Persistiendo en el límite: flujo génico, reproducción y conservación de poblaciones refugio de Quercus robur L.

Tesis doctoral PERSISTIENDO EN EL LÍMITE Flujo génico, reproducción y conservación de ! poblaciones refugio de Quercus robur L. Eva Moracho Martínez 2019 Directores: Arndt Hampe, Pedro Jordano y Gerardo Moreno" Persistiendo en el límite: flujo génico, reproducción y conservación de poblaciones refugio de Quercus robur L. ii CITA RECOMENDADA:! Moracho Martínez E. (2019) Persistiendo en el límite: flujo génico, reproducción y conservación de poblaciones refugio de Quercus robur L. Tesis doctoral. Universidad de Sevilla, Sevilla. Spain DISEÑADO Y MAQUETADO por Eva Moracho Martínez FOTOGRAFÍA DE LA PORTADA por Pablo Villalva Aguilar" iii ESTACIÓN BIOLÓGICA DE DOÑANA (EBD-CSIC) DEPARTAMENTO DE ECOLOGÍA INTEGRATIVA UNIVERSIDAD DE SEVILLA DEPARTAMENTO DE BIOLOGÍA VEGETAL Y ECOLOGÍA PERSISTIENDO EN EL LÍMITE Flujo génico, reproducción y conservación de poblaciones ! refugio de Quercus robur L. Memoria presentada por la Licenciada en Biología Eva Moracho Martínez para optar al Título de Doctora en Biología por la Universidad de Sevilla, Sevilla. Fdo. Eva Moracho Martínez Sevilla, 2019" iv v ESTACIÓN BIOLÓGICA DE DOÑANA (EBD-CSIC) DEPARTAMENTO DE ECOLOGÍA INTEGRATIVA UNIVERSIDAD DE SEVILLA DEPARTAMENTO DE BIOLOGÍA VEGETAL Y ECOLOGÍA El Dr. Arndt Hampe, director de investigación del Institut National de la Recherche Agronomique (INRA), el Dr. Gerardo Moreno Marcos, profesor titular de la Universidad de Extremadura, y el Dr. Pedro Jordano Barbudo, profesor investigador del Departamen- to de Ecología Integrativa de la Estación Biológica de Doñana (EBD-CSIC), y el Dr. Juan Arroyo Marín, catedrático de Botánica del Departamento de Biología Vegetal y Ecología de la Universidad de Sevilla (US) CERTIFICAN:! Que la tesis doctoral titulada Persistiendo en el límite: flujo génico, reproducción y conservación de poblaciones refugio de Quercus robur L. presentada por la Lcda. Eva Moracho Martínez ha sido realizada bajo nuestra dirección y tutoría, y reúne las condiciones de calidad y rigor científico para que pueda ser presentada y defendida ante la comisión nombrada al efecto. Y para que así conste y surta los efectos oportunos, firmamos el presente certificado en Sevilla, 20 Marzo de 2019. Directores Fdo. Arndt Hampe Fdo. Gerardo Moreno Marcos Fdo. Pedro Jordano Barbudo Tutor Fdo. Juan Arroyo Marín" xii individuo hasta el paisaje) e investigados sus principales determinantes ecológicos y genéticos son investigados. El marco conceptual en el que se encuadra la tesis es demostrado empíricamente en un estudio a escala global basado en la síntesis de una colección de datos mediante técnicas de meta-análisis (capítulo 2). Los resultados ponen de manifiesto un cambio asimétrico en el performance de las poblaciones marginales de acuerdo a las predicciones del cambio climático moderno. Las poblaciones del margen de latitudes bajas muestran un performance reducido respecto a otras áreas del rango (centrales y latitudes altas), mientras que aquellas del margen de latitudes altas no sobresalen frente al estimado en áreas centrales pero tampoco quedan atrás (como predice la teoría centro-periferia). Este efecto guarda relación con el diferencial de temperatura entre las zonas de estudio y constituye una tendencia general a través de los reinos y tipos de hábitat. Por tanto, encontramos un impacto del cambio de clima doblemente negativo donde ya se evidencia la vulnerabilidad a la extinción local de las poblaciones del margen de baja latitud, pero además, el retraso de las poblaciones de latitudes altas en su reacción de expansión como respuesta al desplazamiento de las condiciones climáticas. El estudio de los patrones de flujo génico histórico y contemporáneo en poblaciones relictas de Q. robur (capítulo 3) se beneficia del muestreo exhaustivo de todos los adultos presentes en el área de estudio (n = 135 en un total de 10 rodales), y las progenies muestreadas sobre 72 árboles madre (n = 724). Mediante técnicas de clustering genético y de topología de redes caracterizamos la estructura y el flujo génico histórico, mientras que la dispersión de polen contemporáneo se basa en potentes análisis de paternidad. Este estudio ofrece un resultado sobresaliente para una especie anemófila ya que observamos una excepcional escasez de flujo génico y una limitada conectividad poblacional. La fuerte estructura genética de la población adulta salpicada de eventos esporádicos de dispersión histórica es coherente con el limitado flujo de polen que se detecta entre rodales del mismo valle (2.6%) o desde el valle contiguo (4,4%), lo que indica que la débil conectividad no es un fenómeno a corto plazo. El reducido tamaño y el alto aislamiento de los rodales junto a otros aspectos físicos del paisaje refugial suponen un obstáculo a la dispersión de polen. Como consecuencia, el patrón de conectividad es fuertemente heterogéneo espacialmente, con rodales que funcionan predominantemente como fuente o sumidero de polen y otros aislados prácticamente al flujo génico. A pesar de ser escasa, la inmigración de polen puede llegar a tener un impacto exacerbado en la genética de las poblaciones si operan procesos de purga genética. xiii A una escala menor, detectamos una gran variabilidad entre individuos en los patrones de cruzamiento que responde mayoritariamente a diferencias en el ambiente local y rasgos intrínsecos del individuo (capítulo 4). La polinización en la mayoría de los individuos ocurre sólo entre individuos del mismo rodal. Una reducida fracción de los árboles es responsable de los infrecuentes eventos de inmigración procedente de la misma especie (7%) o de la especie hermana Q. pyrenaica con la que coexiste (0.6%). Los eventos de autofecundación son excepcionalmente frecuentes (6.8%) para una especie predominantemente autoincompatible. Este incremento de la tasa de autofecundación observada en nuestras poblaciones relictas ha sido igualmente descrito en poblaciones marginales de otras especies, lo que parece sugerir algún tipo de ventaja asociada a la adquisición de un sistema de cruzamiento mixto en estas situaciones. Los patrones de dispersión de polen (local, inmigrante y propio) son moldeados por un conjunto de factores que operan a escala fina, relacionados con aspectos del vecindario local (densidad de conespecíficos y grado de parentesco entre ellos). La autofecundación se ve además influida por rasgos intrínsecos del individuo (el tamaño y la producción de polen). En resumen, parece que las poblaciones relictas son propensas a albergar individuos de gran tamaño, grandes productores de polen que favorecen la autofecundación (fenómeno a escala fina) cuando no disponen de polen en abundancia y de calidad, mientras que la existencia de núcleos de población densos en las proximidades facilitaría la dispersión de polen inmigrante (fenómeno a escala de paisaje). Adicionalmente, profundizamos en los patrones de cruzamiento dentro de los rodales de mayor tamaño (4 réplicas, n = 22-35 individuos) aplicando modelos de cruzamiento espacialmente explícitos que permiten inferir la red de cruzamiento entre todos los individuos del rodal y caracterizar las componentes femenina y masculina del proceso de polinización (capítulo 4). Existe una red de polinización muy conectada en el interior de rodales, donde los árboles de gran tamaño, productores de vastas cantidades de polen y los individuos que ocupan posiciones centrales del rodal juegan un importante rol como conectores. Además se demuestra una gran variabilidad entre individuos en la fecundidad masculina (liberación de polen que llega a producir frutos) que está íntimamente ligado al tamaño y la capacidad productora del árbol. Como aproximación al potencial de regeneración que presentan las poblaciones relictas de Q. robur estudiamos la fecundidad y el éxito reproductivo (capítulo 5) durante 2 años sobre 159 árboles localizados a través de 12 núcleos de población. La variabilidad en la fecundidad femenina (medida como la producción total de semillas viables y su peso), al igual que la masculina, se debe principalmente a diferencias en el tamaño del árbol. Los árboles mas grandes producen una mayor cantidad de flores femeninas (y masculinas) y xiv bellotas de manera casi invariable a través de los años. Además las bellotas producidas son más grandes y pesadas en los árboles de gran tamaño, lo que puede tener particular relevancia en el contexto de refugios climáticos en ambientes mediterráneos debido a las ventajas asociadas a la germinación, crecimiento y supervivencia de las semillas con más reservas. Por otra parte, el éxito reproductivo (fruit set) en estas poblaciones relictas se ve influido por factores actuando a escala de paisaje (tamaño y aislamiento poblacional), lo que evidencia limitación de polen a esta escala, y no a escala local a pesar de los numerosos indicadores ecológicos y genéticos empleados. La aparente ausencia de restricciones genéticas puede explicar la predominante fertilización local observada en poblaciones relictas. Los patrones de dispersión y cruzamiento parecen garantizar la reproducción exitosa de la especie siempre que se alcance un tamaño poblacional mínimo y/o el aislamiento no sea elevado; de otro modo el escaso flujo de polen no logra compensar la limitación de polen existente. No obstante, la variabilidad interanual en el proceso reproductivo es notable de manera que la limitación de polen tendrá efectos más profundos en años de poca productividad, y los episodios donde la regeneración ocurra se limitarán a años muy productivos asociados a condiciones de benevolencia climática. Las estrategias de conservación en poblaciones relictas deben centrarse en un correcto funcionamiento interno mediante prácticas que aseguren un número y densidad suficiente de individuos reproductivamente muy activos. Especial énfasis debe hacerse en la protección de los árboles de gran tamaño que pueden desempeñar funciones centrales en la conectividad mediada por polen dentro de los rodales y a través del paisaje. Además, poseen múltiples ventajas a nivel de fitness ya que exhiben una mayor éxito reproductivo tanto masculino como femenino, no solo en términos de cantidad sino muy posiblemente en términos de calidad de sus progenies. CAPÍTULO 1 Introducción  | CAPÍTULO 12 Look deep into nature and you will understand everything better (A. Einstein 1879-1955) Few there are [...] who seem to clearly realize how broad a lesson on the life-his- tory of plants is written in the trees that make the great forest regions of the world. (C. B. Clarke 1894) INTRODUCCIÓN | 3 LA RESILIENCIA DE LAS POBLACIONES ANTE EL CAMBIO CLIMÁTICO El cambio climático actual impone alteraciones en el rango de distribución de las especies (IPCC 2014), determinadas en gran medida por la reacción de las poblaciones en los márgenes donde las condiciones ecológicas son próximas a su límite de tolerancia (Parmesan & Yohe 2003). La respuesta local de las poblaciones se define por su capacidad de amortiguar las variaciones ambientales in situ a través de reacciones plásticas o adaptación genética, o mediante movimientos geográficos “persiguiendo” las condiciones óptimas (Aitken et al. 2008). De este modo, el futuro de cada población dependerá de la interacción de una serie de procesos demográficos y evolutivos determinados por la sensibilidad intrínseca de cada especie (ej. rasgos de la historia de vida, límites fisiológicos, etc.) y el grado de cambio climático al que se enfrenta (conocido como “exposición”) (Dawson et al. 2011, Moritz & Agudo 2013). En general, se prevé una respuesta asimétrica en los márgenes frente al cambio climático distinguiéndose un límite septentrional con potencial para la expansión (leading edge) y un límite meridional vulnerable a la extinción de poblaciones locales (rear edge) (Hampe & Petit 2005, Parmesan 2006) (Figura 1.1). Sin embargo, esta tendencia global todavía no ha sido demostrada empíricamente. Además, existe una falta de conocimiento acerca de los procesos biológicos implicados en la respuesta de las poblaciones que limita la capacidad de predecir futuros cambios de rango (Gavin et al. 2014). En el margen de retaguardia (rear edge), las dinámicas de resiliencia (i.e. capacidad de mantener poblaciones viables en situaciones de estrés ambiental) adquieren un papel primordial. Una importante consideración aquí son las áreas topográficamente complejas que sirven y han servido como refugio climático a escala local (lo que se denomina microrefugios) al conservar un clima que se está perdiendo o ya ha desaparecido a escala regional (climas relictos sensu Dobrowski 2011) (Figura 1.2). Presumiblemente, estos refugios pueden funcionar facilitando la migración (stepping-stones), la persistencia a largo plazo (microrefugios) o sosteniendo poblaciones por un tiempo limitado antes de su extinción (holdouts) (Hannah et al. 2014). Se ha reconocido una gran importancia de las poblaciones en refugios (a las que nos referimos como “poblaciones relictas” en la presente tesis) como reservorios de la diversidad genética, la historia evolutiva y el potencial adaptativo cuya pérdida sería irremplazable (Hampe & Petit 2005, Petit et al. 2003) (Figura 1.2). Constituyen, además, valiosos laboratorios naturales para el estudio ecológico y microevolutivo de los procesos que permiten la persistencia de las especies en ambientes abióticos adversos (Woolbright et al. 2014). Gran cantidad de estudios previos se han centrado en caracterizar e identificar refugios que permitan predecir con mayor precisión el impacto del cambio  | CAPÍTULO 14 climático en la distribución de las especies (Ashcroft et al. 2009, Gavin et al. 2014, Storlie et al. 2014), y en última instancia contribuir a la conservación de la biodiversidad (Keppel et al. 2012). Sin embargo, existe un gran desconocimiento sobre el funcionamiento de las poblaciones marginales a pesar de suponer una información clave para predecir su respuesta en el futuro (Thuiller et al. 2008). Figura 1.1. Procesos que emergen durante el cambio de rango direccional para una especie hipotética. Las barras rojas en la izquierda son una representación cuantitativa de la edad y la diversidad genética dentro y entre poblaciones. A ambos lados del mapa de distribución se resumen los procesos dominantes más probables a escala de población (izquierda), comunidad y ecosistema (derecha) en respuesta al cambio climático. A lo largo de un gradiente latitudinal se ilustran el denominado “borde rezagado” reciente y los antiguos y estables relictos climáticos. No obstante, la distribución actual de los relictos climáticos de diferentes edades probablemente varían con la topografía y otras características ambientales (Woolbright et al. 2014) Adaptado de Willis & Birks (2006) y Hampe & Petit (2005). INTRODUCCIÓN | 5 LAS POBLACIONES RELICTAS DE ÁRBOLES COMO MODELO DE RESILIENCIA El impacto del cambio climático a escala global va a depender en gran medida de la capacidad de respuesta de los árboles y los ecosistemas que sustentan (Ozanne et al. 2003, Petit et al. 2004, 2005, Aitken et al. 2008). Los árboles son organismos de un tamaño y una longevidad excepcional que representan iconos de estabilidad y resiliencia entre los seres vivos (Morris et al. 2008). No obstante, sus rangos han sido y continúan siendo extremadamente dinámicos (Petit et al. 2008). El registro paleoecológico (la combinación de evidencias fósiles y filogeografía) ha revelado cambios a gran escala en la distribución de especies arbóreas como respuesta a las oscilaciones climáticas de la transición Pleistoceno-Ho- loceno, pero a su vez sugiere una capacidad robusta (resiliencia) al cambio (Hamrick 2004, Rodríguez-Sánchez et al. 2010). En la actualidad, muchas de las especies de árboles de climas templados y boreales cuentan con poblaciones a lo largo del límite meridional de su distribución que han sido capaces de persistir in situ desde el Cuaternario (Bennett et al. 1991, Gavin et al. 2014), restringidos a islas de hábitat favorable (i.e. microrefugios) más allá de los márgenes del rango de distribución de las especies (Hampe & Jump 2011). Generalmente, configuran una distribución en núcleos pequeños y altamente dispersos a través de un paisaje heterogéneo con profundas consecuencias para los procesos demográficos y genéticos que determinan su viabilidad a largo plazo. La persistencia en estas condiciones depende de estrategias que maximicen el tamaño efectivo poblacional, y minimicen los efectos Allee y la depresión por endogamia. De este modo, los procesos de flujo génico y conectividad a distintas escalas adquieren gran relevancia, junto a mecanismos que promuevan una reproducción exitosa (Ghazoul 2005). No debemos olvidar que en estos refugios las limitaciones climáticas se suman a las restricciones que impone el reducido tamaño poblacional y el aislamiento geográfico, determinando en su conjunto el balance de los procesos ecológicos y evolutivos que experimentan las poblaciones y en última instancia su vulnerabilidad a la extinción (Williams et al. 2008). Existe una gran necesidad por entender los mecanismos que favorecen la viabilidad a largo plazo de poblaciones relictas, pues supone una valiosa información a ser integrada en los planes de manejo que fomenten su conservación (Morelli et al. 2016) y en la predicción de la respuesta que especies similares puedan tener en el futuro.  | CAPÍTULO 16 DIVERSIDAD Y FLUJO GENÉTICO EN POBLACIONES RELICTAS Diversos rasgos relacionados con la historia de vida hacen a los árboles organismos particularmente resistentes a la erosión de la diversidad genética poblacional, incluyendo su longevidad, una prodigiosa fecundidad y su capacidad para dispersar polen y semillas a larga distancia (Petit & Hampe 2006). Existen estudios que demuestran la dispersión anemófila de polen en especies forestales a través de grandes distancias (revisión en Kremer et al. 2012), algunos de ellos en poblaciones de árboles pequeñas y alejadas de núcleos densos de población (ej. Buschbom et al. 2011, Robledo-Arnuncio 2011, Lesser & Jackson 2013). No obstante, en poblaciones relictas las condiciones de aislamiento prolongado y limitado reclutamiento a menudo se imponen al elevado potencial dispersivo propio de las especies arbóreas, dando como resultado diversidades genéticas reducidas dentro de la población (Petit et al. 2003), aunque a escala regional podamos observar valores excepcionalmente altos consecuencia de la diferenciación genética entre poblaciones (Petit et al. 2003, Hampe & Petit 2005). Figura 1.2 Ejemplos de refugios geográficos asociados al ambiente físico cuya probabilidad de verse afectados por el cambio climático es menor (Morelli et al. 2016). INTRODUCCIÓN | 7 Existen evidencias contrastadas, ya descritas por Kramer et al. (2008) en “la paradoja de la genética de los bosques fragmentados”, que demuestran una gran variabilidad en los patrones de diversidad y flujo génico de las poblaciones. Particularidades de cada fenómeno de fragmentación, ya sea natural o antropogénico, reciente o antiguo, relacionadas con su historia demográfica (e.g. cuellos de botella) y los procesos evolutivos que han experimentado (e.g. selección natural, deriva génica…) son clave para comprender los patrones observados. Así mismo las diferencias en la capacidad dispersiva de las especies atribuible a rasgos de la historia de vida como características del polen o la capacidad productora de propágulos (Duminil et al. 2007) generan patrones diversos. Mas allá de los numerosos factores que generalmente afectan la dispersión efectiva de polen (como el tamaño y la viabilidad del grano de polen (Bohrerova et al. 2009, Lindgren et al. 1995, Niklas 1985), las dinámicas de los vientos predominantes (Millerón et al. 2012, Nathan & Katul 2005) y los tipos de vectores de dispersión); en áreas de refugio el flujo génico a distancia puede verse especialmente limitado por: (i) el grado de aislamiento (distancia entre parejas potenciales dentro y entre núcleos de población), (ii) la ausencia de grandes poblaciones actuando como fuente de propágulos, (iii) las características del paisaje relacionadas con una topografía abrupta, una desarrollada cobertura vegetal o un microclima húmedo, y (iv) diferencias fenológicas (Trakhtenbrot et al. 2005, Nathan et al. 2008), entre otros. El flujo génico en poblaciones relictas, aunque previsiblemente sea escaso, puede tener un efecto magnificado sobre la diversidad genética poblacional al producirse una intensa purga genética de individuos consanguíneos durante etapas sucesivas del reclutamiento (Hampe et al. 2013). La acción continuada de los procesos de purga en pequeñas poblaciones fragmentadas desde antiguo probablemente hayan minimizado problemas de incompatibilidad y depresión por endogamia asociados al cruzamiento entre vecinos, que amenazan poblaciones de aislamiento reciente. De cualquier modo, la conectividad y un mínimo número de individuos en estas poblaciones de árboles parecen clave para su mantenimiento a largo plazo (Sexton et al. 2011), más aún cuando el tamaño efectivo poblacional puede estar muy reducido debido a la gran desigualdad en la fecundidad de árboles individuales (Oddou-Muratorio et al. 2005, Moran & Clark 2012). El estudio del flujo génico a distintas escalas en poblaciones relictas es vital ya que constituye un elemento clave para garantizar la diversidad genética, la reproducción y en último término el potencial evolutivo de estas poblaciones."  | CAPÍTULO 114 Figura 1.3 Aspecto de las poblaciones relictas de Q. robur en el refugio del valle del Jerte. A: Panorámica del valle del Jerte desde el Puerto de Tornavacas (1.275 m). B: Barranco de Valdeinfierno (rodal K) donde Q. robur crece inmerso en una matriz dominada por Q. pyrenaica. C: Detalle de la ribera donde crece Q. robur acompañado de alisos, fresnos, sauces, etc. En la parte superior sobresale la copa de un tono verde vivo de uno de los ejemplares. D: Aspecto encajonado del barranco de Nogaledas (rodal J) donde se aprecia la densidad de la cubierta forestal y su estructura. A B D C INTRODUCCIÓN | 15 Figura 1.4 A: Porte de un individuo de Q. robur con un desarrollo foliar incipiente. B: Detalle de las flores masculinas maduras en amentos colgantes, verdes-amarillentos, que nacen solitarios o en grupos de ramillas del año anterior. C: Detalle de las hojas simples, lobuladas y lampiñas de color verde intenso por el haz, y las bellotas colgantes de un largo pedúnculo (rasgo que motiva su nombre común). D: Detalle de las hojas caducas y las bellotas caídas en octubre. D: Detalle de una bellota de Q. robur germinada en la orilla de un arroyo. A C E B D  | CAPÍTULO 116 Figura 1.5 A: Expansión foliar en el robledal de Valdeinfierno durante el mes de abril. Destaca la regeneración de Q. pyrenaica (roble melojo) en este área donde se ha producido el abandono de los cultivos ante su declaración como espacio natural protegido. B: Detalle de la fenología de ambas especies de robles, donde se hace patente la fenología más temprana de Q. robur respecto a la de Q. pyrenaica. CAPÍTULO 2 Asymmetry in marginal population ! performance foreshadows widespread ! species range shifts Pulido, F., Castagneyrol, B., Rodríguez-Sánchez, F., Moracho, E., Cáceres, J., Pardo, A., Kollmann, J., Valladares, F., Ehrlén, J., Jump, A., Svenning, JC., & Hampe, A. In review, Febrero 2019."  | CAPÍTULO 218 PERFORMANCE IN MARGINAL POPULATIONS | 19 ABSTRACT While current climate change is altering the distribution of species worldwide (IPCC 2014), a poor understanding of the mechanisms involved limits our ability to predict future range dynamics. Range shifts are expected to occur when populations at one range margin perform better than those at the other margin (Svenning & Sandel 2013), yet no such global trend has been demonstrated empirically. Here we show that populations at high-latitude range margins generally perform as well as those from the range centre, whereas populations at low-latitude margins perform markedly worse. The trend is moderate but pervasive across plants and animals and terrestrial and marine environments. Such asymmetry in performance between range edges has not been reported before and signals that species are in disequilibrium with current environmental conditions. Our findings are consistent with predicted impacts of a warming climate and imply that the geographic ranges of species are undergoing directional changes. They highlight the pressing need for a more thorough knowledge of population dynamics across species ranges as a mean to forecast climate change impacts on the structure and function of ecosystems across the globe.  | CAPÍTULO 220 ngoing climate changes are anticipated to result in major impacts on life on Earth (IPCC 2014). As these changes continues, increasing mismatches between current conditions and the climate to which populations are adapted are predicted to create range-wide asymmetry in population growth rates (Svenning & Sandel 2013), with positive rates at expanding and negative rates at contracting species range edges. Such an asymmetry has been hypothesized to be the principal biological mechanism driving largescale geographical range shifts (Parmesan et al. 1999, Sexton et al. 2009, Lenoir & Svenning 2015). Yet we do not know how widespread asymmetries are globally. Although population growth rates are difficult to monitor directly, they are the result of demographic processes, such as survival and fecundity, which are possible to observe. Quantifying the global extent of asymmetry in measures of population performance should allow us to assess existing disequilibrium of species ranges with climate and hence the propensity of species to shift their range. Such knowledge is crucial to accurately forecast future climatedriven range shifts (Dullinger et al. 2012, Normand et al. 2013) and changes in ecosystem functioning, and for informing resource use and conservation planning. Changes in the performance of marginal populations should represent a much more direct and immediate indicator of species’ response to climate warming than the more widely monitored distribution changes (Chen et al. 2011). Although the implicit assumption of most studies on range shifts is that climate, either directly or indirectly, is a major determinant of species ranges, range limits can also be triggered by non-climatic constraints, such as habitat availability, dispersal limitation, or, notably, biotic interactions (Louthan et al. 2015), and it is a matter of intense debate to which extent range limits reflect different constraints (Hargreaves et al. 2014, Lee-Yaw et al. 2016, Pironon et al. 2017). Even when range limits are directly determined by contemporary climate, its effects on population dynamics might difficult to detect except in meteorologically extreme years. Detailed real-time observations of marginal population dynamics remain rare, especially for contracting range margins (Hill et al. 2011, Pearce-Higgins et al. 2015). The scant empirical evidence currently prevents wide-ranging comparisons of population dynamics at expanding and retreating range edges. Here, we use the abundant empirical literature spawned by the so-called centreperiphery (CP) paradigm to examine differences in performance between range centres and high- and low-latitude margins for a wide range of taxa. This paradigm states that the size, density and long-term growth rate of populations tend to decrease from the centre towards the periphery of the range as environmental conditions become increasingly less O PERFORMANCE IN MARGINAL POPULATIONS | 21 favourable (Brown 1984, Sagarin et al. 2006, Sexton et al. 2009) (Fig. 2.1). The CP paradigm has spawned hundreds of empirical studies that have compared numerous indicators of population performance (including measures of individual survival or fecundity, population viability and others) in central and marginal populations. We use a comprehensive sample of published studies to compare measures of population performance in sites located at the centre and at the high-latitude margins (HLM) or lowlatitude margins (LLM) of species ranges. We predict that if impacts of ongoing climate change on population performance are widespread, then HLM populations should perform as well as central populations whereas LLM populations should perform worse (Fig. 2.1). To test this prediction, we quantify the empirical support for this hypothesized asymmetry in the performance of HLM and LLM populations compared to central populations, and test if patterns are consistent across taxonomic kingdoms (plants vs. animals) and habitats (marine vs. terrestrial). We also predict that if climate is a real driver of population performance, then the difference in population performance should increase with the difference in climate between central and marginal populations (Fig. 2.1). To test this Figure 2.1 The centre-periphery hypothesis postulates that population performance is maximal around the range centre and decreases towards the margins of the distribution range, as environments become less suitable. Under current climate change, the optimal climate zones would displace polewards so that high-latitude populations (HLM) would increase their fitness whereas low-latitude populations (LLM) would experiment a decrease.  | CAPÍTULO 222 prediction, we relate the observed differences in performance between central and peripheral populations with the actual differences in climate. We searched the scientific literature for peer-reviewed publications published by 23rd May 2017 using keywords related to CP comparisons of population performance, retaining papers that provided data for at least two populations from the range centre and two populations from one latitudinal range margin (HLM or LLM), respectively, each occurring in the species’ natural environment. We only considered primary papers reporting demographic performance metrics that could clearly be identified as estimators of individual fecundity, individual survival, or lifetime fitness. We identified a total of 96 CP comparisons involving 623 populations in the 42 papers that fulfilled our criteria (HLM: n = 58, LLM: n = 38; see Methods and Appendix A). We conducted a mixed-effects meta-analysis on all those studies reporting information that allowed the calculation of effect sizes (i.e., mean, standard deviation and sample size). We used Hedges' d effect sizes for a standardized comparison of performance in central Figure 2.2 Observed differences in performance (Hedges’d effect sizes) between marginal (highlatitude, HLM, and low-latitude, LLM) and central populations across all species and studies. (A) Grand mean (grey) and margin-specific (blue and orange) combined effect sizes. Error bars represent 95% confidence intervals. Numbers in parentheses correspond to the number of case studies. (B) Individual effect sizes for HLM and LLM case studies ranked from the lowest to the highest value. Dot size is proportional to the weight of individual effect sizes in the meta-analysis. Both in (A) and (B), positive and negative values indicate higher and lower performances in marginal than in central populations, respectively. Horizontal dashed lines represent the null hypothesis of no difference in the performance of central vs. marginal populations. PERFORMANCE IN MARGINAL POPULATIONS | 23 Grand mean effect size (± 95% CI) was negative (-0.37; 95% CI: -0.71, -0.04), meaning that marginal populations on average performed worse than central populations. There was a significant amount of total heterogeneity, with 61% of it arising from amongstudy heterogeneity (τ² = 1.49, I² = 0.61, QE = 289.27, P < 0.0001). Performance declined from the range centre towards the LLM (-1.07; 95% CI: -1.67, -0.47), estimated from the model with Margin as the sole moderator but not towards the HLM (-0.14; 95% CI: -0.64, 0.36) (Fig. 2.2). Thus, HLM populations showed overall similar performance as central populations. Margin type was the most important moderator (wH = 0.96) but the best model only explained 4% of total heterogeneity (HLM-LLM difference: z = -2.69, P = 0.007). Residual heterogeneity (best model: QE = 260.63, P < 0.0001) was neither explained by habitat (wH = 0.69, difference between marine and terrestrial habitats in the best model: z = –1.55, P = 0.121) nor kingdom (wH = 0.54; difference between animals and plants in the best model: z = 1.33, P = 0.184) (Appendix B: table B1; Appendix C: figure C1). The differences in performance between marginal and central populations were moderately but significantly related to the difference in their average temperature in the period 1990–2013 (Fig. 2.3; Appendix F: table F1; total deviance explained by an additive mixed model: 24.9%). As expected, performance decreased with increasingly departing temperatures from central populations, while the decline was stronger in LLM than in HLM populations. Thus, HLM populations 5 ºC colder than central populations have similar fitness, whereas 5 °C warmer LLM populations perform worse (Fig. 2.3). These differences in performance were not related to geographical distance between marginal and central populations (Appendix F: figure F1). Overall, our study revealed that populations from the centre of the range tend to outperform those residing at the LLM but not those at the HLM. Albeit indirectly, this implies that HLM populations tend to outperform LLM populations. Such latitudinal asymmetry is predicted when the environmental conditions relevant for population performance are directionally displaced (Fig. 2.1) (Lenoir & Svenning 2015). Global warming has provoked a rapid large-scale poleward displacement of climatic zones since the 1970s, and the trend is predicted to further accelerate through the coming decades (IPCC 2013). Changes in population performance in response to environmental or climate change are expected to occur with no or only short time lags, and the observed difference is therefore likely to largely result from ongoing climate change (although we cannot exclude effects of changes in factors unrelated with current climate) (Normand et al. 2011, Hargreaves et al. 2014, Louthan et al. 2015). Hence, demographic rates could act as early warning signals of impending range shifts. CAPÍTULO 3 Unusually limited pollen dispersal and ! connectivity of Pedunculate oak (Quercus ! robur) refugial populations at the species’ southern range margin Moracho, E., Moreno, G., Jordano, P. & Hampe, A. (2016) Molecular Ecology, 25, 3319-3331. Doi: 10.1111/mec.13692 "  | CAPÍTULO 332 POLLEN DISPERSAL AT REFUGIUM | 33 ABSTRACT Low-latitudinal range margins of temperate and boreal plant species typically consist of scattered populations that persist locally in microrefugia. It remains poorly understood how their refugial habitats affect patterns of gene flow and connectivity, key components for their long-term viability and evolution. We examine landscapescale patterns of historical and contemporary gene flow in refugial populations of the widespread European forest tree Pedunculate oak (Quercus robur) at the species’ southwestern range margin. We sampled all adult trees (n = 135) growing in a 20 km long valley and genotyped 724 acorns from 72 mother trees at 17 microsatellite loci. The ten oak stands that we identified were highly differentiated and formed four distinct genetic clusters, despite sporadic historical dispersal being detectable. By far most contemporary pollination occurred within stands, either between local mates (85.6%) or through selfing (6.8%). Pollen exchange between stands (2.6%) was remarkably rare given their relative proximity and was complemented by longdistance pollen immigration (4.4%) and hybridization with the locally abundant Quercus pyrenaica (0.6%). The frequency of between-stand mating events decreased with increasing size and spatial isolation of stands. Overall, our results reveal outstandingly little long-distance gene flow for a wind-pollinated tree species. We argue that the distinct landscape characteristics of oaks’ refugial habitats, with a combination of a rugged topography, dense vegetation and humid microclimate, are likely to increase plant survival but to hamper effective long-distance pollen dispersal. Moreover, local mating might be favoured by high tree compatibility resulting from genetic purging in these long-term relict populations. KEYWORDS: genetic differentiation, genetic diversity, landscape scale, marginal populations, paternity analysis, rear edge  | CAPÍTULO 334 any temperate and boreal plant species have populations along the low-latitudinal limit of their current distribution range that have persisted roughly in situ through the numerous major climate transitions of the Quaternary (Gavin et al. 2014). These so-called rear-edge populations are important targets for the preservation of the genetic diversity, phylogenetic history and evolutionary potential of species (Hampe & Petit 2005). They are, moreover, excellent models for studying local adaptation under natural conditions and for understanding the processes that help tree populations to successfully persist in an adverse abiotic environment (Woolbright et al. 2014). Rear-edge populations typically survive locally in microrefugia: areas with climatic conditions that have existed for an extended period of time but are currently rare within their surroundings (‘relict climates’ sensu Dobrowski 2010). Hence, populations are typically small, highly scattered and restricted to particularly favourable habitats within heterogeneous landscapes. It remains poorly understood which ecological mechanisms have enabled them to thrive under the constraints of their climate-driven long-term fragmentation (Hampe & Jump 2011). In particular, we know little about how their particular relictual distribution affects patterns of gene flow and population connectivity, a key component for their genetic diversity, longterm resilience and evolution. Trees share life history characteristics that render them particularly resistant to the erosion of population genetic diversity, including a long lifespan, a prodigious fecundity and a great propensity for long-distance gene dispersal (Petit & Hampe 2006). A rapidly accumulating body of evidence indicates that pollenmediated gene flow between tree populations regularly spans various kilometres (reviewed in Kremer et al. 2012). In particular, in wind-pollinated species, small and geographically isolated tree populations commonly experience non-negligible amounts of mating events with immigrant pollen arriving over great distances (e.g. Buschbom et al. 2011; Robledo-Arnuncio 2011; Lesser & Jackson 2013). This pollen-mediated gene immigration can be further exacerbated through the purging of inbred individuals during the subsequent recruit-ment process (Hampe et al. 2013). On the other hand, tree populations also are more dependent on efficient gene exchange than other plants because they are mostly self-incompatible and disproportionately susceptible to inbreeding depression (Petit & Hampe 2006). In addition, the effective population size of trees is often much smaller than apparent in the field because of great inequality in fecundity, with a few individuals contributing very many offspring and the great majority little or nothing (Oddou-Muratorio et al. 2005). Hence, a minimum size and connectivity of relict populations seem crucial for their fitness and long-term persistence in a changing environment (Sexton et al. 2011). M POLLEN DISPERSAL AT REFUGIUM | 35 Relict tree populations at the rear edge of species ranges share many features with other small and isolated tree populations. However, they also