Crop wild relatives range shifts and conservation in Europe under climate change

Climate change is expected to have a great impact on the distribution of wild flora around the world. Wild plant species are an important component of the genetic resources for crop improvement, which is especially important in face of climate change impacts. Still, many crop wild relatives (CWRs) are currently threatened in their natural habitat and are poorly represented in gene bank collections. To guide in situ conservation measures and to prioritize species for ex situ conservation, predictions are needed about future species distributions as a result of climate change.

indispensable source of useful traits for crop improvement when these traits are not found in the cultivated species (Van de Wouw, Kik, van Hintum, van Treuren, & Visser, 2010). For instance, resistance breeding against new pests and diseases relies to a large extent on the presence of traits in crop wild relatives (CWRs) (Hajjar & Hodgkin, 2007; Van Treuren, Van der Arend, & Schut, 2013). Crop productivity and food security may be endangered when wild genetic resources are no longer available; therefore, the safeguarding of CWRs is widely regarded as a high priority (Maxted et al., 2010). Despite their high ecological and economic value for food and agriculture given by their trait availability, wild species are currently severely underrepresented in ex situ genetic resource collections (Castañeda-Álvarez et al., 2016; Commission on Genetic Resources for Food and Agriculture, 2010). In their natural habitats, the survival of many species is at risk due to human influences, such as urbanization and pollution. During the last decades, the in situ survival of species has become a growing concern as a result of climate change (Bilz, Kell, Maxted, & Lansdown, 2011;Dempewolf et al., 2014).
Changes in the earth climate system since the 1950s are undisputed. Increased temperatures have been recorded in nearly all regions of the world and changes in precipitation patterns are now evident (Stocker et al., 2013). It is considered extremely likely that these changes are predominantly human driven and that further changes can be expected in the future due to continued greenhouse gas emissions (Stocker et al., 2013). The magnitude of the expected changes depends on many factors, and therefore, different climate scenarios have been developed, denoted as the Representative Concentration Pathways (RCPs). Four RCPs (2.6, 4.5, 6.0 and 8.5) have been established based on the approximate radiative forcing in the year 2100 relative to the preindustrial conditions with baseline in 1750 (van Vuuren et al., 2011). The radiative forcing is defined as the influence of a factor for disturbing the incoming and outgoing energy in the earth's atmosphere (IPCC, 2007). RCPs are greenhouse gas concentration trajectories of which RCP 2.6 represents an optimistic scenario (greenhouse gas emissions decline after 2020) and RCP 8.5 a pessimistic scenario (greenhouse gas emissions continue to increase). According to the optimistic RCP 2.6, the global mean surface temperature is expected to increase 0.3-1.7°C in the period between 2081 and 2100, whereas for the pessimistic RCP 8.5, the changes in temperature range between 2.6 and 4.8°C (Stocker et al., 2013).
It is expected that climate change will affect agricultural productivity and the survival probability of plant species in their natural habitats (Jarvis, Lane, & Hijmans, 2008). For wild relatives of peanut, potato and cowpea, it has been estimated that due to climate change 16%-22% of their wild relatives will go extinct, while most species are expected to lose more than 50% of their distribution range (Jarvis, Upadhyaya, et al., 2008). Therefore, predictions about the future distribution of wild species are essential for the development of sound conservation strategies (Maxted et al., 2015). Effects of climate change on future species distributions can be estimated through the use of species distribution models (SDMs), in which the presence of a species at geographic locations is related to the environmental conditions at those sites. SDMs can thus render insights into the effects of the underlying environmental conditions on species distribution. Subsequently, this relationship is used to predict the probability of species occurrence at thus far unexplored geographical locations. For this reason, SDMs have been used to identify gaps in genetic resources collections and to support collecting missions (Cobben et al., 2015;Parra-Quijano, Iriondo, & Torres, 2012;Teeling, Maxted, & Ford-Lloyd, 2012). When projections of climate change scenarios are incorporated in SDMs, the probability of occurrence under future environmental conditions can be estimated. Effects of climate change on the species distribution range can then be analysed by comparing present and future projected distributions (Jarvis, Upadhyaya, et al., 2008;Thuiller et al., 2011).
Here, we use a set of spatially explicit species occurrence records of the known natural distribution of eight wild species related to wheat, turnip, rapeseed, mustard, pea, cyprus-vetch, alfalfa, mint, raspberry, blackberry, dewberry, black salsify and corn salad (Table 1).
These species were selected based on their relationship with food crops, their current protection level and their representation of different habitat types, life history, reproduction, pollinator vector, seed dispersal, rarity and availability of species occurrence data (see Section "2"; Table 1). We analysed the current representation of the study species in ex situ collections in Europe. Present and future climatic information was used along with species occurrences to investigate the impact of climate change on their distribution in Europe.
Given the current and predicted changes in climatic conditions in the next decades (Moss et al., 2010;Stocker et al., 2013), we expect that the CWRs show future range contraction of their current distribution if the current environmental conditions to which they are adapted become scarcer (Thuiller, Lavorel, Araujo, Sykes, & Prentice, 2005).
Additionally, given the greater projected warming and dryness in southern European regions in comparison with northern areas (Kovats et al., 2014), we expect range shifts for the species towards northern latitudes following their climatic niche. As we expect climate change to contract the distribution range of the studied species and to shift their spatial distribution, we also address these effects on the representation of the species in European protected areas.

