics”. In: Enﬁeld, NH: Science Publishers, pp. 1–28. Dowkiw, A and C Bastien (2004). “Characterization of two major genetic factors con-

trolling quantitative resistance to Melampsora larici-populina leaf rust in hybrid poplars: strain speciﬁcity, ﬁeld expression, combined eects, and relationship with

a defeated qualitative resistance gene”. In: Phytopathology 94.12, pp. 1358–1367. Eckenwalder, JE (1996). “Systematics and evolution of Populus”. In: Biology of Pop- ulus and its implications for management and conservation. Ed. by R Stettler, T

152 Bradshaw, P Heilman, and T Hinckley. Ottawa: NRC Research Press. Chap. 1, pp. 7–32. Ellstrand, NC and KA Schierenbeck (2000). “Hybridization as a stimulus for the

evolution of invasiveness in plants?” In: Proceedings of the National Academy of Sciences 97.13, pp. 7043–7050. Elshire, RJ, JC Glaubitz, Q Sun, JA Poland, K Kawamoto, ES Buckler, and SE Mitchell (2011). “A robust, simple genotyping-by-sequencing (GBS) approach for

high diversity species”. In: PloS one 6.5, e19379. Evans, LM, GT Slavov, E Rodgers-Melnick, J Martin, P Ranjan, W Muchero, AM Brunner, W Schackwitz, L Gunter, JG Chen, et al. (2014). “Population genomics

of Populus trichocarpa identiﬁes signatures of selection and adaptive trait associations”. In: Nature genetics 46.10, p. 1089. Farquhar, GD, MH O’Leary, and JA Berry (1982). “On the relationship between carbon isotope discrimination and the intercellular carbon dioxide concentration

in leaves”. In: Functional Plant Biology 9.2, pp. 121–137. Feau, N, DL Joly, and RC Hamelin (2007). “Poplar leaf rusts: model pathogens for a

model tree”. In: Botany 85.12, pp. 1127–1135. Fetter, KC, S Eberhardt, RS Barclay, S Wing, and SR Keller (2019). “Stomata- Counter: a neural network for automatic stomata identiﬁcation and counting”. In:

New Phytologist 223.3, pp. 1671–1681. Floate, KD, J Godbout, MK Lau, N Isabel, and TG Whitham (2016). “Plant–

herbivore interactions in a trispeciﬁc hybrid swarm of Populus: assessing support for hypotheses of hybrid bridges, evolutionary novelty and genetic similarity”. In:

New Phytologist 209.2, pp. 832–844.

153 Fournier, DA, HJ Skaug, J Ancheta, J Ianelli, A Magnusson, MN Maunder, A Nielsen, and J Sibert (2012). “AD Model Builder: using automatic dierentiation for sta-

tistical inference of highly parameterized complex nonlinear models”. In: Optim. Methods Softw. 27, pp. 233–249. Glaubitz, JC, TM Casstevens, F Lu, J Harriman, RJ Elshire, Q Sun, and ES Buck- ler (2014). “TASSEL-GBS: A high capacity genotyping by sequencing analysis

pipeline”. In: PLoS ONE 9.2. arXiv: NIHMS150003. Goldberg, TL, EC Grant, KR Inendino, TW Kassler, JE Claussen, and DP Philipp (2005). “Increased infectious disease susceptibility resulting from outbreeding de-

pression”. In: Conservation Biology 19.2, pp. 455–462. Gonzales-Vigil, E, CA Hefer, ME von Loessl, J La Mantia, and SD Mansﬁeld (2017). “Exploiting natural variation to uncover an alkene biosynthetic enzyme in poplar”.

In: The Plant Cell,tpc–00338. Goulet, BE, F Roda, and R Hopkins (2017). “Hybridization in plants: old ideas, new

techniques”. In: Plant physiology 173.1, pp. 65–78. Hadﬁeld, JD (2010). “MCMC Methods for Multi-Response Generalized Linear Mixed

Models: The MCMCglmm R Package”. In: Journal of Statistical Software 33.2, pp. 1–22. Kliebenstein, DJ (2016). “False idolatry of the mythical growth versus immunity

tradeoin molecular systems plant pathology”. In: Physiological and Molecular Plant Pathology 95, pp. 55–59. La Mantia, J, J Klápöt, YA El-Kassaby, S Azam, RD Guy, CJ Douglas, SD Mans- ﬁeld, and R Hamelin (2013). “Association analysis identiﬁes Melampsora columbiana ◊ poplar leaf rust resistance SNPs”. In: PLoS One 8.11, e78423.

154 LeBoldus, JM, N Isabel, KD Floate, P Blenis, and BR Thomas (2013). “Testing the

‘hybrid susceptibility’and ‘phenological sink’ hypotheses using the P. balsamifera– P. deltoides hybrid zone and Septoria leaf spot [Septoria musiva]”. In: PloS one 8.12, e84437. Li, H and R Durbin (2009). “Fast and accurate short read alignment with Burrows-

Wheeler transform”. In: Bioinformatics 25.14, pp. 1754–1760. arXiv: 1303.3997. Lindtke, D and CA Buerkle (2015). “The genetic architecture of hybrid incompatibilities and their eect on barriers to introgression in secondary contact”. In:

Evolution 69.8, pp. 1987–2004. Mallet, J (2007). “Hybrid speciation”. In: Nature 446.7133, p. 279. Martinsen, GD, TG Whitham, RJ Turek, and P Keim (2001). “Hybrid populations

selectively ﬁlter gene introgression between species”. In: Evolution 55.7, pp. 1325– 1335. McKown, AD, RD Guy, L Quamme, J Klápöt, J La Mantia, C Constabel, YA El-Kassaby, RC Hamelin, M Zifkin, and M Azam (2014). “Association genet-

ics, geography and ecophysiology link stomatal patterning in Populus trichocarpa with carbon gain and disease resistance trade-os”. In: Molecular Ecology 23.23, pp. 5771–5790. McKown, AD, J Klápöt, RD Guy, OR Corea, S Fritsche, J Ehlting, YA El-Kassaby, and SD Mansﬁeld (2019). “A role for SPEECHLESS in the integration of leaf

stomatal patterning with the growth vs disease trade-oin poplar.” In: The New phytologist.

155 Melotto, M, W Underwood, J Koczan, K Nomura, and SY He (2006). “Plant stomata

function in innate immunity against bacterial invasion”. In: Cell 126.5, pp. 969– 980. Messina, FJ, SL Durham, JH Richards, and DE McArthur (2002). “Trade-obetween

plant growth and defense? A comparison of sagebrush populations”. In: Oecologia 131.1, pp. 43–51. Muir, CD (2015). “Making pore choices: repeated regime shifts in stomatal ratio”. In:

Proceedings of the Royal Society B: Biological Sciences 282.1813, p. 20151498. Obeso, JR (2002). “The costs of reproduction in plants”. In: New phytologist 155.3, pp. 321–348.

Pinon, J and P Frey (2005). “Interactions between poplar clones and Melampsora populations and their implications for breeding for durable resistance”. In: Rust Diseases of Willow and Poplar. Ed. by MH Pei and AR McCracken. Wallingford, UK: CABI, pp. 139–154.

RCoreTeam(2018).R: A Language and Environment for Statistical Computing.R Foundation for Statistical Computing. Vienna, Austria. Racimo, F, S Sankararaman, R Nielsen, and E Huerta-Sánchez (2015). “Evidence

for archaic adaptive introgression in humans”. In: Nature Reviews Genetics 16.6, p. 359.

Raven, PH and Zy Wu (1999). Flora of China. 4, Cycadaceae Through Fagaceae. Science Press. Roche, B and R Fritz (1998). “Eects of host plant hybridization on resistance to wil-

low leaf rust caused by Melampsora sp.” In: European Journal of Forest Pathology 28.4, pp. 259–270.

156 Rohart, F, B Gautier, A Singh, and KAL Cao (2017). “mixOmics: An R package for

’omics feature selection and multiple data integration”. In: PLoS computational biology 13.11, e1005752. Schluter, D, TD Price, and L Rowe (1991). “Conﬂicting selection pressures and life

history trade-os”. In: Proc. R. Soc. Lond. B 246.1315, pp. 11–17. Schneider, CA, WS Rasband, and KW Eliceiri (2012). “NIH Image to ImageJ: 25

years of image analysis”. In: Nature methods 9.7, p. 671. Skaug, H, D Fournier, B Bolker, A Magnusson, and A Nielsen (2016). Generalized Linear Mixed Models using ’AD Model Builder’. R package version 0.8.3.3. Stearns, SC (1989). “Trade-os in life-history evolution”. In: Functional ecology 3.3, pp. 259–268. Suarez-Gonzalez, A, CA Hefer, C Lexer, QC Cronk, and CJ Douglas (2018). “Scale

and direction of adaptive introgression between black cottonwood (Populus trichocarpa) and balsam poplar (P. balsamifera)”. In: Molecular ecology 27.7, pp. 1667– 1680. Suarez-Gonzalez, A, CA Hefer, C Lexer, CJ Douglas, and QC Cronk (2018). “Intro-

gression from Populus balsamifera underlies adaptively signiﬁcant variation and range boundaries in P. trichocarpa”. In: New Phytologist 217.1, pp. 416–427. Suarez-Gonzalez, A, C Lexer, and QC Cronk (2018). “Adaptive introgression: a plant

perspective”. In: Biology letters 14.3, p. 20170688. Tian, D, M Traw, J Chen, M Kreitman, and J Bergelson (2003). “Fitness costs of

R-gene-mediated resistance in Arabidopsis thaliana”. In: Nature 423.6935, p. 74.

157 Tuskan, G, S Difazio, S Jansson, J Bohlmann, I Grigoriev, U Hellsten, N Putnam, S Ralph, S Rombauts, A Salamov, et al. (2006). “The genome of black cottonwood,

Populus trichocarpa (Torr. & Gray)”. In: Science 313.5793, p. 1596. \ Ullah, C, CJ Tsai, SB Unsicker, L Xue, M Reichelt, J Gershenzon, and A Ham- merbacher (2018). “Salicylic acid activates poplar defense against the biotrophic

rust fungus Melampsora larici-populina via increased biosynthesis of catechin and proanthocyanidins”. In: New Phytologist. Van Kraayenoord, C, G Laundon, and A Spiers (1974). “Poplar rusts invade New

Zealand.” In: Plant disease reporter 58.5, pp. 423–427. Wang, ZY (2012). “Brassinosteroids modulate plant immunity at multiple levels”. In:

Proceedings of the National Academy of Sciences 109.1, pp. 7–8. White, J, B Vaughn, and S Michel (2011). “Stable Isotopic Composition of Atmo- spheric Carbon Dioxide (13C and 18O) from the NOAA ESRL Carbon Cycle Cooperative Global Air Sampling Network, 1990–2014, Version: 2015-10-26.” In:

University of Colorado, Institute of Arctic and Alpine Research (INSTAAR). Whitham, TG (1989). “Plant hybrid zones as sinks for pests”. In: Science, pp. 1490– 1493. Whitney, KD, RA Randell, and LH Rieseberg (2006). “Adaptive introgression of

herbivore resistance traits in the weedy sunﬂower Helianthus annuus”. In: The American Naturalist 167.6, pp. 794–807.

158 Chapter 4

Guard cell regulation, the plant immune system, and growth transcription factors are under selection and associated with fungal disease variation in advanced generation P. balsamifera x trichocarpa hybrids.

Karl C. Fettera,1, Stephen R. Kellera,1 aDepartment of Plant Biology, University of Vermont

159 1Corresponding authors: Stephen R. Keller (e-mail: [email protected], tel: +1 802 656 2930) 1Karl C. Fetter (e-mail: [email protected])

160 Abstract

Balsam poplars (Populus balsamifera) generally have stomata on only the lower leaf surface and are largely resistant to the local strain of the leaf rust Melampsora medusae in New England forests. The black balsam poplar (P. trichocarpa) has stomata on both leaf surfaces (amphistomy) and hybrids formed from advanced generation backcrosses with P. trichocarpa x balsamifera hybrids are amphistomatous and susceptible to Melampsora. We used admixture mapping, a genome-wide association method ideal for hybrid populations, to detect loci under positive selection that pleiotropically contribute to trait variation via genetic associations with multiple stomata patterning and ecophysiology traits. We detect genes with functions for guard cell homeostasis, constituent genes of the plant immune system, growth related transcription factors, and components of constitutive defenses that likely dier between P. balsamifera and P. trichocarpa. Our results indicate selection is acting on these genes, which regulate suites of traits that contibute to disease variation in natural populations. 4.1 Introduction

The consequences of hybridization across landscapes has received considerable attention in plants, particularly in sunflowers (Heiser Jr, 1951; Whitton et al., 1997;Yatabe et al., 2007; Rieseberg, SC Kim, et al., 2007; Hübner et al., 2019), monkey-flowers (Macnair and Cumbes, 1989; Fishman and Willis, 2006; Stankowski and Streisfeld, 2015; Chase et al., 2017), and poplars (RL Smith and Sytsma, 1990; Floate et al., 2016; Chhatre et al., 2018; Bresadola et al., 2018). Hybridization studies in poplars have been particularly influential in developing our understanding of how admixture between foundation tree species can alter biotic interactions between herbivores (Whitham, Floate, et al., 1996; Evans, Allan, Shuster, et al., 2008), fungi (Bailey, Deckert, et al., 2005), understory plant species (Lamit et al., 2011), birds (Adams and TM Burg, 2015), and indeed, entire ecosystems (Schweitzer et al., 2008; Bailey, Schweitzer, et al., 2009). Research from poplar hybrid zones indicates their genomes are porous to movement of adaptive species from one species to another (Stölting et al., 2013). Intro- gression can occur when a gene evolved in one species is introduced to another via reproduction, and subsequently experiences selection to maintain the original allele of the gene in its new genomic context (Barton and Hewitt, 1985). Along with mutations and standing genetic variation, introgression is an additional source of adaptive genetic variation in a population (De Carvalho et al., 2010; Suarez-Gonzalez, Lexer, et al., 2018). Genomes may be resistant to introgression if there are numerous reproductive incompatibility loci preventing interspecific mating from occurring, and thus, inhibiting recombination from disrupting linkage between adaptive loci and condi-

162 tionally neutral or maladaptive loci (Rieseberg and Carney, 1998). In an ecological context, adaptive introgression has been demonstrated to explain clinal variation of phenology in European oaks, with genes mapping to phenology traits exhibiting patterns of clinal variation between Quercus petraea and Q. robur (Leroy et al., 2019). In a novel common garden study after repeated controlled matings between Helianthus annuus ssp. annuus and H. debilis, multiple traits representing ecophysiology, defense against predators, phenology, and plant architecture showed adaptive changes leading to an increase in ﬁtness above an unadmixed control population in as few as seven generations of backcrossing (Mitchell et al., 2019). These studies demonstrate the potential of populations to use standing genetic variation from one species for adaptive evolution in another species in similar or novel environments. A major goal of ecological genomics is to link phenotypic variation to genetic variation to ﬁnd genes underlying biological or ecological processes, such as, the evolution of sexes (Geraldes, Hefer, et al., 2015), climatically adaptive genetic variation (Martins et al., 2018), or batesian mimicry (Iijima et al., 2018). Advances in genome sequencing have enabled the discovery of tens of thousands (to millions) of genetic polymorphisms within a species. Statistical methods which allow for the association of phenotypic to genotypic variation routinely come in the form of mixed-models that contain covariates for population ancestry that serve to condition allele frequency variation given historical patterns of allele frequency change (Ingvarsson and Street, 2011; Sul et al., 2018). While association genetic methods are robust to demography, populations formed from hybrids or from genetically divergent parental populations within a species, will contain genomes with relatively few recombination events yield- ing large regions of high linkage between loci on a chromosome. Admixture mapping

163 was invented to circumvent this problem by exchanging the identify of a base pair at a locus with the probability that the locus is a derived from species/population A or B. Thus, the association of phenotypic variance to a local ancestry genotype is estimated, rather than to a base pair (MW Smith and O’Brien, 2005; Buerkle and Lexer, 2008; Shriner et al., 2011).

Populus is a genus of long-lived, wind-pollinated trees which exhibit remarkable ecological amplitude. Species in the genus occur beyond the tundra-boreal forest ecotone in northern Alaska and Canada (e.g. P. balsamifera, Breen, 2014), to humid subtropical forests in the southeastern American coastal plain (e.g. P. heterophylla, Fowells et al., 1965), and the Arid deserts of the the Middle East (e.g. P. euphratica, Rajput et al., 2016). Populus species will readily form fertile hybrids, and bi- directional crossing is frequent (DiFazio, GT Slavov, et al., 2011; Suarez-Gonzalez,

Lexer, et al., 2018), although not all species can do this, (e.g. P. balsamifera x deltoides hybrids will only cross into P. balsamifera, SL Thompson et al., 2010a). Large eective population sizes (Savolainen et al., 2007; Evans, GT Slavov, et al., 2014; J Wang et al., 2016) and strong local adaptation increase the probability that adaptive genetic variation will evolve and can be shared among populations. Given the promiscuity of poplars and their ability to persist as clones on a landscape for thousands of years (Ally et al., 2008), introgression is expected to occur; creating the biological conditions to employ admixture mapping to identify phenotype-genotype associations in hybrid populations (Lexer et al., 2007; Lindtke, S Gonzalez-Martinez, et al., 2013; Bresadola et al., 2018). In a previous study, we found evidence of hybrids expressing a strong genetic correlation between stomatal traits and resistance to the basidiomycete leaf rust, Melampsora medusae (Fetter, 2019, Ch.3). M. medusaue is

164 a foliar pathogen with two obligate hosts: a poplar sp. as the telial host, and a larch (Larix) sp. as the aecial host (Feau et al., 2007). In this study, we use admixture mapping to identify if associations between stomata and disease co-localize to the same genomic regions, indicating a shared genomic architecture and possibly pleiotropy. Furthermore, we test if candidate associations to disease and stomatal traits show evidence of positive selection in one or both parental species, as expected if introgression in hybrid zones is responsible for the movement of adaptive alleles from one species’ genome into another.

4.2 Materials & Methods

4.2.1 Plant collections, traits, and sequencing

Plant material was collected from across the western range of P. balsamifera (Fig. 4.1a) and planted in a common garden in Vermont, USA in June, 2014. After an overwintering, ecophysiology, and bud ﬂush traits were measured in the common garden (Table 4.2). Disease severity to M. medusae was measured in 2015 using the ordinal scale of La Mantia et al. (2013), and in 2016, both the ordinal scales of La Mantia et al. (2013) and Dowkiw and Bastien (2004) were used. Genotyping- by-sequencing (GBS) (Elshire et al., 2011b) was used to create sequence reads, and a modiﬁed Tassel GBS Pipeline (Glaubitz et al., 2014b) was used to process raw sequence reads and call variants and genotypes. For more details regarding collection materials, phenotypes, and genotyping, refer to Fetter (2019, Ch.3).

165 4.2.2 Admixture mapping

Admixture mapping requires phased reference haplotypes from unadmixed parental populations to estimate locus speciﬁc ancestry in a test population. To choose reference individuals for P. balsamifera, we estimated global ancestry of the 534 individuals we sequenced with ADMIXTURE (Alexander et al., 2009b), and chose 25 individuals from the western populations of Duck Mountain, Manitoba (DCK) and Hudson Bay, Ontario (HBY) that exhibited minimal signs of admixture with other

Populus species (i.e. ADMIXTURE q-matrix > 0.95 P. balsamifera at K = 2) (Fig. S4.1). For the P. trichocarpa reference set, 25 individuals with whole genome sequences publicly available were chosen from western Washington which were known to lack admixture with P. balsamifera (Evans, GT Slavov, et al., 2014), and downloaded from https://phytozome.jgi.doe.gov/. 117 hybrid and non-hybrid genotypes from the Canadian prairie provinces and the western and mid-western United States were chosen to form the test set (Table 4.1, Fig. 4.1a).

The locus positions of the test vcf files were used to filter the P. trichocarpa reference set, and after discarding flipped and multi-allelic sites, 74,878 homologous sites remained. Haplotypes of the reference populations were jointly estimated with fast- PHASE (Scheet and Stephens, 2006), setting the following parameters: 20 random starts; 45 EM iterations per run; 5000 samples of the posterior haplotype distribution; the K-selection function was limited between 5 and 30, at 5 unit intervals; and loci with genotype probability <90 were flagged. Locus specific ancestry was determined for the test set via recombination ancestry switchpoint probability estimation with the program RASPberry (Wegmann et al., 2011). The input data sets for RASPberry

166 were: a test set of 485 individuals with 227,607 SNP loci and no site-wise missingness; two reference populations of 25 individuals each with 74,878 SNP loci; and the ADMIXTURE q-matrix at K = 2. The default recombination rate of 5 cM was used and population recombination rates set to 120 and 173 for P. balsamifera and P. trichocarpa. The mutation rate was set to 0.0079365, and miscopy rate to 0.01. Mixed eects models were ﬁt to identify association between 24 phenotypes (Table 4.2) and local ancestry genotypes corrected by the global ancestry of the individual using the bayesian mixed model (BMIX) model of Shriner et al. (2011):

f(yi)=—0 + —1Aij + —2A¯i + ‘i

Where, yi is a vector of phenotypes, —N are regression coecients to estimate,

th Aij is a vector of local ancestries for the j locus, Ai is a vector of global ancestries

th th for the i individual, and ‘i is error variance for the i individual. Global ancestry was again estimated for each individual with the RASPberry data by summing the frequency of the local ancestries of the homozygous and half the heterozyous genotypes corresponding to the P. balsamifera allele. Association probabilities were estimated from chi-sq test statistics of converted model p-values. The significance level for true associations was set to –/admixture burden (0.05/237.5), which represents the number of independent tests in the sample. Admixture burden was estimated from the first-order autoregressive (AR(1)) models for each locus summed across the genome for all individuals. Manhattan plots were used to visualize p-values of tests across the genome for each trait. Intersections between candidate gene lists were identified with UpSetR (Conway et al., 2017).

167 4.2.3 Selection Scans

Patterns of positive selection were identiﬁed with RAiSD (Alachiotis and Pavlidis, 2018) using the µ statistic, a composite statistic of the product of µSFS which measured shifts in the site frequency spectrum, µLD measuring linkage disequilibrium, and µVAR which measured genetic polymorphism. These statistics were calculated in overlapping 50 SNP sliding windows. RAiSD is ideal for identifying hard sweeps, but false positives can be generated by population bottlenecks, background selection, and population structure (Alachiotis and Pavlidis, 2018). As a result, we sub-sampled the

P. balsamifera data set to a single deme in the western core which was identiﬁed by running ADMIXTURE (Alexander et al., 2009b) on all 534 P. balsamifera samples from K = 2 until resolution of known demes (sensu Keller, Olson, et al., 2010) within P. balsamifera was possible at K = 7 (Fig. S4.1). Individuals > 0.98 ancestry in the western core deme at K = 7 were selected. The hybrid set was selected from individuals with RASPberry global ancestry < 0.99. RAiSD was run separately for the P. trichocarpa reference inds (N = 46), Western core P. balsamifera (N = 81), and hybrids (N = 39) (Table S4.2). The input sequence data set contained 31,545

SNPs common to both P. trichocarpa and P. balsamiera with no missingness. We considered windows in the top 1% of µ statistics as evidence of a hard selective sweep.