assemble some specific characteristics that could render them particularly prone to experiencing lower levels of effective pollen flow. First, they typically grow far away (i.e. tens to hundreds of kilometres) from large pollen sources such as extensive populations belonging to the continuous distribution range. Second, their continued persistence at relatively low population size implies that they probably have undergone extensive genetic purging and may be suffering less from incompatibility and inbreeding depression than more recently isolated populations (such as pioneer stands at the leading edge or those resulting from recent anthropogenic fragmentation). Third, their particular relictual habitats usually are located in topographic settings that help maintain a constant minimum humidity (Hampe & Jump 2011), such as shady valleys, gorges or ravines; these habitats often occur in rugged terrain and contain a dense vegetation that tends to represent an obstacle to long-distance pollen flow (Damschen et al. 2014; Shohami & Nathan 2014). Hence, relict populations residing at the rear edge of species ranges could experience less long-distance gene flow than many other tree populations that have experienced fragmentation and isolation as a consequence of human activity (Kramer et al. 2008). Yet we still have a limited understanding of how landscape complexity combines with individual tree traits to result in the distinct fecundity patterns of relictual scenarios (Bacles & Jump 2011). One of the most direct and powerful methods for studying the movement of genes within and among small, isolated populations consists in performing population-specific paternity tests. The precise detection of male parents and their spatial position allows to retrace pollen movements, providing information about the spatial patterns of mating across complex landscapes, and the potential effects of the landscape context on spatial genetic structure and contemporary gene flow (Klein et al. 2011). Documenting these patterns across networks of tree stands connected at a landscape scale can unveil demographic and environmental correlates of gene flow patterns (Dyer & Nason 2004; Cheptou & Schoen 2007; DiLeo et al. 2014) and ultimately mechanisms of isolation associated to their relictual situation. Finally, complementing these analyses with an in-depth assessment of historical gene flow patterns as reflected by the adult tree populations can reveal how consistent through time and hence biologically relevant observed patterns of pollen flow are. Here, we use a particularly suited model system to examine landscape-scale patterns of historical and contemporary gene flow and connectivity in relict tree populations residing at the southwestern range limit of the species. This particular setting provides a rare opportunity for comprehensively assessing landscape-scale mating patterns in a major European forest tree. Specifically, we (i) describe the genetic structure of adult trees and patterns of historical gene flow based on an exhaustive sampling of individuals, (ii) test hypotheses  | CAPÍTULO 336 0 4 Km Jerte river N I J ML K E F C D N 0.00 0.25 0.50 0.75 1.00 N I JML K EF CD Ancestry proportion A B C Figure 3.1 (a) Distribution area of Quercus robur according to EUFORGEN with a black dot indicating the study area. (b) Map of target stands. Shading indicates altitudinal ranges in 500 m steps from <500 m asl (white) to >2000 m asl (dark grey). Study stands are indicated by circles with circle size indicating the number of adult trees and circle colour the dominant genetic cluster as identified by STRUCTURE. Dark grey lines indicate significant genetic covariance (0.01 ≤ α ≤ 0.1) between stands according to Population Graph analysis and their thickness is proportional to the cGD value, i.e. the genetic similarity of stands. (c) Bar plots showing STRUCTURE ancestry proportions for K = 4 clusters. Each individual is represented as a line segment, which is vertically partitioned with different colours representing the individual’s estimated proportions of ancestry in each cluster (cluster 1 in yellow; 2 in red; 3 in green and 4 in blue). Letters below the graph refer to stands as shown in plate b. POLLEN DISPERSAL AT REFUGIUM | 37 about contemporary pollen flow limitation within and among stands as well as limited pollen immigration (either through long-distance pollen dispersal or through hybridization with the locally abundant sister species Quercus pyrenaica), and (iii) test the effects of stand characteristics (size, geographic isolation and phenology) on patterns of within and among-stand gene exchange. METHODS Study species Pedunculate oak (Quercus robur L.) is one of the most prominent and widely distributed European forest tree species. Its southwestern range margin is located in the mountain ranges of central Spain, where scattered populations occur along water courses and in other environments that mitigate the summer drought characteristic of the regional Mediterranean climate. Quercus robur is an almost exclusively outcrossing, wind-pollinated species. Extensive long-distance pollen flow into highly disjunct stands has been documented in this and other oak species (e.g. Buschbom et al. 2011; Hampe et al. 2013; Gerber et al. 2014). In the study area, it flowers from early to late April and acorns ripen in late September and early October. Hybridization with the sister species Pyrenean oak (Quercus pyrenaica) is possible due to partial overlap of flowering periods (Lepais et al. 2009). Quercus robur shows alternate fruit bearing in the area (G. Moreno, unpublished) as it does in many other parts of its range. Study area and field sampling The study was conducted in the Jerte valley (40° 13’ N, 5° 44’ W; Caceres province), ca. 25 km northeast of the town of Plasencia in western Spain. The valley stretches almost linearly over ca. 30 km descending from an elevation of 1275 m a.s.l. at the header (Fig. 3.1a,b). Its relatively temperate and humid local climate has favoured the development of an extensive fruticulture in the smoother, more accessible slopes while steeper and higher slopes are covered by broadleaved forests dominated by Q. pyrenaica. Streamsides and gorges harbour mesic and riparian tree species including Q. robur, Celtis australis, Castanea sativa, Fraxinus angustifolia and Alnus glutinosa among others. Our study benefited from a comprehensive survey of adult and close to adult Q. robur trees throughout the Jerte valley conducted between 2002 and 2003 (Pulido et al. 2007). The species’ earlier budburst and later shedding of leaves allows, during certain periods of the year, spotting and distinguishing individuals over hundreds of metres from the far more abundant sister species Q. pyrenaica. The survey detected a total of ten small Q. robur stands (n = 2–35 adult trees) spread over ca. 20 km of the Jerte valley (Fig. 3.1b). Somew-  | CAPÍTULO 338 hat larger populations (i.e. with a few hundred trees) are located at ca. 10–15 km, although outside the valley and therefore separated by a mountain chain with an altitudinal differential of up to 1000 m. We mapped all adult and close to adult Q. robur trees known to occur in the Jerte valley (n = 135) and collected several leaves from each tree that were stored in silica gel until genetic analyses. We collected up to 20 acorns (mean: 19.0) in October 2009 from or, if impossible, beneath the canopy of 72 trees from all ten stands (i.e. virtually all individuals reproducing this year in the Jerte valley). Acorns were weighted, frozen and stored until being processed. We extracted the embryo from each acorn and dried it for genotyping. A pilot analysis of acorns from 10 randomly chosen mother trees using genetic diversity accumulation curves of multilocus genotypes indicated that a sample of 8–11 acorns per mother adequately describes the genetic diversity of seed families (see Appendix G: Fig. G1). We hence decided to analyse a minimum of 10 acorns per mother tree, resulting in a total of 724 acorns from 72 seed families. Finally, we characterized the flowering phenology of all adult trees in the Jerte valley on two dates somewhat before and after the peak of the flowering period, respectively (28 March 2015 and 4 April 2015). Four phenological stages of male flowers were distinguished during each survey: (i) swelling buds (score 0); (ii) emerging and immature catkins (score 1); (iii) mature catkins (score 2); and (iv) old, dried catkins (score 3). The phenological stage of the upper and the lower part of each tree was separately recorded and subsequently averaged. A similar survey in the previous year (2014) allowed us to corroborate that the phenological sequence of trees within stands remains reasonably consistent from year to year (Spearman rank correlation, rs = 0.38, P = 0.004 (see also Bacilieri et al. 1995). Microsatellite genotyping All adult trees and acorns were genotyped using the 20 nuclear microsatellite markers described in Guichoux et al. (2011), integrated in two multiplex kits of twelve and eight markers each. DNA extraction, PCR amplifications and genotype scoring with an ABI 3130xl Genetic Analyzer (Applied Biosystems, Foster City, CA, USA) were performed following the protocols described by these authors. Electropherograms were independently scored by two persons to minimize genotyping error. We actually quantified the error rate using 44 randomly chosen acorns as blind samples. Three independent extractions, amplifications and analyses of each blind sample, respectively, resulted in an estimated overall genotyping error rate of 0.85% (n = 880 scorings), thus confirming the high marker quality pointed out by Guichoux et al. (2011). Two loci (S19 and AB) did not fully amplify across all acorns and another one (Qr20) was found to strongly deviate from Hardy–Weinberg equilibrium due to allele dropout. All POLLEN DISPERSAL AT REFUGIUM | 39 three loci were discarded, leaving 17 loci for the paternity analyses (while we used all 20 loci for the analyses of adult trees). Data analysis Adult genetic diversity and relatedness The following statistics were calculated overall and for each stand using the GSTUDIO package in R version 3.1.0 (R Development Core Team 2014): mean number of alleles (Na), effective number of alleles (Nae), observed and expected heterozygosity (Ho and He) and fixation index (FIS). We tested the null hypothesis H0: FIS = 0 using 10 000 randomizations of alleles among individuals. The effective number of alleles was estimated according to Nielsen et al. (2003) to account for the unequal sample sizes among stands. Gene diversity (He) was also corrected for sample size (Nei 1978). We estimated genetic relatedness (Queller & Goodnight 1989) among all within-stand adult tree pairs with SPAGEDI 1.4 (Hardy & Vekemans 2002). The expected relatedness values (R) are 0.5 among full-sibs, 0.25 among half-sibs and 0.0 for unrelated individuals. To calculate within-stand pairwise relatedness we used the allele frequency of the total, pooled 10 stands in the valley. Adult genetic differentiation and historical migration We assessed overall and pairwise genetic differentiation among stands computing FST (Weir & Cockerham 1984) in SPAGEDI 1.4 (Hardy & Vekemans 2002). FST was tested against the null expectation of absence of population structure based on 10 000 permutations. We also tested for isolation by distance between stands by means of a Mantel test with 9999 permutations in GENALEX 6.5 (Peakall & Smouse 2012). For this purpose, we calculated the geographic distance among stands (taking their altitude into account). Bayesian clustering of the genetic data was performed using STRUCTURE v.2.3.4 (Pritchard et al. 2000). We ran STRUCTURE with K ranging from 1 to 10, and with 10 runs for each K value. A burn-in period of 50 000 iterations was followed by 200 000 MCMC repetitions assuming allele frequencies to be correlated among populations and an admixture model of population structure. No prior information was used to assist clustering in the first 10 x 10 runs, whereas we included stand identity as prior information in a second series of 10 x 10 runs. Results were highly consistent between both approaches, and we report the results including stand identity. We selected the K value that best described the data from the change in likelihood (delta K) as proposed by Evanno et al. (2005) and the highest posterior probability (LnP(D|K)). We further explored patterns of historical gene exchange between stands using Population Graphs (Dyer & Nason 2004) as implemented in the popgraph library in R version  | CAPÍTULO 346 hybridization event with Q. pyrenaica (i.e. probability of assignment to cluster ‘Q. pyrenaica’ around 0.5, see Appendix I: Table I1).On the other hand, a few of our adult Quercus robur trees show signs of being hybrids in first or second generation (see Appendix I: Fig. I1). We observed a marked directionality of pollen movements between stands (Fig. 3.3): Of the 18 events detected overall, 11 (61%) were from upstream to downstream stands, six (33%) involved stands growing nearby (<1 km) across the valley and only one long-distan- ce pollination event (5%) occurred in upstream direction. Correlates of variation in pollen dispersal among stands As revealed by the GLMM model fit, stand characteristics influenced the proportions of between-stand mating events and slightly with local pollinations (see Fig. 3.3), whereas no effect on the proportions of selfed and unassigned acorns was evident (Table 3.2). The frequency of local pollinations tended to increase with stand size, whereas that of betweenstand pollinations decreased with increasing stand size and in particular isolation. We also found a marked effect of phenology on the proportion of between-stand pollinations, with lateflowering stands receiving more immigrant pollen than early-flowering stands (Table 3.2). CDE FI J LMN 1 km K <0.01 0.01-0.05 0.05-0.10 >0.10 Figure 3.3 Contemporary pollen-mediated connectivity of the investigated Quercus robur stands. Stands are drawn as nodes with circle size being proportional to stand size. Arrows represent pollen dispersal probability from source to sink according to the proportion of between-stand pollen dispersion inferred by paternity analysis. Letters beneath circles indicate the identity of stands while horizontal lines below stand codes indicate those stands that belong to the same genetic cluster. POLLEN DISPERSAL AT REFUGIUM | 47 DISCUSSION Adult population structure and historical gene flow Our study stands showed a relatively high level of genetic diversity given their geographical isolation and relatively small population size [global HE = 0.71, compared to 0.75–0.85 (Buschbom et al. 2011), 0.79 (Hampe et al. 2010), 0.79 (Muir & Schlötterer 2005;), or 0.87 (Streiff et al. 1998) for SSR-based estimates from other Quercus robur populations]. This diversity was strongly structured, indicating that gene flow and population connectivity along the Jerte valley have been weak through an extended period of time. We observed no <4 well-supported genetic clusters and an overall FST (0.12) that clearly exceeds those of far more geographically separated oak populations from other ecological settings (e.g. Bruschi et al. 2003; Muir et al. 2004; Craft & Ashley 2007, 2010; Cavender-Bares & Pahlich 2009). Even though the FST estimate might be somewhat inflated due to the small size of some stands, our study system clearly is at odds with the widespread notion that disjunct populations of wind-pollinated temperate trees tend to experience extensive longdistance gene exchange (Kramer et al. 2008; Kremer et al. 2012). Stands of the upper and those of the lower valley actually were largely disconnected, although no apparent topographic or other landscape-related barriers to gene flow exist between the upstream and the downstream stands (except for population K, which is located in a lateral valley; see Fig. 3.1b). The observed disconnection should hence largely be driven by the mere spatial distance between the two groups (ca. 8 km). Note also, however, that the STRUCTURE analysis identified a certain number of trees (ca. 10%) that stem from between-cluster pollination or acorn dispersal events. At a finer spatial scale, the genetic differentiation of stands was not homogeneous throughout the valley but much more pronounced in the upper than in the lower part. This difference is remarkable because the two valley parts do not appreciably differ in the spatial distribution of stands, the topography or greater landscape structure. The strong genetic divergence of stands in the upper part of the valley suggests that the species must have persisted more or less independently at different sites in the valley. Such circumstances and the limited size of these stands probably have led to their rather low genetic diversity (HE = 0.61). Note that high population differentiation combined with low within-population diversity is a key characteristic of stable rear-edge populations (Hampe & Petit 2005); thus, our study stands are likely to be real long-term relicts with a history of persistence in situ through many generations. On the other hand, the relatively weak differentiation of the lower valley stands compared to their upper valley counterparts could reflect that they are remainders of a once larger and more continuous stand that has undergone shrinking and fragmentation in relatively recent time. Such a process would explain that the lower valley stands show  | CAPÍTULO 348 higher levels of diversity than their upper valley counterparts (0.67 vs. 0.61; paired t test: t = 2.42, df = 19, P = 0.03), and that most historical long-distance pollination events occurred from lower towards upper parts of the valley (i.e. from cluster 1 towards the other clusters, see Fig. 3.1c). It is well documented that human land use in the Jerte valley steadily increased in intensity through the 20th century (Cruz 1983), although the remotest horticultural areas are now being abandoned and recolonized by scrub and forest (Ezquerra & Gil 2008). A concomitant size reduction and fragmentation of Q. robur stands, especially in the more populated and readily transformable lower part of the valley, appears hence realistic. Contemporary pollen dispersal By far most mating events involved trees from the same stand (>90% when adding local pollinations and selfing). Larger stands tended to experience somewhat more local pollinations than smaller ones, although the tendency was weak. The dispersal kernel (Fig. 3.2) and the estimated median pollen dispersal distance of only 30 m underpin moreover that mating occurred primarily between neighbouring trees within each stand. We also detected a remarkably high number of selfing events for a strongly outcrossing species such as Q. robur (Bacilieri et al. 1996). These stemmed mostly from a few heavily selfing trees (i.e. five individuals accounted for more than half of all selfed offspring), suggesting that breakdown of self-incompatibility has occurred several times in our study system (see Hampe et al. 2013 for a similar case in a marginal oak population). Overall, the probability to receive pollen from other stands within the valley decreased with increasing size of stands and especially with their geographical isolation, whereas it increased from early- to late-flowering stands. The two lowermost stands I and N did not participate at all in the observed pollen flow, whereas the central stand D exchanged pollen with no <6 other stands. The small and late-flowering stands L and M served as sinks for 39% of all observed between-stand pollination events, whereas the more isolated, earlyflowering stands J and C served exclusively as pollen sources (see Fig. 3.3). Overall, these patterns point to great spatial heterogeneity in the network of pollen gene flow. The detected phenology effect indicates that between-stand mating tends to increase with decreasing pollen abundance, suggesting that competition for access to ovules could to some extent trigger the observed mating patterns (see also Lagache et al. 2013). Interestingly, patterns of contemporary pollen flow showed two major inconsistencies with those of historical gene flow as reflected by the adult trees. First, we detected several instances of ongoing pollen dispersal across the major genetic disconnection between upper valley and lower valley stands (see Figs 3.1b and 3.3). Second, contemporary pollen movements occurred markedly more often downstream than upstream, whereas the STRUCTURE analysis of the adult trees indicated that historical pollen flow has primarily occurred POLLEN DISPERSAL AT REFUGIUM | 49 in the opposite direction (from cluster 1 towards the other clusters). The directionality of contemporary pollen flow is in agreement with the dominant wind direction during the flowering season of our study year (Sistema de Informacion Agroclimatica para el Regadío, http://eportal.