Red clover (Trifolium pratense L.) is the most important perennial forage legume in Northern and Central Europe. It is either grown in pure stands or in mixtures with tall grasses. Due to its ability to fix atmospheric nitrogen, red clover produces protein-rich forage and, when grown in mixtures, can provide additional nitrogen to companion grasses (Broderick, 1995; Taylor and Quesenberry, 1996; Halling et al., 2004; Nyfeler et al., 2011). Grass-legume mixtures have the additional benefit of reducing nitrogen leaching, increasing forage quality, and improving drought resilience (Lüscher et al., 2014; Hofer et al., 2016). Red clover is, therefore, an important component of sustainable grassland-based animal production and can substantially contribute to the protein self-sufficiency of Europe.

Red clover is native to Europe, the Middle East, North Africa and Central Asia, and was introduced to most temperate regions in the world as a forage crop (Taylor and Quesenberry, 1996; Boller et al., 2010). The benefits of red clover were readily appreciated by farmers, when they realized that the replacement of fallow by clover led to higher yields and better nutrient availability in the fields (Weir, 1926; Taylor and Quesenberry, 1996). Consequently, farmers started to harvest red clover seeds on farm for sowing in the crop rotation or for improving existing grasslands. This has led to the development of numerous landraces adapted to the specific management regimes and environmental conditions of the sites they emerged from (Taylor, 2008). Many of these landraces survived as naturalized populations, were integrated in breeders’ collections, or were conserved in genebanks. Together with other germplasm collections, these landraces laid the foundation for systematic breeding of red clover, which started in the second half of the 20^th century and is nowadays conducted in research organizations or small to large multinational companies throughout Europe.

Red clover is a naturally diploid (2n = 2x = 14), insect pollinated, outbreeding species with a high degree of self-incompatibility. Tetraploid red clover accessions have been produced through colchicine treatment, but despite their higher yield, their market success is hampered by low seed yield (Boller et al., 2010). In order to maintain a certain level of genetic diversity to maximize adaptability to a broad range of environmental conditions and due to its self-incompatibility, red clover is usually bred as population-based cultivars. Breeding schemes are mainly based on recurrent phenotypic selection either to directly develop open pollinated cultivars through population-improvement or to select suitable parental plants for mutual intermating (poly-crossing) and the development of synthetic cultivars (Posselt, 2010). The base populations used for breeding may consist of ecotype or wild populations, landraces, breeding material as well as commercial cultivars and are continuously complemented to ensure sufficient diversity in target traits. Specific breeding goals may differ between individual programs, but general breeding targets include high biomass and seed yield, persistence over the desired cultivation period and disease resistance (Abberton and Marshall, 2005; Taylor, 2008; Riday, 2010). In the northern regions of Europe, particular emphasis is put on improving freezing tolerance and resistance to clover rot, two factors strongly associated with persistence. Clover rot, caused by the ascomycete Sclerotinia trifoliorum Eriks., is favored by cooler temperatures, high humidity and long snow cover, and can lead to severe overwintering damage (Saharan and Mehta, 2008). Freezing tolerance is associated with winter survival and is particularly pronounced in wild populations from northern regions (Zanotto et al., 2021). In the warmer climates of Central Europe, persistence is an important breeding target with the specific aim to extend the period of cultivation for red clover cultivars over several growing seasons. Based on landraces developed through decades of on-farm seed production, a particular type of red clover (‘Mattenklee’) evolved and a number of highly persistent and high yielding cultivars were developed from this type (Kölliker et al., 2003; Boller et al., 2010). In the same region, resistance to southern anthracnose has gained importance as a breeding target. The disease is caused by Colletotrichum trifolii Bain & Essary and has benefitted from warmer summer temperatures, which has made it a limiting factor of red clover production in the warmer regions of Europe (Boller et al., 2010). The disease has long been recognized as a major threat in Southern USA, where targeted selection resulted in cultivars largely resistant to southern anthracnose (Taylor, 2008). North American red clover accessions are also often distinguished by their characteristic hairiness that is thought to aid in preventing leafhopper damage (Pieters and Hollowell, 1937). To account for different utilization requirements, different growth types have been developed. Early flowering clovers are capable of vigorous regrowth and remain generative throughout the growing period. Late flowering clover, termed ‘single-cut’ or ‘mammoth’ clover in the USA, remain vegetative after the first cut and are generally more persistent (Bird, 1948). Other types, such as the Swiss ‘Mattenklee’, combine the rapid regrowth of early flowering red clover with the persistence of the late flowering types (Boller et al., 2010). Thus, the various breeding activities combined with the use of landraces and the introgression of wild populations for specific characteristics resulted in very diverse genetic resources, adapted to vastly different environments – ranging from arctic to mediterranean and continental climates. However, most of these breeding activities were focused on creating cultivars optimized for local or regional needs, and little effort was spent on breeding for a broader range of environments. The latter may be particularly important in the future in view of rapidly changing environmental conditions through climate change and shifts in priorities for land-use.