168 4.2.4 Gene annotation and candidate gene fil-

tering

Gene annotations for the 227,607 SNPs were pulled from the P. trichocarpa v3.0 genome and used to annotate significant loci. After the top 1% of µ statistic windows were identified, the overlap of BMIX candidate genes and selection windows was determined using GenomicRanges (Lawrence et al., 2013). Local ancestry sites from genes that passed the BMIX/selection filter were identified, and monomorphic sites removed. With the remaining polymorphic local ancestry sites, boxplots comparing local ancestry genotype and SR, D1’, and G were made to evaluate distributions for signs of false positives. Sites that passed this final filter were mapped to P. trichocarpa gene annotations and the genes were manually investigated for gene function using www.popgenie.org (Sjödin et al., 2009a; Sundell et al., 2015), atgenie.org (Sundell et al., 2015), and TAIR (Berardini et al., 2015). Populus gene orthologies to Arabidop- sis were determined by the best BLAST hit on www.popgenie.org or via manually BLAST on The Arabidopsis Information Resource (www.arabidopsis.org).

4.3 Results

4.3.1 Global and local ancestry of hybrids

Sequence ﬁltering and merging of the test individuals and the 50 reference individuals yielded 31,523 SNPs for global and local ancestry estimation. Global ancestry calculated from the local ancestries indicate the individuals we sampled span a range

169 of hybrid ancestries from unadmixed (RASPberry K = 2 q-matrix = 0.9996) to admixed (RASPberry K = 2 q-matrix 0.5754) (Fig. 4.1c). Filial generation of the test individuals was estimated with NewHybrids (E Anderson and E Thompson, 2002) in Fetter (2019, Ch.3), and this set of individuals included 18 P1.F2, 6 P1.P1F1,10

P1.P1F1 and 83 unadmixed P. balsamifera. The ancestry of each locus estimated with RASPberry (Wegmann et al., 2011) revealed a patchwork of introgressing chromosomal regions across the genome (Fig. 4.1d). Among the hybrid individuals, heterozygous sites predominated, with relatively few sites homozygous for the P. trichocarpa parent. The predominance of heterozygous local ancestry sites is consistent with our expectations, given that the majority of samples are derived from advanced generation backcrosses into P. balsamifera.

4.3.2 Trait variation

The traits we measured fall into three general patterns of distributions: approximately normal, left-skewed and zero inﬂated (Fig. 4.1b). Traits with approximately normal distributions included the ecophysiology, budﬂush, relative growth rate and abaxial stomata traits. Left-skewed traits include the three stomatal ratio traits (SR, LR, PR) and the adaixal stomatal traits which are left-skewed as a result of many unadmixed

P. balsamifera lacking adaxial stomata. The disease phenotypes we measured are all zero-inﬂated and the tail of the distribution contains most of the hybrid genotypes. A detailed discussion of these traits and their covariation to each other and with global ancestry of the hybrids is presented in Fetter (2019, Ch.3). We ﬁt a logistic model of the presence/absence of disease to the 19 traits and global ancestry (Fig. 4.2a). Total porosity (TO_PO) was excluded from the analysis, as that

170 trait is the sum of the abaxial and adaxial porosity, which were included separately. SR had the largest odds ratio (9.8e14), and PR had the smallest (1.8e-16), indicating variation of these traits contributed the most variance to the presence or absence of disease. Ecophysiology and bud ﬂush traits explained little variance of disease presence. Predicted values from the logistic model for SR indicated the probability of disease increased strongly with non-zero values of SR (Fig. 4.2b). Predicted values of disease presence from global ancestry indicated a decrease in the probability of disease with increasing values of P. balsamifera ancestry (Fig. 4.2c). A biplot of the SR and global ancestry demonstrated a decrease of SR with increasing P. balsamifera ancestry.

4.3.3 Admixture mapping

Admixture mapping was performed with BMIX (Shriner et al., 2011) using the 31,523 locus-specific ancestry estimates and 24 traits (Fig. 4.3, Table 4.2). Out of 746,928 tests, 3.2% (23,670 tests) had p-values larger than the admixture burden corrected p-value threshold (– = 0.0.5 / 237.50) (Fig. 4.3a, Fig S4.2, S4.3, S4.4). Significant loci from multiple tests colocalized into chromosomal regions, possibly indicative of pleiotropy or clustering of genes as a result of selection. Based on the colocalizaton of results, we grouped manhattan plots into three groups: disease and stomatal patterning traits (Fig. 4.3b), ecophysiology traits (Fig. 4.3c), and stomata traits that did not colocalize with significant results from disease tests (Fig. 4.3d). To identify shared signals of association between traits, significant loci were mapped to 13,997 genes, of which less than 10% (3,877 genes) were uniquely identified. The largest intersection contained 1,142 unique candidate genes identified in the BMIX

171 tests for SR, LR, PR, PL_AD, PO_AB, SD_AD, dis_pres, D1’ and D2’ (Fig. 4.4). Concatenating the list of genes uniquely found in association to the four disease traits and not in association to non-disease traits resulted in 1,562 candidate genes.

4.3.4 Selection scans

Positive selection was inferred using a sliding windows approach implemented with RAiSD (Alachiotis and Pavlidis, 2018). 31,545 SNPs were input into the selection scans of sliding windows of 50 SNPs in size yielded 13,395 windows in the hybrid set,

13,863 windows in the P. trichocarpa set, and 5,559 windows in the western core P. balsamifera set. The median value of the µ statistic for the P. trichocarpa, hybrid, and P. balsamifera data sets were 1.13, 2.42, and 3.97, respectively (Fig. 4.5). µ statistics were plotted by window position (Fig. 4.5) Nine chromosomal regions contained candidate genes from BMIX analyses that overlapped with the top 1% of µ statistic windows and contained 271 unique genes (Table 4.3, Fig. 4.5).

4.3.5 Candidate genes and local ancestry class

phenotypic distributions

To further investigate the 271 genes we identiﬁed with BMIX that are in the top 1% of selections windows, we determined the local ancestry sites mapping to those genes and ﬁltered them for monomorphic sites. We had 535 local ancestry sites that mapped to the 271 genes, and 221 of those sites were polymorphic. Boxplots of the local ancestry genotypes for SR, D1’ and G for each site were made (results not shown), and the distributions for were evaluated for phenotypic patterns among

172 the loci to remove sites that had either no phenotypes for a the heterozygote local ancestry genotype (53 sites), or had only one individual in the heterozygote local ancestry genotype (5 sites). After ﬁltering for phenotypic distributions, 163 sites remained which shared one of ﬁve patterns of phenotypic distributions (Fig. 4.6).

Substituting a P. trichocarpa ancestry allele had a large average allelic eect for SR and D1’, and to a lesser extent a negative eect for G for sites on chromosome 1 and 11 (Table 4.4). We mapped the 163 sites to 99 genes and investigated each gene with popgenie.org (Sjödin et al., 2009a), atgenie.org (Sundell et al., 2015), and TAIR (Berardini et al., 2015) for gene families, functions, and gene expression profiles that were relevant to disease or stomata patterning. Among the 99 candidate genes, we identified genes involved in guard cell functioning, the immune system, detoxicants, lipid biosynthesis and tracking, growth related proteins, cell wall production, abiotic/biotic stress response, epigenetic regulation, ubiquination, membrane transport, transcription factors, DNA replication, signal transduction, and floral development related genes (Table 4.5).

4.4 Discussion

We conducted an admixture mapping study using a collection of P. balsamifera and P. trichocarpa hybrids predominated by advanced generation backcrosses to P. balsamifera. Our primary interest was to deconstruct the genetic variation of disease susceptibility to a fungal pathogen (Melampsora medusae). Previous analyses of the phenotypic correlations between disease, stomatal, and ecophysiology traits indicated

173 high covariance between disease and stomatal patterning traits, in particular SR (Fet- ter, 2019, Ch3). Using the results of admixture mapping, we were able to find a common set of genes associated with both stomatal patterning and disease trait variation. We identified patterns of selection in the genome with sliding window selection scans, and used the top 1% of genes under selection as a filter for identifying candidate genes. Filtering candidate genes based on the polymorphism of local ancestry sites mapped to each gene allowed us to identify a core set of genes contributing to the signal of shared genomic associations from multiple phenotypes (Fig. 4.4). We identified 99 candidate genes underlying the shared genetic variation of stomatal patterning and disease severity traits in P. balsamifera x trichocarpa hybrids. Many of the candidate genes we identified have been studied in Arabidopsis and collectively have functions at the intersection of stomata guard cell regulation, pathogen identification, signal transduction, constitutive defenses (e.g. lipid and cell wall synthesis), inducible defenses, and regulation of growth. The molecular functions of the genes we identified indicate these genes may form a core network of genes under pathogen-induced selection. The proteins coded by these genes may interact directly with pathogens, for example if they are a component of innate immunity, or indirectly through their role in regulation of stomatal functioning. Interactions between the candidate genes are likely, given their shared co-regulation of several important defensive responses in plants.

174 4.4.1 Candidate genes

Guard cell functioning

We identified several genes with known functions for guard cell regulation. Guard cells open and close a stoma aperture pore via reversible changes in the concentration of ions and subsequent turgor pressure changes in the cell (Pandey et al., 2007). Calcium ions can function as messenger molecules in plant signalling pathways, and can be pumped into guard cells to change the ionization of the cell (Lecourieux et al., 2006). We identified Potri.016G115500, a putative Ca2+/H+ ion transmembrane transporter CATION EXCHANGER 3 (CAX3) that forms homo- and heterodimers with CAX1 in guard cells which are required to regulate Ca2+ homeostasis and to open and close the aperture pore in Arabidopsis (Hocking et al., 2017). CAX3 expression can be induced in Arabidopsis guard cells by inoculating leaves with Psudomonas and Botrytis (eFP BAR; http://bbc.botany.utoronto.ca, Winter et al., 2007). In cax1-1 and cax3- 1 stomata aperture pores were smaller than wildtype (Col-0) and mutants showed increased transcript levels of pathogenesis responsive genes (e.g PR1 and PR5), indicating interactions between the immune system and CAX1 and CAX3 (Hocking et al., 2017). The nearby intracellular signal transduction kinase Potri.016G117200, CALCIUM DEPENDENT PROTEIN KINASE 13-2 (CPK13-2), was identified. In

Arabidopsis, this kinase propagates a defensive response signal to close stomata, and overexpression inhibits light induced stomatal opening (Ronzier et al., 2014). Reac- tive oxygen species (ROS) are involved in calcium dependent guard cell homeostasis, and we identiﬁed Potri.011G112700, GLYCOLATE OXIDASE 1 (GOX1), which encodes an enzyme that modulates ROS-mediated signal transduction during pathogen-

175 induced defensive responses (Rojas et al., 2012).

Immune system

A successful host defensive response to a pathogen is initiated by recognizing the presence of pathogens, and plants have evolved a variety of resistance (R) genes which detect their environment for pathogen associated molecular patterns (PAMPs) indicative of pathogen presence (Jones and Dangl, 2006). PAMP-triggered immunity (PTI) can be induced by the detection of PAMPs, which results in a signaling cascade to initiate a broad-spectrum defensive response by the host plant (Dodds and Rathjen, 2010). We detected several components of the plant immune system. Potri.011G116200, PRENYLATED RAB ACCEPTOR 1 (PRA1), is involved in immune system signal transduction (Berardini et al., 2015). In tomato, PRA1 interacts with the pattern recognition receptor (PRR) LeEIX2, which recognizes the fungal

PAMP ethylene-inducing xylanase (EIX) (Pizarro et al., 2018). In P. trichocarpa, PRA1 has increased expression in mature leaves, and beetle, and mechanically in- jured leaves (Sjödin et al., 2009a). The AtPRA1 is highly expressed in guard cells (log2-FC = 4.54, eFP), responsive to most abiotic stresses, and responsive to inoculation with Pseudomonas, Phytophthora, and Hyaloperonospora (an oomycete) Winter et al., 2007. Potri.011G113000, GLUTATHIONE S-TRANSFERASE TAU 19 (GSTU19) is responsive to oxidative stress and is involved in detoxyﬁng herbicides (Hara et al., 2010).

Expression of GSTU19 in drought and mechanically damaged leaves is elevated in P. trichocarpa (Sjödin et al., 2009a), and in Arabidopsis it shows increased expression after inoculation with Psudomonas or Botrytis, and after osmotic and heat stress

176 Winter et al., 2007. We identiﬁed four tandomly arrayed MULTI ANTIMICROBIAL EXTRUSION (MATE)/DETOXIFICATION EFFLUX CARRIERS (DTX) proteins (Potri. 011G117100, Potri.011G117200, Potri.011G117300, Potri.011G117400). These proteins are transporters of toxins (Diener et al., 2001), xenobiotic molecules (Miyauchi et al., 2017), and drugs such as salicylic acid (Nawrath et al., 2002). Some MATE/DTX proteins will also transport ﬂavonoids (Buer et al., 2010; Marinova et al., 2007), which, in poplars, are inducible defensive molecules which inhibit the growth of

Melampsora sp. (Ullah, Unsicker, et al., 2017). METALlOTHIONEIN 3 (MT3), Potri.011G118900, is involved in the detoxiﬁcation of metals, and guard cells expressing MT3 have been demonstrated to detoxify cadmium (J Lee et al., 2004). The plant immune system can initiate regulated cell death after detecting a pathogen, and is generally an eective response to inhibit pathogen growth. How- ever, controlled cell death is a costly reaction, and the initiation of apoptosis is tightly regulated by negative immune system regulatory proteins (Jones and Dangl, 2006). We identiﬁed Potri.011G121200, CONSTITUTIVE EXPRESSER OF PR GENES 1 (CPR1), an F-box associated binding protein that is a negative regulator of the immune system (Gou et al., 2012).

Constitutive defenses

The plant cuticle is composed of lipids and is an eective barrier to pathogen colonization, and is a constitutive defense (Serrano et al., 2014). Dierences in cuticle lipid chemistry have been linked to variation of Melampsora infection in P. trichocarpa (Gonzales-Vigil et al., 2017). We observed six genes involved in lipid biosynthesis or

177 transport, including Potri.001G317400, CERAMIDASE 1 (CES1); Potri.016G115800, GDSL esterase/lipase CPRD49 (CPRD49-1); Potri. 016G116400, PYRIDOXINE BIOSYNTHESIS 1-2 (PDX1-2); Potri.016G113800, MULTIPLE C2 DOMAIN AND TRANSMEMBRANE REGION PROTEIN 15 (MCTP15); Potri.016G118000, ACYL- COA OXIDASE 4 (ACX4); and Potri.001G316600, a GDSL-motif type esterase/lipase. CES1 encodes a ceramide synthase which metabolizes sphingolipids and is involved in disease resistance and response to salt stress in Arabidopsis (Berardini et al., 2015). Loss of CES1 inhibits the autophagy response, and its overexpression promotes autophagy (Zheng et al., 2018). PDX1-2 functions in tandem with PDX2 to form a complex that synthesizes pyridoxal 5-phosphate (PLP) (Ristilä et al., 2011), a metabolically active form of vitamin B6 that is used by ceramidases to synthesize long-chain fatty acids (Dunn et al., 2004). Arabidopsis pdx1.2 and pdx1.3 have been demonstrated to have lower levels of vitamin B6 that correlated to increased infection by Pseudomonas (Y Zhang et al., 2015). Increased expression of PDX 1-2 in Arabidopsis after wounding is correlated with increased CHALCONE SYNTHASE (CHS) expression, a flavonoid biosynthesis protein, resulting in anthocyanin accumulation (S Li and Strid, 2005). ACX4 is involved early in the pathway of long-chain fatty acid biosynthesis in seeds of Arabidopsis (Rylott et al., 2003), and its gene expression is responsive to inoculation with Pseudomonas, Phytophthora, and Botrytis in Arabidopsis (Winter et al., 2007). In P. trichocrapa, AXC4 was identified to be introgressing into P. balsamifera (Suarez-Gonzalez, Hefer, Christe, et al., 2016). Fi- nally, we identified a GDSL-motif esterase/acyltransferase/lipase that is involved in catalyzing acyltransfer reactions during lipid biosynthesis (Berardini et al., 2015). Two proteins involved in cell wall homeostasis were identified by our scans. Potri.

178 013G056800, TUBULIN ALPHA 5 (TUA5) is a cytoskeleton protein that is involved in cell wall development and responsive to cadmium ions (Berardini et al., 2015). In both Arabidopsis and P. trichocarpa, gene expression of TUA5 is responsive to biotic and abiotic stress (Winter et al., 2007; Sjödin et al., 2009a). Potri.016G114300, AROMATIC ALDEHYDE SYNTHASE (AAS) is a gene which makes precursors to numerous plant secondary metabolites (Gutensohn et al., 2011), including hydroxycinnamic acid-conjugated amides, which are thought to be deposited on cell walls to act as a barrier to pathogens (Facchini et al., 2002). In P. trichocarpa, AAS gene expression is increased in leaves in drought and wounding treatments (Sjödin et al.,

2009a), and in Arabidopsis, increased gene expression after ﬂooding the leaf with Pseudomonas and under abiotic stress is observed Winter et al., 2007. Gutensohn et al. (2011) further document the role of AAS in defense by demonstrating gene expression is responsive to jasmonic acid treatment and wounding.

Growth regulation

We identified several genes involved in transcriptional regulation of growth and cell boundary specification. Potri.011G115400 is a transcription factor with several names in Arabidopsis: ARABIDOPSIS NAC DOMAIN CONTAINING PROTEIN 98 (ANAC098), CUP-SHAPED COTYLEDON 2 (CUC2), and NO APICAL MERIS- TEM/ CUP-SHAPED COTYLEDONS 2 (NAC2). The ANAC098/CUC2/NAC2 transcription factor is involved in cell boundary specification in the meristem, which is essential for forming lateral organs in plants (Maugarny-Calès et al., 2019). anac098 lack meristems and will grow deformed cotyledons and eventually die (Berardini et al.,

2015). In P. trichocarpa, NAC2 is expressed in dormant vegetative and ﬂoral buds

179 and was identiﬁed in a genome-wide association of height growth, %C, %N, C:N,

13C, and 15N in P. trichocarpa (Suren, 2017). NAC86 (Potri.011G121300) was also identified by our methods, and has increased expression in dormant buds, roots, mature leaves, and drought and beetle damaged leaves in P. trichocarpa (Sjödin et al., 2009a). An auxin transmembrane transporter Potri.016G113600, AUXIN RESIS- TANT 1 (AUX1), was identified by our scans, and is a gene with numerous growth and organ specificity roles in the plant (Berardini et al., 2015), and shows decreased expression in Arabidopsis after exposure to nematodes (Winter et al., 2007).

4.4.2 Limitations of our study

There are limitations inherent in our sample and methodology worth considering. The

P. trichocarpa genome is approximately 422 Mb in size and our admixture mapping is based on a data set of 31,523 local ancestry sites. Linkage disequilibrium between sites decays rapidly in the poplar genome (GA Tuskan et al., 2006), and even with 31K sites, we can cover much of the relevant gene space of the genome. However, it is worth noting that these results are likely sensitive to low sampling density, and we would likely ﬁnd dierent candidate genes with a dierent sequence data set. A further limitation of our data is from sampling disease for only two years, and from sampling the other traits in only one year. Our results indicate that genomic associations for the same trait can change from year to year (e.g. D1’, and D2’, Fig. 4.5B), and associations on the same trait measured with dierent scales can also generate dierent lists of signiﬁcant loci (e.g. D2’ and D2”). We assume that pathogen pressure was homogeneous in our common garden, but we have no data to test that assumption. There was a single large larch tree 100 m from our garden, and

180 dierences in trait variation due to the position in the garden where the tree grew were controlled for by our quantitative genetic mixed models. Nevertheless, variation due to genetic dierences within M. medusae are unknown. Finally, we only had access to advanced generation backcrosses, and likely missed much of the segregating genetic variation that contributes to disease variance. Fortunately, the patterns of disease variance between unadmixed and admixed poplars was large, and we recovered loci with high statistical conﬁdence that are relevant to disease and stomatal processes in

Arabidopsis and poplar.

4.4.3 Conclusion

We have used admixture mapping to identify genes under selection that are associated with disease severity to a fungal pathogen, stomata patterning traits, and growth. These result provide evidence that admixture mapping can be used to ﬁnd ecologically relevant genes, and supports the hypothesis that variation of loci within genes can have eects that cascade to the ecological relationships and the broader environment (AS Wymore et al., 2011). In this study, hybrid genotypes serve as a reservoir of disease, an observation shared by other studies in poplars (Whitham, 1989). These genes have likely experienced selection to produce phenotypes that exhibit growth-defense trade-os demonstrated from the set of genotypes under study here (Fetter, 2019, Ch.3). The colocalization of genomic associations we observe to shared chromosomal regions between disease and stomatal patterning traits likely indicates that stomata are under strong positive selection. In particular the SR, a trait which should have a much broader phenotypic distribution in nature than what is observed (Muir, 2015), is likely under selection. These results suggests natural

181 selection can eectively purge maladaptive genetic variation in hybrid populations, and that stomata are integral to an integrated network of genes regulating growth and defense.

182 4.5 Acknowledgements

This study was funded by a National Science Foundation Award 1461868. Matt Fitz- patrick, Andrew Elmore, Raju Soolanayakanahally, Steve Guinn aided in collections of dormant cuttings. Dave Nelson provided isotope data. Terrence Delaney provided access to his microscopy facilities. Sven Eberhardt helped create an early version of StomataCounter used to phenotype stomata density. Madie Hassett collected disease phenotypes in 2015. Many volunteers from the Plant Biology department planted cuttings in 2014 to establish the Spear Street common garden.

4.6 Data Accessibility

Cuticle micrographs are deposited on Dryad (doi:10.5061/dryad.kh2gv5f). Raw sequence reads are available for download at NCBI SRA accession no. SRP070954.

4.7 Author contributions

KCF conceived the study, phenotyped accessions, performed the analyses, and wrote the manuscript. SRK provided extensive feedback on earlier drafts and acquired plant material and DNA sequences, and phenotyped accessions for study. Both authors read and approved the ﬁnal manuscript.

183 4.8 Tables

184 Table 4.1: Population locality and sample size summaries for individuals used in admixture mapping. ref-Pb. and ref-Pt. are abbreviations for Populus balsamifera and P. trichocarpa that formed the reference sets.