| Species distribution data
Geographic occurrence data of the eight selected species for Central-Western Europe were obtained from the database of the Global Biodiversity Information Facility (GBIF, 2015). Records with missing geographic information were discarded. All species occurrence records were compiled at a resolution of 5 × 5 km grid cells to accommodate the higher uncertainty in geographic coordinates of the older records relative to the higher location accuracy of the more recent records. Multiple observations for the same grid cell were reduced to one, in order to obtain only unique records per species per grid cell. To account for possible spatial autocorrelation between closely located species presence records, only presence locations that were separated from each other by at least one grid cell were used. A total of 28,494 presence records were obtained for the eight species, ranging from 656 to 10,645 records per species (Table 1).

| Environmental data
Current bioclimatic conditions related to temperature and precipitation at a grid size resolution of 5 × 5 km were obtained for Europe from the WorldClim dataset (Hijmans, Cameron, Parra, Jones, & Jarvis, 2005). The bioclimatic variables represent annual trends in climatic conditions, seasonality and climate extremes, which may impact the reproduction and survival of vegetation (Hijmans et al., 2005). In addition, data on soil pH and topsoil organic carbon were obtained. These The results of the different climate models may differ (Murphy et al., 2004;Pierce, Barnett, Santer, & Gleckler, 2009), so to account for their variability, an ensemble of average values from 14 different climate models for the RCP 2.6 and 8.5 was used (Table S1). Details on the different climate models are provided by Flato et al. (2013). The same set of bioclimatic variables as used for predicting current distributions was used to predict future distributions. The soil-related variables were assumed to be constant across time and were only used to delimit the potential distribution of the species under current and future conditions. All analyses were carried out in R (https://www.r-project. org) using the RAsTER package.

| Species distribution modelling
We used an ensemble modelling approach based on a recent study comparing modelling algorithms and measures of model performance  (Phillips & Dudik, 2008) and Random Forest (Breiman, 2001). Single and quadratic terms were included for GLM, while linear and quadratic features were used for mAxEnT to avoid overparameterization (Merow, Smith, & Silander, 2013). Five hundred trees were used for Random Forest. To account for the within-algorithm model variation when different sets of data are used for model fitting, we computed distribution models for each species using ten model repetitions with a bootstrap approach where 80% of the presence data were used for model training and 20% for model testing.
The models' performance was assessed by the area under the curve (AUC) value (Hanley & McNeil, 1982). The obtained ensemble model was then used to predict species distribution under the future climatic conditions (RCP 2.6 and 8.5). All analyses were carried out in R (https:// www.r-project.org) with the bIomod2 package (Thuil l er et al ., 2014).

| Species range shifts
To investigate possible range changes between the current and future distributions (RCP 2.6 and 8.5), the species ensemble models were converted to a presence/absence binary prediction using the thresh-

| Species representation in ex situ collections
The representation of wild or weedy accessions of European origin in ex situ collections in Europe varied among the study species, ranging from 0 for E. gallicum to 504 for M. polymorpha (Table 2). Only for the latter species, a relatively high number of accessions was observed, mainly originating from southern European countries. Five of the eight study species were represented in EURISCO with <10 accessions and only three species presented information about their geographic collection location (Tables 2 and S2).  (Table 3). However, the threshold at which each variable delimited the distributions varied per species, underpinning their adaptation to often dissimilar environmental conditions (Fig. S2).

| Species range changes
Range contractions were predicted for six of the eight study species under RCP 2.6 (optimistic scenario), ranging from almost 21% loss for V. rimosa to 61% for B. secalinus under the full dispersal assumption.
Only M. polymorpha and M. pulegium, representing two of the most widely distributed study species under the current conditions, showed an increase in distribution area of almost 22% and 1%, respectively (Table 4)  climate change, especially in southern Europe (Stocker et al., 2013).