- magrama.gob.es/websiar/Inicio.aspx). We do not know whether our study year was an exception and upstream winds have been historically prevailing in the Jerte valley. But the dominance of upstream pollen dispersal in the past would be in line with our hypothesis that Q. robur could have been more abundant in the lower Jerte valley during past decades than it is today. Pollen inflow from other sources than our target stands (i.e. either Q. robur populations outside the Jerte valley or Quercus pyrenaica) was also infrequent. Even the smallest and most isolated stands did not show any signs of being particularly susceptible to experiencing long-distance gene inflow or hybridization. Our STRUCTURE analysis indicated that only a tiny fraction of all mating events (0.6%) involves hybridization with Q. pyrenaica. Given the great abundance of this species in the Jerte valley, our finding implies that interspecific gene flow played a very minor role. We observed, however, signs of hybridization in some of our adult Q. robur trees, which are roughly in line with the hybridization rate of 5.9% that Lepais et al. (2009) reported from a mixed oak stand with Q. robur and Q. pyrenaica. This discrepancy suggests that hybridization might be more frequent in our study system in years of less abundant flowering and concomitant stronger pollen limitation (cf. Lepais et al. 2009; Lagache et al. 2013). Mechanisms behind restricted gene flow in relict Quercus robur stands We observed only marginal effects of isolation by distance in explaining genetic differentiation; thus, other processes must be triggering levels of gene flow in addition to simple geographic distance effects. Quercus robur stands in the Jerte valley exchanged little pollen, and the marked genetic structure of the adult population clearly indicates that this weak connectivity is not only a short-term phenomenon. It is particularly remarkable since oaks and other wind-pollinated forest tree species are known for their great ability to disseminate pollen over great distances (Schueler & Schlünzen 2006; Kremer et al. 2012). For instance, Hampe et al. (2013) observed 6% of pollen inflow into a small, marginal holm oak (Quercus ilex) population 30 km ahead of the nearest larger stands. Buschbom et al. (2011) reported that no <35% of all mating events involved immigrant pollen in a small Q. robur stand whose closest conspecifics grow >80 km away. And Robledo- Arnun- cio & Gil (2005) detected 4.3% of pollen immigration in a Pinus sylvestris stand 30 km apart from other populations. So why is gene flow so much more restricted in our study system? At least two mechanisms arising from the relictual character of our stands could help explain the phenomenon. The first is related to their refugial environment: almost all stands  | CAPÍTULO 350 grow within dense riparian forests along water courses that are surrounded by rugged terrain (typically gorges). This setting provides the humidity required by Q. robur to withstand the dry Mediterranean summers typical of the region. But it also represents a significant obstacle to both the departure and the arrival of oak pollen. The effect of this refugial habitat is probably twofold. First, by creating abundant small-scale turbulences and weakening thermal air uplift, fine-scale topography and the dense canopy reduce the probability of pollen grains to enter free atmosphere layers where directed air flows could transport them over longer distances (cf. Dupont & Brunet 2008; Dobrowski 2010). Second, the humid environment of these riparian forests renders floating pollen likely to be ‘captured’ by sticky surfaces such as leaves or branches. This phenomenon is known from tropical rain forests and has been used to explain the scarcity of wind-pollinated species in this biome (e.g. Turner 2001). Our study suggests that it could also constrain gene flow when wind-pollinated species are confined to other humid habitats. Relict trees that persist in these refugial microenvironments may hence result environmentally ‘trapped’ in terms of dispersal limitation and realized gene flow. The second mechanism combines environmental and genetic components and refers to the observed high rates of mating between neighbours. Our study stands have apparently persisted through extended periods of time in isolation and at small population size under the constraints of their scattered refugial habitat. This situation has provided opportunities for genetic purging, while the great ability of trees to maintain within-population diversity (Petit & Hampe 2006) would have limited negative genetic effects on population regeneration and viability. This suggests that incompatibility and inbreeding depression do not profoundly hinder extensive mating within refugial populations. An abundant local pollen production, such as in the year of our study, can hence easily saturate stigmas and thwart the arrival of foreign pollen. This so-called pollen ‘swamping’ from nearby trees has been described from closed oak forests (Lagache et al. 2013) and should be even much stronger in the highly disjunct stands of the Jerte valley. On the contrary, cases of extensive pollen inflow into geographically isolated populations typically come from genetically strongly impoverished, little stands where adult trees are likely to experience compatibility issues (e.g. Buschbom et al. 2011; Hampe et al. 2013). Our findings of moderate genetic diversity and relatively low extent of family structures give support to this mechanism as a plausible cause of the extensive within-stand mating in these relict populations. In addition, two lines of evidence indicate that inbreeding depression is not a major issue in our study system: (i) the size of acorns is well known to trigger seedling establishment and is hence used as a proxy for fitness (e.g. Hampe et al. 2013); yet the size and the progeny inbreeding of our acorns were uncorrelated (Pearson r = 0.16, t = 1.38, df = 70, P = 0.17); (ii) a concomitant greenhouse experiment with >1500 seedlings revealed no differences in the POLLEN DISPERSAL AT REFUGIUM | 51 performance of individuals from small and from large stands (G. Moreno, unpublished). Our peculiar case study and its unusual results hence suggest that compatibility issues and inbreeding depression could be a key driver shaping long-distance dispersal patterns of small populations of wind-pollinated trees (see also Kremer et al. 2012). Insights into organism–environment relationships within microrefugia The functioning of microrefugia and their role for the survival of species in a rapidly changing climate are the object of rapidly growing attention (e.g. Hampe & Jump 2011; Gavin et al. 2014; Woolbright et al. 2014). Refugia offer pockets of suitable climate space where climatic conditions in the surroundings do not permit the existence of the species (Dobrowski 2010). Long-term persistence in strong isolation and resulting microevolutionary differentiation has converted many refugial populations in conservation targets of high priority (Hampe & Petit 2005). Their isolation is usually attributed to their highly scattered distribution within a climatically adverse landscape matrix. Our study suggests that not only the climatic conditions around refugia but also those within them can constrain the connectivity of the populations that they harbour. This notion adds an important piece to our understanding of organism–environment relationships and population dynamics at species’ low-latitudinal range margins. ACKNOWLEDGEMENTS We thank Juan Carlos Benito and Cristina Rigueiro for their valuable help with the field and laboratory work, and Francisco Rodríguez for his insightful statistical advice. This study was funded by an FPI predoctoral fellowship to EM and the project PERSLIM (CGL2010-18381) from the Spanish Ministerio de Economía y Competitividad, as well as a Severo Ochoa Excellence Award (SEV-2012-0262) and a Junta de Andalucía Excellence Grant (RNM-5731); by the EU ERA-NET project TipTree (BiodivERsA2012-15) and the INRA EFPA project FORADAPT. CAPÍTULO 4 Ecological drivers of individual mating ! patterns in refugial populations of ! Pedunculate oak (Quercus robur) Moracho, E., Moreno, G., Jordano, P. & Hampe, A. In Review, Febrero 2019."  | CAPÍTULO 454 INDIVIDUAL MATING PATTERNS | 55 ABSTRACT The low-latitude range margins of many temperate and boreal plant species consist of scattered populations that persist locally in microrefugia. It remains poorly understood the mechanisms that allow their long-term persistence such as the maintenance of genetic diversity and adaptive potential. We examined fine-scale patterns of mating and pollen dispersal across ten refugial stands of Pedunculate oak (Quercus robur) scattered over 20 km along a valley located at the species’ southwestern range margin. We genotyped progenies from 72 mother trees and assessed individual variation in mating using categorical paternity assignments, a spatially explicit mating model and mating network analysis. We then examined how this variation was shaped by numerous features of the tree and its ecological neighbourhood. At valley-wide scale, trees showed unusually strong variation in selfing (0-80%, average 6.8%) and immigrant pollinations (0-89%, average 7.3%). Both were most prevalent in large trees surrounded by few and genetically related neighbours. At within-stand scale, extensive among-tree variation existed in all mating parameters, but consistently more accused for the male function than the female one. Large and centrally-located trees assumed a key role for pollen-mediated connectivity. Overall, our results revealed that the character and situation of individual trees within their forest stands clearly surpass the role of the stands themselves in shaping the mating system. The strong dominance of small-scale phenomena for mating implies that even small changes in the structure of extant stands, such as the removal of key pollen donors, could have severe consequences for their conservation. KEYWORDS: individual variation, ecological and genetic correlates, gene flow, landscape scale, marginal populations, male fecundity, paternity analysis, mating network, rear edge  | CAPÍTULO 462 Tree size. We constructed a compound measure by mutiplying the dbh and the crown projection of each tree. Male flower production. Contrary to common expectations, male flowering was not very tightly related with tree size (Pearson r = 0.65). Therefore, we directly included our half-quantitative estimates of male flowering (scores 1 to 8) as a surrogate of potential fecundity. Neighbour density. This variable was estimated as the number of Q. robur trees within radius of 25 m. This radius represents the finest spatial scale that could be addressed for all extrinsic predictor variables (see below). In addition, it is close to the median pollen distance in the study area (30.6 m, Moracho et al. 2016). Neighbour genetic relatedness. We computed the mean genetic relatedness of each mother tree with its conspecifics neighbours within the 25 m radius using SPAGeDi 1.4. Topographic position. The growing site of mother trees was characterized using a topographic position index (TPI) that measures the difference between the elevation of the focal tree and the mean elevation of the surrounding terrain within a radius of 25 m. Positive TPI values indicate a tree that is located on a top while negative TPI values indicate a tree that is growing in a hollow. We used a digital elevation model with a resolution of 5 m (Instituto Geográfico Nacional, distributed by the Centro Nacional de Información Geográfica; http://centrodedescargas.cnig.es) to deduce terrain elevation. The index was computed with the function tpi in package spatialEco in R 3.1.0. Forest cover. The forest cover was estimated in a circular plot of 25 m radius around the mother tree using the Google Earth base map. We derived a semi-quantitative variable with two classes: below and above 50%. Analysis of individual variation in mating and pollen dispersal To identify intrinsic and extrinsic ecological determinants of among-tree variation in mating and pollen dispersal, we performed analyses at two complementary levels: across all stands distributed along the Jerte valley (landscape scale) and within stands (local scale). Valley-wide analysis We constructed generalized linear mixed models (GLMM) using the proportion of pollen dispersal types (selfing, local pollination and pollen immigration) as binomially distributed response variables. The variables tree size, male flower production, neighbour density, neighbour genetic relatedness, TPI and forest cover were included as fixed effects and the stand of origin as random term. First, we tested the effect of each ecological variable in the saturated model for the different pollen dispersal types. Then, we performed model selection based on the Akaike Information Criterion (AIC). To restrict the total number of models compared, we pooled our explanatory variables in the three groups described above: those intrinsic to the focal tree (I), those related with the conspecific neighbours (N) and those related with the physical environment (E). Then, all possible combinations of these three groups of variables were modelled: I, N, E, I + N, I + E, N + E, I + N + E. We also included a null model without explanatory variables in this procedure. All GLMM analyses were performed using the lme4 package in R 3.1.0. Within-stand analysis INDIVIDUAL MATING PATTERNS | 63 This analysis was performed for the four largest stands only to take full advantage of the information obtained through the spatially explicit mating model. We used canonical correlation analysis (CCA) to examine relationships between two multivariate datasets: the set of mating variables obtained from the pollination matrices estimated with MEMM and the set of ecological variables for the individual trees. Two independent CCAs addressed the maternal and the paternal component of mating patterns, respectively. The CCA for the maternal component included the number of pollen donors, the mean kinship of them, the mean genetic distance between pollen donors and the focal mother tree, the median geographic distance between mates, and their network degree centrality. The CCA for the paternal component considered the number of female mates, the median geographical distance between mates, the relative male fecundity and their degree centrality. Both analyses included moreover the previously described ecological variables tree size, male flower production, neighbour density, neighbour genetic relatedness and TPI. We did not use forest cover because of the binary nature of the variable. Instead, we also considered neighbour density within a larger distance (100 m radius) to explore the effect of this variable for more than one spatial scale. Note that neighbour densities at 25 m and at 100 m were only moderately correlated (Pearson r = 0.63). The CCA aims at generating composite variables (canonical variables) that maximize correlations among the multivariate sets of mating and ecological variables. The squared canonical correlation (R2) corresponds to the percentage of variance in the dependent set explained by the independent set of variables along a given dimension. We assessed whether this relationship is significantly different from zero by means of Bartlett's chi-squared test. We used the cca and yacca packages in R 3.1.0 for this analysis. RESULTS Individual variation in mating and pollen dispersal: valley-wide analysis By far most mating events involved a father tree located within the same forest stand (mean 84.4%, range 0 - 100%; Fig. 4.1). Similar proportions were observed for outcrossing with immigrant pollen (mean 7.6%, range 0 - 89%) and for self-fertilization (mean 6.8%, range 0 – 80%). We detected great variation among individuals. Thus, 19 of the 73 trees analysed experienced at least one selfing event and nine counted with >25% of selfed progenies (Fig. 4.1). A total of 21 trees experienced mating events with pollen from outside the forest stand and six of them displayed pollen immigration rates above 25%. Trees with high rates of selfing or pollen immigration grew across the valley without showing any aggregation in particular stands (Fig. 4.1). Individual variation in mating and pollen dispersal: within-stand analysis Table S1 in the Online Appendix provides the average estimates for mating system in each of the four target stands according to MEMM. This analysis produced an adjacency matrix describing within-population patterns of mating, in which each pair of trees was assigned an estimate of the fraction of pollen received by the mother tree from the corresponding father. The spatially explicit representation of this matrix is shown in Fig. 4.2. We generally observed high connectivity among individuals and no major gaps were present among groups of trees except for stand DF. Mating occurred most often among nearby trees. The degree of individual trees within the network was related with their size and  | CAPÍTULO 464 200m200m 200m E J K DF 100m Figure 4.2 Mating networks for the four largest Quercus robur forest stands based on the spatially explicit mating model derived with MEMM. Mating events among the mapped individuals of each stand are presented as lines with different colours and thickness according to the probability for a given mother to be sired by a given father. White < 0.05; 0.05 < light blue < 0.1; 0.1 < blue < 0.5; 0.5 < black < 1. Circle size indicates the diameter at breast height (dbh) of the tree and the circle colour represents its degree in the mating network (with darker shading indicating higher degree). INDIVIDUAL MATING PATTERNS | 65 spatial centrality within the stand (GLMM with poisson errors: β= 0.025, SE= 0.002, P < 0.0001; β= -1.280, SE= 0.291, P < 0.0001, respectively). Also, the degree was highly correlated to the number of mates for the father perspective (Pearson r = 0.73, t = 10.78, df = 105, P < 0.0001), suggesting that network connectivity is mainly due to the paternal function.All mating parameters inferred from the pollination matrix showed extensive variation among individuals, both for maternal and paternal progenies (Table 4.1). While the average number of mates was the same for maternal and paternal progenies (4.4), the inequality among pollen donors was much stronger than that among mother trees. Accordingly, MEMM estimated a high variance in male fecundity with most trees contributing little or nothing to effective pollination whilst only 4-18% of the trees in the stand showed fecundities larger than twice the mean value (Fig. S1). Ecological correlates of mating and pollen