Pop. Set N Lat. Lon. BESC ref-Pt. 15 47.81 -122.01 GW ref-Pt. 9 47.30 -122.58 Nisqually ref-Pt. 1 47.03 -122.67 DCK ref-Pb. 20 51.60 -101.73 CLK test 6 54.22 -110.08 HBY ref-Pb./test 23 52.89 -102.39 CYH test 4 49.64 -109.98 FNO test 8 58.50 -122.37 JKH test 6 43.83 -110.47 MMT test 6 49.88 -102.59 MSG test 4 44.31 -106.90 OFR test 14 53.14 -101.10 OUT test 4 51.14 -106.20 SKN test 4 52.33 -106.27 SSR test 12 44.46 -109.61 TUR test 7 53.20 -108.32 USDA12 test 6 46.58 -88.030 USDA13 test 2 46.08 -88.030 USDA14 test 1 45.58 -88.030 USDA15 test 1 46.42 -86.870 USDA18 test 1 45.42 -84.500 USDA3 test 1 47.17 -91.670 USDA7 test 3 48.42 -92.980 USDA8 test 1 48.33 -94.520 USDA9 test 4 47.92 -94.520 WLK test 4 60.05 -128.44 Total 167

185 Table 4.2: Trait deﬁnitions and abbreviations.

Trait Abbr. Disease severity 2015 scale 1 D1’ Disease severity 2016 scale 1 D2’ Disease severity 2016 scale 2 D2” Disease presence (converted from D2”) dis_pres Relative growth rate (cm) G Stomatal ratio SR Pore length ratio LR Porosity ratio PR Abaxial stomata density (mm2) SD_AB Adaxial stomata density (mm2) SD_AD Abaxial pore length (µm) PL_AB Adaxial pore length (µm) PL_AD Abaxial porosity PO_AB Adaxial porosity PO_AD Total porosity TO_PO Carbon isotope discrimination (%) 13C Nitrogen isotope fractionation (%) ”15N Carbon:Nitrogen CN Carbon(%) C Nitrogen(%) N Specific leaf area SLA Chlorophyll content index CCI Cumulative degree growing days to bud flush cGDD-15 Cumulative degree growing days to bud flush cGDD-16

Table 4.3: Summary of genes associated to stomatal patterning and disease traits also under positive selection. The number of genes found in each chromosomal region under selection is given beneath each chromosome.

BMIX Gene Set P. trichocarpa Hybrids P. balsamifera Stomata + Disease Chr11 Chr11 (46) (46) Disease Chr9, Chr11 Chr11 Chr1, Chr13, Chr16 (2, 57) (18) (21, 12, 28)

186 Table 4.4: Average allelic eects for substituting a P. balsamifera ancestry site (0) for P. trichocarpa (1). Phenotypes were scaled and standardized before calculating the allelic eect. The allelic eects for each site in the chromosomal region under consideration was identical. The phenotypic distributions for these sites are in Fig. 4.6.

Trait Chr1 Chr9 Chr11 Chr13 Chr16 SR 1.87 1.61 2.00 1.17 1.17 D1’ 2.21 1.26 2.21 1.02 1.02 G -1.88 -1.61 -2.00 -1.16 -1.16

187 Table 4.5: Descriptions of candidate genes. Arabidopsis orthologs were identiﬁed by the best BLAST hit from www.popgenie.org or through BLAST results from The Arabidopsis Information Resource (www.arabidopsis.org). See text for details regarding candidate gene selection.

Potri ID At-ortholog Pfam description

Guard cell function 001G317800 AT1G15690 NA 001G318800 AT4G30990 Down-regulated in metastasis 011G112700 AT3G14420 oxidoreductase activity 013G057500 AT5G56540 NA 016G115500 AT3G51860 transmembrane transport 016G117200 AT3G51850 Protein tyrosine kinase, protein phosphorylation EF hand

Immune system & detoxicants 011G113000 AT1G78380 protein binding, Glutathione S-transferase, C-terminal domain 011G116200 AT4G27540 PRA1 family protein 011G116900 AT5G53890 Leucine rich repeat N-terminal domain 011G117100 AT3G23560 MatE (multi antimicrobial extrusion protein) 011G117200 AT3G23550 MatE (multi antimicrobial extrusion protein) 011G117300 AT3G23560 MatE (multi antimicrobial extrusion protein) 011G117400 AT3G23550 MatE (multi antimicrobial extrusion protein) 011G118200 AT1G28280 VQ motif 011G118900 AT3G15353 NA 011G121200 AT4G12560 F-box associated, protein binding 013G058500 AT3G03960 cellular protein metabolic process 016G118100 AT3G51830 SacI homology domain

Lipid biosyntheis & transport 001G316600 AT3G05180 NA 001G317400 AT4G22330 ceramide metabolic process 016G113800 AT1G74720 protein binding 016G115800 AT2G38180 lipid metabolic process Continued on next page 188 Table 4.5 – continued from previous page Potri ID At-ortholog Pfam description 016G116400 AT5G01410 pyridoxal phosphate biosynthetic process 016G118000 AT3G51840 acyl-CoA dehydrogenase activity

Growth related 011G121300 AT5G17260 no apical meristem (NAM) protein 011G115400 AT5G53950 regulation of transcription, DNA-dependent 016G113600 AT2G38120 transmembrane amino acid transporter protein 016G114600 AT5G01270 double-stranded RNA binding 016G118400 AT4G33270 WD domain, G-beta

Cell wall related 013G056800 AT5G19780 Tubulin/FtsZ family, GTPase domain 016G114300 AT2G20340 carboxylic acid metabolic process

Abiotic/biotic stress responsive 001G316900 AT4G04980 NA 001G317000 AT4G04980 NA 001G317300 AT5G49210 NA 001G318900 NA NA 011G115200 AT3G30390 NA 013G056900 AT5G18100 superoxide metabolic process 013G057700 AT3G03890 FMN binding

Epigenetics & DNA replication 001G316200 AT3G01320 nucleus, Histone deacetylase (HDAC) interacting 001G316300 AT3G01320 nucleus 001G316500 AT1g04840 PPR repeat 001G317500 AT4G13780 tRNA binding, aminoacyl-tRNA ligase activity 001G317700 AT2G31740 methyltransferase activity 009G085400 AT1G44910 protein binding, FF domain 011G114100 AT1G17160 pfkB family carbohydrate kinase 011G116000 AT4G13870 nucleobase, nucleoside, nucleotide and nucleic acid metabolic process 011G116100 AT5G53920 protein methyltransferase activity 013G056700 AT5G19790 apetella 2 domian (AP2 domain) transcription factor 013G057000 AT5G18110 translation initiation factor activity 013G058800 AT1G54390 protein binding Continued on next page

189 Table 4.5 – continued from previous page Potri ID At-ortholog Pfam description 013G058900 AT4G13650 PPR repeat 016G116900 AT5G05610 PHD-ﬁnger 016G117300 At5g01380 Myb/SANT-like DNA-binding domain

Ubiquination 001G316400 AT1G04850 protein binding, PUB domain 011G112800 AT3G14400 ubiquitin thiolesterase activity 016G115300 AT5G01520 zinc ﬁnger, C3HC4 type (RING ﬁnger)

Flower related 001G316800 AT1G04910 GDP-fucose protein O-fucosyltransferase 011G112500 AT1G31660 bystin 011G115000 AT5G57850 catalytic activity 016G116300 AT5G01450 NA 016G117400 AT5G01370 NA

Membrane transporters 001G316700 NA Rab GTPase activator activity 001G317100 AT4G13750 NA 001G317200 AT4G13750 NA 001G318700 AT1G71900 NA 016G115400 AT5G01500 Mitochondrial carrier protein

Secondary metabolism related 001G317600 AT1G04920 sucrose metabolic process 016G115600 AT2G25300 NA

Signal transduction 011G114200 NA protein transport, Plug domain of Sec61p 016G114800 AT3G09010 protein phosphorylation 016G119300 AT2G38280 purine ribonucleoside monophosphate biosynthetic process

Unknown genes, functions, or enigmatic 001G317900 NA NA 001G318000 NA NA Continued on next page

190 Table 4.5 – continued from previous page Potri ID At-ortholog Pfam description 009G085500 AT2G20240 NA 011G112600 NA NA 011G115100 AT5G53970 transferase activity, transferring nitrogenous groups 011G115300 NA Plant mobile domain 011G115900 NA NA 011G117000 NA NA 011G118000 AT5G53860 NA 011G118100 NA metal ion binding 013G057100 AT3G03860 cell redox homeostasis 013G057200 AT5G18130 NA 013G058600 AT1G64770 NA 016G113900 NA NA 016G114000 NA NA 016G114100 NA NA 016G114700 AT2G40060 clathrin coat of trans-Golgi network vesicle 016G115700 NA NA 016G116200 AT5G01460 NA 016G116800 NA NA 016G117900 NA NA

191 4.9 Figures

Figure 4.1: Map of collection localities in western North America with the range of P. balsamifera and P. trichocarpa colored in blue and green, respectively (a). Ranges from (Little, 1971). Traits distributions fall into three categories: approximately normal, left- skewed, and zero-inﬂated distributions (b). Global ancestry estimates from locus-speciﬁc ancestries estimated by RASPberry (c). Local ancestry for each individual (in rows) and for every site (in columns) colored as gray if the locus is homozygous for the P. balsamifera allele, blue for heterozygous loci, and green for homozygous for the P. trichocrapa allele (d).

192 a Y = Disease presence b c d SR ● 100% 100% ● LR ● PO AD ● SD AB ● 0.3 ● PL AB ● ● ● N ● 75% 75% ● ● ● G ● ● ● ● ● ● C ● ● ● ● CN 0.2 ● ● ● D 13 C ● SR ● 50% 50% ● ● ● bfvt 16 ● ● ● ● ● bfvt 15 ● ● ● d 15 N ● ● ● global ancestry ● ● 0.1 ● Disease presence Disease presence CCI ● 25% 25% ● SLA ● ● ● ● ● PO AB ● ● SD AD ● PL AD ● ● ● ● ●●●●●● PR ● 0% 0% 0.0 1e−34 1e−19 1e−04 1e+11 1e+26 0.0 0.1 0.2 0.3 0.4 0.4 0.6 0.8 1.0 0.6 0.7 0.8 0.9 1.0 Odds Ratios SR global ancestry Global ancestry

Figure 4.2: Summary of logistic model and trait variation. Forest plot of standardized regression coecients from a logistic model with disease presence as the response. Blue and red points are positive and negative odds ratios, respectively, and red line is an odds ratio of 1 (a). See Table S4.1 for logistic model output. Predicted values of disease presence for stomata ratio (SR) (b) and global ancestry (c). Biplot of SR and global ancestry (d). Global ancestry of 1 indicates an unadmixed P. balsamifera individual.

193 194

Figure 4.3: Output of BMIX models represented in manhattan plots. All traits are plotted in the same space in a. Disease and stomata traits grouped in b, ecophysiology traits in c, stomata traits in d. Red points in b, c, and d indicate signiﬁcant loci at –=0.05/237.5. 1200 1142

800

558

Intersection Size 505

400

225

180 168 149146

103 92 90 58 53 45 37 36 32 31 31 27 27 25 22 18 16 12 9 6 6 5 5 4 4 4 3 1 1 1 0 10 PL_ABpl_ab 59 sd_abSD_AB 98 cC 139 cGDD-15bfvt15 274 cGDD-16bfvt16 275 slaSLA 341 d2_dowD2’’ 449 D13C∆13C 1142 lr

LR 1142 PL_ADpl_ad 1143 srSR 1160 PO_ABpo_ab 1202 SD_ADsd_ad 1291 prPR 1506 dis_presdis_pres 1770 d1_lamD1’ 1996 d2_lamD2’ 2000 1500 1000 500 0

Set Size Disease + Stomata Disease

Figure 4.4: Intersection plot of signiﬁcant genes from BMIX results.

195 1 1 2 2 3 3 4 4 5 5 10.0 10.0 10.0 10.0 10.0 10.0 10.0 10.0 10.0 10.0

7.5 7.5 7.5 7.5 7.5 7.5 7.5 7.5 7.5 7.5

5.0 5.0 5.0 5.0 5.0 5.0 5.0 5.0 5.0 5.0

2.5 2.5 2.5 2.5 2.5 2.5 2.5 2.5 2.5 2.5

0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0

6 6 7 7 8 8 9 9 10 10 10.0 10.0 10.0 10.0 10.0 10.0 10.0 10.0 10.0 10.0

7.5 7.5 7.5 7.5 7.5 7.5 7.5 7.5 7.5 7.5

5.0 5.0 5.0 5.0 5.0 5.0 5.0 5.0 5.0 5.0

2.5 2.5 2.5 2.5 2.5 2.5 2.5 2.5 2.5 2.5

0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 µ µ 11 11 12 12 13 13 14 14 15 15 10.0 10.0 10.0 10.0 10.0 10.0 10.0 10.0 10.0 10.0

7.5 7.5 7.5 7.5 7.5 7.5 7.5 7.5 7.5 7.5

5.0 5.0 5.0 5.0 5.0 5.0 5.0 5.0 5.0 5.0

2.5 2.5 2.5 2.5 2.5 2.5 2.5 2.5 2.5 2.5

0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0

16 16 17 17 18 18 19 19 0.8 10.0 10.0 10.0 10.0 10.0 10.0 10.0 10.0 0.6 7.5 7.5 7.5 7.5 7.5 7.5 7.5 7.5 0.4 5.0 5.0 5.0 5.0 5.0 5.0 5.0 5.0 density 0.2 2.5 2.5 2.5 2.5 2.5 2.5 2.5 2.5 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0 3 6 9 POS POS µ

set hybridset balsamhybrid balsamtricho overlapoverlaptricho overlapgenestomata set 1 + diseasegenegene set 1set 2 genedisease set 2

Figure 4.5: Plot of µ statistics for three sets of individuals by chromosome. The overlap of BMIX candidate genes and the top 1% of µ statistic outliers are indicated by the solid and dashed boxes for the disease + stomata gene set, and disease only gene set, respectively. The density distribution of µ-statistics is presented by set, and the vertical colored ticks on x-axis indicate the top 1% of values for each set.

196 Selection/BMIX/Polymorphic/Phenotypes-GoodCHR01 CHR11 CHR13 OutlierCHR16 Loci CHR09 27,849,546 - 12,838,087 - 4,259,351 - 10,335,133 - 6,190,343 - S1_3217280435,094,368 S11_1375805714,647,856 S13_42590308,055,615 S16_1177197218,699,424 S9_80047268,067,988

0.3

0.2 SR

0.1

0.0 01 01 01 01 01

S1_32172804 S11_13758057 S13_4259030 S16_11771972 S9_8004726

3 D1'

01 01 01 01 01

S1_32172804 S11_13758057 S13_4259030 S16_11771972 S9_8004726

30 G

15 01 01 01 01 01 Local ancestry genotype

Figure 4.6: Patterns of stomata ratio (SR), disease severity (D1’), and relative growth rate (G) by local ancestry for genomic regions identiﬁed to be associated with stomata patterning and disease traits, in the top 1% of RAiSD selection windows, polymorphic, and with phenotypic distributions indicative of true positive results. Local ancestry genotype 0 indicates site is homozygous for P. balsamﬁera ancestry, and 1 indicates the site is heterozygous with P. trichocarpa contributing the other copy of the locus.

197 Supporting Information

Table S4.1: Output of logistic model of disease presence and traits.

term estimate p.value SR 9.8e14 0.08 LR 2.4e10 0.21 PO_AD 3.7e8 0.18 SD_AB 1.e5 0.05 PL_AB 684.74 0.12 N 66.89 0.09 G 17.26 0.04 C 6.05 0.31 CN 3.84 0.49 13C 1.53 0.71 bfvt16 1.27 0.82 bfvt15 0.98 0.99 ”15N 0.69 0.70 global_ancestry 0.51 0.62 CCI 0.07 0.10 SLA 0.00 0.01 PO_AB 0.00 0.06 SD_AD 0.00 0.19 PL_AD 0.00 0.25 PR 0.00 0.06

198 Table S4.2: Summary of populations used in the RAiSD selection scan analysis. Pb. is an abbreviation for Populus balsamifera.

Pop. Set N Lat. Lon. CLK P. balsamifera 12 54.22 -110.08 DCK P. balsamifera 12 51.60 -101.73 HBY P. balsamifera 6 52.90 -102.39 MMT P. balsamifera 13 49.88 -102.59 OFR P. balsamifera 12 53.14 -101.10 OUT P. balsamifera 5 51.15 -106.26 SKN P. balsamifera 10 52.35 -106.64 TUR P. balsamifera 10 53.20 -108.36 CYH Hybrid 4 49.64 -109.98 FNO Hybrid 8 58.51 -122.38 HBY Hybrid 1 52.93 -102.39 JKH Hybrid 6 43.84 -110.47 MSG Hybrid 4 44.32 -106.91 SKN Hybrid 1 52.22 -106.27 SSR Hybrid 7 44.46 -109.60 USDA12 Hybrid 3 46.58 -88.03 USDA18 Hybrid 1 45.42 -84.50 WLK Hybrid 4 60.05 -128.44 13 P. trichocarpa 2 54.15 -128.60 31 P. trichocarpa 8 49.71 -125.06 Nisqually P. trichocarpa 9 47.10 -122.64 Nooksack P. trichocarpa 8 48.80 -122.17 Olympic Penninsula P. trichocarpa 3 47.59 -122.90 Puyallup P. trichocarpa 5 47.09 -122.20 Skagit P. trichocarpa 5 48.51 -122.07 Skykomish P. trichocarpa 6 47.82 -121.86 Total 165

199 iueS.:Goa netyetmtdfo DITR Aeadre l,20)a K at 2009) al., et (Alexander ADMIXTURE from estimated =2(top)andK=7(bottom). ancestry Global S4.1: Figure

WLK

FNO

JKH

DMTCYH

SSR

TUR

MSG

CLK

SKN

OUT

RMP

MMT

HBY

DCK

OFR

USDA8 USDA9 USDA1 USDA2 USDA5 USDA7USDA6 USDA3 USDA4 200 USDA10 USDA21USDA11 USDA12NIC USDA13 USDA14 USDA15 USDA16 USDA17 USDA18 USDA19 USDA20WTR

HST

CPL

TIM

KAP

NBY

DPR

LON

RAD

GAM

UMI CHL SLC

LSMLPD

VER NEG

CBI FIS

HWK

MBK Figure S4.2: Individual manhattan plots of p-values from BMIX tests. See Table 4.2 for trait abbreviation deﬁnitions.

201 Figure S4.3: Individual manhattan plots of p-values from BMIX tests. See Table 4.2 for trait abbreviation deﬁnitions.

202 Figure S4.4: Individual manhattan plots of p-values from BMIX tests. See Table 4.2 for trait abbreviation deﬁnitions.

203 References

Adams, RV and TM Burg (2015). “Gene ﬂow of a forest-dependent bird across a

fragmented landscape”. In: PloS one 10.11, e0140938. Alachiotis, N and P Pavlidis (2018). “RAiSD detects positive selection based on mul-

tiple signatures of a selective sweep and SNP vectors”. In: Communications biology 1.1, p. 79. Alexander, DH, J Novembre, and K Lange (2009b). “Fast model-based estimation of

ancestry in unrelated individuals”. In: Genome research 19.9, pp. 1655–1664. Ally, D, K Ritland, and S Otto (2008). “Can clone size serve as a proxy for clone

age? An exploration using microsatellite divergence in Populus tremuloides”. In: Molecular Ecology 17.22, pp. 4897–4911. Anderson, E and E Thompson (2002). “A model-based method for identifying species

Populus, condensed tannins, and fungal endophyte infection”. In: Canadian Jour- nal of Botany 83.4, pp. 356–361. Bailey, JK, JA Schweitzer, F Ubeda, J Koricheva, CJ LeRoy, MD Madritch, BJ Rehill, RK Bangert, DG Fischer, GJ Allan, et al. (2009). “From genes to ecosystems: a synthesis of the eects of plant genetic factors across levels of organization”. In:

Philosophical Transactions of the Royal Society B: Biological Sciences 364.1523, pp. 1607–1616.

204 Barton, NH and GM Hewitt (1985). “Analysis of hybrid zones”. In: Annual review of Ecology and Systematics 16.1, pp. 113–148. Berardini, TZ, L Reiser, D Li, Y Mezheritsky, R Muller, E Strait, and E Huala (2015). “The Arabidopsis information resource: making and mining the “gold standard”

annotated reference plant genome”. In: genesis 53.8, pp. 474–485. Breen, AL (2014). “Balsam poplar (Populus balsamifera L.) communities on the Arctic Slope of Alaska”. In: Phytocoenologia 44.1-2, pp. 1–24. Bresadola, L, C Caseys, S Castiglione, CA Buerkle, D Wegmann, and C Lexer (2018). “Admixture mapping in interspeciﬁc Populus hybrids identiﬁes classes of genomic

architectures for phytochemical, morphological and growth traits”. In: BioRxiv, p. 497685. Buer, CS, N Imin, and MA Djordjevic (2010). “Flavonoids: new roles for old molecules”.

In: Journal of integrative plant biology 52.1, pp. 98–111. Buerkle, CA and C Lexer (2008). “Admixture as the basis for genetic mapping”. In:

Trends in ecology & evolution 23.12, pp. 686–694. Chase, MA, S Stankowski, and MA Streisfeld (2017). “Genomewide variation provides insight into evolutionary relationships in a monkeyﬂower species complex

(Mimulus sect. Diplacus)”. In: American journal of botany 104.10, pp. 1510–1521. Chhatre, VE, LM Evans, SP DiFazio, and SR Keller (2018). “Adaptive introgression and maintenance of a trispecies hybrid complex in range-edge populations of

Populus”. In: Molecular ecology 27.23, pp. 4820–4838. Conway, JR, A Lex, and N Gehlenborg (2017). “UpSetR: an R package for the vi-

sualization of intersecting sets and their properties”. In: Bioinformatics 33.18, pp. 2938–2940.

205 De Carvalho, D, PK Ingvarsson, J Joseph, L Suter, C Sedivy, D MACAYA-SANZ, J Cottrell, B Heinze, I Schanzer, and C Lexer (2010). “Admixture facilitates adap-

tation from standing variation in the European aspen (Populus tremula L.), a widespread forest tree”. In: Molecular Ecology 19.8, pp. 1638–1650. Diener, AC, RA Gaxiola, and GR Fink (2001). “Arabidopsis ALF5, a multidrug eux

transporter gene family member, confers resistance to toxins”. In: The Plant Cell 13.7, pp. 1625–1638.