| Role of protected areas
The increase in temperature in the current distribution area is ex-  (Minteer & Collins, 2010). In case of limited natural migration, we expect the future distribution ranges to be considerably smaller than predicted in the present study for the Apart from dispersal aspects, other population biological factors may play a role in the future distribution of species. Crop wild relatives may show high phenotypic plasticity, which means that species may be able to cope with higher environmental variation than expected based on observed distribution patterns (Merilä & Hendry, 2014).
Moreover, the successful establishment of a species in new areas may depend on the interaction with other species occurring at suitable sites. In case of limited competitive abilities, migrating species may be out-competed by the indigenous species at new locations, preventing their establishment. Vice versa, the survival of indigenous flora may be jeopardized when migrating species possess strong colonizing abilities (Pyšek et al., 2012). A species may also rely on specific insects for its reproduction, which means that successful establishment at new locations depends on the presence or comigration of suitable pollinators.
Largely due to the limited availability of data, species interactions are lacking from most modelling frameworks (but see Giannini, Chapman, Saraiva, Alves-dos-Santos, & Biesmeijer, 2013). Notwithstanding the uncertainties related to niche modelling and future climate change scenarios, it remains undisputed that the climate is changing and that this will impact the distribution and survival of species. Considering the importance of wild relatives for crop improvement, we cannot afford to await actual distribution changes and associated loss of diversity. Instead, we need to anticipate the changes by developing sound conservation strategies.    Table S2) may suggest a lack of protection for these genetic resources in face of future environmental changes. To maintain the role protected areas carry out, they should be of sufficient size and provide a large variety in suitable habitats (Thomas et al., 2012).

| Implications for conservation
Moreover, for protected areas to work as in situ reservoirs of CWR, one of their main considerations should be to preserve the species genetic diversity based on the development and implementation of a set of sound management and conservation plans (see Iriondo et al., 2012).
In Europe, the Natura 2000 and CDDA network of protected sites currently covers over 25% of the EU's land territory. Although not all human activities are excluded from these sites, the network provides a sustainable approach for biodiversity conservation (European Commission, 2016). While at least 10% of the distribution of the analysed species is predicted to be within protected areas under current conditions, climate change, especially under RCP 8.5, is expected to reduce this representation for most species. Thus, current protected areas are no guarantee for future species survival. Species may even go extinct on a regional scale, such as predicted for B. secalinus and R. saxatilis in the Netherlands (Figure 1). Therefore, our results call for the application of national-level CWR genetic resources protection strategies that encompass in situ and ex situ conservation programmes. The adopted national-level conservation strategies should not be forgotten as it is first of national interest to conserve the genetic resources that are fundamental for food security (see Phillips et al., 2017). Moreover, we also show the need of applying an integrated transnational conservation strategy for important CWRs as that presented by the European Cooperative Programme for Plant Genetic Resources (ECPGR, Maxted et al., 2015). This strategy should not only encompass in situ conservation measures in protected areas (Meilleur & Hodgkin, 2004) but also ex situ conservation measures to guarantee the survival of these genetic resources.
To allow for the development of effective conservation strategies, more insight in the probability of survival of CWRs in their natural habitats under climate change conditions is needed. Therefore, the impact of climate change on the distribution of CWRs that may be used to improve the adaptation of crops to environmental changes, as climate change, should be determined. This approach should be combined with long-term monitoring programmes. This is especially important for rare and narrowly distributed species that due to their small range size and limited data availability are not well suited for range change analyses. Based on the outcome, genetic reserves can be identified where relatively stable populations occur (Maxted et al., 2015). For all species that appear vulnerable, seed samples should be collected and stored ex situ as backup collections to secure the option of future restoration and use. The preservation of these genetic resources is of primary importance as CWRs are valuable genetic resources that will help to improve our crops (Dempewolf et al., 2014;Nevo & Chen, 2010).

| CONCLUDING REMARKS
Our study predicts substantial changes in the distribution of eight CWRs in Europe for the 2070s based on two climate change scenarios. These changes include pronounced range contractions and range shifts towards northern areas for most of the studied species.
These changes can be expected to result in loss of genetic diversity.
We therefore suggest to investigate the impact of climate change on CWRs and to combine the results with long-term monitoring programmes and collecting of CWRs for ex situ backing up. These backup collections will secure the option of future restoration of wild populations and the use of these valuable resources in research and plant breeding.