dispersal We observed a series of consistent relationships between pollination types and the quantified ecological correlates (Table 4.2). The propensity of trees to self was affected by intrinsic as well as extrinsic drivers. Selfing was most common in large trees with abundant male flower production that were surrounded by few, closely related conspecifics and grew in areas with high forest cover. The probability of local mating was likewise linked with intrinsic and extrinsic parameters. Small trees with many, little related neighbours showed the highest rates. Finally, the probability of pollination with immigrant pollen was only triggered by extrinsic variables related with the conspecific neighbourhood. Most immigration occurred on trees with few and closely related neighbours. In summary, model selection showed that neighbourhood-related variables Table 4.1 Summary of predicted mating system (maternal and paternal component) for the Quercus robur trees in the four largest forest stands of the study area (n = 107). PD refers to pollen donors and FM to female mates. Mating variables Mean (SD) Range Maternal progenies Number of PD 4.3 (1.5) 1 - 8 Genetic distance to PD 22.8 (5.2) 11.3 - 36.3 Kinship of PD 0.11 (0.07) -0.07 - 0.44 Median distance to PD (m) 100 (87) 6 - 418 Paternal progenies Number of FM 4.3 (6.9) 0 - 32 Genetic distance to FM 22.6 (5.5) 12 - 37 Kinship of FM 0.10 (0.06) -0.04 – 0.26 Median distance to FM (m) 73 (84) 0 - 405 Male fecundity 0.8 (1.8) 0 -13.6 Both Degree 4.3 (4.2) 1 - 30  | CAPÍTULO 466 Table 4.2 Parameter estimates (β, standard error and P) for the complete mixed-effect model tested for the proportion of each type of pollen dispersal: selfing, local pollination and pollen immigration. ns, not significant at P < 0.05. To test different candidate models, we pooled the explanatory variables in groups of predictors intrinsic to the tree (I), those related with the conspecific neighbours (N) and those related with the physical environment (E). All possible combinations of these 3 groups of variables plus the null model were modelled (see Table S2). ∆AIC, difference between the best and the second model; AICwt, probability that the model will be the best model in the set. A. Full model Response variable β S.E. P Selfing Size (I) 0.482 0.193 0.012 Male flower production (I) 1.052 0.332 0.002 Neighbour density (N) -1.423 0.42 <0.001 Neighbour genetic relatedness (N) 0.532 0.225 0.018 Topographic Position (E) 0.02 0.196 ns Forest cover (E) 0.785 0.244 0.001 Local pollination Size -0.676 0.205 <0.001 Male flower production -0.011 0.204 ns Neighbour density 0.821 0.249 <0.001 Neighbour genetic relatedness -0.39 0.153 0.011 Topographic Position -0.115 0.164 ns Forest cover -0.043 0.179 ns Pollen immigration Size 0.203 0.277 ns Male flower production -0.41 0.265 ns Neighbour density -0.758 0.289 0.009 Neighbour genetic relatedness 0.445 0.190 0.019 Topographic Position 0.299 0.245 ns Forest cover -0.268 0.255 ns B. Models with grouped explanatory variables Response variable Best model ∆ AIC AICwt Selfing I + N + E 1.33 0.64 Local pollination I + N 1.94 0.73 Pollen immigration N 2.19 0.63 INDIVIDUAL MATING PATTERNS | 67 consistently explained pollen dispersal patterns as the only category common to all best models (Tables 4.2, S2). The CCA conducted for the four largest stands likewise revealed several consistent relationships between individual variation in the mating patterns predicted by MEMM and the measured ecological correlates. The first and the first three pairs of canonicalvariates (CV) were significantly correlated for the fathers and the mothers, respectively (Table S3). We only retained the results for the first CV pair for further interpretation, however, because they explained a far greater fraction of the total variance than the following CVs (Table S3). The best fit model for the mother tree perspective explained 59% of the overall variation (Table S3). The mating canonical variate (Mating) showed a strong positive relation with the density of conspecifics within a 100 m radius and a weaker negative relation with the genetic relatedness and the topographic exposition of the tree (Table 4.3). Its relationship with the mating parameters was characterized by a positive link with the number and the genetic distance to pollen donors, and a negative one with the kinship of mates (Table 4.3). Taken together, the result suggests that the density of conspecific neighbours within 100 m was the most consistent trigger of mother trees’ mating system with a positive effect on the number of (male) mates and a negative effect on their kinship. The best fit model for the father perspective explained 41% of the total variation (Table S3). The mating canonical variable (Mating) was strongly related with tree size and male flower production. This canonical variable defines an overall measure of mating system including all mating parameters, being particularly strong the correlation of the number of female mates, male fecundity and degree. Overall, our results indicate that tree size and flower production were the dominant drivers of male mating system in its different dimensions. DISCUSSION A previous study on our refugial oak forest stands (Moracho et al. 2016) revealed that they have maintained very little gene exchange - both historically and at present - contradicting the widespread notion that wind-pollinated tree populations tend to regularly experience effective pollen dispersal spanning many kilometers (Kremer et al. 2012). The study argued that the low connectivity of the stands would have been caused by their refugial habitats whose topography, humidity and dense vegetation would hamper (especially longdistance) pollen movements. Additionally, an efficient genetic purge of recruiting oaks, maintained over many generations, would allow the majority of pollinations to occur among nearby, little genetically-related trees. Here, we advance on the ecological causal mechanisms underlying the limited pollen dispersal in our refugial stands by providing detailed insights into individual-level variation of mating and pollen dispersal. More generally, this study is one of the first to decompose individual-level mating patterns of forest trees from a local (within-stand) to a landscape (valley-wide) scale (but see also Gaüzère et al. 2013). Our results show that different maternal and paternal components of the mating system respond to varying drivers depending on the scale of observation. Thus, they illustrate the great complexity of individual-level mating patterns that underlie the more commonly quantified population-level mating systems (Friedman & Barrett 2009; see also Chybicki & Burczyk 2013; Gaüzère et al. 2013).  | CAPÍTULO 468 Original variables Canonical variables Mother tree perspective Ecological variables Ecological - DBH 0.12 - Male flower production 0.31 - Neighbour density 25 m 0.18 - Neighbour density 100 m 0.77 - Neighbour genetic relatedness -0.39 - Topographic Position -0.44 - Mating variables Mating Ecological Number of pollen donors 0.73 0.56 Genetic distance to pollen donors 0.44 0.34 Kinship of pollen donors -0.78 -0.60 Median intermate distance 0.06 0.05 Degree 0.19 0.15 Father tree perspective Ecological variables Ecological - DBH 0.89 - Male flower production 0.85 - Neighbour density 25 m 0 - Neighbour density 100 m 0.05 - Neighbour genetic relatedness -0.07 - Topographic Position -0.32 - Mating variables Mating Ecological Number of female mates 0.92 0.59 Median intermate distance 0.5 0.32 Male fecundity 0.73 0.47 Degree 0.69 0.44 Table 4.3 Structural correlations of CCA analysis involving mating variables and ecological correlates of individual Quercus robur trees, calculated once from a maternal and once from a paternal perspective. Results from the first canonical variate pair are presented with the correlation of each original variable with its corresponding canonical variate (Ecological or Mating) and the original mating variables with the Ecological one. Bold numbers indicate correlation values >0.40. INDIVIDUAL MATING PATTERNS | 69 Among-individual variation in mating: from landscape to local scale Valley-wide analysis In accordance with our expectation, trees displayed extensive variation in their proportions of selfing, local pollination and pollen immigration. Moracho et al. (2016) had not detected any stand effect on levels of selfing or local pollination, whereas pollen immigration had been triggered to some extent by the size and geographical isolation of stands. Neither selfing nor pollen immigration showed, however, a noteworthy aggregation in particular stands (Fig. 4.1). Selfing was overall frequent (6.8%) compared to levels reported from other oak populations (Nakanishi et al. 2005: 0.1%; Pluess et al. 2009: 3.5%; Chybicki & Burczyk 2013: 0.5%; Gerber et al. 2014: 2.5%), even small and isolated ones (Ortego et al. 2013: 1.4%; Vranckx et al. 2014: 0.7%). A closer inspection revealed, however, that 54 trees experienced no selfing at all whereas the remaining 19 trees had up to 80% selfed seeds, and only five individuals produced more than half of all selfed offspring. Such an inequality exceeds by far those observed in other oak populations (maximum individual selfing estimates: Q. robur: 2.6%, Q. petraea: 7.4% [Chybicki & Burczyk 2013]; Q. lobata: 9.8% [Pluess et al. 2009]; Q. salicina: 0.7% [Nakanishi et al. 2005]). One could hence speculate whether long-term refugial populations might be disproportionately prone to sustaining individuals capable of selfing (see also Moracho et al. 2016). Relictual settings characterized by low density, low pollen competition, and extreme variability of selfing levels, may result in low effective population size and high variation in individual tree contributions to progeny. Interestingly, Gaüzère et al. (2013) also reported a high overall frequency and extensive individual variation in selfing (23 of 60 trees affected with maximum values up to 48%) from a southern marginal population of beech (Fagus sylvatica). Yet Ouayjan & Hampe (2018) observed no selfing at all in another long-term refugial population of the same species. On the other hand, pollen migration showed levels of among-individual variation that were curiously similar to selfing (mean: 7.6%, maximum: 89%, distribution: 21 trees with and 52 trees without events), although they concerned a different set of trees (Fig. 4.1). We are not aware of similar studies assessing individual variation in pollen immigration across disjunct broadleaf forest stands (for studies on conifers see e.g. Robledo-Arnuncio & Gil 2005; O’Connell et al. 2007; Lesser & Jackson 2013), but the distribution we observed appears to be unusually unbalanced in comparison with other studies on tree stands within continuous populations (e.g. Chybicki & Burczyk 2013; Gaüzère et al. 2013; Lagache et al. 2013; Gerber et al. 2014), single isolated stands (Buschbom et al. 2011; Hampe et al. 2013), or single isolated trees (Ortego et al. 2011). Note, however, that direct comparisons of migration rates and their individual variation are quite challenging because of the idiosyncrasy of case studies. Within-stand analysis We also detected extensive among-tree variation in all mating parameters estimated for the four largest stands (Table 4.1). Interestingly, this variation was consistently higher for the male mating components than for the female components. Thus, female and male trees shared the same average number of mates (4.4) yet the variance was much larger for the male component. This trend went along with a highly unbalanced distribution of male fecundities, with a few trees dominating the effective pollen pool and numerous trees  | CAPÍTULO 470 contributing nothing at all (see Fig. S1). Pronounced inequality in male fecundity is a widespread phenomenon in forest trees (e.g. Klein et al. 2008; Chybicki & Burczyk 2013; Gerber et al. 2014; Ouayjan & Hampe 2018). The resulting reduction in effective population size should however be of special concern for small and isolated tree populations such as those found in long-term refugia. The only mating parameter involving both sexes, the degree within mating networks, showed intermediate levels of individual variation. The connectivity of within-stand mating networks was relatively high (average 4.3 mates) and the tendency to form local mating clusters not particularly pronounced (Fig. 4.2). Intrinsic and extrinsic determinants of individual mating patterns Valley-wide analysis We found substantial evidence indicating that among-tree variation in the frequency of selfing, local pollination and pollen immigration was driven by a combination of intrinsic and extrinsic determinants related with fine-scale pollen availability. Local pollinations were most prevalent in small trees surrounded by many, genetically little related neighbours. On the contrary, both selfing and pollen immigration occurred most commonly in trees that were surrounded by few and genetically related neighbours. Selfing was moreover most frequent in large trees that produced themselves many male flowers. All these trends suggest that both selfing and pollen immigration are most likely to occur in those trees that are not in reach of an abundant and genetically diverse local pollen cloud. High male flower abundance on the tree itself (i.e. a fine-scale phenomenon) would then favour selfing while the existence of large stands in the vicinity (i.e., a landscape-scale phenomenon) would favour pollen immigration. The observed trend fully corresponds to expectations for a mating system driven by pollen competition, where chance pollination events start to predominate over deterministic competition under low pollen load on stigmas (El-Kassaby & Ritland 1992). Oaks, like many other plants, are known to experience pollen limitation (Koenig & Ashley 2003; Pesendorfer et al. 2016). Independent field data actually indicate that the fruit set of our oak stands underlies pollen limitation at the landscape (but not at the local) scale (Moracho et al. 2018). Several other studies have shown that even small and fine-scale differences in pollen availability can have an important impact on a tree’s mating system. Thus, Lagache et al. (2013) demonstrated its effects on hybridization in white oaks (see also Chybicki & Burczyk 2013). Quite in line with our results, Gaüzère et al. (2013) observed highest levels of pollen immigration in large mother trees located in low-density patches of Fagus sylvatica. On the contrary, they also found that selfing was lowest (and not highest) in large trees and not influenced by the density of conspecific neighbours. This counterintuitive result could arise from an overarching effect of the vegetation surrounding trees that acts as a barrier to pollen flow (Gaüzère et al. 2013; see also Millerón et al. 2012). We detected a similar vegetation effect; yet in our case it did not counteract but paralleled the described effect of tree size (Table 4.2). Within-stand analysis All four stands were characterized by a high connectivity of individuals, in line with the absence of pollen limitation within stands (Moracho et al. 2018). However, the amount of connections established by individual trees varied among them being related with their size INDIVIDUAL MATING PATTERNS | 71 and spatial location within the stand, suggesting that big and centrally-located trees assumed a main role in the pollen-mediated connectivity. The CCA on female trees revealed that the density of conspecifics within 100 m - but not within 25 m - was the most consistent trigger of the mating system, with a positive effect on the number of (male) mates and a negative effect on their kinship. This relationship is likely to directly influence the reproductive success of mother trees, because their fruit set tends to increase with their number of mates (Moracho et al. 2018). The greater neighourhood appeared to be more influential than the direct neighbourhood in the largest stands, whereas we did not detect such an effect in the valley-wide analysis (results not shown) probably as a simple sampling effect: No or very few individuals grow at greater distances in many of the smaller oak stands. The observed inconsistency hence suggests that the importance of the small-scale neighbourhood for individual mating patterns might arise from the small size of the stands, rather than from biological features inherent to the studied species (see Lagache et al. 2014). On the contrary, the CCA on male trees revealed that their mating system was primarily determined by their size and flower production. This dominance of intrinsic tree features contradicts our prediction that the heterogeneous refugial environment should exert a significant impact on pollen mobilization and resulting mating success (Moracho et al. 2016). It also is at odds with observations of Chybicki & Burczyk (2013) and Lagache et al. (2014) that, in mixed Q. robur/Q. petraea forest stands, male fecundity decreased systematically from the higher towards the lower parts of the study plots, suggesting a major influence of environmental triggers. This could be a particular effect in relict populations where the outstanding presence of huge old trees, producing a vast amount of pollen relative to medium and small sized one, would have a preponderant effect on determining effective pollination over the environment-related issues. Oak reproduction and conservation in long-term refugia Our detailed analysis showed that the intrinsic character and the immediate conspecific environment of individual trees are the dominant drivers of the mating system – in other words: the situation of trees within their forest stands overwhelms the situation of the stands themselves. The only exception was (unsurprisingly) the rate of pollen immigration into stands determined by both local and landscape drivers, but even here the effect became only relevant when both local or self pollen were little abundant. We have to admit that our study was carried out in a year of an abundant fruit crop and that the observed trends could well differ in years with less abundant fructification (Nakanishi et al. 2005; Pesendorfer et al. 2016). However, the great importance of the local environment for mating and reproductive success is in line with what would be expected for long-term refugial populations (Hampe & Jump 2011). And in any case, high productive years are likely the most determinant for the population dynamics. In this sense, the observed strong dominance of local pollination and absence of localscale pollen limitation might even be taken as a reassuring phenomenon as it indicates that our study populations could be less susceptible to negative genetic effects of small population size than many more recently fragmented tree populations (Jump & Peñuelas 2006; Aguilar et al. 2008). The mechanism of being fertilized by abundant, local but genetically distant pollen, and otherwise by immigrant or even self pollen may help explain the successful long-term persistence of sparse refugial tree populations in an  | CAPÍTULO 578 METHODS Study system The southwestern range limit of Q. robur is located in the mountain ranges of central Spain. A comprehensive survey across the north of the autonomous region Extremadura (W Spain, ca. 20,000 km2) identified a total of 18 refugial Q. robur forest stands (Pulido et al. 2007). Large and continuous populations of the species occur in northern and northwestern Spain. The present study was conducted across 12 of the detected stands, including six from the ten investigated by Moracho et al. (2016). They occur in locally humid environments where the summer dry regional Mediterranean climate is mitigated, most typically along streams where Q. robur coexists with other mesic tree species such as Celtis australis, Castanea