DiFazio, SP, GT Slavov, and CP Joshi (2011). “Populus: a premier pioneer system for plant genomics”. In: Enﬁeld, NH: Science Publishers, pp. 1–28. Dodds, PN and JP Rathjen (2010). “Plant immunity: towards an integrated view of

plant–pathogen interactions”. In: Nature Reviews Genetics 11.8, p. 539. Dowkiw, A and C Bastien (2004). “Characterization of two major genetic factors con-

trolling quantitative resistance to Melampsora larici-populina leaf rust in hybrid poplars: strain speciﬁcity, ﬁeld expression, combined eects, and relationship with

a defeated qualitative resistance gene”. In: Phytopathology 94.12, pp. 1358–1367. Dunn, TM, DV Lynch, LV Michaelson, and JA Napier (2004). “A post-genomic ap-

proach to understanding sphingolipid metabolism in Arabidopsis thaliana”. In: Annals of Botany 93.5, pp. 483–497. Elshire, RJ, JC Glaubitz, Q Sun, JA Poland, K Kawamoto, ES Buckler, and SE Mitchell (2011b). “A robust, simple genotyping-by-sequencing (GBS) approach

for high diversity species”. In: PloS one 6.5, e19379. Evans, LM, GJ Allan, SM Shuster, SA Woolbright, and TG Whitham (2008). “Tree hybridization and genotypic variation drive cryptic speciation of a specialist mite

206 herbivore”. In: Evolution: International Journal of Organic Evolution 62.12, pp. 3027– 3040. Evans, LM, GT Slavov, E Rodgers-Melnick, J Martin, P Ranjan, W Muchero, AM Brunner, W Schackwitz, L Gunter, JG Chen, et al. (2014). “Population genomics

of Populus trichocarpa identiﬁes signatures of selection and adaptive trait associations”. In: Nature genetics 46.10, p. 1089. Facchini, PJ, J Hagel, and KG Zulak (2002). “Hydroxycinnamic acid amide metabolism:

physiology and biochemistry”. In: Canadian Journal of Botany 80.6, pp. 577–589. Feau, N, DL Joly, and RC Hamelin (2007). “Poplar leaf rusts: model pathogens for a

model tree”. In: Botany 85.12, pp. 1127–1135. Fetter, KC (2019). “Natural selection for disease resistance in hybrid poplars targets stomata patterning traits and genes”. PhD thesis. University of Vermont. Fishman, L and JH Willis (2006). “A cytonuclear incompatibility causes anther steril-

ity in Mimulus hybrids”. In: Evolution 60.7, pp. 1372–1381. Floate, KD, J Godbout, MK Lau, N Isabel, and TG Whitham (2016). “Plant–

herbivore interactions in a trispeciﬁc hybrid swarm of Populus: assessing support for hypotheses of hybrid bridges, evolutionary novelty and genetic similarity”. In:

New Phytologist 209.2, pp. 832–844. Fowells, HA et al. (1965). “Silvics of forest trees of the United States.” In: Agric. Handb. US Dep. Agric. 271. Geraldes, A, CA Hefer, A Capron, N Kolosova, F Martinez-Nuñez, RY Soolanayakana- hally, B Stanton, RD Guy, SD Mansﬁeld, CJ Douglas, et al. (2015). “Recent Y

chromosome divergence despite ancient origin of dioecy in poplars (Populus)”. In: Molecular ecology 24.13, pp. 3243–3256.

207 Glaubitz, JC, TM Casstevens, F Lu, J Harriman, RJ Elshire, Q Sun, and ES Buck- ler (2014b). “TASSEL-GBS: a high capacity genotyping by sequencing analysis

pipeline”. In: PloS one 9.2, e90346. Gonzales-Vigil, E, CA Hefer, ME von Loessl, J La Mantia, and SD Mansﬁeld (2017). “Exploiting natural variation to uncover an alkene biosynthetic enzyme in poplar”.

In: The Plant Cell,tpc–00338. Gou, M, Z Shi, Y Zhu, Z Bao, G Wang, and J Hua (2012). “The F-box protein

CPR1/CPR30 negatively regulates R protein SNC1 accumulation”. In: The Plant Journal 69.3, pp. 411–420. Gutensohn, M, A Klempien, Y Kaminaga, DA Nagegowda, F Negre-Zakharov, JH Huh, H Luo, R Weizbauer, T Mengiste, D Tholl, et al. (2011). “Role of aromatic aldehyde synthase in wounding/herbivory response and ﬂower scent production in

dierent Arabidopsis ecotypes”. In: The Plant Journal 66.4, pp. 591–602. Hara, M, Y Yatsuzuka, K Tabata, and T Kuboi (2010). “Exogenously applied isoth-

iocyanates enhance glutathione S-transferase expression in Arabidopsis but act as herbicides at higher concentrations”. In: Journal of plant physiology 167.8, pp. 643–649. Heiser Jr, CB (1951). “Hybridization in the annual sunﬂowers: Helianthus annuus ◊ H. debilis var. cucumerifolius”. In: Evolution, pp. 42–51. Hocking, B, SJ Conn, M Manohar, B Xu, A Athman, MA Stancombe, AR Webb, KD Hirschi, and M Gilliham (2017). “Heterodimerization of Arabidopsis calcium/pro- ton exchangers contributes to regulation of guard cell dynamics and plant defense

responses”. In: Journal of experimental botany 68.15, pp. 4171–4183.

208 Hübner, S, N Bercovich, M Todesco, JR Mandel, J Odenheimer, E Ziegler, JS Lee, GJ Baute, GL Owens, CJ Grassa, et al. (2019). “Sunﬂower pan-genome analysis

shows that hybridization altered gene content and disease resistance”. In: Nature plants 5.1, p. 54. Iijima, T, R Kajitani, S Komata, CP Lin, T Sota, T Itoh, and H Fujiwara (2018).

“Parallel evolution of Batesian mimicry supergene in two Papilio butterﬂies, P. polytes and P. memnon”. In: Science advances 4.4, eaao5416. Ingvarsson, PK and NR Street (2011). “Association genetics of complex traits in

plants”. In: New Phytologist 189.4, pp. 909–922. Jones, JD and JL Dangl (2006). “The plant immune system”. In: nature 444.7117, p. 323. Keller, SR, MS Olson, S Silim, W Schroeder, and P Tin (2010). “Genomic diversity, population structure, and migration following rapid range expansion in the Balsam

Poplar, Populus balsamifera”. In: Molecular Ecology 19.6, pp. 1212–1226. La Mantia, J, J Klápöt, YA El-Kassaby, S Azam, RD Guy, CJ Douglas, SD Mans- ﬁeld, and R Hamelin (2013). “Association analysis identiﬁes Melampsora columbiana ◊ poplar leaf rust resistance SNPs”. In: PLoS One 8.11, e78423. Lamit, L, T Wojtowicz, Z Kovacs, S Wooley, M Zinkgraf, TG Whitham, R Lin- droth, and CA Gehring (2011). “Hybridization among foundation tree species

inﬂuences the structure of associated understory plant communities”. In: Botany 89.3, pp. 165–174. Lawrence, M, W Huber, H Pages, P Aboyoun, M Carlson, R Gentleman, MT Morgan, and VJ Carey (2013). “Software for computing and annotating genomic ranges”.

In: PLoS computational biology 9.8, e1003118.

209 Lecourieux, D, R Ranjeva, and A Pugin (2006). “Calcium in plant defence-signalling

pathways”. In: New Phytologist 171.2, pp. 249–269. Lee, J, S Donghwan, S Won-yong, H Inhwan, and L Youngsook (2004). “Arabidopsis metallothioneins 2a and 3 enhance resistance to cadmium when expressed in Vicia

faba guard cells”. In: Plant Molecular Biology 54.6, pp. 805–815. Leroy, T, JM Louvet, C Lalanne, G Le Provost, K Labadie, JM Aury, S Delzon, C Plomion, and A Kremer (2019). “Adaptive introgression as a driver of local

adaptation to climate in European white oaks”. In: bioRxiv, p. 584847. Lexer, C, C Buerkle, J Joseph, B Heinze, and M Fay (2007). “Admixture in Euro-

pean Populus hybrid zones makes feasible the mapping of loci that contribute to reproductive isolation and trait dierences”. In: Heredity 98.2, p. 74. Li, S and Å Strid (2005). “Anthocyanin accumulation and changes in CHS and PR-5

gene expression in Arabidopsis thaliana after removal of the inﬂorescence stem (decapitation)”. In: Plant Physiology and Biochemistry 43.6, pp. 521–525. Lindtke, D, S Gonzalez-Martinez, D Macaya-Sanz, and C Lexer (2013). “Admixture

mapping of quantitative traits in Populus hybrid zones: power and limitations”. In: Heredity 111.6, p. 474. Macnair, MR and Q Cumbes (1989). “The genetic architecture of interspeciﬁc varia-

tion in Mimulus.” In: Genetics 122.1, pp. 211–222. Marinova, K, L Pourcel, B Weder, M Schwarz, D Barron, JM Routaboul, I Debeaujon, and M Klein (2007). “The Arabidopsis MATE transporter TT12 acts as a vacuolar ﬂavonoid/H+-antiporter active in proanthocyanidin-accumulating cells of the seed

coat”. In: The Plant Cell 19.6, pp. 2023–2038.

210 Martins, K, PF Gugger, J Llanderal-Mendoza, A González-Rodríguez, ST Fitz-Gibbon, JL Zhao, H Rodríguez-Correa, K Oyama, and VL Sork (2018). “Landscape genomics provides evidence of climate-associated genetic variation in Mexican pop-

ulations of Quercus rugosa”. In: Evolutionary Applications 11.10, pp. 1842–1858. Maugarny-Calès, A, M Cortizo, B Adroher, N Borrega, B Gonçalves, G Brunoud, T Vernoux, N Arnaud, and P Laufs (2019). “Dissecting the pathways coordinat-

ing patterning and growth by plant boundary domains”. In: PLoS genetics 15.1, e1007913. Mitchell, N, GL Owens, SM Hovick, LH Rieseberg, and KD Whitney (2019). “Hy-

bridization speeds adaptive evolution in an eight-year ﬁeld experiment”. In: Sci- entiﬁc reports 9.1, p. 6746. Miyauchi, H, S Moriyama, T Kusakizako, K Kumazaki, T Nakane, K Yamashita, K Hirata, N Dohmae, T Nishizawa, K Ito, et al. (2017). “Structural basis for xenobi-

otic extrusion by eukaryotic MATE transporter”. In: Nature communications 8.1, p. 1633. Muir, CD (2015). “Making pore choices: repeated regime shifts in stomatal ratio”. In:

Proceedings of the Royal Society B: Biological Sciences 282.1813, p. 20151498. Nawrath, C, S Heck, N Parinthawong, and JP Métraux (2002). “EDS5, an essential component of salicylic acid–dependent signaling for disease resistance in Ara-

bidopsis, is a member of the MATE transporter family”. In: The Plant Cell 14.1, pp. 275–286. Pandey, S, W Zhang, and SM Assmann (2007). “Roles of ion channels and transporters

in guard cell signal transduction”. In: FEBS letters 581.12, pp. 2325–2336.

211 Pizarro, L, M Leibman-Markus, S Schuster, M Bar, T Meltz, and A Avni (2018). “Tomato prenylated RAB acceptor protein 1 modulates tracking and degra- dation of the pattern recognition receptor LeEIX2, aecting the innate immune

response”. In: Frontiers in plant science 9, p. 257. Rajput, VD, T Minkina, C Yaning, S Sushkova, VA Chapligin, and S Mandzhieva

(2016). “A review on salinity adaptation mechanism and characteristics of Populus euphratica, a boon for arid ecosystems”. In: Acta Ecológica Sinica 36.6, pp. 497– 503.

Rieseberg, LH and SE Carney (1998). “Plant hybridization”. In: The New Phytologist 140.4, pp. 599–624. Rieseberg, LH, SC Kim, RA Randell, KD Whitney, BL Gross, C Lexer, and K Clay (2007). “Hybridization and the colonization of novel habitats by annual sunﬂow-

ers”. In: Genetica 129.2, pp. 149–165. Ristilä, M, H Strid, LA Eriksson, Å Strid, and H Sävenstrand (2011). “The role of the pyridoxine (vitamin B6) biosynthesis enzyme PDX1 in ultraviolet-B radiation

responses in plants”. In: Plant Physiology and Biochemistry 49.3, pp. 284–292. Rojas, CM, M Senthil-Kumar, K Wang, CM Ryu, A Kaundal, and KS Mysore (2012). “Glycolate oxidase modulates reactive oxygen species–mediated signal transduc-

tion during nonhost resistance in Nicotiana benthamiana and Arabidopsis”. In: The Plant Cell 24.1, pp. 336–352. Ronzier, E, C Corratgé-Faillie, F Sanchez, K Prado, C Brière, N Leonhardt, JB Thibaud, and TC Xiong (2014). “CPK13, a noncanonical Ca2+-dependent protein kinase, speciﬁcally inhibits KAT2 and KAT1 shaker K+ channels and reduces

stomatal opening”. In: Plant physiology 166.1, pp. 314–326.

212 Rylott, EL, CA Rogers, AD Gilday, T Edgell, TR Larson, and IA Graham (2003). “Arabidopsis mutants in short-and medium-chain acyl-CoA oxidase activities accumulate acyl-CoAs and reveal that fatty acid —-oxidation is essential for embryo

development”. In: Journal of Biological Chemistry 278.24, pp. 21370–21377. Savolainen, O, T Pyhäjärvi, and T Knürr (2007). “Gene ﬂow and local adaptation in

trees”. In: Annu. Rev. Ecol. Evol. Syst. 38, pp. 595–619. Scheet, P and M Stephens (2006). “A fast and ﬂexible statistical model for large- scale population genotype data: applications to inferring missing genotypes and

haplotypic phase”. In: The American Journal of Human Genetics 78.4, pp. 629– 644. Schweitzer, JA, MD Madritch, JK Bailey, CJ LeRoy, DG Fischer, BJ Rehill, RL Lin- droth, AE Hagerman, SC Wooley, SC Hart, et al. (2008). “From genes to ecosystems: the genetic basis of condensed tannins and their role in nutrient regulation

in a Populus model system”. In: Ecosystems 11.6, pp. 1005–1020. Serrano, M, F Coluccia, M Torres, F L’Haridon, and JP Métraux (2014). “The cuticle

and plant defense to pathogens”. In: Frontiers in plant science 5, p. 274. Shriner, D, A Adeyemo, and CN Rotimi (2011). “Joint ancestry and association test-

ing in admixed individuals”. In: PLoS computational biology 7.12, e1002325. Sjödin, A, NR Street, G Sandberg, P Gustafsson, and S Jansson (2009a). “The Pop- ulus Genome Integrative Explorer (PopGenIE): a new resource for exploring the Populus genome”. In: New phytologist 182.4, pp. 1013–1025. Smith, MW and SJ O’Brien (2005). “Mapping by admixture linkage disequilibrium:

advances, limitations and guidelines”. In: Nature Reviews Genetics 6.8, p. 623.

213 Smith, RL and KJ Sytsma (1990). “Evolution of Populus nigra (sect. Aigeiros): introgressive hybridization and the chloroplast contribution of Populus alba (sect. Populus)”. In: American journal of botany 77.9, pp. 1176–1187. Stankowski, S and MA Streisfeld (2015). “Introgressive hybridization facilitates adap-

tive divergence in a recent radiation of monkeyﬂowers”. In: Proceedings of the Royal Society B: Biological Sciences 282.1814, p. 20151666. Stölting, KN, R Nipper, D Lindtke, C Caseys, S Waeber, S Castiglione, and C Lexer (2013). “Genomic scan for single nucleotide polymorphisms reveals patterns of

divergence and gene ﬂow between ecologically divergent species”. In: Molecular ecology 22.3, pp. 842–855. Suarez-Gonzalez, A, CA Hefer, C Christe, O Corea, C Lexer, QC Cronk, and CJ Douglas (2016). “Genomic and functional approaches reveal a case of adaptive

introgression from Populus balsamifera (balsam poplar) in P. trichocarpa (black cottonwood)”. In: Molecular ecology 25.11, pp. 2427–2442. Suarez-Gonzalez, A, C Lexer, and QC Cronk (2018). “Adaptive introgression: a plant

perspective”. In: Biology letters 14.3, p. 20170688. Sul, JH, LS Martin, and E Eskin (2018). “Population structure in genetic studies:

confounding factors and mixed models”. In: PLoS genetics 14.12, e1007309. Sundell, D, C Mannapperuma, S Netotea, N Delhomme, YC Lin, A Sjödin, Y Van de Peer, S Jansson, TR Hvidsten, and NR Street (2015). “The Plant Genome Integra-

tive Explorer Resource: PlantGen IE. org”. In: New Phytologist 208.4, pp. 1149– 1156. Suren, H (2017). “Sequence capture as a tool to understand the genomic basis for adaptation in angiosperm and gymnosperm trees”. PhD thesis. Virginia Tech.

214 Thompson, SL, M Lamothe, PG Meirmans, P Perinet, and N Isabel (2010a). “Re- peated unidirectional introgression towards Populus balsamifera in contact zones

of exotic and native poplars”. In: Molecular Ecology 19.1, pp. 132–145. Tuskan, GA et al. (2006). “The genome of black cottonwood, Populus trichocarpa (Torr. & Gray)”. In: science 313.5793, pp. 1596–1604. Ullah, C, SB Unsicker, C Fellenberg, CP Constabel, A Schmidt, J Gershenzon, and A Hammerbacher (2017). “Flavan-3-ols are an eective chemical defense against

rust infection”. In: Plant physiology 175.4, pp. 1560–1578. Wang, J, NR Street, DG Scoﬁeld, and PK Ingvarsson (2016). “Natural selection and recombination rate variation shape nucleotide polymorphism across the genomes

of three related Populus species”. In: Genetics 202.3, pp. 1185–1200. Wegmann, D, DE Kessner, KR Veeramah, RA Mathias, DL Nicolae, LR Yanek, YV Sun, DG Torgerson, N Rafaels, T Mosley, et al. (2011). “Recombination rates in

admixed individuals identiﬁed by ancestry-based inference”. In: Nature genetics 43.9, p. 847.

Whitham, TG (1989). “Plant hybrid zones as sinks for pests”. In: Science, pp. 1490– 1493. Whitham, TG, KD Floate, GD Martinsen, EM Driebe, and P Keim (1996). “Eco-

logical and evolutionary implications of hybridization: Populus–herbivore interactions”. In: Biology of Populus and its implications for management and conservation Part I, pp. 247–275. Whitton, J, D Wolf, D Arias, A Snow, and L Rieseberg (1997). “The persistence of cultivar alleles in wild populations of sunﬂowers ﬁve generations after hybridiza-

tion”. In: Theoretical and Applied Genetics 95.1-2, pp. 33–40.

215 Winter, D, B Vinegar, H Nahal, R Ammar, GV Wilson, and NJ Provart (2007). “An “Electronic Fluorescent Pictograph” browser for exploring and analyzing large-

scale biological data sets”. In: PloS one 2.8, e718. Wymore, AS, AT Keeley, KM Yturralde, ML Schroer, CR Propper, and TG Whitham (2011). “Genes to ecosystems: exploring the frontiers of ecology with one of the

smallest biological units”. In: New Phytologist 191.1, pp. 19–36. Yatabe, Y, NC Kane, C Scotti-Saintagne, and LH Rieseberg (2007). “Rampant gene

exchange across a strong reproductive barrier between the annual sunﬂowers, He- lianthus annuus and H. petiolaris”. In: Genetics 175.4, pp. 1883–1893. Zhang, Y, X Jin, Z Ouyang, X Li, B Liu, L Huang, Y Hong, H Zhang, F Song, and

D Li (2015). “Vitamin B6 contributes to disease resistance against Pseudomonas syringae pv. tomato DC3000 and Botrytis cinerea in Arabidopsis thaliana”. In: Journal of plant physiology 175, pp. 21–25. Zheng, P, JX Wu, SK Sahu, HY Zeng, LQ Huang, Z Liu, S Xiao, and N Yao (2018). “Loss of alkaline ceramidase inhibits autophagy in Arabidopsis and plays an im-

portant role during environmental stress response”. In: Plant, cell & environment 41.4, pp. 837–849.

216 Complete Bibliography

Abràmo, MD, PJ Magalhães, and SJ Ram (2004). “Image processing with ImageJ”. In: Biophotonics international 11.7, pp. 36–42. Adams, RV and TM Burg (2015). “Gene ﬂow of a forest-dependent bird across a fragmented landscape”. In: PloS one 10.11, e0140938. Aitken, SN and JB Bemmels (2016). “Time to get moving: assisted gene ﬂow of forest trees”. In: Evolutionary Applications 9.1, pp. 271–290. Alachiotis, N and P Pavlidis (2018). “RAiSD detects positive selection based on multiple signatures of a selective sweep and SNP vectors”. In: Communications biology 1.1, p. 79. Alberto, FJ, J Derory, C Boury, JM Frigerio, NE Zimmermann, and A Kremer (2013). “Imprints of natural selection along environmental gradients in phenology-related genes of Quercus petraea”. In: Genetics 195.2, pp. 495–512. Alexander, DH, J Novembre, and K Lange (2009a). “Fast model-based estimation of ancestry in unrelated individuals”. In: Genome Research 19.9, pp. 1655–1664. — (2009b). “Fast model-based estimation of ancestry in unrelated individuals”. In: Genome research 19.9, pp. 1655–1664. Allen, E, B Hazen, H Hoch, Y Kwon, G Leinhos, R Staples, M Stumpf, B Terhune, et al. (1991). “Appressorium formation in response to topographical signals by 27 rust species.” In: Phytopathology 81.3, pp. 323–331. Ally, D, K Ritland, and S Otto (2008). “Can clone size serve as a proxy for clone age? An exploration using microsatellite divergence in Populus tremuloides”. In: Molecular Ecology 17.22, pp. 4897–4911. Altizer, S, R Bartel, and BA Han (2011). “Animal migration and infectious disease risk”. In: science 331.6015, pp. 296–302. Anderson, E and E Thompson (2002). “A model-based method for identifying species hybrids using multilocus genetic data”. In: Genetics 160.3, pp. 1217–1229. Anderson, MK (2013). Tending the wild: Native American knowledge and the management of California’s natural resources. University of California Press.

217 Aono, A, J Nagai, G Dickel, R Marinho, P Oliveira, and F Faria (2019). “A stomata classiﬁcation and detection system in microscope images of maize cultivars”. In: bioRxiv, p. 538165. Arens, P, H Coops, J Jansen, and B Vosman (1998). “Molecular genetic analysis of black poplar (Populus nigra L.) along Dutch rivers”. In: Molecular Ecology 7.1, pp. 11–18. Augspurger, CK (1984). “Seedling survival of tropical tree species: interactions of dispersal distance, light-gaps, and pathogens”. In: Ecology 65.6, pp. 1705–1712. Avelino, J, A Romero-Gurdián, HF Cruz-Cuellar, and FA Declerck (2012). “Landscape context and scale dierentially impact coee leaf rust, coee berry borer, and coee root-knot nematodes”. In: Ecological applications 22.2, pp. 584–596. Bagchi, R, RE Gallery, S Gripenberg, SJ Gurr, L Narayan, CE Addis, RP Freckleton, and OT Lewis (2014). “Pathogens and insect herbivores drive rainforest plant diversity and composition”. In: Nature 506.7486, p. 85. Bailey, JK, R Deckert, JA Schweitzer, BJ Rehill, RL Lindroth, C Gehring, and TG Whitham (2005). “Host plant genetics aect hidden ecological players: links among Populus, condensed tannins, and fungal endophyte infection”. In: Canadian Jour- nal of Botany 83.4, pp. 356–361. Bailey, JK, JA Schweitzer, F Ubeda, J Koricheva, CJ LeRoy, MD Madritch, BJ Rehill, RK Bangert, DG Fischer, GJ Allan, et al. (2009). “From genes to ecosystems: a synthesis of the eects of plant genetic factors across levels of organization”. In: Philosophical Transactions of the Royal Society B: Biological Sciences 364.1523, pp. 1607–1616. Barclay, RS, R Bush, AA Baczynski, C France, and S Wing (2016). “Constraining the drivers of variance in leaf ”13C in fossil Ginkgo adiantoides and extant Ginkgo biloba: canopy eect or diagenesis?” In: Geological Society of America Annual Meeting and Exposition Program, p. 241. Barclay, R, J McElwain, D Dilcher, and B Sageman (2007). “The cuticle database: developing an interactive tool for taxonomic and paleoenvironmental study of the fossil cuticle record”. In: Courier-Forschungsinstitut Senckenberg 258, p. 39. Barclay, R and S Wing (2016). “Improving the Ginkgo CO2 barometer: implications for the early Cenozoic atmosphere”. In: Earth and Planetary Science Letters 439, pp. 158–171. Barton, NH and GM Hewitt (1985). “Analysis of hybrid zones”. In: Annual review of Ecology and Systematics 16.1, pp. 113–148. Bates, DM, M Maechler, BM Bolker, and S Walker (2014). “Fitting Linear Mixed- Eects Models Using lme4”. In: Journal of Statistical Software arXiv, p. 1406.5823. Bates, D, M Mächler, B Bolker, and S Walker (2015). “Fitting Linear Mixed-Eects Models Using lme4”. In: Journal of Statistical Software 67.1, pp. 1–48.

218 Bates, D, M Maechler, B Bolker, S Walker, et al. (2014). “lme4: Linear mixed-eects models using Eigen and S4”. In: R package version 1.7, pp. 1–23. Becker, C, J Hagmann, J Müller, D Koenig, O Stegle, K Borgwardt, and D Weigel (2011). “Spontaneous epigenetic variation in the Arabidopsis thaliana methylome”. In: Nature 480.7376, p. 245. Becker, C and D Weigel (2012). “Epigenetic variation: origin and transgenerational inheritance”. In: Current opinion in plant biology 15.5, pp. 562–567. Berardini, TZ, L Reiser, D Li, Y Mezheritsky, R Muller, E Strait, and E Huala (2015). “The Arabidopsis information resource: making and mining the “gold standard” annotated reference plant genome”. In: genesis 53.8, pp. 474–485. Berg, JJ and G Coop (2014). “A population genetic signal of polygenic adaptation”. In: PLoS genetics 10.8, e1004412. Bergmann, DC and FD Sack (2007). “Stomatal development”. In: Annual Review of Plant Biology 58, pp. 163–181. Bernard, E, L Jacob, J Mairal, E Viara, and JP Vert (2015). “A convex formulation for joint RNA isoform detection and quantification from multiple RNA-seq samples”. In: BMC bioinformatics 16.1, p. 262. Bhugra, S, D Mishra, A Anupama, S Chaudhury, B Lall, A Chugh, and V Chinnusamy (2018). “Deep convolutional neural networks based framework for estimation of stomata density and structure from microscopic images”. In: European Conference on Computer Vision. Springer, pp. 412–423. Block, WM, ML Morrison, and J Verner (1990). “Wildlife and oak-woodland inter- dependency”. In: Fremontia 18.3, pp. 72–76. Bodénès, C, E Chancerel, O Gailing, GG Vendramin, F Bagnoli, J Durand, PG Goicoechea, C Soliani, F Villani, C Mattioni, et al. (2012). “Comparative mapping in the Fagaceae and beyond with EST-SSRs”. In: BMC plant biology 12.1, p. 153. Böhlenius, H, T Huang, L Charbonnel-Campaa, AM Brunner, S Jansson, SH Strauss, and O Nilsson (2006). “CO/FT regulatory module controls timing of flowering and seasonal growth cessation in trees”. In: Science 312.5776, pp. 1040–1043. Bomblies, K, J Lempe, P Epple, N Warthmann, C Lanz, JL Dangl, and D Weigel (2007). “Autoimmune response as a mechanism for a Dobzhansky-Muller-type incompatibility syndrome in plants”. In: PLoS biology 5.9, e236. Bono, JM, EC Olesnicky, and LM Matzkin (2015). “Connecting genotypes, phenotypes and fitness: harnessing the power of CRISPR/Cas9 genome editing”. In: Molecular ecology 24.15, pp. 3810–3822. Bragg, JG, MA Supple, RL Andrew, and JO Borevitz (2015). “Genomic variation across landscapes: insights and applications”. In: New Phytologist 207.4, pp. 953– 967. Breen, AL (2014). “Balsam poplar (Populus balsamifera L.) communities on the Arctic Slope of Alaska”. In: Phytocoenologia 44.1-2, pp. 1–24.

219 Breen, AL, DF Murray, and MS Olson (2012). “Genetic consequences of glacial survival: the late Quaternary history of balsam poplar (Populus balsamifera L.) in North America”. In: Journal of Biogeography 39.5, pp. 918–928. Bresadola, L, C Caseys, S Castiglione, CA Buerkle, D Wegmann, and C Lexer (2018). “Admixture mapping in interspecific Populus hybrids identifies classes of genomic architectures for phytochemical, morphological and growth traits”. In: BioRxiv, p. 497685. Brooks, ME, K Kristensen, KJ van Benthem, A Magnusson, CW Berg, A Nielsen, HJ Skaug, M Maechler, and BM Bolker (2017). “Modeling Zero-Inflated Count Data With glmmTMB”. In: bioRxiv. eprint: https://www.biorxiv.org/content/ early/2017/05/01/132753.full.pdf. Brown, TB, R Cheng, XR Sirault, T Rungrat, KD Murray, M Trtilek, RT Furbank, M Badger, BJ Pogson, and JO Borevitz (2014). “TraitCapture: genomic and environment modelling of plant phenomic data”. In: Current opinion in plant biology 18, pp. 73–79. Browning, BL and SR Browning (2016). “Genotype imputation with millions of reference samples”. In: The American Journal of Human Genetics 98.1, pp. 116– 126. Browning, BL, Y Zhou, and SR Browning (2018). “A one-penny imputed genome from next-generation reference panels”. In: The American Journal of Human Genetics 103.3, pp. 338–348. Bruns, EL, J Antonovics, and M Hood (2018). “Is there a disease-free halo at species range limits? The co-distribution of anther-smut disease and its host species”. In: Journal of Ecology. Buer, CS, N Imin, and MA Djordjevic (2010). “Flavonoids: new roles for old molecules”. In: Journal of integrative plant biology 52.1, pp. 98–111. Buerkle, CA and C Lexer (2008). “Admixture as the basis for genetic mapping”. In: Trends in ecology & evolution 23.12, pp. 686–694. Burns, RM and BH Honkala (1990). Silvics of north America. Vol. 2. United States Department of Agriculture Washington, DC. Busby, PE, N Zimmerman, DJ Weston, SS Jawdy, J Houbraken, and G Newcombe (2013). “Leaf endophytes and Populus genotype aect severity of damage from the necrotrophic leaf pathogen, Drepanopeziza populi”. In: Ecosphere 4.10, pp. 1–12. Bylesjö, M, V Segura, RY Soolanayakanahally, AM Rae, J Trygg, P Gustafsson, S Jansson, and NR Street (2008). “LAMINA: a tool for rapid quantification of leaf size and shape parameters”. In: BMC Plant Biology 8.1, p. 82. Cai, Q, L Qiao, M Wang, B He, FM Lin, J Palmquist, SD Huang, and H Jin (2018). “Plants send small RNAs in extracellular vesicles to fungal pathogen to silence virulence genes”. In: Science 360.6393, pp. 1126–1129.

220 ∆aliÊ, I, F Bussotti, PJ Martínez-García, and DB Neale (2016). “Recent landscape genomics studies in forest trees—what can we believe?” In: Tree genetics & genomes 12.1, p. 3. Callahan, CM, CA Rowe, RJ Ryel, JD Shaw, MD Madritch, and KE Mock (2013). “Continental-scale assessment of genetic diversity and population structure in quaking aspen (Populus tremuloides)”. In: Journal of biogeography 40.9, pp. 1780– 1791. Casasoli, M, J Derory, C Morera-Dutrey, O Brendel, I Porth, JM Guehl, F Villani, and A Kremer (2006). “Comparison of quantitative trait loci for adaptive traits between oak and chestnut based on an expressed sequence tag consensus map”. In: Genetics 172.1, pp. 533–546. Chae, E, K Bomblies, ST Kim, D Karelina, M Zaidem, S Ossowski, C Martín-Pizarro, RA Laitinen, BA Rowan, H Tenenboim, et al. (2014). “Species-wide genetic incompatibility analysis identifies immune genes as hot spots of deleterious epistasis”. In: Cell 159.6, pp. 1341–1351. Chandran, D, J Rickert, Y Huang, MA Steinwand, SK Marr, and MC Wildermuth (2014). “Atypical E2F transcriptional repressor DEL1 acts at the intersection of plant growth and immunity by controlling the hormone salicylic acid”. In: Cell host & microbe 15.4, pp. 506–513. Chase, MA, S Stankowski, and MA Streisfeld (2017). “Genomewide variation provides insight into evolutionary relationships in a monkeyflower species complex (Mimulus sect. Diplacus)”. In: American journal of botany 104.10, pp. 1510–1521. Chhatre, VE (2018). Distruct 2.3: A modified admixture plotting script. Chhatre, VE, LM Evans, SP DiFazio, and SR Keller (2018). “Adaptive introgression and maintenance of a trispecies hybrid complex in range-edge populations of Populus”. In: Molecular ecology 27.23, pp. 4820–4838. Christe, C, KN Stölting, L Bresadola, B Fussi, B Heinze, D Wegmann, and C Lexer (2016). “Selection against recombinant hybrids maintains reproductive isolation in hybridizing Populus species despite F1 fertility and recurrent gene flow”. In: Molecular ecology 25.11, pp. 2482–2498. Clark, RM, TN Wagler, P Quijada, and J Doebley (2006). “A distant upstream enhancer at the maize domestication gene tb1 has pleiotropic eects on plant and inflorescent architecture”. In: Nature genetics 38.5, p. 594. Cokus, SJ, PF Gugger, and VL Sork (2015). “Evolutionary insights from de novo transcriptome assembly and SNP discovery in California white oaks”. In: BMC genomics 16.1, p. 552. Collaboration, OS et al. (2015). “Estimating the reproducibility of psychological science”. In: Science 349.6251, aac4716.

221 Conway, JR, A Lex, and N Gehlenborg (2017). “UpSetR: an R package for the vi- sualization of intersecting sets and their properties”. In: Bioinformatics 33.18, pp. 2938–2940. Cutter, AD (2012). “The polymorphic prelude to Bateson–Dobzhansky–Muller incompatibilities”. In: Trends in ecology & evolution 27.4, pp. 209–218. Dahlgren, RA, MJ SINGER, and X Huang (1997). “Oak tree and grazing impacts on soil properties and nutrients in a California oak woodland”. In: Biogeochemistry 39.1, pp. 45–64. Danecek, P, A Auton, G Abecasis, CA Albers, E Banks, MA DePristo, RE Handsaker, G Lunter, GT Marth, ST Sherry, et al. (2011a). “The variant call format and VCFtools”. In: Bioinformatics 27.15, pp. 2156–2158. — (2011b). “The variant call format and VCFtools”. In: Bioinformatics 27.15, pp. 2156– 2158. Dangl, JL and JD Jones (2001). “Plant pathogens and integrated defence responses to infection”. In: nature 411.6839, p. 826. De Carvalho, D, PK Ingvarsson, J Joseph, L Suter, C Sedivy, D MACAYA-SANZ, J Cottrell, B Heinze, I Schanzer, and C Lexer (2010). “Admixture facilitates adaptation from standing variation in the European aspen (Populus tremula L.), a widespread forest tree”. In: Molecular Ecology 19.8, pp. 1638–1650. Delﬁno-Mix, A, JW Wright, PF Gugger, C Liang, and VL Sork (2015). “Establishing a range-wide provenance test in valley oak (Quercus lobata Née) at two Califor- nia sites”. In: Gen. Tech. Rep. PSW-GTR-251. Berkeley, CA: US Department of Agriculture, Forest Service, Paciﬁc Southwest Research Station: 413-424 251, pp. 413–424. Deng, J, W Dong, R Socher, LJ Li, K Li, and L Fei-Fei (2009). “Imagenet: A large- scale hierarchical image database”. In: 2009 IEEE conference on computer vision and pattern recognition. Ieee, pp. 248–255. Derory, J, C Scotti-Saintagne, E Bertocchi, L Le Dantec, N Graignic, A Jaures, M Casasoli, E Chancerel, C Bodénès, F Alberto, et al. (2010). “Contrasting relationships between the diversity of candidate genes and variation of bud burst in natural and segregating populations of European oaks”. In: Heredity 104.5, p. 438. Derory, J, P Léger, V Garcia, J Schaeer, MT Hauser, F Salin, C Luschnig, C Plomion, J Glössl, and A Kremer (2006). “Transcriptome analysis of bud burst in sessile oak (Quercus petraea)”. In: New Phytologist 170.4, pp. 723–738. Di Lonardo, S, M Capuana, M Arnetoli, R Gabbrielli, and C Gonnelli (2011). “Ex- ploring the metal phytoremediation potential of three Populus alba L. clones using an in vitro screening”. In: Environmental Science and Pollution Research 18.1, pp. 82–90.

222 Diener, AC, RA Gaxiola, and GR Fink (2001). “Arabidopsis ALF5, a multidrug eux transporter gene family member, confers resistance to toxins”. In: The Plant Cell 13.7, pp. 1625–1638. DiFazio, SP, S Leonardi, GT Slavov, SL Garman, WT Adams, and SH Strauss (2012). “Gene flow and simulation of transgene dispersal from hybrid poplar plantations”. In: New Phytologist 193.4, pp. 903–915. DiFazio, SP, GT Slavov, and CP Joshi (2011). “Populus: a premier pioneer system for plant genomics”. In: Enfield, NH: Science Publishers, pp. 1–28. Dillon, S, R McEvoy, DS Baldwin, S Southerton, C Campbell, Y Parsons, and GN Rees (2015). “Genetic diversity of Eucalyptus camaldulensis Dehnh. following population decline in response to drought and altered hydrological regime”. In: Austral Ecology 40.5, pp. 558–572. Dittberner, H, A Korte, T Mettler-Altmann, A Weber, G Monroe, and J de Meaux (2018). “Natural variation in stomata size contributes to the local adaptation of water-use eciency in Arabidopsis thaliana”. In: bioRxiv, p. 253021. Dodds, PN and JP Rathjen (2010). “Plant immunity: towards an integrated view of plant–pathogen interactions”. In: Nature Reviews Genetics 11.8, p. 539. Dosskey, M, R Schultz, and T Isenhart (1997). “Riparian Buers for Agricultural Land in Agroforestry Notes”. In: National Agroforestry Center, USDA Forest Service, Rocky Mountain Station USDA Natural Resources Conservation Service, East Campus-UNL, Lincoln, Nebraska. Dowkiw, A and C Bastien (2004). “Characterization of two major genetic factors controlling quantitative resistance to Melampsora larici-populina leaf rust in hybrid poplars: strain specificity, field expression, combined eects, and relationship with a defeated qualitative resistance gene”. In: Phytopathology 94.12, pp. 1358–1367. Du, Q, B Xu, C Gong, X Yang, W Pan, J Tian, B Li, and D Zhang (2014). “Variation in growth, leaf, and wood property traits of Chinese white poplar (Populus tomentosa), a major industrial tree species in Northern China”. In: Canadian journal of forest research 44.4, pp. 326–339. Duarte, KT, MAG de Carvalho, and PS Martins (2017). “Segmenting high-quality digital images of stomata using the wavelet spot detection and the watershed transform.” In: VISIGRAPP (4: VISAPP). Setubal, Portugal: Science and Tech- nology Publications, Lda, pp. 540–547. Dunn, TM, DV Lynch, LV Michaelson, and JA Napier (2004). “A post-genomic approach to understanding sphingolipid metabolism in Arabidopsis thaliana”. In: Annals of Botany 93.5, pp. 483–497. Durand, J, C Bodénès, E Chancerel, JM Frigerio, G Vendramin, F Sebastiani, A Buonamici, O Gailing, HP Koelewijn, F Villani, et al. (2010). “A fast and cost- eective approach to develop and map EST-SSR markers: oak as a case study”. In: BMC genomics 11.1, p. 570.

223 Eckenwalder, JE (1996). “Systematics and evolution of Populus”. In: Biology of Pop- ulus and its implications for management and conservation. Ed. by R Stettler, T Bradshaw, P Heilman, and T Hinckley. Ottawa: NRC Research Press. Chap. 1, pp. 7–32. Eisele, JF, F Fäßler, PF Bürgel, and C Chaban (Oct. 2016). “A rapid and simple method for microscopy-based stomata analyses”. In: PLOS ONE 11.10, pp. 1–13. Ellstrand, NC and KA Schierenbeck (2000). “Hybridization as a stimulus for the evolution of invasiveness in plants?” In: Proceedings of the National Academy of Sciences 97.13, pp. 7043–7050. Elmore, A, C Stylinski, and K Pradhan (2016). “Synergistic use of citizen science and remote sensing for continental-scale measurements of forest tree phenology”. In: Remote Sensing 8.6, p. 502. Elshire, RJ, JC Glaubitz, Q Sun, JA Poland, K Kawamoto, ES Buckler, and SE Mitchell (2011a). “A robust, simple genotyping-by-sequencing (GBS) approach for high diversity species”. In: PloS one 6.5, e19379. — (2011b). “A robust, simple genotyping-by-sequencing (GBS) approach for high diversity species”. In: PloS one 6.5, e19379. Espinoza, O, U Buehlmann, M Bumgardner, and B Smith (2011). “Assessing changes in the US hardwood sawmill industry with a focus on markets and distribution”. In: BioResources 6.3, pp. 2676–2689. Evans, LM, GJ Allan, SP DiFazio, GT Slavov, JA Wilder, KD Floate, SB Rood, and TG Whitham (2015). “Geographical barriers and climate influence demographic history in narrowleaf cottonwoods”. In: Heredity 114.4, p. 387. Evans, LM, GJ Allan, SM Shuster, SA Woolbright, and TG Whitham (2008). “Tree hybridization and genotypic variation drive cryptic speciation of a specialist mite herbivore”. In: Evolution: International Journal of Organic Evolution 62.12, pp. 3027– 3040. Evans, LM, GT Slavov, E Rodgers-Melnick, J Martin, P Ranjan, W Muchero, AM Brunner, W Schackwitz, L Gunter, JG Chen, et al. (2014). “Population genomics of Populus trichocarpa identifies signatures of selection and adaptive trait associations”. In: Nature genetics 46.10, p. 1089. Facchini, PJ, J Hagel, and KG Zulak (2002). “Hydroxycinnamic acid amide metabolism: physiology and biochemistry”. In: Canadian Journal of Botany 80.6, pp. 577–589. Fan, D, T Liu, C Li, B Jiao, S Li, Y Hou, and K Luo (2015). “Ecient CRISPR/Cas9- mediated targeted mutagenesis in Populus in the first generation”. In: Scientific reports 5, p. 12217. Fang, GC, BP Blackmon, ME Staton, CD Nelson, TL Kubisiak, BA Olukolu, D Henry, T Zhebentyayeva, CA Saski, CH Cheng, et al. (2013). “A physical map of the Chinese chestnut (Castanea mollissima) genome and its integration with the genetic map”. In: Tree genetics & genomes 9.2, pp. 525–537.

224 Farquhar, GD, MH O’Leary, and JA Berry (1982). “On the relationship between carbon isotope discrimination and the intercellular carbon dioxide concentration in leaves”. In: Functional Plant Biology 9.2, pp. 121–137. Feau, N, DL Joly, and RC Hamelin (2007). “Poplar leaf rusts: model pathogens for a model tree”. In: Botany 85.12, pp. 1127–1135. Fetter, KC (2019). “Natural selection for disease resistance in hybrid poplars targets stomata patterning traits and genes”. PhD thesis. University of Vermont. Fetter, KC, S Eberhardt, RS Barclay, S Wing, and SR Keller (2019). “Stomata- Counter: a neural network for automatic stomata identification and counting”. In: New Phytologist 223.3, pp. 1671–1681. Fetter, KC and A Weakley (2019). “Reduced Gene Flow from Mainland Populations of Liriodendron tulipifera into the Florida Peninsula Promotes Diversification”. In: International Journal of Plant Sciences 180.3, pp. 253–269. Fetter, KC (2014). “Migration, adaptation, and speciation-A post-glacial history of the population structure, phylogeography, and biodiversity of Liriodendron tulipifera l.(Magnoliaceae)”. PhD thesis. The University of North Carolina at Chapel Hill. Feurdean, A, SA Bhagwat, KJ Willis, HJB Birks, H Lischke, and T Hickler (2013). “Tree migration-rates: narrowing the gap between inferred post-glacial rates and projected rates”. In: PLoS One 8.8, e71797. Fischer, R, D Rees, K Sayre, Z Lu, A Condon, and L Saavedra (1998). “Wheat yield progress associated with higher stomatal conductance and photosynthetic rate, and cooler canopies”. In: Crop science 38.6, pp. 1467–1475. Fishman, L and JH Willis (2006). “A cytonuclear incompatibility causes anther steril- ity in Mimulus hybrids”. In: Evolution 60.7, pp. 1372–1381. Fitzpatrick, MC and SR Keller (2015). “Ecological genomics meets community-level modelling of biodiversity: Mapping the genomic landscape of current and future environmental adaptation”. In: Ecology letters 18.1, pp. 1–16. Floate, KD (2004). “Extent and patterns of hybridization among the three species of Populus that constitute the riparian forest of southern Alberta, Canada”. In: Canadian Journal of Botany 82.2, pp. 253–264. Floate, KD, J Godbout, MK Lau, N Isabel, and TG Whitham (2016). “Plant– herbivore interactions in a trispecific hybrid swarm of Populus: assessing support for hypotheses of hybrid bridges, evolutionary novelty and genetic similarity”. In: New Phytologist 209.2, pp. 832–844. Foll, M and O Gaggiotti (2008). “A genome-scan method to identify selected loci appropriate for both dominant and codominant markers: a Bayesian perspective”. In: Genetics 180.2, pp. 977–993.