sativa and Alnus glutinosa among others. The landscape surrounding this habitat is largely covered by broadleaved forests dominated by Q. pyrenaica where it has not been transformed for horticulture or livestock exploitations. The investigated Q. robur stands vary considerably in size (n= 3-340 adult trees) and the level of geographical isolation (Fig. 5.1). Figure 5.1 (a) Distribution area of Quercus robur according to EUFORGEN with a black dot indicating the study area. (b) Map of the investigated refugial stands, represented by circles with circle size indicating the number of adult trees. TREE FECUNDITY AT REFUGIUM | 79 Quercus robur is a strongly outcrossing, wind-pollinated species with interannual variation in seed production that can range from alternating to classical masting. In the study area, the species flowers between March and early April, shortly before bud burst, and acorns ripen in late September and early October. Oak populations exhibit a great ability for long-distance pollen flow (e.g. Craft & Ashley 2007; Buschbom et al. 2011; Hampe et al. 2013). However, different lines of evidence also indicate that pollen limitation could be a widespread phenomenon in oak populations (Knapp et al. 2001; Koenig et al. 2012; Lagache et al. 2013; Pearse et al. 2015). Field sampling Female flower production: We selected a total of 159 adult trees from across the 12 stands for our field survey of flower and acorn production (Fig. 5.1). This sample included 65% of all the adult trees growing in small stands (i.e., with up to 35 individuals) and a minimum of 20 trees in each of the larger stands. Counts were performed in the first week of November 2012 and 2013, respectively, when virtually all female flower structures had fallen to the ground but remained mostly intact. The number of female flowers produced by a given target tree was estimated by counting the number of aborted and of fully developed acorn cupules in 0.5 x 0.5 m quadrats randomly placed beneath the tree canopy. The number of quadrats sampled was roughly proportional to the total crown projection of the target tree (range: 2-8 quadrats per tree) and their summed surface spanned at least 15% of the projection area. Total flower production per tree was then computed by extrapolating the number of cupules counted in quadrats to the full crown projection. We inferred the overall production of female flowers as the sum of aborted and of fully developed acorn cupules, while fruit set was estimated as the proportion of the overall female flower production that arrived to produce fully developed acorns. Note that our fruit set estimates could be somewhat inflated due to the difficulty in accurately detecting the earliest flower abortions (even though we spent a great effort in identifying them, a task that was facilitated by the conspicuous peduncules of Q. robur). Male flower production: In early April 2013 (that is, near the peak flowering period), we counted the number of catkins on two randomly selected branches from each of the 159 trees also used for the female flower survey. Counts were performed using binoculars and extrapolated to the total crown surface. The validity of this sampling procedure had previously been tested and corroborated by directly counting the total number of catkins on 37 trees that covered the entire range of flower production in the area. Besides, we collected 10 catkins from 12 trees in order to determine the average number of male flowers per cat-  | CAPÍTULO 580 kin (18.5±3.5, mean ±SD) and computed the number of male flowers per tree as the product of the number of catkins and the average number of flowers per catkin. Floral sex ratio: We used the 2013 field data on female and male flower production to derive the functional gender (Gi) of individual trees. This was done through the function Gi = gi/(gi + ai x E), where gi is the number of female flowers produced by the i tree, ai is the number of male flowers, and E = ∑ gi / ∑ ai . E is an equivalence factor that equates the probability of androecial units and gynoecial units contributing genes to the next generation (Lloyd 1979, 1980). By this procedure, Gi characterizes the functional “femaleness” of an individual plant. Acorn mass: We haphazardly collected 11 full-sized, mature acorns from each tree in 2013. Their fresh mass was determined by weighing them to the nearest 0.1 mg. Tree size and flowering phenology: We measured the diameter at breast height (DBH) of all the trees sampled and characterized their flowering phenology on two dates somewhat before and after the peak of the flowering period 2013 (5 and 23 April, respectively). Four phenological stages of male flowers were distinguished during each survey: (i) swelling buds (score 0); (ii) emerging and immature catkins (score 1); (iii) mature catkins (score 2); and (iv) old, dried catkins (score 3). The phenological stage of the upper and the lower part of each tree was recorded separately and averaged afterwards. Repeating the survey in the following year (2014) allowed us to corroborate that the phenological ranking of trees within stands remains reasonably consistent between years (Spearman rank correlation, rs = 0.38, P = 0.004; see also Bacilieri et al. 1995). Ecological and genetic correlates of flowering and acorn production For each of our 12 target stands, we obtained the following variables: (i) size, calculated as the number of adult trees belonging to the stand and (ii) geographical isolation, measured as the average distance to the other 18 stands. Within stands, we measured the following variables for each target tree: (i) elevation, according to a Digital Elevation Model with a resolution of 5 m (Instituto Geográfico Nacional, distributed by Centro Nacional de Información Geográfica CNIG; http://centrodedescargas.cnig.es); (ii) topographic index (TPI), computed as the difference between the elevation of a focal tree and the mean value of the surrounding landscape (25 m radius) as implemented in package spatialEco function tpi (R Development Core Team 2014). This measure describes whether a tree is located on a top (TPI > 0) or in a hollow (TPI < 0); (iii) percentage of forest cover within a 25 m radius around each focal tree (with the two classes <50% and >50 %); and (iv) population density, measured as the number of conspecific trees within the same radius. TREE FECUNDITY AT REFUGIUM | 81 All target trees had been genotyped using 20 nuclear microsatellite markers and the procedures are described in detail in Moracho et al. (2016). Based on the trees’ multilocus genotypes, we computed the mean genetic relatedness of each target tree with its conspecific neighbours growing within a 25 m radius using the coefficient of Queller & Goodnight (1989). In addition, we used the individual genotypes of the acorn families from 47 trees reported in Moracho et al. (2016) to calculate, for each mother tree’s progeny, the following variables: (i) gene diversity (HE), (ii) mean kinship (computed using the coancestry coefficient of Loiselle et al. 1995), (iii) inbreeding coefficient (FIS), and (iv) number of pollen donors (computed as described in Moracho et al. 2016). These variables can be considered to characterize the ‘pollen cloud’ of individual mother trees. All genetic parameters were obtained with SPAGeDi 1.4 (Hardy & Vekemans 2002) except for the number of pollen donors that was based on paternity analyses conducted in CERVUS 3.0 (Kalinowski et al. 2007). Statistical analyses Flowering and acorn production We first tested for correlations between our different response variables (numbers of male and of female flowers, acorn number, fruit set, acorn mass, floral sex ratio). Then we estimated variance components of the female fecundity parameters (female flower number, fruit set and acorn mass) considering four levels: among stands, among individuals within stands, within individuals, and between years. Finally, we performed linear mixed models considering a hierarchical nested structure of random effects to assess the statistical support of our variance components analysis. No fixed terms where included in these models. These analyses were performed in R v.3.0.2 (R Development Core Team 2014) using the packages ‘ape’ and ‘lme’, respectively. Ecological and genetic correlates of flowering and acorn production We attempted to identify both extrinsic and intrinsic determinants of among-tree variation in the three response variables female flower number, fruit set and acorn mass. Two analyses focused on extrinsic determinants. The first model examined among-stand variation considering stand size and geographical isolation as predictor variables. The second model examined within-stand variation with elevation, topographic index, forest cover, and population density as predictors. Two further analyses focused on trees’ intrinsic determinants (again for the three response variables female flower number, fruit set and acorn mass). The first model included DBH, flowering phenology, and mean genetic relatedness with conspecific neighbours as predictor variables. The second model used only the subset of  | CAPÍTULO 582 trees for which we had analysed acorn progenies and included the (progeny-based) variables gene diversity (HE), mean progeny kinship, inbreeding coefficient, and number of pollen donors. Finally, we constructed two more models with the floral sex ratio as response variable. The first model considered the extrinsic variables elevation, topographic index, forest cover, and population density as predictors, while the second model included the intrinsic variables DBH, phenology and relatedness. We employed general linear mixed models (GLMMs) in a maximum likelihood framework (Burnham & Andersen 2002) using the lme function (package ‘nlme’) to specify Gaussian errors and the glmer function (package ‘lme4’) to specify Poisson errors when estimating models with the variable female flower number. We considered the stand as a random factor and also the individual tree when fecundity data were available for two years (e.g. for female flower number; see Table 5.1). All predictor variables of the six models were considered as fixed factors. RESULTS Patterns of flowering and acorn production Average values for flowering and acorn production are summarized in Table 5.1. Male flower number was positively related with both female flower number (Spearman rs = 0.68, n = 156, P < 0.0001) and acorn number (rs = 0.64, n = 155, P < 0.0001). Even after accounting for tree size, male and female flower number retained a clear positive relationship (Spearman rs = 0.62, n = 156, P < 0.001). Female flower number closely matched acorn production in both years (2012: rs = 0.92, n = 157, P < 0.001; 2013: rs = 0.93, n = 157, P < 0.001) and floral sex ratio (measured as ‘femaleness’: rs = 0.37, n = 156, P < 0.001) showed a similar albeit weaker trend. Finally, fruit set was unrelated with either fe- 2012 2013 Male flower number - 3,387,400 ± 3,106,100 Female flower number 17,000 ± 23,900 10,100 ± 15,700 Floral sex ratio - 0.455 ± 0.232 Acorn number 7,400 ± 11,600 3,700 ± 7,500 Fruit set (%) 42.6 ± 18.9 33.5 ± 15.6 Acorn mass (g) 5.9 ± 1.7 - Table 5.1 Summary statistics for the surveyed fecundity parameters of individual Quercus robur trees. Integers are rounded to the nearest hundred. TREE FECUNDITY AT REFUGIUM | 83 male flower number (rs = -0.02, n = 157, P > 0.05), male flower number (rs = 0.13, n = 154, P > 0.05) or floral sex ratio (rs = 0.03, n = 154, P > 0.05). Despite marked between-year variation in the number of female flowers, the fertility ranking of individual trees in 2012 and 2013 remained highly consistent (Spearman rs = 0.84, n = 156, P < 0.001). Accordingly, the variance component analysis revealed that the main source of overall variation in female flower number were differences among trees (CV = 0.59; see Fig. 5.2). Fruit set also varied substantially between the two years of study (CV = 0.52), being higher in the year of more abundant female flower production (2012). The ranking of individual trees regarding fruit set was rather consistent between years (Pearson r = 0.60, n = 155, P < 0.001) albeit less than female flower number. Acorn weight varied primarily among trees (Fig. 5.2). Among-stand variation was low (although statistically significant) for all fecundity parameters (Fig. 5.2). AIC-based tests indicated that including all these sources of variation as random effects in the models of ecological and genetic correlates significantly improved their fit with respect to the null model. 0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00 Female flower number Fruit set Acorn mass Variance components Stand Year Tree Within tree Figure 5.2 Distribution of overall variation in three measures of Quercus robur tree fecundity and offspring fitness in refugial forest stands. Four hierarchical sources of variance are considered (within-tree < tree < year < stand). Note that between-year variation among individual trees was not available for acorn mass.  | CAPÍTULO 584 Table 5.2 Effects of different grouped predictor variables (‘extrinsic’ versus ‘intrinsic’, each at two different scales) on female flower number, fruit set and acorn mass. The SD of random effects (year/stand) and the sample size are shown for each model. Models for female flower number and fruit set are based on 2 years of sampling (and hence the twofold number of observations); the fact that trees were repeatedly censused was taken into account by including their identity as random effect in the model. Results of GLMM are shown: – P < 0.1, *P < 0.05, **P < 0.01, ***P < 0.001. TREE FECUNDITY AT REFUGIUM | 85 Determinants of flowering and acorn production Extrinsic determinants Stand characteristics influenced the fruit set (Table 5.2), which increased with greater stand size and decreased with stronger isolation. Stand size also had a marginal effect on female flower number (P = 0.07), with trees in larger stands producing more flowers. Acorn mass was unaffected by stand identity. Within stands, forest cover was the only relevant variable in predicting female flower number, with more fecund trees occurring in less forested vicinities. We detected no effects on fruit set, while a tendency of larger acorns to occur in less forested sites was only marginally significant (P = 0.1). Finally, floral sex ratio variation was affected by the elevation of trees’ growing sites, although the slope of the relationship was virtually horizontal (slope = 0.001; Table 5.3). Intrinsic determinants Unsurprisingly, female flower number was strongly predicted by the DBH of target trees (Table 5.2). The DBH was moderately but significantly related with female flower number (rs 2012 = 0.35, rs 2013 = 0.34), male flower number (rs = 0.32) and acorn number (rs 2012 = 0.31, rs 2013 = 0.26; P < 0.01 for all variables). Acorn mass also increased with the DBH, whereas fruit set was unaffected by any of the tested predictor variables (Table 5.2). Our analysis of the trees with available progeny data revealed instead that fruit set tended to be predicted by the number of pollen donors (P = 0.06; Table 5.2). On the contrary, neither female flower number nor acorn mass showed any relationship with the genetic properties of the progenies. Finally, the floral sex ratio was marginally affected by trees’ flowering phenology (P = 0.08; Table 5.3). Table 5.3 Effect of different predictor variables on floral sex ratio (measured as ‘femaleness’). Stand is included as a random factor and sample size is n = 110.  | CAPÍTULO 586 DISCUSSION Patterns of flowering and acorn production We observed ample variation in flower and acorn production as well as in acorn mass. The principal source of variation were differences among individual trees, whereas differences among stands were throughout very minor. Both flower and acorn numbers also differed markedly between the two years of study. These patterns of variation largely correspond to those reported from various other oak species in similar environments (e.g. Sork et al. 1993; Pérez-Ramos et al. 2015; Pesendorfer et al. 2016). At the level of individual trees, the initial number of female flowers was a powerful predictor of the resulting acorn crop (rs = 0.92-0.93), implying that fruit set in a given year was remarkably consistent among trees (Fig. 5.2; see also Feret et al. 1982; Sork et al. 1993). At the level of the entire population, however, fruit set was about a third higher in the year with a higher initial abundance of female flowers (2012), indicating considerably more frequent pollination and resulting in a twofold acorn crop. Although based on only two years of data, our observation is fully in line with reports indicating that the mating environment largely triggers interannual variation in oak acorn crops (“wind pollination hypothesis”: Koenig et al. 1994; see also García-Mozo et al. 2007; Koenig et al. 2015; Pearse et al. 2015; Pesendorfer et al. 2016). Meteorological conditions may have further supported the observed trend (Bell & Clark 2016) as the accumulated precipitation during the acorn maturation period was three times higher in 2012 than in 2013 (Agencia Estatal de Meteorología; www.aemet.es). Finally, the strong effect of female flower number on acorn crop size probably triggered also the observed (weaker) relationship between floral sex ratio and acorn production. Determinants of flowering and acorn production Unsurprisingly, the production of flowers and acorns was consistently related with tree size (estimated via the DBH). This relationship remained virtually unaltered by the observed between-year variation in flowering and fruiting, suggesting that trees of different sizes tended to behave similarly across years (Sork et al. 1993; but see Hirayama et al. 2008). We also observed larger female flower numbers in less forested sites; however, this trend is most probably just a secondary effect that arises from a tendency of larger trees to grow in open sites (E. Moracho, unpubl. data). Male and female flower number retained a remarkably strong positive relationship (rs = 0.62) even after accounting for tree size, indicating that some trees invested systematically more in reproductive structures than others. This phenomenon has repeatedly been shown in oaks and has been related with small-scale variation in growing site productivity (Knops TREE FECUNDITY AT REFUGIUM | 87 & Koenig 2012; Pérez-Ramos et al. 2014) or microclimate (Koenig et al. 2015). It is in line with the notion that, for long-lived organisms, a plastic resource tracking response to environmental fluctuations may be more adaptive than directly linking life-history traits through trade-offs such as well-defined functional genders (Knops & Koenig 2012; see also Petit & Hampe 2006). Fruit set outstood from the other tree fecundity measures as the only parameter that was primarily determined by stand-level characteristics. Both the size and the geographical isolation of stands affected fruit set in a way that points to the existence of pollen limitation at a landscape scale (cf Fig. 5.1). Fruit set hence appears to underlie the same constraints as previously shown for the rate of between-stand pollen flow (Moracho et al. 2016). On the contrary, we found very little evidence for pollen limitation at finer spatial scales despite using several ecological and genetic indicators (population density, tree phenology, genetic relatedness, number of mates, progeny kinship and inbreeding). The only eventual and very indirect indication consists in the fact that fruit set showed a marginally significant trend (P = 0.06) of being higher in those trees that received pollen from a greater pool of mates. Our inability to detect local-scale pollen limitation is at odds with several other studies in oaks (Knapp et al. 2001; Koenig et al. 