225 Foster, AJ, G Pelletier, P Tanguay, and A Séguin (2015). “Transcriptome Analysis of Poplar during Leaf Spot Infection with Sphaerulina spp.” In: PLoS One 10.9, e0138162. Fournier, DA, HJ Skaug, J Ancheta, J Ianelli, A Magnusson, MN Maunder, A Nielsen, and J Sibert (2012). “AD Model Builder: using automatic dierentiation for statistical inference of highly parameterized complex nonlinear models”. In: Optim. Methods Softw. 27, pp. 233–249. Fowells, HA et al. (1965). “Silvics of forest trees of the United States.” In: Agric. Handb. US Dep. Agric. 271. Fralish, JS (2004). “The keystone role of oak and hickory in the central hardwood forest”. In: Gen. Tech. Rep. SRS-73. Asheville, NC: US Department of Agriculture, Forest Service, Southern Research Station. pp. 78-87. Frichot, E, SD Schoville, G Bouchard, and O François (2013). “Testing for associations between loci and environmental gradients using latent factor mixed models”. In: Molecular biology and evolution 30.7, pp. 1687–1699. Gailing, O (2014). “Strategies to identify adaptive genes in hybridizing trees like oaks and poplars”. In: Notulae Botanicae Horti Agrobotanici Cluj-Napoca 42.2, pp. 299–309. Gailing, O and AL Curtu (2014). “Interspecific gene flow and maintenance of species integrity in oaks”. In: Annals of Forest Research 57.1, pp. 5–18. Gelasca, ED, J Byun, B Obara, and BS Manjunath (2008). “Evaluation and bench- mark for biological image segmentation”. In: Proceedings - International Confer- ence on Image Processing, ICIP, pp. 1816–1819. Gentry, AH (1988). “Tree species richness of upper Amazonian forests”. In: Proceed- ings of the National Academy of Sciences 85.1, pp. 156–159. Geraldes, A, S Difazio, G Slavov, P Ranjan, W Muchero, J Hannemann, L Gunter, A Wymore, C Grassa, N Farzaneh, et al. (2013). “A 34K SNP genotyping array for Populus trichocarpa: design, application to the study of natural populations and transferability to other Populus species”. In: Molecular Ecology Resources 13.2, pp. 306–323. Geraldes, A, N Farzaneh, CJ Grassa, AD McKown, RD Guy, SD Mansfield, CJ Dou- glas, and QC Cronk (2014). “Landscape genomics of Populus trichocarpa: the role of hybridization, limited gene flow, and natural selection in shaping patterns of population structure”. In: Evolution 68.11, pp. 3260–3280. Geraldes, A, CA Hefer, A Capron, N Kolosova, F Martinez-Nuñez, RY Soolanayakana- hally, B Stanton, RD Guy, SD Mansfield, CJ Douglas, et al. (2015). “Recent Y chromosome divergence despite ancient origin of dioecy in poplars (Populus)”. In: Molecular ecology 24.13, pp. 3243–3256.

226 Glaubitz, JC, TM Casstevens, F Lu, J Harriman, RJ Elshire, Q Sun, and ES Buck- ler (2014a). “TASSEL-GBS: A high capacity genotyping by sequencing analysis pipeline”. In: PLoS ONE 9.2. arXiv: NIHMS150003. — (2014b). “TASSEL-GBS: a high capacity genotyping by sequencing analysis pipeline”. In: PloS one 9.2, e90346. Goldberg, TL, EC Grant, KR Inendino, TW Kassler, JE Claussen, and DP Philipp (2005). “Increased infectious disease susceptibility resulting from outbreeding depression”. In: Conservation Biology 19.2, pp. 455–462. Gonzales-Vigil, E, CA Hefer, ME von Loessl, J La Mantia, and SD Mansfield (2017). “Exploiting natural variation to uncover an alkene biosynthetic enzyme in poplar”. In: The Plant Cell,tpc–00338. Gonzalez-Martinez, SC, KV Krutovsky, and DB Neale (2006). “Forest-tree population genomics and adaptive evolution”. In: New Phytologist 170.2, pp. 227–238. Gou, M, Z Shi, Y Zhu, Z Bao, G Wang, and J Hua (2012). “The F-box protein CPR1/CPR30 negatively regulates R protein SNC1 accumulation”. In: The Plant Journal 69.3, pp. 411–420. Goulet, BE, F Roda, and R Hopkins (2017). “Hybridization in plants: old ideas, new techniques”. In: Plant physiology 173.1, pp. 65–78. Graham, EA, EC Riordan, EM Yuen, D Estrin, and PW Rundel (2010). “Public internet-connected cameras used as a cross-continental ground-based plant phenology monitoring system”. In: Global Change Biology 16.11, pp. 3014–3023. Grivet, D, MF Deguilloux, RJ Petit, and VL Sork (2006). “Contrasting patterns of historical colonization in white oaks (Quercus spp.) in California and Europe”. In: Molecular Ecology 15.13, pp. 4085–4093. Gugger, PF, SJ Cokus, and VL Sork (2016). “Association of transcriptome-wide sequence variation with climate gradients in valley oak (Quercus lobata)”. In: Tree Genetics & Genomes 12.2, p. 15. Gugger, PF, S Fitz-Gibbon, M Pellegrini, and VL Sork (2016). “Species-wide patterns of DNA methylation variation in Quercus lobata and their association with climate gradients”. In: Molecular Ecology 25.8, pp. 1665–1680. Gugger, PF, JM Peñaloza-Ramírez, JW Wright, and VL Sork (2017). “Whole-transcriptome response to water stress in a California endemic oak, Quercus lobata”. In: Tree physiology 37.5, pp. 632–644. Günther, T and G Coop (2013). “Robust identification of local adaptation from allele frequencies”. In: Genetics 195.1, pp. 205–220. Gutensohn, M, A Klempien, Y Kaminaga, DA Nagegowda, F Negre-Zakharov, JH Huh, H Luo, R Weizbauer, T Mengiste, D Tholl, et al. (2011). “Role of aromatic aldehyde synthase in wounding/herbivory response and flower scent production in dierent Arabidopsis ecotypes”. In: The Plant Journal 66.4, pp. 591–602.

227 Hacquard, S and CW Schadt (2015). “Towards a holistic understanding of the beneficial interactions across the Populus microbiome”. In: New Phytologist 205.4, pp. 1424–1430. Hadfield, JD (2010). “MCMC Methods for Multi-Response Generalized Linear Mixed Models: The MCMCglmm R Package”. In: Journal of Statistical Software 33.2, pp. 1–22. Haldane, J (1948). “The theory of a cline”. In: Journal of genetics 48.3, pp. 277–284. Hall, D, V Luquez, VM Garcia, KR St Onge, S Jansson, and PK Ingvarsson (2007). “Adaptive population dierentiation in phenology across a latitudinal gradient in European aspen (Populus tremula, L.): a comparison of neutral markers, candidate genes and phenotypic traits”. In: Evolution 61.12, pp. 2849–2860. Hall, D, XF MA, and PK Ingvarsson (2011). “Adaptive evolution of the Populus tremula photoperiod pathway”. In: Molecular ecology 20.7, pp. 1463–1474. Hamilton, MB (1999). “Tropical tree gene flow and seed dispersal”. In: Nature 401.6749, p. 129. Hamzeh, M and S Dayanandan (2004). “Phylogeny of Populus (Salicaceae) based on nucleotide sequences of chloroplast TRNT-TRNF region and nuclear rDNA”. In: American journal of botany 91.9, pp. 1398–1408. Hancock, AM, B Brachi, N Faure, MW Horton, LB Jarymowycz, FG Sperone, C Toomajian, F Roux, and J Bergelson (2011). “Adaptation to climate across the Arabidopsis thaliana genome”. In: Science 334.6052, pp. 83–86. Hara, M, Y Yatsuzuka, K Tabata, and T Kuboi (2010). “Exogenously applied isoth- iocyanates enhance glutathione S-transferase expression in Arabidopsis but act as herbicides at higher concentrations”. In: Journal of plant physiology 167.8, pp. 643–649. Hartfield, M (2016). “Evolutionary genetic consequences of facultative sex and out- crossing”. In: Journal of evolutionary biology 29.1, pp. 5–22. He, K, X Zhang, S Ren, and J Sun (2016). “Deep residual learning for image recognition”. In: Proceedings of the IEEE conference on computer vision and pattern recognition. Los Alamitos, California: IEEE Computer Society, pp. 770–778. Heiser Jr, CB (1951). “Hybridization in the annual sunflowers: Helianthus annuus ◊ H. debilis var. cucumerifolius”. In: Evolution, pp. 42–51. Herman, DJ, LJ Halverson, and MK Firestone (2003). “Nitrogen dynamics in an annual grassland: oak canopy, climate, and microbial population eects”. In: Eco- logical Applications 13.3, pp. 593–604. Hetherington, AM and FI Woodward (2003). “The role of stomata in sensing and driving environmental change”. In: Nature 424.6951, p. 901. Hewitt, GM (1999). “Post-glacial re-colonization of European biota”. In: Biological journal of the Linnean Society 68.1-2, pp. 87–112.

228 Higaki, T, N Kutsuna, and S Hasezawa (2014). “CARTA-based semi-automatic detection of stomatal regions on an Arabidopsis cotyledon surface”. In: Plant Mor- phology 26.1, pp. 9–12. Hocking, B, SJ Conn, M Manohar, B Xu, A Athman, MA Stancombe, AR Webb, KD Hirschi, and M Gilliham (2017). “Heterodimerization of Arabidopsis calcium/pro- ton exchangers contributes to regulation of guard cell dynamics and plant defense responses”. In: Journal of experimental botany 68.15, pp. 4171–4183. Holeski, LM, MS Zinkgraf, JJ Couture, TG Whitham, and RL Lindroth (2013). “Transgenerational eects of herbivory in a group of long-lived tree species: maternal damage reduces ospring allocation to resistance traits, but not growth”. In: Journal of Ecology 101.4, pp. 1062–1073. Holliday, JA, L Zhou, R Bawa, M Zhang, and RW Oubida (2016). “Evidence for extensive parallelism but divergent genomic architecture of adaptation along altitudinal and latitudinal gradients in Populus trichocarpa”. In: New Phytologist 209.3, pp. 1240–1251. Homolova, L, Z Malenovsk˝, JG Clevers, G García-Santos, and ME Schaepman (2013). “Review of optical-based remote sensing for plant trait mapping”. In: Ecological Complexity 15, pp. 1–16. Huang, YN, H Zhang, S Rogers, M Coggeshall, and K Woeste (2015). “White oak growth after 23 years in a three-site provenance/progeny trial on a latitudinal gradient in Indiana”. In: Forest Science. Hübner, S, N Bercovich, M Todesco, JR Mandel, J Odenheimer, E Ziegler, JS Lee, GJ Baute, GL Owens, CJ Grassa, et al. (2019). “Sunflower pan-genome analysis shows that hybridization altered gene content and disease resistance”. In: Nature plants 5.1, p. 54. Hudson, ME (2008). “Sequencing breakthroughs for genomic ecology and evolutionary biology”. In: Molecular ecology resources 8.1, pp. 3–17. Hut, G (1987). “Consultants’ group meeting on stable isotope reference samples for geochemical and hydrological investigations, Report to the Director General.” In: International Atomic Energy Agency. Vienna. Iijima, T, R Kajitani, S Komata, CP Lin, T Sota, T Itoh, and H Fujiwara (2018). “Parallel evolution of Batesian mimicry supergene in two Papilio butterflies, P. polytes and P. memnon”. In: Science advances 4.4, eaao5416. Imbert, E and F Lefévre (2003). “Dispersal and gene flow of Populus nigra (Salicaceae) along a dynamic river system”. In: Journal of Ecology 91.3, pp. 447–456. Ingvarsson, PK (2008). “Multilocus patterns of nucleotide polymorphism and the demographic history of Populus tremula”. In: Genetics 180.1, pp. 329–340. Ingvarsson, PK, MV García, D Hall, V Luquez, and S Jansson (2006). “Clinal variation in phyB2, a candidate gene for day-length-induced growth cessation and

229 bud set, across a latitudinal gradient in European aspen (Populus tremula)”. In: Genetics 172.3, pp. 1845–1853. Ingvarsson, PK, MV Garcia, V Luquez, D Hall, and S Jansson (2008). “Nucleotide polymorphism and phenotypic associations within and around the phytochrome B2 locus in European aspen (Populus tremula, Salicaceae)”. In: Genetics 178.4, pp. 2217–2226. Ingvarsson, PK and NR Street (2011). “Association genetics of complex traits in plants”. In: New Phytologist 189.4, pp. 909–922. Isabel, N, M Lamothe, and SL Thompson (2013). “A second-generation diagnostic single nucleotide polymorphism (SNP)-based assay, optimized to distinguish among eight poplar (Populus L.) species and their early hybrids”. In: Tree genetics & genomes 9.2, pp. 621–626. Ispolatov, I and M Doebeli (2009). “Speciation due to hybrid necrosis in plant- pathogen models”. In: Evolution: International Journal of Organic Evolution 63.12, pp. 3076–3084. IvetiÊ, V, S StjepanoviÊ, J DevetakoviÊ, D StankoviÊ, and M äkoriÊ(2014). “Re- lationships between leaf traits and morphological attributes in one-year bareroot Fraxinus angustifolia Vahl. seedlings”. In: Annals of Forest Research 57.2, pp. 197– 203. Järvinen, P, J Lemmetyinen, O Savolainen, and T Sopanen (2003). “DNA sequence variation in BpMADS2 gene in two populations of Betula pendula”. In: Molecular ecology 12.2, pp. 369–384. Jayakody, H, S Liu, M Whitty, and P Petrie (2017). “Microscope image based fully automated stomata detection and pore measurement method for grapevines”. In: Plant methods 13.1, p. 94. Jia, Y, E Shelhamer, J Donahue, S Karayev, J Long, R Girshick, S Guadarrama, and T Darrell (2014). “Cae: convolutional architecture for fast feature embedding”. In: arXiv preprint arXiv:1408.5093. Jian, S, C Zhao, and Y Zhao (2011). “Based on remote sensing processing technol- ogy estimating leaves stomatal density of Populus euphratica”. In: 2011 IEEE International Geoscience and Remote Sensing Symposium. IEEE, pp. 547–550. Johnson, AC and MJ Cipolla (2017). “Altered hippocampal arteriole structure and function in a rat model of preeclampsia: potential role in impaired seizure-induced hyperemia”. In: Journal of Cerebral Blood Flow & Metabolism 37.8, pp. 2857–2869. Jones, JD and JL Dangl (2006). “The plant immune system”. In: nature 444.7117, p. 323. Keller, SR, N Levsen, PK Ingvarsson, MS Olson, and P Tin (2011). “Local selection across a latitudinal gradient shapes nucleotide diversity in balsam poplar, Populus balsamifera L”. In: Genetics 188.4, pp. 941–952.

230 Keller, SR, N Levsen, MS Olson, and P Tin (2012). “Local adaptation in the flowering-time gene network of balsam poplar, Populus balsamifera L.” In: Molec- ular biology and evolution 29.10, pp. 3143–3152. Keller, SR, MS Olson, S Silim, W Schroeder, and P Tin (2010). “Genomic diversity, population structure, and migration following rapid range expansion in the Balsam Poplar, Populus balsamifera”. In: Molecular Ecology 19.6, pp. 1212–1226. Kempes, CP, GB West, K Crowell, and M Girvan (2011). “Predicting maximum tree heights and other traits from allometric scaling and resource limitations”. In: PLoS One 6.6, e20551. Kim, TH, M Böhmer, H Hu, N Nishimura, and JI Schroeder (2010). “Guard cell signal transduction network: advances in understanding abscisic acid, CO2, and Ca2+ signaling”. In: Annual review of plant biology 61, pp. 561–591. Kim, YH, MD Kim, YI Choi, SC Park, DJ Yun, EW Noh, HS Lee, and SS Kwak (2011). “Transgenic poplar expressing Arabidopsis NDPK2 enhances growth as well as oxidative stress tolerance”. In: Plant biotechnology journal 9.3, pp. 334– 347. Kleinknecht, GJ, HE Lintz, A Kruger, JJ Niemeier, MJ Salino-Hugg, CK Thomas, CJ Still, and Y Kim (2015). “Introducing a sensor to measure budburst and its environmental drivers”. In: Frontiers in Plant Science 6, p. 123. Kliebenstein, DJ (2016). “False idolatry of the mythical growth versus immunity tradeoin molecular systems plant pathology”. In: Physiological and Molecular Plant Pathology 95, pp. 55–59. Koho, S, E Fazeli, JE Eriksson, and PE Hänninen (2016). “Image quality ranking method for microscopy”. In: Scientific reports 6, p. 28962. Kremer, A, AG Abbott, JE Carlson, PS Manos, C Plomion, P Sisco, ME Staton, S Ueno, and GG Vendramin (2012). “Genomics of Fagaceae”. In: Tree Genetics & Genomes 8.3, pp. 583–610. Krizhevsky, A, I Sutskever, and GE Hinton (2012). “Imagenet classification with deep convolutional neural networks”. In: Advances in neural information processing systems, pp. 1097–1105. Kroeger, T, F Casey, P Alvarez, M Cheatum, and L Tavassoi (2010). An Economic Analysis of the Benefits of Habitat Conservation on California Rangelands. Wash- ington, D.C.: Conservation Economics Program, Defenders of Wildlife. Krutovsky, KV and DB Neale (2005). “Nucleotide diversity and linkage disequilibrium in cold-hardiness-and wood quality-related candidate genes in Douglas fir”. In: Genetics 171.4, pp. 2029–2041. La Mantia, J, J Klápöt, YA El-Kassaby, S Azam, RD Guy, CJ Douglas, SD Mans- field, and R Hamelin (2013). “Association analysis identifies Melampsora columbiana ◊ poplar leaf rust resistance SNPs”. In: PLoS One 8.11, e78423.

231 Laga, H, F Shahinnia, and D Fleury (2014). “Image-based plant stornata phenotyping”. In: 2014 13th International Conference on Control Automation Robotics and Vision, ICARCV 2014, pp. 217–222. Lamit, L, T Wojtowicz, Z Kovacs, S Wooley, M Zinkgraf, TG Whitham, R Lin- droth, and CA Gehring (2011). “Hybridization among foundation tree species inﬂuences the structure of associated understory plant communities”. In: Botany 89.3, pp. 165–174. Langlet, O (1971). “Two hundred years genecology”. In: Taxon, pp. 653–721. Larcher, W (2003). Physiological plant ecology: ecophysiology and stress physiology of functional groups. Springer Science & Business Media. Lawrence, M, W Huber, H Pages, P Aboyoun, M Carlson, R Gentleman, MT Morgan, and VJ Carey (2013). “Software for computing and annotating genomic ranges”. In: PLoS computational biology 9.8, e1003118. Le Corre, V, N Machon, RJ Petit, and A Kremer (1997). “Colonization with long- distance seed dispersal and genetic structure of maternally inherited genes in forest trees: a simulation study”. In: Genetics Research 69.2, pp. 117–125. LeBoldus, JM, N Isabel, KD Floate, P Blenis, and BR Thomas (2013). “Testing the ‘hybrid susceptibility’and ‘phenological sink’ hypotheses using the P. balsamifera– P. deltoides hybrid zone and Septoria leaf spot [Septoria musiva]”. In: PloS one 8.12, e84437. Lecourieux, D, R Ranjeva, and A Pugin (2006). “Calcium in plant defence-signalling pathways”. In: New Phytologist 171.2, pp. 249–269. LeCun, Y, Y Bengio, and G Hinton (2015). “Deep learning”. In: Nature 521.7553, p. 436. Lee, JH, ER Daugharthy, J Scheiman, R Kalhor, TC Ferrante, R Terry, BM Tur- czyk, JL Yang, HS Lee, J Aach, et al. (2015). “Fluorescent in situ sequencing (FISSEQ) of RNA for gene expression proﬁling in intact cells and tissues”. In: Nature protocols 10.3, p. 442. Lee, J, S Donghwan, S Won-yong, H Inhwan, and L Youngsook (2004). “Arabidopsis metallothioneins 2a and 3 enhance resistance to cadmium when expressed in Vicia faba guard cells”. In: Plant Molecular Biology 54.6, pp. 805–815. Lepoittevin, C, C Bodénés, E Chancerel, L Villate, T Lang, I Lesur, C Boury, F Ehrenmann, D Zelenica, A Boland, et al. (2015). “Single-nucleotide polymorphism discovery and validation in high-density SNP array for genetic analysis in Euro- pean white oaks”. In: Molecular ecology resources 15.6, pp. 1446–1459. Leroy, T, JM Louvet, C Lalanne, G Le Provost, K Labadie, JM Aury, S Delzon, C Plomion, and A Kremer (2019). “Adaptive introgression as a driver of local adaptation to climate in European white oaks”. In: bioRxiv, p. 584847. Lesur, I, G Le Provost, P Bento, C Da Silva, JC Leplé, F Murat, S Ueno, J Bartholomé, C Lalanne, F Ehrenmann, et al. (2015). “The oak gene expression atlas: insights

232 into Fagaceae genome evolution and the discovery of genes regulated during bud dormancy release”. In: BMC genomics 16.1, p. 112. Levsen, ND, P Tin, and MS Olson (2012). “Pleistocene speciation in the genus Populus (Salicaceae)”. In: Systematic biology 61.3, p. 401. Lexer, C, C Buerkle, J Joseph, B Heinze, and M Fay (2007). “Admixture in Euro- pean Populus hybrid zones makes feasible the mapping of loci that contribute to reproductive isolation and trait dierences”. In: Heredity 98.2, p. 74. Li, H and R Durbin (2009a). “Fast and accurate short read alignment with Burrows– Wheeler transform”. In: bioinformatics 25.14, pp. 1754–1760. — (2009b). “Fast and accurate short read alignment with Burrows-Wheeler transform”. In: Bioinformatics 25.14, pp. 1754–1760. arXiv: 1303.3997. Li, S and Å Strid (2005). “Anthocyanin accumulation and changes in CHS and PR-5 gene expression in Arabidopsis thaliana after removal of the inﬂorescence stem (decapitation)”. In: Plant Physiology and Biochemistry 43.6, pp. 521–525. Liang, H, S Ayyampalayam, N Wickett, A Barakat, Y Xu, L Landherr, PE Ralph, Y Jiao, T Xu, SE Schlarbaum, et al. (2011). “Generation of a large-scale genomic resource for functional and comparative genomics in Liriodendron tulipifera L.” In: Tree Genetics & Genomes 7.5, pp. 941–954. Lindtke, D and CA Buerkle (2015). “The genetic architecture of hybrid incompatibilities and their eect on barriers to introgression in secondary contact”. In: Evolution 69.8, pp. 1987–2004. Lindtke, D, Z Gompert, C Lexer, and CA Buerkle (2014). “Unexpected ancestry of Populus seedlings from a hybrid zone implies a large role for postzygotic selection in the maintenance of species”. In: Molecular ecology 23.17, pp. 4316–4330. Lindtke, D, S Gonzalez-Martinez, D Macaya-Sanz, and C Lexer (2013). “Admixture mapping of quantitative traits in Populus hybrid zones: power and limitations”. In: Heredity 111.6, p. 474. Linhart, YB and MC Grant (1996). “Evolutionary signiﬁcance of local genetic dierentiation in plants”. In: Annual review of ecology and systematics 27.1, pp. 237– 277. Little Elbert L., J (1971). Atlas of United States trees. Volume 1. Conifers and important hardwoods. 1146. Washington, D.C.: U.S. Department of Agriculture, Forest Service., p. 320. Liu, C, N He, J Zhang, Y Li, Q Wang, L Sack, and G Yu (2018). “Variation of stomatal traits from cold temperate to tropical forests and association with water- use eciency”. In: Functional Ecology 32.1, pp. 20–28. Long, J, E Shelhamer, and T Darrell (2015). “Fully convolutional networks for seman- tic segmentation”. In: Proceedings of the IEEE conference on computer vision and pattern recognition. Los Alamitos, California: IEEE Computer Society, pp. 3431– 3440.

233 Lotterhos, KE and MC Whitlock (2014). “Evaluation of demographic history and neutral parameterization on the performance of FST outlier tests”. In: Molecular ecology 23.9, pp. 2178–2192. Luppold, W and M Bumgardner (2008). “Regional analysis of hardwood lumber production: 1963–2005”. In: Northern journal of applied forestry 25.3, pp. 146–150. Luquez, V, D Hall, BR Albrectsen, J Karlsson, P Ingvarsson, and S Jansson (2008). “Natural phenological variation in aspen (Populus tremula): the SwAsp collection”. In: Tree Genetics & Genomes 4.2, pp. 279–292. Ma, XF, D Hall, KRS Onge, S Jansson, and PK Ingvarsson (2010). “Genetic dierentiation, clinal variation and phenotypic associations with growth cessation across the Populus tremula photoperiodic pathway”. In: Genetics 186.3, pp. 1033–1044. Macnair, MR and Q Cumbes (1989). “The genetic architecture of interspecific variation in Mimulus.” In: Genetics 122.1, pp. 211–222. Madritch, MD, CC Kingdon, A Singh, KE Mock, RL Lindroth, and PA Townsend (2014). “Imaging spectroscopy links aspen genotype with below-ground processes at landscape scales”. In: Philosophical Transactions of the Royal Society B: Bio- logical Sciences 369.1643, p. 20130194. Mallet, J (2007). “Hybrid speciation”. In: Nature 446.7133, p. 279. Manel, S, C Perrier, M Pratlong, L Abi-Rached, J Paganini, P Pontarotti, and D Aurelle (2016). “Genomic resources and their influence on the detection of the signal of positive selection in genome scans”. In: Molecular ecology 25.1, pp. 170– 184. Manel, S, MK Schwartz, G Luikart, and P Taberlet (2003). “Landscape genetics: combining landscape ecology and population genetics”. In: Trends in ecology & evolution 18.4, pp. 189–197. Marinova, K, L Pourcel, B Weder, M Schwarz, D Barron, JM Routaboul, I Debeaujon, and M Klein (2007). “The Arabidopsis MATE transporter TT12 acts as a vacuolar flavonoid/H+-antiporter active in proanthocyanidin-accumulating cells of the seed coat”. In: The Plant Cell 19.6, pp. 2023–2038. Martins, K, PF Gugger, J Llanderal-Mendoza, A González-Rodríguez, ST Fitz-Gibbon, JL Zhao, H Rodríguez-Correa, K Oyama, and VL Sork (2018). “Landscape genomics provides evidence of climate-associated genetic variation in Mexican populations of Quercus rugosa”. In: Evolutionary Applications 11.10, pp. 1842–1858. Martinsen, GD, TG Whitham, RJ Turek, and P Keim (2001). “Hybrid populations selectively filter gene introgression between species”. In: Evolution 55.7, pp. 1325– 1335. Maugarny-Calès, A, M Cortizo, B Adroher, N Borrega, B Gonçalves, G Brunoud, T Vernoux, N Arnaud, and P Laufs (2019). “Dissecting the pathways coordinat- ing patterning and growth by plant boundary domains”. In: PLoS genetics 15.1, e1007913.

234 McCauley, DE (1995). “The use of chloroplast DNA polymorphism in studies of gene flow in plants”. In: Trends in ecology & evolution 10.5, pp. 198–202. McElwain, JC and M Steinthorsdottir (2017). “Paleoecology, ploidy, paleoatmospheric composition, and developmental biology: a review of the multiple uses of fossil stomata”. In: Plant Physiology 174.2, pp. 650–664. McKown, AD, J Klápöt, RD Guy, A Geraldes, I Porth, J Hannemann, M Friedmann, W Muchero, GA Tuskan, J Ehlting, QCB Cronk, YA El-Kassaby, SD Mansfield, and CJ Douglas (2014). “Genome-wide association implicates numerous genes underlying ecological trait variation in natural populations of Populus trichocarpa”. In: New Phytologist 203.2, pp. 535–553. McKown, AD, RD Guy, J Klápöt, A Geraldes, M Friedmann, QC Cronk, YA El- Kassaby, SD Mansfield, and CJ Douglas (2014). “Geographical and environmental gradients shape phenotypic trait variation and genetic structure in Populus trichocarpa”. In: New Phytologist 201.4, pp. 1263–1276. McKown, AD, RD Guy, L Quamme, J Klápöt, J La Mantia, C Constabel, YA El-Kassaby, RC Hamelin, M Zifkin, and M Azam (2014a). “Association genetics, geography and ecophysiology link stomatal patterning in Populus trichocarpa with carbon gain and disease resistance trade-os”. In: Molecular Ecology 23.23, pp. 5771–5790. — (2014b). “Association genetics, geography and ecophysiology link stomatal patterning in Populus trichocarpa with carbon gain and disease resistance trade-os”. In: Molecular Ecology 23.23, pp. 5771–5790. McKown, AD, J Klápöt, RD Guy, OR Corea, S Fritsche, J Ehlting, YA El-Kassaby, and SD Mansfield (2019). “A role for SPEECHLESS in the integration of leaf stomatal patterning with the growth vs disease trade-oin poplar.” In: The New phytologist. McLachlan, JS, JS Clark, and PS Manos (2005). “Molecular indicators of tree migration capacity under rapid climate change”. In: Ecology 86.8, pp. 2088–2098. McLean, EH, SM Prober, WD Stock, DA Steane, BM Potts, RE Vaillancourt, and M Byrne (2014). “Plasticity of functional traits varies clinally along a rainfall gradient in Eucalyptus tricarpa”. In: Plant, Cell & Environment 37.6, pp. 1440– 1451. McShea, WJ, WM Healy, P Devers, T Fearer, FH Koch, D Stauer, and J Waldon (2007). “Forestry matters: decline of oaks will impact wildlife in hardwood forests”. In: The Journal of Wildlife Management 71.5, pp. 1717–1728. Meijón, M, SB Satbhai, T Tsuchimatsu, and W Busch (2014). “Genome-wide association study using cellular traits identifies a new regulator of root development in Arabidopsis”. In: Nature genetics 46.1, p. 77. Meirmans, PG, M Lamothe, MC Gros-Louis, D Khasa, P Périnet, J Bousquet, and N Isabel (2010). “Complex patterns of hybridization between exotic and native

235 North American poplar species”. In: American Journal of Botany 97.10, pp. 1688– 1697. Melotto, M, W Underwood, J Koczan, K Nomura, and SY He (2006). “Plant stomata function in innate immunity against bacterial invasion”. In: Cell 126.5, pp. 969– 980. Menon, M, WJ Barnes, and MS Olson (2015). “Population genetics of freeze tolerance among natural populations of Populus balsamifera across the growing season”. In: New Phytologist 207.3, pp. 710–722. Messina, FJ, SL Durham, JH Richards, and DE McArthur (2002). “Trade-obetween plant growth and defense? A comparison of sagebrush populations”. In: Oecologia 131.1, pp. 43–51. Michonneau, F, JW Brown, and DJ Winter (2016). “rotl: an R package to interact with the Open Tree of Life data”. In: Methods in Ecology and Evolution 7.12, pp. 1476–1481. Mishra, P and KC Panigrahi (2015). “GIGANTEA–an emerging story”. In: Frontiers in plant science 6, p. 8. Mitchell-Olds, T (1996). “Genetic constraints on life-history evolution: quantitative- trait loci influencing growth and flowering in Arabidopsis thaliana”. In: Evolution 50.1, pp. 140–145. Mitchell, N, GL Owens, SM Hovick, LH Rieseberg, and KD Whitney (2019). “Hy- bridization speeds adaptive evolution in an eight-year field experiment”. In: Sci- entific reports 9.1, p. 6746. Miyauchi, H, S Moriyama, T Kusakizako, K Kumazaki, T Nakane, K Yamashita, K Hirata, N Dohmae, T Nishizawa, K Ito, et al. (2017). “Structural basis for xenobiotic extrusion by eukaryotic MATE transporter”. In: Nature communications 8.1, p. 1633. Mock, KE, CM Callahan, MN Islam-Faridi, JD Shaw, HS Rai, SC Sanderson, CA Rowe, RJ Ryel, MD Madritch, RS Gardner, and PG Wolf (2012). “Widespread triploidy in western North American aspen (Populus tremuloides)”. In: PLoS One 7.10, e48406. Mock, KE, C Rowe, MB Hooten, J Dewoody, and V Hipkins (2008). “Clonal dynamics in western North American aspen (Populus tremuloides)”. In: Molecular Ecology 17.22, pp. 4827–4844. Morales-Navarro, S, R Pérez-Díaz, A Ortega, A de Marcos, M Mena, C Fenoll, E González-Villanueva, and S Ruiz-Lara (2018). “Overexpression of a SDD1-Like gene from wild tomato decreases stomatal density and enhances dehydration avoidance in Arabidopsis and cultivated tomato”. In: Frontiers in plant science 9.

236 Moran, EV, J Willis, and JS Clark (2012). “Genetic evidence for hybridization in red oaks (Quercus sect. Lobatae, Fagaceae)”. In: American Journal of Botany 99.1, pp. 92–100. Muir, CD (2015). “Making pore choices: repeated regime shifts in stomatal ratio”. In: Proceedings of the Royal Society B: Biological Sciences 282.1813, p. 20151498. Muller, CH (1952). “Ecological control of hybridization in Quercus: a factor in the mechanism of evolution”. In: Evolution, pp. 147–161. Myburg, AA, D Grattapaglia, GA Tuskan, U Hellsten, RD Hayes, J Grimwood, J Jenkins, E Lindquist, H Tice, D Bauer, et al. (2014). “The genome of Eucalyptus grandis”. In: Nature 510.7505, p. 356. Myneni, R, J Dong, C Tucker, R Kaufmann, P Kauppi, J Liski, L Zhou, V Alexeyev, and M Hughes (2001). “A large carbon sink in the woody biomass of Northern forests”. In: Proceedings of the National Academy of Sciences 98.26, pp. 14784– 14789. Nathan, R, GG Katul, HS Horn, SM Thomas, R Oren, R Avissar, SW Pacala, and SA Levin (2002). “Mechanisms of long-distance dispersal of seeds by wind”. In: Nature 418.6896, p. 409. Nawrath, C, S Heck, N Parinthawong, and JP Métraux (2002). “EDS5, an essential component of salicylic acid–dependent signaling for disease resistance in Ara- bidopsis, is a member of the MATE transporter family”. In: The Plant Cell 14.1, pp. 275–286. Neale, DB and A Kremer (2011). “Forest tree genomics: growing resources and applications”. In: Nature Reviews Genetics 12.2, p. 111. Neiman, M, MS Olson, and P Tin (2009). “Selective histories of poplar protease in- hibitors: elevated polymorphism, purifying selection, and positive selection driving divergence of recent duplicates”. In: New Phytologist 183.3, pp. 740–750. Obeso, JR (2002). “The costs of reproduction in plants”. In: New phytologist 155.3, pp. 321–348. Oliverio, KA, M Crepy, EL Martin-Tryon, R Milich, SL Harmer, J Putterill, MJ Yanovsky, and JJ Casal (2007). “GIGANTEA regulates phytochrome A-mediated photomorphogenesis independently of its role in the circadian clock”. In: Plant Physiology 144.1, pp. 495–502. Oliviera, MWdS, NR da Silva, D Casanova, LFS Pinheiro, RM Kolb, and OM Bruno (2014). “Automatic counting of stomata in epidermis microscopic images”. In: X Workshop de Visao Computacional 3, pp. 253–257. Olson, MS, N Levsen, RY Soolanayakanahally, RD Guy, WR Schroeder, SR Keller, and P Tin (2013). “The adaptive potential of Populus balsamifera L. to phenology requirements in a warmer global climate”. In: Molecular Ecology 22.5, pp. 1214–1230.

237 Ordonez, N, MF Seidl, C Waalwijk, A Drenth, A Kilian, BP Thomma, RC Ploetz, and GH Kema (2015). “Worse comes to worst: bananas and Panama disease—when plant and pathogen clones meet”. In: PLoS pathogens 11.11, e1005197. Ortego, J, PF Gugger, EC Riordan, and VL Sork (2014). “Influence of climatic niche suitability and geographical overlap on hybridization patterns among southern Californian oaks”. In: Journal of Biogeography 41.10, pp. 1895–1908. Pandey, S, W Zhang, and SM Assmann (2007). “Roles of ion channels and transporters in guard cell signal transduction”. In: FEBS letters 581.12, pp. 2325–2336. Papanatsiou, M, A Amtmann, and MR Blatt (2016). “Stomatal spacing safeguards stomatal dynamics by facilitating guard cell ion transport independent of the epidermal solute reservoir”. In: Plant physiology 172.1, pp. 254–263. Pavlik, BM, SG Johnson, and P M. (1991). Oaks of California. Los Olivos, CA: Cachuma Press. Peel, JR, MC Mandujano Sánchez, J López Portillo, and J Golubov (2017). “Stomatal density, leaf area and plant size variation of Rhizophora mangle (Malpighiales: Rhizophoraceae) along a salinity gradient in the Mexican Caribbean”. In: Revista de Biología Tropical 65.2, pp. 701–712. Petit, RJ, S Brewer, S Bordács, K Burg, R Cheddadi, E Coart, J Cottrell, UM Csaikl, B van Dam, JD Deans, et al. (2002). “Identification of refugia and post-glacial colonization routes of European white oaks based on chloroplast DNA and fossil pollen evidence”. In: Forest ecology and management 156.1-3, pp. 49–74. Petit, RJ, J Carlson, AL Curtu, ML Loustau, C Plomion, A González-Rodríguez, V Sork, and A Ducousso (2013). “Fagaceae trees as models to integrate ecology, evolution and genomics”. In: New Phytologist 197.2, pp. 369–371. Petit, RJ and A Hampe (2006). “Some evolutionary consequences of being a tree”. In: Annu. Rev. Ecol. Evol. Syst. 37, pp. 187–214. Petit, RJ, FS Hu, and CW Dick (2008). “Forests of the past: a window to future changes”. In: Science 320.5882, pp. 1450–1452. Petit, RJ, E Pineau, B Demesure, R Bacilieri, A Ducousso, and A Kremer (1997). “Chloroplast DNA footprints of postglacial recolonization by oaks”. In: Proceed- ings of the National Academy of Sciences 94.18, pp. 9996–10001. Pettorelli, N, K Safi, and W Turner (2014). Satellite remote sensing, biodiversity research and conservation of the future. Pinon, J and P Frey (2005). “Interactions between poplar clones and Melampsora populations and their implications for breeding for durable resistance”. In: Rust Diseases of Willow and Poplar. Ed. by MH Pei and AR McCracken. Wallingford, UK: CABI, pp. 139–154. Pizarro, L, M Leibman-Markus, S Schuster, M Bar, T Meltz, and A Avni (2018). “Tomato prenylated RAB acceptor protein 1 modulates tracking and degra-

238 dation of the pattern recognition receptor LeEIX2, aecting the innate immune response”. In: Frontiers in plant science 9, p. 257. Platt, A, PF Gugger, M Pellegrini, and VL Sork (2015). “Genome-wide signature of local adaptation linked to variable CpG methylation in oak populations”. In: Molecular Ecology 24.15, pp. 3823–3830. Platt, A, BJ Vilhjálmsson, and M Nordborg (2010). “Conditions under which genome- wide association studies will be positively misleading”. In: Genetics 186.3, pp. 1045– 1052. Plomion, C, JM Aury, J Amselem, T Alaeitabar, V Barbe, C Belser, H Bergès, C Bodénès, N Boudet, C Boury, A Canaguier, A Couloux, C Da Silva, S Duplessis, F Ehrenmann, B Estrada-Mairey, S Fouteau, N Francillonne, C Gaspin, C Guichard, C Klopp, K Labadie, C Lalanne, I Le Clainche, JC Leplé, G Le Provost, T Leroy, I Lesur, F Martin, J Mercier, C Michotey, F Murat, F Salin, D Steinbach, P Faivre- Rampant, P Wincker, J Salse, H Quesneville, and A Kremer (2016). “Decoding the oak genome: public release of sequence data, assembly, annotation and publication strategies”. In: Molecular ecology resources 16.1, pp. 254–265. Poggio, T, K Kawaguchi, Q Liao, B Miranda, L Rosasco, X Boix, J Hidary, and H Mhaskar (2017). “Theory of deep learning III: explaining the non-overfitting puzzle”. In: arXiv preprint arXiv:1801.00173. Porth, I, J Klápöt, AD McKown, J La Mantia, RD Guy, PK Ingvarsson, R Hamelin, SD Mansfield, J Ehlting, CJ Douglas, et al. (2015). “Evolutionary quantitative genomics of Populus trichocarpa”. In: PloS one 10.11, e0142864. Porth, I, J Klapöte, O Skyba, J Hannemann, AD McKown, RD Guy, SP DiFazio, W Muchero, P Ranjan, GA Tuskan, MC Friedmann, J Ehlting, QCB Cronk, YA El-Kassaby, CJ Douglas, and SD Mansfield (2013). “Genome-wide association mapping for wood characteristics in Populus identifies an array of candidate single nucleotide polymorphisms”. In: New Phytologist 200.3, pp. 710–726. Porth, I, M Koch, M Berenyi, A Burg, and K Burg (2005). “Identification of adaptation- specific dierences in mRNA expression of sessile and pedunculate oak based on osmotic-stress-induced genes”. In: Tree physiology 25.10, pp. 1317–1329. Purcell, S, B Neale, K Todd-Brown, L Thomas, MA Ferreira, D Bender, J Maller, P Sklar, PI De Bakker, MJ Daly, et al. (2007). “PLINK: a tool set for whole-genome association and population-based linkage analyses”. In: The American journal of human genetics 81.3, pp. 559–575. RCoreTeam(2013).R: A Language and Environment for Statistical Computing.R Foundation for Statistical Computing. Vienna, Austria. —(2018).R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing. Vienna, Austria.

239 Racimo, F, S Sankararaman, R Nielsen, and E Huerta-Sánchez (2015). “Evidence for archaic adaptive introgression in humans”. In: Nature Reviews Genetics 16.6, p. 359. Raj, A, M Stephens, and JK Pritchard (2014). “FastSTRUCTURE: Variational inference of population structure in large SNP data sets”. In: Genetics 197.2, pp. 573– 589. arXiv: arXiv:1309.5056v3. Rajput, VD, T Minkina, C Yaning, S Sushkova, VA Chapligin, and S Mandzhieva (2016). “A review on salinity adaptation mechanism and characteristics of Populus euphratica, a boon for arid ecosystems”. In: Acta Ecológica Sinica 36.6, pp. 497– 503. Raven, PH and Zy Wu (1999). Flora of China. 4, Cycadaceae Through Fagaceae. Science Press. Rieseberg, LH and SE Carney (1998). “Plant hybridization”. In: The New Phytologist 140.4, pp. 599–624. Rieseberg, LH, SC Kim, RA Randell, KD Whitney, BL Gross, C Lexer, and K Clay (2007). “Hybridization and the colonization of novel habitats by annual sunﬂow- ers”. In: Genetica 129.2, pp. 149–165. Ristilä, M, H Strid, LA Eriksson, Å Strid, and H Sävenstrand (2011). “The role of the pyridoxine (vitamin B6) biosynthesis enzyme PDX1 in ultraviolet-B radiation responses in plants”. In: Plant Physiology and Biochemistry 49.3, pp. 284–292. Roche, B and R Fritz (1998). “Eects of host plant hybridization on resistance to willow leaf rust caused by Melampsora sp.” In: European Journal of Forest Pathology 28.4, pp. 259–270. Rockman, MV (2012). “The QTN program and the alleles that matter for evolution: all that’s gold does not glitter”. In: Evolution: International Journal of Organic Evolution 66.1, pp. 1–17. Rohart, F, B Gautier, A Singh, and KAL Cao (2017). “mixOmics: An R package for ’omics feature selection and multiple data integration”. In: PLoS computational biology 13.11, e1005752. Rojas, CM, M Senthil-Kumar, K Wang, CM Ryu, A Kaundal, and KS Mysore (2012). “Glycolate oxidase modulates reactive oxygen species–mediated signal transduction during nonhost resistance in Nicotiana benthamiana and Arabidopsis”. In: The Plant Cell 24.1, pp. 336–352. Ronzier, E, C Corratgé-Faillie, F Sanchez, K Prado, C Brière, N Leonhardt, JB Thibaud, and TC Xiong (2014). “CPK13, a noncanonical Ca2+-dependent protein kinase, speciﬁcally inhibits KAT2 and KAT1 shaker K+ channels and reduces stomatal opening”. In: Plant physiology 166.1, pp. 314–326. Royer, DL (2001). “Stomatal density and stomatal index as indicators of paleoatmospheric CO2 concentration”. In: Review of Palaeobotany and Palynology 114.1-2, pp. 1–28.