2012; Lagache et al. 2013; Pearse et al. 2015) that have led to increasing awareness of the phenomenon (Koenig & Ashley 2003; Friedman & Barrett 2009). The absence of local-scale pollen limitation in our system could also explain why we failed to detect any influence of phenology on fruit set, whereas Koenig et al. (2012) reported clear evidence for stabilizing selection on flowering time that arises from pollen limitation of very early or late flowering individuals. Acorn size also was positively related with tree size, a phenomenon that has previously been reported in detail by Koenig et al. (2009). This relationship could be of particular relevance in our refugial context because larger acorns are known to produce more performant offspring (Gómez 2004; Sage et al. 2011), an important fitness advantage in Mediterranean-type environments where early germination and rapid growth can greatly enhance seedling survival to summer drought. Hence large trees would have a double fitness advantage in our context, given that their reproductive success exceeds that of smaller conspecifics both in terms of offspring quantity and quality. Consequences for tree reproductive success in a refugial context Overall, the reproductive success of our refugial oak stands appears to be triggered by a combination of local-scale and landscape-scale drivers. Flower and acorn production depend basically on individual tree size, whereas mating success in terms of fruit set is primarily determined by the size and geographical isolation of the stand. The fact that pollen  | CAPÍTULO 694 Figura 6.1 Resumen gráfico de los resultados de los modelos que explican los patrones de dispersión de polen en poblaciones relictas de Q. robur a escala local (Capítulo 4) y de paisaje (Capítulo 3). Domina la polinización local siempre que los árboles dispongan de un vecindario denso y poco emparentado. La autofecundación se ve favorecida en árboles grandes productores de polen con una baja densidad de conspecíficos emparentados a su alrededor (A). En un ambiente similar pero con núcleos de población actuando como fuentes de polen en las proximidades la frecuencia de inmigración aumenta. A. Proceso dominante: AUTOFECUNDACIÓN Árboles de gran tamaño Grandes productores de polen Baja densidad vecinos Vecinos emparentados C. Proceso dominante: INMIGRACIÓN Baja densidad vecinos o rodales muy pequeños Vecinos poco emparentados Núcleos de población en las proximidades B. Proceso dominante: POLINIZACIÓN LOCAL Alta densidad vecinos Vecinos poco emparentados A B C DISCUSION GENERAL | 95 parece que las poblaciones relictas son propensas a albergar individuos longevos grandes productores de polen que favorecen la autofecundación (fenómeno a escala fina) cuando no disponen de polen en abundancia y de calidad, mientras que la existencia de núcleos de población densos en las proximidades facilitaría la dispersión de polen inmigrante (fenómeno a escala de paisaje). El análisis de los patrones de cruzamiento dentro de los rodales de mayor tamaño (4 réplicas, n = 22-35 individuos) revela dos resultados interesantes: (i) Existe una alta conectividad vía polen en el interior de rodales contraria a la observada a escala de paisaje (Capítulo 4: figura 2). El rol de conectores dentro de la red de polinización es asumido por árboles que ocupan posiciones centrales del parche y principalmente, por individuos de gran tamaño, productores de vastas cantidades de polen que son muy efectivos en su función paternal (donando polen) (ver Capítulo 4: fig. 4.2). Éstos, además son una fuente de polen importante para favorecer la dispersión de polen a larga distancia y la conectividad a través del paisaje (Figura 6.2 y Apéndice M). (ii) Existe una gran variabilidad en los patrones de cruzamientos para ambas funciones sexuales (la donación y la recepción de polen), especialmente notable para la componente masculina (Capítulo 4: tabla 1). El gran desequilibrio encontrado en la fecundidad masculina (liberación de polen que llega a producir frutos) ha sido ampliamente demostrado en poblaciones de especies forestales (p. ej., Chybicki & Burczyk 2013, Gerber et al. 2014, Klein et al. 2008). Encontramos que la fecundidad, y en general, los patrones de cruzamiento masculino guardan relación con rasgos intrínsecos de los árboles como su tamaño y la producción de polen (Capítulo 4: tabla 3), y no con aspectos del ambiente físico como han demostrado otros estudios (Chybicki & Burczyk 2013, Lagache et al. 2014). Los enormes árboles viejos en poblaciones relictas adquieren un papel preponderante para los patrones de polinización (Petit & Hampe 2006), cuyo efecto se sobrepone a aspectos relacionados con el ambiente refugial (ver la distribución del tamaño de los árboles en el sistema de estudio, Figura 6.1). Por otra parte, el estudio de fecundidad en poblaciones relictas de Q. robur (Capítulo 5) revela que la fecundidad femenina (i.e. la producción de semillas viables y su peso), al igual que la masculina, debe su principal fuente de variación a diferencias en el tamaño del árbol (Capítulo 5: figura 5.2 y tabla 5.2). Los árboles mas grandes producen una mayor cantidad de flores tanto masculinas como femeninas de manera casi invariable a través de los años (Sork et al. 1993; pero ver Hirayama et al. 2008). No obstante, al eliminar el efecto del tamaño, encontramos que algunos árboles invierten sistemáticamente más en su función reproductora que otros. Este fenómeno puede adquirir un carácter adaptativo si constituye una respuesta plástica frente a las fluctuaciones ambientales (Knops & Koe-  | CAPÍTULO 696 nig 2012, Pérez-Ramos et al. 2014, Koenig et al. 2015). Además, encontramos una mayor producción de bellotas más grandes y pesadas en árboles de gran tamaño (ver Koenig et al. 2009). Este efecto puede tener particular relevancia en el contexto de refugios climáticos, dada la ventaja para el desarrollo que supone disponer de una mayor cantidad de reservas (Gómez 2004, Sage et al. 2011), especialmente en ambientes mediterráneos donde la germinación temprana y el rápido crecimiento aumenta la probabilidad de supervivencia de las plántulas durante la sequía estival. En resumen, los árboles grandes tienen una múltiple ventaja a nivel de fitness: su éxito reproductivo supera al de conespecíficos de menor tamaño tanto en su función masculina y femenina, y muy posiblemente en términos de calidad de sus progenies. El éxito reproductivo (fruit set) es la única medida de fecundidad influida por el escenario relicto (Capítulo 5: tabla 5.2), lo que relacionamos con el escaso flujo de polen a este nivel. El efecto del tamaño del rodal y el grado de aislamiento sobre el fruit set pone de manifiesto la limitación de polen a escala de paisaje, mientras que a escala fina las evidencias que encontramos son mínimas a diferencia de otros estudios (Knapp et al. 2001, Koenig et al. 2012, Lagache et al. 2013, Pearse et al. 2015), a pesar de haber usado nume- Figura 6.2 Relación existente entre el tamaño del árbol donador de polen (medido como DBH) y la distancia de dispersión efectiva, exclusivamente para los eventos de flujo polínico entre núcleos de población del valle del Jerte. El DBH promedio (SD) de los árboles responsables de estos eventos es 51,7 (18,8). Se representa la línea de tendencia con mayor ajuste (R2 = 0,35; y = 1,499 x1,86). Distancia de polinización (m) 0 1500 3000 4500 6000 7500 9000 DBH del árbol donador de polen 0 10 20 30 40 50 60 70 80 90 DISCUSION GENERAL | 97 0 10 20 30 0 10 20 30 40 50 60 70 80 90 DBH Frecuencia Valle del Jerte 0 2 4 6 0 10 20 30 40 50 60 70 80 90 DBH Frecuencia Rodales de hasta 10 árboles Figura 6.3 Distribución de frecuencias de DBH para el conjunto de árboles distribuidos a lo largo del Valle del Jerte, en el rodal de mayor tamaño (E con 35 individuos), y en el conjunto de rodales más pequeños (compuestos de hasta 10 árboles). La estructura de edades en las poblaciones relictas de Q. robur revela la abundancia de árboles de gran tamaño, especialmente en rodales pequeños donde persisten árboles muy longevos. Las poblaciones mayores presentan indicios de cierto reclutamiento a largo plazo con una mayor proporción de individuos jóvenes. 0.0 2.5 5.0 7.5 0 10 20 30 40 50 60 70 80 90 DBH Frecuencia Rodal de mayor tamaño  | CAPÍTULO 698 rosos indicadores tanto ecológicos como genéticos. La ausencia de restricciones genéticas que afecten el sistema de cruzamiento en nuestras poblaciones relictas puede explicar que la mayor parte de la dispersión de polen ocurra a cortas distancias entre conjuntos de árboles en suficiente densidad, como para que existan oportunidades para un polinización y reproducción efectiva. ¿QUE MECANISMOS FACILITAN LA PERSISTENCIA EN POBLACIONES RELICTAS DE ÁRBOLES? El flujo génico representa el foco principal de esta tesis ya que constituye un elemento clave para la conectividad y diversidad genética de las poblaciones, el éxito reproductivo, el potencial adaptativo en ambientes cambiantes y en definitiva la persistencia a largo plazo (Sexton et al. 2011). Sorprendentemente, los núcleos poblacionales de Q. robur en el valle del Jerte intercambian poco polen, y la marcada estructura genética de la población adulta sugiere que la débil conectividad no es solo un fenómeno a corto plazo (véase también Ouayjan & Hampe 2018). El enclave y aislamiento de las poblaciones en refugios jun- to a la baja densidad poblacional suponen un obstáculo a la dispersión de polen. A pesar de ello, la predominante fertilización entre individuos del mismo rodal garantiza una reproducción exitosa siempre que se alcance un cierto tamaño poblacional, de otro modo el escaso flujo de polen no compensa la limitación de polen existente. El mecanismo de ser fertilizado por polen local abundante y poco emparentado, y en su ausencia, por polen propio o inmigrante parece contribuir al éxito reproductivo de estas poblaciones (Figura 6.1). Además la escasa inmigración de polen puede llegar a tener un impacto exacerbado en la genética de las poblaciones si operan procesos de purga genética (Hampe et al. 2013), y la salvaguarda de la autofecundación permitiría la reproducción en circunstancias extremas, aunque la viabilidad de los descendientes procedentes queda por confirmar. Por último, el flujo génico interespecífico no parece tener un papel notable en los patrones de dispersión de Q. robur, a pesar de que a baja densidad poblacional o en situaciones de limitación de polen cabría esperar una mayor frecuencia (Lagache et al. 2013, Lepais et al. 2009). A menudo, el flujo génico entre especies de robles en simpatría es escaso tanto en poblaciones centrales como del margen de la distribución (Chybicki et at. 2012, Curtu et al. 2007, Lepais et al. 2009). Parece que principalmente barreras sexuales prezigóticas incluyendo el desacople fenológico de la floración, la competencia del polen y las interacciones polenpistilo, limitan el cruzamiento interespecífico entre estas 2 especies. No obstante, aunque la tasa anual de hibridación sea baja, no es desdeñable el rol que puede desempeñar en la adaptación local de las poblaciones. De hecho, la frecuencia de individuos con apariencia DISCUSION GENERAL | 99 híbrida es notable en algunas poblaciones del área de estudio donde la regeneración es patente, lo que podría sugerir un establecimiento ventajoso de los híbridos (E. Moracho, obs. pers). Las poblaciones relictas de Q. robur parecen poco afectadas por los procesos genéticos que amenazan otras poblaciones de pequeño tamaño (Aguilar et al. 2008, Jump & Peñuelas 2006). Los siguientes evidencias apoyan esta afirmación: (i) no se demuestra una depresión por endogamia notable en base a datos de campo y experimentales (ver Capítulo 3: página 53); (ii) el grado de parentesco es moderado-bajo (Capítulo 3: tabla 3.1) resultado de la acción de mecanismos prezigóticos (se evita la fertilización con genotipos emparentados) y postzigóticos (por procesos de purga genética); y (iii) las poblaciones preservan en su conjunto una diversidad genética relativamente elevada, aunque inferior a la encontrada en el rango continuo de la distribución de la especie (Hampe et al. 2010, Muir & Schloetterer, Streiff et al. 1998) o en poblaciones aisladas con altas tasas de flujo génico (Buschbom et al. 2011). Dentro de los rodales encontramos valores de diversidad menores y combinaciones de alelos únicas (Capítulo 3: tabla 1) como es de esperar bajo condiciones de aislamiento continuado y tamaño reducido de las poblaciones. Esto junto a la elevada diversidad genética regional, la fuerte diferenciación y las inferencias de la distribución en el pasado (Benito et al. 2007), sugieren que nuestras poblaciones son realmente relictos con una historia de persistencia a través de muchas generaciones (Hampe & Petit 2005) (ver Capítulo 1: figura 1.1). En contraposición, los rasgos de las poblaciones de fragmentación reciente o núcleos de nueva colonización determinan unos patrones de flujo génico sustancialmente distintos. En el margen en expansión, las poblaciones muestran una dependencia especialmente alta por el flujo génico al ser comúnmente originadas por efectos fundadores (Pujol & Pannell 2008, Lesser & Jackson 2013) hasta que alcanzan un cierto tamaño poblacional. Por su parte, las poblaciones fragmentadas por causas antrópicas son susceptibles a procesos de incompatibilidad y depresión por endogamia al cruzarse con árboles vecinos, ya que representan los parches resultantes de procesos de cuello de botella no sometidos a purga por selección. Por tanto, dependen de una alta tasa de flujo génico que a menudo se ve favorecida por la proximidad a núcleos de población mayores y por la apertura del hábitat degradado que facilita la dispersión por el viento entre parches dispersos (Kramer et al. 2008). La estructura de edades en base al dbh de los árboles en el valle del Jerte demuestra la gran frecuencia de árboles de medio y gran tamaño, especialmente sesgada en poblaciones pequeñas (Figura 6.3). En núcleos de mayor tamaño se observa una mayor proporción de árboles más pequeños lo que sugiere que aquí prospera cierto reclutamiento a lo largo del  | CAPÍTULO 6100 tiempo, así mismo la densidad de plántulas es notablemente mayor (E. Moracho, pers. obs). Los resultados obtenidos sugieren que las poblaciones de Q. robur de mayor tamaño o menos aisladas mantienen potencial de regeneración en lo que se refiere a su fecundidad y éxito reproductivo (fruit set). Existe una variabilidad interanual no desdeñable (p. ej., Pérez-Ramos et al. 2015 Pesendorfer et al. 2016, Sork et al. 1993) con una producción de semillas mayor en años con altas frecuencias de polinización (capítulo 5), mientras que la limitación de polen tendrá efectos más profundos en años de poca productividad. Numerosos estudios indican que el ambiente de polinización está entre los principales determinantes de la variación interanual en la producción de semillas en robles (‘hipótesis de la polinización por el viento’: Koenig et al. 1994; ver también García-Mozo et al. 2007, Koenig et al. 2015, Pearse et al. 2015, Pesendorfer et al. 2016). De todos modos, los episodios de mayor éxito reproductivo y reclutamiento en poblaciones relictas es probable que se limiten a años de benevolencia climática debido a la gran influencia de las condiciones meteorológicas a lo largo del proceso reproductivo, desde la producción de flores y el éxito de la polinización (Bell & Clark 2016, Koenig et al. 1996) hasta la maduración correcta del fru- to (Espelta et al. 2008, Pérez-Ramos et al. 2010). Las evidencias presentadas en esta tesis y los mecanismos discutidos ayudan a explicar la alta resiliencia de las poblaciones relictas de árboles. No podemos olvidar que hablamos de persistencia referida a una escala de tiempo y bajo unas condiciones ambientales determinadas (Hanna et al. 2014) y que por ello, mantener la capacidad adaptativa es necesario para asegurar la persistencia a largo plazo. Sería interesante continuar la investigación en esta línea para averiguar si las poblaciones relictas de Q. robur pueden experimentar procesos rápidos de adaptación local que les permitan afrontar las condiciones ambientales cada vez más adversas en que crecen, a pesar de sus reducidos niveles de flujo génico. IMPLICACIONES PARA LA CONSERVACIÓN DE POBLACIONES RELICTAS La gran relevancia de los fenómenos a escala local para los sistemas de cruzamiento y probablemente para la reproducción de plantas en general, así como la desigualdad existente entre individuos en la fecundidad femenina y sobretodo masculina, implica que incluso pequeños cambios en la estructura de los núcleos de población (como la eliminación de unos pocos árboles de gran tamaño con una función central como donantes de polen y altamente fecundos), pueden tener consecuencias muy severas en poblaciones relictas. Por ello, el manejo y las estrategias de conservación en estos sistemas deben centrarse en asegurar el funcionamiento interno de los propios rodales mediante prácticas que aseguren un número y densidad suficiente de individuos reproductivamente muy activos. Así mismo, es DISCUSION GENERAL | 101 fundamental garantizar la supervivencia de individuos grandes y longevos, que además poseen potencialidad para favorecer la conectividad entre rodales. " CONCLUSIONES  | APÉNDICES110 APÉNDICE A LISTADO Y DISTRIBUCIÓN DE LOS CASOS DE ESTUDIO # Figure A1. Map of the 623 populations included in this study, classified as ‘High-Latitude Margin’ (HLM), ‘Central’ populations, or ‘Low-Latitude Margin’ (LLM). Note that HLM populations for some organisms can be at lower latitudes than LLM populations of other species (and vice versa). The following list includes the total of 42 retained papers with CP comparisons: 1 Aikens, M. L. & Roach, D. A. Population dynamics in central and edge populations of a narrowly endemic plant. Ecology 95, 1850-1860 (2014). 2 Angilletta Jr, M. J., Niewiarowski, P. H., Dunham, A. E., Leaché, A. D., & Porter, W. P. Bergmann´s clines in ectotherms: illustrating a life-history perspective with sceloporine lizards. Am. Nat. 164, E168-E183 (2004). 3 Araújo, R.M., Serrao, E.A. & Sousa-Pinto I. & Aberg, P. Spatial and temporal dynamics of fucoid populations (Ascophyllum nodosum and Fucus serratus): a comparison between central and range edge populations. PLoS ONE 9, e92177 (2014). 4 Brante, A., Cifuentes, S., Pörtner, H. O., Arntz, W., & Fernández, M. Latitudinal comparisons of reproductive traits in five brachyuran species along the Chilean coast. Rev. Chil. Hist. Nat. 77, 15-27 (2004). 5 Cardoso, R.S. & Defeo, O. Geographical patterns in reproductive biology of the Pan-American sandy beach isopod Excirolana braziliensis. Mar. Biol. 143, 573-581 (2003). 6 Carey, P. D., Watkinson, A.R. & Gerard, F.F.O. The determinante of the distribution and abundance of the winter annual grass Vulpia ciliata ssp. ambigua. J. Ecol. 83, 177-187 (1995). CAPÍTULO 2 | 111 7 Dixon, A. L., Herlihy, C. R., & Busch, J. W. Demographic and population-genetic tests provide mixed support for the abundant centre hypothesis in the endemic plant Leavenworthia stylosa. Mol. Ecol. 22, 1777-1791 (2013). 8 Doak, D. F., & Morris, W. F. Demographic compensation and tipping points in climate-induced range shifts. Nature, 467, 959-962 (2010). 9 Ebert, T. A. Demographic patterns of the purple sea urchin Strongylocentrotus purpuratus along a latitudinal gradient, 1985-1987. Mar. Ecol. Prog. Ser. 406, 105-120 (2010). 10 Ebert, T.A., Dixon, J.D., Schroeter, S.C., Kalvass, P.E., Richmond, N.T., Bradbury, W.A. & Woodby, D.A. Growth and mortality of red sea urchins Strongylocentrotus franciscanus across a latitudinal gradient. Mar. Ecol. Prog. Ser. 190, 189-209 (1999). 11 Fargallo, J. A. Latitudinal trends of reproductive traits in the Blue Tit Parus caeruleus. Ardeola 51, 177-190 (2004). 