240 Russakovsky, O, J Deng, H Su, J Krause, S Satheesh, S Ma, Z Huang, A Karpathy, A Khosla, M Bernstein, et al. (2015). “Imagenet large scale visual recognition challenge”. In: International journal of computer vision 115.3, pp. 211–252. Rylott, EL, CA Rogers, AD Gilday, T Edgell, TR Larson, and IA Graham (2003). “Arabidopsis mutants in short-and medium-chain acyl-CoA oxidase activities accumulate acyl-CoAs and reveal that fatty acid —-oxidation is essential for embryo development”. In: Journal of Biological Chemistry 278.24, pp. 21370–21377. Santos-del-Blanco, L, J Climent, S González-Martínez, and J Pannell (2012). “Genetic dierentiation for size at first reproduction through male versus female functions in the widespread Mediterranean tree Pinus pinaster”. In: Annals of botany 110.7, pp. 1449–1460. Savolainen, O, T Pyhäjärvi, and T Knürr (2007). “Gene flow and local adaptation in trees”. In: Annu. Rev. Ecol. Evol. Syst. 38, pp. 595–619. Scheet, P and M Stephens (2006). “A fast and flexible statistical model for large- scale population genotype data: applications to inferring missing genotypes and haplotypic phase”. In: The American Journal of Human Genetics 78.4, pp. 629– 644. Schluter, D, TD Price, and L Rowe (1991). “Conflicting selection pressures and life history trade-os”. In: Proc. R. Soc. Lond. B 246.1315, pp. 11–17. Schmitz, RJ, MD Schultz, MA Urich, JR Nery, M Pelizzola, O Libiger, A Alix, RB McCosh, H Chen, NJ Schork, and J Ecker (2013). “Patterns of population epige- nomic diversity”. In: Nature 495.7440, p. 193. Schneider, CA, WS Rasband, and KW Eliceiri (2012a). “NIH Image to ImageJ: 25 years of image analysis”. In: Nature methods 9.7, p. 671. — (2012b). “NIH Image to ImageJ: 25 years of image analysis”. In: Nature methods 9.7, p. 671. Scholthof, KBG (2007). “The disease triangle: pathogens, the environment and society”. In: Nature Reviews Microbiology 5.2, p. 152. Schumer, M, R Cui, GG Rosenthal, and P Andolfatto (2015). “Reproductive isolation of hybrid populations driven by genetic incompatibilities”. In: PLoS genetics 11.3, e1005041. Schweitzer, JA, MD Madritch, JK Bailey, CJ LeRoy, DG Fischer, BJ Rehill, RL Lin- droth, AE Hagerman, SC Wooley, SC Hart, et al. (2008). “From genes to ecosystems: the genetic basis of condensed tannins and their role in nutrient regulation in a Populus model system”. In: Ecosystems 11.6, pp. 1005–1020. Serrano, M, F Coluccia, M Torres, F L’Haridon, and JP Métraux (2014). “The cuticle and plant defense to pathogens”. In: Frontiers in plant science 5, p. 274. Shahinnia, F, J Le Roy, B Laborde, B Sznajder, P Kalambettu, S Mahjourimajd, J Tilbrook, and D Fleury (2016). “Genetic association of stomatal traits and yield in wheat grown in low rainfall environments”. In: BMC plant biology 16.1, p. 150.

241 Sheehan, MJ, LM Kennedy, DE Costich, and TP Brutnell (2007). “Subfunctional- ization of PhyB1 and PhyB2 in the control of seedling and mature plant traits in maize”. In: The Plant Journal 49.2, pp. 338–353. Shen, D, G Wu, and HI Suk (2017). “Deep learning in medical image analysis”. In: Annual review of biomedical engineering 19, pp. 221–248. Shimazaki, Ki, M Doi, SM Assmann, and T Kinoshita (2007). “Light regulation of stomatal movement”. In: Annual Review of Plant Biology 58, pp. 219–247. Shriner, D (2013). “Overview of admixture mapping”. In: Current protocols in human genetics 76.1, pp. 1–23. Shriner, D, A Adeyemo, and CN Rotimi (2011). “Joint ancestry and association testing in admixed individuals”. In: PLoS computational biology 7.12, e1002325. Simonyan, K and A Zisserman (2014). “Very deep convolutional networks for large- scale image recognition”. In: arXiv preprint arXiv:1409.1556. Sjödin, A, NR Street, G Sandberg, P Gustafsson, and S Jansson (2009a). “The Pop- ulus Genome Integrative Explorer (PopGenIE): a new resource for exploring the Populus genome”. In: New phytologist 182.4, pp. 1013–1025. — (2009b). “The Populus Genome Integrative Explorer (PopGenIE): a new resource for exploring the Populus genome”. In: New phytologist 182.4, pp. 1013–1025. Skaug, H, D Fournier, B Bolker, A Magnusson, and A Nielsen (2016). Generalized Linear Mixed Models using ’AD Model Builder’. R package version 0.8.3.3. Skaug, H, D Fournier, A Nielsen, A Magnusson, B Bolker, et al. (2013). “Generalized linear mixed models using AD model builder”. In: R package version 0.7 7. Slavov, GT, SP DiFazio, J Martin, W Schackwitz, W Muchero, E Rodgers-Melnick, MF Lipphardt, CP Pennacchio, U Hellsten, LA Pennacchio, LE Gunter, P Ran- jan, K Vining, KR Pomraning, LJ Wilhelm, M Pellegrini, TC Mockler, M Freitag, A Geraldes, YA El-Kassaby, SD Mansﬁeld, QCB Cronk, CJ Douglas, SH Strauss, D Rokhsar, and GA Tuskan (2012). “Genome resequencing reveals multiscale geographic structure and extensive linkage disequilibrium in the forest tree Populus trichocarpa”. In: New Phytologist 196.3, pp. 713–725. Smith, MW and SJ O’Brien (2005). “Mapping by admixture linkage disequilibrium: advances, limitations and guidelines”. In: Nature Reviews Genetics 6.8, p. 623. Smith, RL and KJ Sytsma (1990). “Evolution of Populus nigra (sect. Aigeiros): introgressive hybridization and the chloroplast contribution of Populus alba (sect. Populus)”. In: American journal of botany 77.9, pp. 1176–1187. Soler, M, O Serra, M Molinas, E García-Berthou, A Caritat, and M Figueras (2008). “Seasonal variation in transcript abundance in cork tissue analyzed by real time RT-PCR”. In: Tree Physiology 28.5, pp. 743–751. Soltis, DE, AB Morris, JS McLachlan, PS Manos, and PS Soltis (2006). “Compara- tive phylogeography of unglaciated eastern North America”. In: Molecular ecology 15.14, pp. 4261–4293.

242 Soolanayakanahally, RY, RD Guy, SN Silim, EC Drewes, and WR Schroeder (2009). “Enhanced assimilation rate and water use eciency with latitude through increased photosynthetic capacity and internal conductance in balsam poplar (Pop- ulus balsamifera L.)” In: Plant, Cell & Environment 32.12, pp. 1821–1832. Soolanayakanahally, RY, RD Guy, SN Silim, and M Song (2013). “Timing of photoperiodic competency causes phenological mismatch in balsam poplar (Populus balsamifera L.)” In: Plant, cell & environment 36.1, pp. 116–127. Soolanayakanahally, RY, RD Guy, NR Street, KM Robinson, SN Silim, BR Albrect- sen, and S Jansson (2015). “Comparative physiology of allopatric Populus species: geographic clines in photosynthesis, height growth, and carbon isotope discrimination in common gardens”. In: Frontiers in plant science 6, p. 528. Sork, VL, FW Davis, PE Smouse, VJ Apsit, RJ Dyer, J Fernandez-M, and B Kuhn (2002). “Pollen movement in declining populations of California Valley oak, Quer- cus lobata: where have all the fathers gone?” In: Molecular Ecology 11.9, pp. 1657– 1668. Sork, VL, ST Fitz-Gibbon, D Puiu, M Crepeau, PF Gugger, R Sherman, K Stevens, CH Langley, M Pellegrini, and SL Salzberg (2016). “First draft assembly and annotation of the genome of a California endemic oak Quercus lobata Née (Fagaceae)”. In: G3: Genes, Genomes, Genetics 6.11, pp. 3485–3495. Sork, VL and PE Smouse (2006). “Genetic analysis of landscape connectivity in tree populations”. In: Landscape ecology 21.6, pp. 821–836. Sork, VL, PE Smouse, VJ Apsit, RJ Dyer, and RD Westfall (2005). “A two-generation analysis of pollen pool genetic structure in flowering dogwood, Cornus florida (Cornaceae), in the Missouri Ozarks”. In: American journal of botany 92.2, pp. 262– 271. Sork, VL, K Squire, PF Gugger, SE Steele, ED Levy, and AJ Eckert (2016). “Land- scape genomic analysis of candidate genes for climate adaptation in a California endemic oak, Quercus lobata”. In: American Journal of Botany 103.1, pp. 33–46. Sork, V, S Aitken, R Dyer, A Eckert, P Legendre, and D Neale (2013). “Putting the landscape into the genomics of trees: approaches for understanding local adaptation and population responses to changing climate”. In: Tree Genetics & Genomes 9.4, pp. 901–911. Sorrie, BA and AS Weakley (2001). “Coastal plain vascular plant endemics: phyto- geographic patterns”. In: Castanea, pp. 50–82. Standiford, RB and RE Howitt (1993). “Multiple use management of California’s hardwood rangelands.” In: Rangeland Ecology & Management/Journal of Range Management Archives 46.2, pp. 176–182. Stankowski, S, JM Sobel, and MA Streisfeld (2015). “The geography of divergence with gene flow facilitates multitrait adaptation and the evolution of pollinator isolation in Mimulus aurantiacus”. In: Evolution 69.12, pp. 3054–3068.

243 Stankowski, S and MA Streisfeld (2015). “Introgressive hybridization facilitates adaptive divergence in a recent radiation of monkeyflowers”. In: Proceedings of the Royal Society B: Biological Sciences 282.1814, p. 20151666. Steane, DA, BM Potts, E McLean, SM Prober, WD Stock, RE Vaillancourt, and M Byrne (2014). “Genome-wide scans detect adaptation to aridity in a widespread forest tree species”. In: Molecular Ecology 23.10, pp. 2500–2513. Stearns, SC (1989). “Trade-os in life-history evolution”. In: Functional ecology 3.3, pp. 259–268. Stinchcombe, JR and HE Hoekstra (2008). “Combining population genomics and quantitative genetics: finding the genes underlying ecologically important traits”. In: Heredity 100.2, p. 158. Stölting, KN, R Nipper, D Lindtke, C Caseys, S Waeber, S Castiglione, and C Lexer (2013). “Genomic scan for single nucleotide polymorphisms reveals patterns of divergence and gene flow between ecologically divergent species”. In: Molecular ecology 22.3, pp. 842–855. Strei, R (1999). “Pollen dispersal inferred from paternity analysis in a mixed oak stand of Quercus robur L. and Q. petraea (Matt.) Liebl”. In: Molecular Ecology 8, pp. 831–841. Suarez-Gonzalez, A, CA Hefer, C Christe, O Corea, C Lexer, QC Cronk, and CJ Douglas (2016). “Genomic and functional approaches reveal a case of adaptive introgression from Populus balsamifera (balsam poplar) in P. trichocarpa (black cottonwood)”. In: Molecular ecology 25.11, pp. 2427–2442. Suarez-Gonzalez, A, CA Hefer, C Lexer, QC Cronk, and CJ Douglas (2018). “Scale and direction of adaptive introgression between black cottonwood (Populus trichocarpa) and balsam poplar (P. balsamifera)”. In: Molecular ecology 27.7, pp. 1667– 1680. Suarez-Gonzalez, A, CA Hefer, C Lexer, CJ Douglas, and QC Cronk (2018). “Intro- gression from Populus balsamifera underlies adaptively significant variation and range boundaries in P. trichocarpa”. In: New Phytologist 217.1, pp. 416–427. Suarez-Gonzalez, A, C Lexer, and QC Cronk (2018). “Adaptive introgression: a plant perspective”. In: Biology letters 14.3, p. 20170688. Sul, JH, LS Martin, and E Eskin (2018). “Population structure in genetic studies: confounding factors and mixed models”. In: PLoS genetics 14.12, e1007309. Sumathi, M, V Bachpai, B Deeparaj, A Mayavel, MG Dasgupta, B Nagarajan, D Rajasugunasekar, V Sivakumar, and R Yasodha (2018). “Quantitative trait loci mapping for stomatal traits in interspecific hybrids of Eucalyptus”. In: Journal of genetics, pp. 1–7. Sundell, D, C Mannapperuma, S Netotea, N Delhomme, YC Lin, A Sjödin, Y Van de Peer, S Jansson, TR Hvidsten, and NR Street (2015). “The Plant Genome Integra-

244 tive Explorer Resource: PlantGen IE. org”. In: New Phytologist 208.4, pp. 1149– 1156. Suren, H (2017). “Sequence capture as a tool to understand the genomic basis for adaptation in angiosperm and gymnosperm trees”. PhD thesis. Virginia Tech. Swarbreck, D, C Wilks, P Lamesch, TZ Berardini, M Garcia-Hernandez, H Foerster, D Li, T Meyer, R Muller, L Ploetz, et al. (2007). “The Arabidopsis Informa- tion Resource (TAIR): gene structure and function annotation”. In: Nucleic acids research 36.suppl_1, pp. D1009–D1014. Sweigart, AL, AR Mason, and JH Willis (2007). “Natural variation for a hybrid incompatibility between two species of Mimulus”. In: Evolution 61.1, pp. 141– 151. Takahashi, S, K Monda, J Negi, F Konishi, S Ishikawa, M Hashimoto-Sugimoto, N Goto, and K Iba (2015). “Natural variation in stomatal responses to environmental changes among Arabidopsis thaliana ecotypes”. In: PloS one 10.2, e0117449. Tarkka, MT, S Herrmann, T Wubet, L Feldhahn, S Recht, F Kurth, S Mailänder, M Bönn, M Neef, O Angay, M Bacht, M Graf, H Maboreke, F Fleischmann, TEE Grams, L Ruess, M Schädler, R Brandl, S Scheu, SD Schrey, I Grosse, and F Buscot (2013). “OakContig DF 159.1, a reference library for studying dierential gene expression in Quercus robur during controlled biotic interactions: use for quantitative transcriptomic proﬁling of oak roots in ectomycorrhizal symbiosis”. In: New Phytologist 199.2, pp. 529–540. Taylor, G (2002). “Populus: Arabidopsis for forestry. Do we need a model tree?” In: Annals of Botany 90.6, pp. 681–689. Thompson, SL, M Lamothe, PG Meirmans, P Perinet, and N Isabel (2010a). “Re- peated unidirectional introgression towards Populus balsamifera in contact zones of exotic and native poplars”. In: Molecular Ecology 19.1, pp. 132–145. — (2010b). “Repeated unidirectional introgression towards Populus balsamifera in contact zones of exotic and native poplars”. In: Molecular Ecology 19.1, pp. 132– 145. Tian, D, M Traw, J Chen, M Kreitman, and J Bergelson (2003). “Fitness costs of R-gene-mediated resistance in Arabidopsis thaliana”. In: Nature 423.6935, p. 74. Tin, P and J Ross-Ibarra (2014). “Advances and limits of using population genetics to understand local adaptation”. In: Trends in ecology & evolution 29.12, pp. 673– 680. Toomey, M, MA Friedl, S Frolking, K Hufkens, S Klosterman, O Sonnentag, DD Baldocchi, CJ Bernacchi, SC Biraud, G Bohrer, et al. (2015). “Greenness indices from digital cameras predict the timing and seasonal dynamics of canopy-scale photosynthesis”. In: Ecological Applications 25.1, pp. 99–115. Tuskan, GA et al. (2006). “The genome of black cottonwood, Populus trichocarpa (Torr. & Gray)”. In: science 313.5793, pp. 1596–1604.

245 Tuskan, G, S Difazio, S Jansson, J Bohlmann, I Grigoriev, U Hellsten, N Putnam, S Ralph, S Rombauts, A Salamov, et al. (2006). “The genome of black cottonwood, Populus trichocarpa (Torr. & Gray)”. In: Science 313.5793, p. 1596. \ Ubbens, JR and I Stavness (2017). “Deep plant phenomics: a deep learning platform for complex plant phenotyping tasks”. In: Frontiers in plant science 8, p. 1190. Ueno, S, G Le Provost, V Léger, C Klopp, C Noirot, JM Frigerio, F Salin, J Salse, M Abrouk, F Murat, O Brendel, J Derory, P Abadie, P Léger, C Cabane, A Barré, A de Daruvar, A Couloux, P Wincker, MP Reviron, A Kremer, and C Plomion (2010). “Bioinformatic analysis of ESTs collected by Sanger and pyrosequencing methods for a keystone forest tree species: oak”. In: BMC genomics 11.1, p. 650. Ullah, C, CJ Tsai, SB Unsicker, L Xue, M Reichelt, J Gershenzon, and A Ham- merbacher (2018). “Salicylic acid activates poplar defense against the biotrophic rust fungus Melampsora larici-populina via increased biosynthesis of catechin and proanthocyanidins”. In: New Phytologist. Ullah, C, SB Unsicker, C Fellenberg, CP Constabel, A Schmidt, J Gershenzon, and A Hammerbacher (2017). “Flavan-3-ols are an eective chemical defense against rust infection”. In: Plant physiology 175.4, pp. 1560–1578. Van Kraayenoord, C, G Laundon, and A Spiers (1974). “Poplar rusts invade New Zealand.” In: Plant disease reporter 58.5, pp. 423–427. Van Valen, L (1976). “Ecological species, multispecies, and oaks”. In: Taxon, pp. 233– 239. Verhoeven, KJ, JJ Jansen, PJ van Dijk, and A Biere (2010). “Stress-induced DNA methylation changes and their heritability in asexual dandelions”. In: New Phy- tologist 185.4, pp. 1108–1118. Vialet-Chabrand, S and O Brendel (2014). “Automatic measurement of stomatal density from microphotographs”. In: Trees 28.6, pp. 1859–1865. Vieland, VJ (2001). “The replication requirement”. In: Nature Genetics 29.3, p. 244. Wang, J, NR Street, DG Scoﬁeld, and PK Ingvarsson (2016). “Natural selection and recombination rate variation shape nucleotide polymorphism across the genomes of three related Populus species”. In: Genetics 202.3, pp. 1185–1200. Wang, L, P Tin, and MS Olson (2014). “Timing for success: expression phenotype and local adaptation related to latitude in the boreal forest tree, Populus balsamifera”. In: Tree genetics & genomes 10.4, pp. 911–922. Wang, R, G Yu, N He, Q Wang, N Zhao, and Z Xu (2015). “Latitudinal variation of leaf stomatal traits from species to community level in forests: linkage with ecosystem productivity”. In: Nature Scientiﬁc Reports 5.14454, pp. 1–11. Wang, ZY (2012). “Brassinosteroids modulate plant immunity at multiple levels”. In: Proceedings of the National Academy of Sciences 109.1, pp. 7–8. Wegmann, D, DE Kessner, KR Veeramah, RA Mathias, DL Nicolae, LR Yanek, YV Sun, DG Torgerson, N Rafaels, T Mosley, et al. (2011). “Recombination rates in

246 admixed individuals identified by ancestry-based inference”. In: Nature genetics 43.9, p. 847. White, J, B Vaughn, and S Michel (2011). “Stable Isotopic Composition of Atmo- spheric Carbon Dioxide (13C and 18O) from the NOAA ESRL Carbon Cycle Cooperative Global Air Sampling Network, 1990–2014, Version: 2015-10-26.” In: University of Colorado, Institute of Arctic and Alpine Research (INSTAAR). Whitham, TG (1989). “Plant hybrid zones as sinks for pests”. In: Science, pp. 1490– 1493. Whitham, TG, SP DiFazio, JA Schweitzer, SM Shuster, GJ Allan, JK Bailey, and SA Woolbright (2008). “Extending genomics to natural communities and ecosystems”. In: Science 320.5875, pp. 492–495. Whitham, TG, KD Floate, GD Martinsen, EM Driebe, and P Keim (1996). “Eco- logical and evolutionary implications of hybridization: Populus–herbivore interactions”. In: Biology of Populus and its implications for management and conservation Part I, pp. 247–275. Whitney, KD, RA Randell, and LH Rieseberg (2006). “Adaptive introgression of herbivore resistance traits in the weedy sunflower Helianthus annuus”. In: The American Naturalist 167.6, pp. 794–807. Whitton, J, D Wolf, D Arias, A Snow, and L Rieseberg (1997). “The persistence of cultivar alleles in wild populations of sunflowers five generations after hybridization”. In: Theoretical and Applied Genetics 95.1-2, pp. 33–40. Winter, D, B Vinegar, H Nahal, R Ammar, GV Wilson, and NJ Provart (2007). “An “Electronic Fluorescent Pictograph” browser for exploring and analyzing large- scale biological data sets”. In: PloS one 2.8, e718. Wu, CI (2001). “The genic view of the process of speciation”. In: Journal of evolutionary biology 14.6, pp. 851–865. Wu, JX, J Li, Z Liu, J Yin, ZY Chang, C Rong, JL Wu, FC Bi, and N Yao (2015). “The Arabidopsis ceramidase At ACER functions in disease resistance and salt tolerance”. In: The Plant Journal 81.5, pp. 767–780. Wymore, AS, AT Keeley, KM Yturralde, ML Schroer, CR Propper, and TG Whitham (2011). “Genes to ecosystems: exploring the frontiers of ecology with one of the smallest biological units”. In: New Phytologist 191.1, pp. 19–36. Wysoker, A, T Fennell, J Ruan, N Homer, G Marth, G Abecasis, R Durbin, et al. (2009). “The Sequence alignment/map (SAM) format and SAMtools”. In: Bioin- formatics 25, pp. 2078–2079. Yang, WY, J Novembre, E Eskin, and E Halperin (2012). “A model-based approach for analysis of spatial structure in genetic data”. In: Nature genetics 44.6, p. 725. Yatabe, Y, NC Kane, C Scotti-Saintagne, and LH Rieseberg (2007). “Rampant gene exchange across a strong reproductive barrier between the annual sunflowers, He- lianthus annuus and H. petiolaris”. In: Genetics 175.4, pp. 1883–1893.

247 Yu, F and V Koltun (2015). “Multi-scale context aggregation by dilated convolutions”. In: arXiv preprint arXiv:1511.07122. Zanewich, KP, DW Pearce, and SB Rood (2018). “Heterosis in poplar involves phenotypic stability: cottonwood hybrids outperform their parental species at subop- timal temperatures”. In: Tree physiology 38.6, pp. 789–800. Zhang, P, F Wu, X Kang, C Zhao, and Y Li (2015). “Genotypic variations of biomass feedstock properties for energy in triploid hybrid clones of Populus tomentosa”. In: BioEnergy Research 8.4, pp. 1705–1713. Zhang, Y, X Jin, Z Ouyang, X Li, B Liu, L Huang, Y Hong, H Zhang, F Song, and D Li (2015). “Vitamin B6 contributes to disease resistance against Pseudomonas syringae pv. tomato DC3000 and Botrytis cinerea in Arabidopsis thaliana”. In: Journal of plant physiology 175, pp. 21–25. Zheng, P, JX Wu, SK Sahu, HY Zeng, LQ Huang, Z Liu, S Xiao, and N Yao (2018). “Loss of alkaline ceramidase inhibits autophagy in Arabidopsis and plays an important role during environmental stress response”. In: Plant, cell & environment 41.4, pp. 837–849. Zhou, L, R Bawa, and J Holliday (2014). “Exome resequencing reveals signatures of demographic and adaptive processes across the genome and range of black cottonwood (Populus trichocarpa)”. In: Molecular ecology 23.10, pp. 2486–2499.

248