12 García, D., Zamora, R., Gómez, J.M., Jordano, P, & Hódar, J.A. Geographical variation in seed production, predation and abortion in Juniperus communis througthout its range in Europe. J. Ecol. 88, 436-446 (2000). 13 García, M. B., Goñi, D. & Guzmán, D. Living at the edge: local versus positional factors in the long-term population dynamics of an endangered orchid. Cons. Biol. 24, 1219-1229 (2010). 14 Graves, G. Geographic clines of age ratios of black-throated blue warblers (Dendroica caerulescens). Ecology 78, 2524-2531 (1997). 15 Hidas, E. Z., Russell, K. G., Ayre, D. J., & Minchinton, T. E. Abundance of Tesseropora rosea at the margins of its biogeographic range is closely linked to recruitment, but not fecundity. Mar. Ecol. Prog. Ser. 483, 199-208 (2013). 16 Jump, A. S. & Woodward, F. I. Seed production and population density decline approaching the range-edge of Cirsium species. New Phytol. 160, 349-358 (2003). 17 Lammi, A., Siikamäki, P. & Mustajärvi, K. Genetic diversity, population size, and fitness in central and peripheral populations of a rare plant Lychnis viscaria. Cons. Biol. 13, 1069-1078 (1999). 18 Lardies, M. A., Arias, M. B., & Bacigalupe, L. D. Phenotypic covariance matrix in life-history traits along a latitudinal gradient: a study case in a geographically widespread crab on the coast of Chile. Mar. Ecol. Prog. Ser. 412, 179-187 (2010). 19 Lathlean, J.A., Ayre, D.J., Minchinton, T.E. Supply-side biogeography: Geographic patterns of settlement and early mortality for a barnacle approaching its range limit. Mar. Ecol. Prog. Ser. 412, 141-150 (2010). 20 Lester, S. E., Gaines, S. D., & Kinlan, B. P. Reproduction on the edge: large-scale patterns of individual performance in a marine invertebrate. Ecology 88, 2229-2239 (2007). 21 Matías, L., & Jump, A. S. Asymmetric changes of growth and reproductive investment herald altitudinal and latitudinal range shifts of two woody species. Glob. Change Biol. 21, 882-896 (2015). 22 Nantel, P. & Gagnon, D. Variability in the dynamics of northern peripheral versus southern populations of two clonal plant species, Helianthus divaricatus and Rhus aromatica. J. Ecol. 87, 748-760 (1999). 50 m  | APÉNDICES112 23 Parry, D., Goyer, R. A., & Lenhard, G. J. Macrogeographic clines in fecundity, reproductive allocation, and offspring size of the forest tent caterpillar Malacosoma disstria. Ecol. Entomol. 26, 281-291 (2001). 24 Paul, V., Bergeron, Y. & Tremblay, F. Does climate control the northern range limit of eastern white cedar (Thuja occidentalis L.)? Plant Ecol. 215, 181-194 (2014). 25 Pfeifer, M., Passalacqua, N. G., Bartram, S., Schatz, B., Croce, A., Carey, P. D., Kraudelt, H., et al. Conservation priorities differ at opposing species borders of a European orchid. Biol. Cons. 143, 2207-2220 (2010). 26 Rhainds, M. & Fagan, W.F. Broad-scale latitudinal variation in female reproductive success contributes to the maintenance of a geographic range boundary in bagworms (Lepidoptera: Psychidae). PLoS ONE 5, e14166 (2010). 27 Rivadeneira, M. M., Hernández, P., Antonio Baeza, J., Boltana, S., Cifuentes, M. et al. Testing the abundant-centre hypothesis using intertidal porcelain crabs along the Chilean coast: linking abundance and life-history variation. J. Biogeogr. 37, 486-498 (2010). 28 Sanz, J. J. Geographic variation in breeding parameters of the Pied Flycatcher Ficedula hypoleuca. Ibis 139, 107-114 (1997). 29 Sanz, R., Pulido, F., & Nogués-Bravo, D. Predicting mechanisms across scales: amplified effects of abiotic constraints on the recruitment of yew Taxus baccata. Ecography 32, 993-1000 (2009). 30 Silva-Montellano, A., & Eguiarte, L. E. Geographic patterns in the reproductive ecology of Agave lechuguilla (Agavaceae) in the Chihuahuan desert. I. Floral characteristics, visitors, and fecundity. Am. J. Bot. 90, 377-387 (2003). 31 Simoncini, M., Piña, C.I. & Siroski, P.A. Clutch size of Caiman latirostris (Crocodylia: Alligatoridae) varies on a latitudinal gradient. North-West. J. Zool. 5, 191-196 (2009). 32 Stanton-Gedes, J., Tiffin, P. & Shaw, R.G. Role of climate and competitors in limiting fitness across range edges of an annual plant. Ecology 93, 1604-1613 (2012). 33 Starmer, W. T., Wolf, L. L., Barker, J. S. F., Bowles, J. M., & Lachance, M. A. Reproductive characteristics of the flower breeding Drosophila hibisci Bock (Drosophilidae) along a latitudinal gradient in eastern Australia: relation to flower and habitat features. Biol. J. Linn. Soc. 62, 459-473 (1997). 34 Stocks, J. R., Gray, C. A. and Taylor, M. D. Intra-population trends in the maturation and reproduction of a temperate marine herbivore Girella elevata across latitudinal clines. J. Fish Biol. 86: 463-483 (2015). 35 Viejo, R. M., Martínez, B., Arrontes, J., Astudillo, C., & Hernández, L. Reproductive patterns in central and marginal populations of a large brown seaweed: drastic changes at the southern range limit. Ecography 34, 75-84 (2011). 36 Villellas, J., Ehrlén, J., Olesen, J.M., Braza, R. & García, M.B. Plant performance in central and northern peripheral populations of the widespread Plantago coronopus. Ecography 35, 001-010 (2012). 37 Villellas, J., Morris, W. F., & García, M. B. (2013). Variation in stochastic demography between and within central and peripheral regions in a widespread short-lived herb. Ecology 94, 1378-1388 (2013). CAPÍTULO 2 | 113 38 Vogler, F., & Reisch, C. Vital survivors: low genetic variation but high germination in glacial relict populations of the typical rock plant Draba aizoides. Biodiv. Cons. 22, 1301-1316 (2013). 39 Willemsen, R. E., & Hailey, A. Variation in adult survival rate of the tortoise Testudo hermanni in Greece: implications for evolution of body size. J. Zool. 255, 43-53 (2001). 40 Wilson, B. S. & Cooke, D.E. Latitudinal variation in rates of overwinter mortality in the lizard Uta stansburiana. Ecology 85, 3406-3417 (2004). 41 Yakimowski, S. B. & Eckert, C. G. Threatened peripheral populations in context: geographical variation in population frequency and size and sexual reproduction in a clonal woody shrub. Cons. Biol. 21, 811-822 (2007). 42 Zardi, G. I., Nicastro, K. R., Serrao, E. A., Jacinto, R., Monteiro, C. A., & Pearson, G. A. Closer to the rear edge: ecology and genetic diversity down the core-edge gradient of a marine macroalga. Ecosphere 6, 1-25 (2015).  | APÉNDICES114 APÉNDICE B RESUMEN DE LOS MODELOS Y SELECCIÓN DE LOS MEJORES ! MODELOS Table B1. Summary of models with ΔAICc, wi and heterogeneity. Model selection procedure retained five models within two units of ∆AICc of the best model (bold characters). All included margin type as a moderator and the null model (i.e., intercept only) was excluded. Only models with ΔAICc < 10 are presented. Pseudo R² was calculated as 1 – LLR where LLR is the ratio between the log-likelihood of model i and the log-likelihood of the null model. Model AICc wi delta QM P(QM) QE P(QE) Pseudo- R2 Habitat + Margin 381.39 0.17 0.00 9.70 0.008 413.28 < 0.001 0.04 Margin 381.47 0.17 0.08 7.32 0.007 431.87 < 0.001 0.03 Habitat + Margin + Kingdom 382.03 0.13 0.64 11.57 0.009 396.76 < 0.001 0.05 Habitat + Margin + Kingdom + Kingdom × Margin 382.55 0.10 1.16 13.69 0.008 394.03 < 0.001 0.06 Habitat + Margin + Margin × Habitat 383.12 0.07 1.74 10.30 0.016 413.2 < 0.001 0.05 Margin + Kingdom + Kingdom × Margin 383.45 0.06 2.06 10.21 0.017 425.68 < 0.001 0.05 Margin + Kingdom 383.56 0.06 2.18 7.44 0.024 430.11 < 0.001 0.04 Habitat + Margin + Kingdom + Margin × Habitat 383.57 0.06 2.19 12.40 0.015 396.62 < 0.001 0.06 Habitat + Margin + Kingdom + Kingdom × Habitat 384.36 0.04 2.97 11.57 0.021 396.2 < 0.001 0.06 Habitat + Margin + Kingdom + Margin × Habitat + Kingdom × Margin 384.83 0.03 3.44 13.77 0.017 393.8 < 0.001 0.07 Habitat + Margin + Kingdom + Kingdom × Habitat + Kingdom × Margin 384.92 0.03 3.53 13.70 0.018 393.2 < 0.001 0.07 Habitat 385.92 0.02 4.53 2.59 0.108 427.56 < 0.001 0.02 Habitat + Margin + Kingdom + Margin × Habitat + Kingdom × Habitat 385.94 0.02 4.55 12.41 0.03 395.96 < 0.001 0.07 Null 386.17 0.02 4.79 - - 449.01 < 0.001 0.00 Habitat + Kingdom 386.68 0.01 5.30 4.51 0.105 412.59 < 0.001 0.03 Habitat + Margin + Kingdom + Margin × Habitat + Kingdom × Habitat + Kingdom × Margin 387.26 0.01 5.88 13.77 0.032 393.06 < 0.001 0.08 Kingdom 388.29 0.01 6.90 0.07 0.798 448.06 < 0.001 0.01 Habitat + Kingdom + Kingdom × Habitat 388.84 0.00 7.45 4.60 0.204 412.12 < 0.001 0.04 CAPÍTULO 2 | 115 Table B2. Summary of the five models retained in the set of best models (i.e., with AICc < 2, Appendix G.). Margin explained a significant amount of heterogeneity in each of the five competing best models whereas neither Kingdom nor Habitat explained a significant amount of heterogeneity in any of the five models retained in the set of best models. QM and associated P-values represent the test associated with each moderator, separately. Pseudo R² were calculated as 1 – LLR, where LLR is the ratio between the log-likelihood of model i and the log-likelihood of the null model. Model Moderators QM (P-value) Pseudo-R² Model 1 Margin 7.21 (0.007) 0.040 Habitat 2.25 (0.134) Model 2 Margin 7.23 (0.007) 0.026 Model 3 Margin 7.40 (0.007) 0.053 Habitat 3.72 (0.054) Kingdom 0.14 (0.705) Model 4 Margin 7.65 (0.006) 0.065 Kingdom 3.1 (0.78) Habitat 0.14 (0.705) M a r g i n × Kingdom 1.83 (0.176) Model 5 Margin 4.23 (0.040) 0.049 Habitat 2.74 (0.098) M a r g i n × Habitat 0.55 (0.458)  | APÉNDICES116 APÉNDICE C EFECTO DE LOS MODERADORES Y SUS INTERACCIONES Figure C1. Sum of weights of moderators quantifying the relative importance of individual moderators and their interactions. Values are interpreted as the probability that a given variable is retained in the best model. Among tested moderators, margin type was the most important predictor (wH, margin = 0.96), whereas Habitat (wH,habitat = 0.69) and Kingdom (wH, kingdom = 0.54) received only marginal support, and interactions (Margin × Habitat and Margin × Kingdom) were even less relevant. Fully in line with this result, neither Kingdom nor Habitat explained a significant amount of heterogeneity in any of the five models retained in the set of best models (Appendix H). CAPÍTULO 2 | 117 Figure C2. Asymmetry in population performance at High Latitude Margins (HLM) and Low Latitude Margins (LLM) for each Kingdom and Habitat. Symbols representing Habitats and Kingdoms are centered on the mean estimate. Vertical bars represent 95% CI estimated from the multi-level meta-analysis. Negative and positive values indicate lower and higher performance of marginal populations as compared to central populations, respectively. Numbers within parentheses indicate the number of case studies for each category.  | APÉNDICES118 APÉNDICE D META-ANÁLISIS ACUMULATIVO CAPÍTULO 2 | 119 Figure D1. Cumulative meta-analysis. Grand mean effect sizes (dots), 95% CI (bars) and sample sizes (k) are shown for each year, including all previous years. Plate (A) depicts the global data set, plates (B) and (C) the datasets for HLM and LLM populations, respectively. Only significant relationships between publication year and effect sizes are shown by a regression line (continuous) and its 95% CI (dotted).  | APÉNDICES126 Figure G1. Genetic diversity accumulation curves for sampled acorn progenies of some illustrative mother trees (Q11, Q12, Q3 y Q4). The number of embryos at which the number of alleles stabilizes (ca. 10) represents a sufficiently large sample size for adequately characterizing the genetic diversity of acorn families. Number of embryos Cumulative number of alleles CAPÍTULO 3 | 127 APÉNDICE H DIFERENCIACIÓN ENTRE NÚCLEOS POBLACIONALES Y ! ESTRUCTURA GENÉTICA DE LOS ADULTOS Table H1. Pairwise differentiation (FST, below diagonal) and pairwise conditional genetic distance (cGD, upper diagonal) among stands. FST values different from zero according to permutation tests are shown in bold (when P < 0.01) or bold italics (when P < 0.05). C D E F I J K L M N C 1.590 1.721 1.663 1.471 1.996 0.762 1.536 1.661 1.050 D 0.180 2.061 0.560 1.835 2.027 1.978 1.933 1.989 1.270 E 0.149 0.140 1.684 1.589 2.010 2.027 1.668 1.704 1.584 F 0.161 0.005 0.103 1.661 1.867 2.074 1.779 1.821 1.219 I 0.194 0.196 0.126 0.150 0.742 1.400 0.580 0.557 0.700 J 0.168 0.135 0.110 0.121 0.103 1.882 0.815 0.655 1.097 K 0.114 0.138 0.115 0.138 0.128 0.098 1.432 1.529 1.143 L 0.200 0.198 0.154 0.160 0.129 0.120 0.139 0.589 0.857 M 0.235 0.218 0.156 0.175 0.186 0.108 0.130 0.097 0.901 N 0.135 0.093 0.093 0.087 0.069 0.073 0.079 0.086 0.123  | APÉNDICES128 Figure H1. Posterior probability and change in likelihood used for detecting the K value that best described the adult genotypes. Plot of mean likelihood L(K) and variance per K value from STRUCTURE on a dataset containing 135 adult Quercus robur trees genotyped at 20 polymorphic microsatellite loci (upper panel). Change in likelihood (delta K) plot as proposed by Evanno et al. (2005) points to k = 4 as the best number of groups fitting the data (lower panel). Results generated by STRUCTURE have been collated and summarized using the program STRUCTURE HARVESTED. K Delta K K Mean of est. Ln prob of data CAPÍTULO 3 | 129 APÉNDICE I ANÁLISIS DE LOS EVENTOS DE HIBRIDACIÓN Figure I1. Bar plots of STRUCTURE ancestry proportions for K = 2 clusters including unassigned acorns (n = 35, individuals indicated as group 1 below the plot), adult Q. pyrenaica trees (n = 109, group 2) and adult Q. robur trees (n = 135, group 3). Each individual is represented as a line segment, which is vertically partitioned with different colours representing the individual’s assignment to one of the two clusters (green or red). In the figure below, change in likelihood (delta K) is presented, which point to K = 2 as the best value describing the data. K Delta K Ancestry proportion Unassigned acorns Q. pyrenaica adults Q. robur adults  | APÉNDICES130 Table I1. Probability of non-assigned acorns to stem in first generation from a hybridization event with Q. pyrenaica. Assignment probability of acorns to clusters “Q. robur” and Q. pyrenaica” was computed in STRUCTURE. From 35 acorns without identified fathers, only 4 are likely to be first generation hybrids (i.e., probability of assignment to cluster “Q. pyrenaica” 0.4 < value < 0.6). Acorn ID Qrob_cluster Qpyr_cluster Resulting assignment Unass_1 0.996 0.004 Qrob Unass_2 0.995 0.005 Qrob Unass_3 0.995 0.005 Qrob Unass_4 0.995 0.005 Qrob Unass_5 0.996 0.004 Qrob Unass_6 0.995 0.005 Qrob Unass_7 0.995 0.005 Qrob Unass_8 0.995 0.005 Qrob Unass_9 0.996 0.004 Qrob Unass_10 0.995 0.005 Qrob Unass_11 0.994 0.006 Qrob Unass_12 0.996 0.004 Qrob Unass_13 0.994 0.006 Qrob Unass_14 0.997 0.003 Qrob Unass_15 0.588 0.412 Hybrid Unass_16 0.696 0.304 Qrob Unass_17 0.758 0.242 Qrob Unass_18 0.872 0.128 Qrob Unass_19 0.994 0.006 Qrob Unass_20 0.995 0.005 Qrob Unass_21 0.994 0.006 Qrob Unass_22 0.986 0.014 Qrob Unass_23 0.825 0.175 Qrob Unass_24 0.846 0.154 Qrob Unass_25 0.981 0.019 Qrob Unass_26 0.913 0.087 Qrob Unass_27 0.557 0.443 Hybrid Unass_28 0.996 0.004 Qrob Unass_29 0.996 0.004 Qrob Unass_30 0.992 0.008 Qrob Unass_31 0.967 0.033 Qrob Unass_32 0.499 0.501 Hybrid Unass_33 0.320 0.680 Hybrid Unass_34 0.743 0.257 Qrob CAPÍTULO 4 Mating patterns and their ecological drivers in long-term refugial populations of Pedunculate oak (Quercus robur)  | APÉNDICES132 APÉNDICE J ESTIMAS BASADAS EN LOS MODELOS DE CRUZAMIENTO (MEMM) Table J1. Summary descriptors of mating patterns in the four largest forest stands according to MEMM simulations: rates of immigration (m) and selfing (s), median dispersal distance (δ) and shape parameter (b), as well as the ratio of observed (dobs) and effective (dep) densities of male reproductive trees. The 95% confidence intervals [shown in brackets] were obtained by computing the 2.5% and 97.5% quantiles from 50,000 retained values of the parameter in the MCMC simulation. Forest stands showing significant differences (i.e., non-overlapping CI) are identified by different superscript letters. The selfing rates were sJ = 9%, sK = 9.9%, sDF = 1.1%, sE = 1.8% and significantly differed among stands with J and K having higher rates than DF and E, according to the CI. Similarly, the average pollen migration rates (mK = 7%, mDF = 15%, mJ = 19% and mE = 21%) were significantly different between stands K and E, and between stands K and J. The average pollen dispersal distance, δ, ranged from 138 m (stand DF) to 672 m (stand E). MEMM yielded longer dispersal distance estimates than previous empirically-estima- ted frequency distributions of pollen dispersal distances obtained by CERVUS in the study area (ranged from 59.6 to 185.4 in stand J and DF, respectively). This is expected given that MEMM estimates include all trees in each population and the effect of cryptic gene flow, contrary to CERVUS estimates. The shape parameter, b, of the pollen dispersal kernel (b<1; range: 0.33 - 0.57) indicated a non-negligible amount of long distance dispersal events. Stand DF tended to exhibit a higher b (bDF = 0.57) and a lower δ (δDF = 138 m) than the other three populations, but pollen dispersal parameters did not significantly differ between the stands (the credibility intervals, CI, overlapped). E J K DF m 0.21 [0.15, 0.27]a 0.19 [0.13, 0.26]a 0.075 [0.03, 0.13]b 0.15 [0.10, 0.22]ab s 0.018 [0.004, 0.048]a 0.090 [0.051, 0.150]b 0.099 [0.051, 0.170]b 0.011 [0.001, 0.041]a δ 668 [286, 982]a 570 [172, 975]a 595 [258, 974]a 142 [95, 503]a b 0.34 [0.24, 0.52]a 0.35 [0.23, 0.68]a 0.36 [0.21, 0.65]a 0.58 [0.26, 0.92]a dobs/dep 16 [4.5, 33]a 8.6 [2.6, 32]a 23 [7, 34]a 22 [6.9, 34]a CAPÍTULO 4 | 133 There is a extremely high variance of individual male fecundity. This resulted in highly skewed distributions of relative fecundity in all populations (Fig. L1). Figure J1. Individual variation on male fecundity for each stand Individual relative fecundity Frequency  | APÉNDICES134 APÉNDICE K RESUMEN DE LOS MODELOS DE DISPERSIÓN DE POLEN EMPÍRICOS Table K1. Ranking of candidate mating models of empirically-estimated pollen dispersal types based on the Akaike information criterion (AIC). To configure the candidate models, we pooled the explanatory variables in groups of predictors intrinsic to the tree (I), those related with the conspecific neighbors (N) and those related with the physical environment (E). Then, all possible combinations of these 3 groups of variables were modeled (including the null model): I, N, E, I + N, I + E, N + E, I + N + E. ∆ AIC, difference between the best and the second model. AICwt, probability that the model will be the best model in the set. Response variable Candidate model AIC ∆ AIC AICwt Selfing I + N + E 152.46 0.00 0.64 I + N 153.79 1.33 0.33 I 159.22 6.76 0.02 I + E 163.00 10.54 0 N 209.16 56.70 0 N + E 209.95 57.50 0 null 232.55 80.10 0 E 236.24 83.78 0 Local pollination I + N 248.83 0.00 0.73 I + N + E 250.77 1.94 0.27 I 269.70 20.88 0 I + E 272.69 23.86 0 N 278.16 29.34 0 N + E 281.75 32.93 0 null 325.42 76.59 0 E 325.89 77.07 0 Pollen immigration N 172.96 0.00 0.63 N + E 175.15 2.19 0.21 I + N 176.29 3.33 0.12 I + N + E 178.46 5.50 0.04 I 194.51 21.55 0 I + E 195.56 22.60 0 E 195.76 22.80 0 null 197.60 24.64 0 CAPÍTULO 4 | 135 APÉNDICE L ANÁLISIS DE CORRESPONDENCIA CANÓNICA BASADOS EN LOS MODELOS DE CRUZAMIENTO (MEMM) Table L1. Test of canonical dimensions. Two independent CCA were applied to mating variables from a mother and father perspective. Four and five dimensions respectively, were inferred following the number of mating variables in the analysis. CC, canonical correlations; R2, squared canonical correlations; CR, canonical redundancies. Perspective Dimension CC R2 CR Bartlett's test (ρ2) χ2 df P Mother 1 0.77 0.59 0.16 0.59 170.76 30 <0.001 2 0.62 0.39 0.06 0.39 93.85 20 <0.001 3 0.58 0.34 0.07 0.34 51.67 12 <0.001 4 0.40 0.16 0.04 0.16 15.79 6 0.015 5 0.90 0.01 0.00 0.01 0.70 2 0.704 Father 1 0.64 0.41 0.22 0.41 69.64 24 <0.001 2 0.43 0.19 0.03 0.19 22.38 15 0.098 3 0.17 0.03 0.00 0.03 3.83 8 0.872