Plant genetic resources, i.e., the genetic material contained in wild relatives, ecotypes, landraces, and cultivars, are crucial for the continued improvement of modern forage crop cultivars (Boller and Greene, 2010). For red clover, the European Search Catalogue for Plant Genetic Resources (EURISCO; https://eurisco.ipk-gatersleben.de/apex/eurisco_ws/r/eurisco/home) currently lists more than 10,600 red clover accessions and many more are likely to be found in non-European or local germplasm collections. For targeted utilization in breeding programs, detailed phenotypic and genotypic characterization is indispensable, but is often not available for many of the current genetic resources. Molecular genetics or DNA-based markers allow a rapid characterization of genetic diversity in a large number of samples, independent of environmental effects (Bachmann, 1994). PCR-amplification based marker systems targeting either many unspecific loci (e.g., amplified fragment length polymorphism (AFLP) markers; Vos et al., 1995) or single highly variable loci (e.g., simple sequence repeat (SSR) markers; Tautz, 1989) have long been the methods of choice for analyzing genetic diversity in forage crop species where genome sequence information has been scarce. More recently, highly cost-effective genotyping methods based on high-throughput sequencing of restriction site-associated DNA such as RAD-seq (Baird et al., 2008) or genotyping-by-sequencing (GBS; Elshire et al., 2011) have become available. These methods allow an even larger number of samples to be efficiently genotyped at several thousand loci. In red clover, molecular markers have been successfully used to investigate the origin of red clover types and cultivars (Semerikov et al., 2002; Kölliker et al., 2003), the genetic structure of germplasm collections (Herrmann et al., 2005; Dias et al., 2008; Gupta et al., 2017) and the distribution of genetic diversity over a broad geographic range (Jones et al., 2020). As expected for an outbreeding species, a large proportion of the genetic variation is observed between individuals within populations (Kölliker et al., 2003; Dias et al., 2008), making it necessary to analyze multiple individuals to capture the genetic diversity present in a population. Pooling leaves of 20 individual red clover plants prior to DNA extraction has been proven effective to analyze genetic diversity on a population level (Herrmann et al., 2005). In perennial ryegrass, GBS on pooled leaf samples has proven useful to establish genome wide allele frequency fingerprints and to differentiate populations (Byrne et al., 2013). Despite the large number of accessions available in genebanks, most studies have used a rather restricted number of accessions and/or individual plants or focused on specific geographic regions. Jones et al. (2020) investigated genetic variability of 75 red clover accessions using eight to 16 individual plants per accession and more than 8,000 single nucleotide polymorphism (SNP) markers. The accessions were clearly grouped into four groups, corresponding to their geographic origin (Asia, Iberia, UK and Central Europe). The largest study so far investigated genetic diversity of 382 Nordic red clover accessions using 661 SNPs (Osterman et al., 2022).

Although studies on genetic diversity can substantially assist breeding decisions, knowledge about phenotypic diversity is key to efficiently exploit red clover germplasm for cultivar development. While some studies have investigated the variability and inheritance of specific traits such as disease resistance (Frey et al., 2022), seed yield (Herrmann et al., 2005) or flowering time (Jones et al., 2020), only a few studies have investigated phenotypic diversity of several traits within a region or country and even fewer have studied diversity across regions. A study by Zanotto et al. (2021) that was conducted with 48 ecotypes and six cultivars from Norway, Sweden, and Finland found large variation in winter survival and yield between red clover accessions. Wild accessions sometimes outperformed commercial cultivars, indicating their value for improving adaptation to colder climates. Winter survival was found to vary considerably between the four locations and could only partially be predicted by a test under controlled conditions (Zanotto et al., 2021). Furthermore, in a meta-analysis of legume yields in trials conducted in several locations ranging from Southern Germany to Northern Norway, Halling et al. (2004) found large differences in forage yield not only across different environments but also between early, intermediate and late flowering accessions. These findings highlight the need to account for accession x location interactions and to characterize breeding germplasm in their respective environments.

In order to provide a solid foundation for the utilization of red clover genetic resources in breeding programs, we aimed at (i) establishing a representative, diverse collection of red clover accessions, (ii) characterizing the genetic diversity and structure present in this collection, (iii) assessing the phenotypic diversity of agriculturally relevant traits at multiple locations and (iv) getting an insight into the extent of accession x location interaction for future adaptive breeding efforts.

This study, to the best of our knowledge, characterised the most comprehensive collection of red clover accessions to date. By evaluating the collection on field plot level at diverse locations, a wide range of environmental conditions across Europe were covered. Combined with population-level allele frequency genotyping, this allowed to relate genetic diversity to phenotypic performance.

The genetic structure of the EUCLEG panel largely reflected the geographical origin of the accessions. Ecotypes from Southern Europe and some cultivars from non-European countries were clearly separated from the remaining accessions ( Figure 2 ). A similar pattern was observed in an analysis of a global red clover collection, where accessions could be grouped into four regional groups (Asia, Iberia, UK and Central Europe) based on GBS of single plants (Jones et al., 2020). In the present study, accessions from Central and Eastern Europe clustered together, while accessions of Northern Europe and Switzerland each formed distinct clusters. Within the Northern European and Swiss clusters, sub-clusters were identified that distinguished breeding material from landraces ( Figure 3 ). The genetic similarity of landraces and breeding material is in agreement with the historic origin of breeding programs that mostly started with improving local germplasm (Boller et al., 2010). Still, systematic improvement through breeding for traits like disease resistance or yield has led to populations that are distinctly different from the original landraces as indicated by the separate clustering of breeding material and landraces within the Swiss and Nordic accessions ( Figure 3 ). A separation of ecotypes from cultivars has also been demonstrated recently in a study on 29 Nordic red clover accessions using genetic markers (Osterman et al., 2021).

Closer inspection of the genetic structure of the EUCLEG red clover diversity panel reveals a few particularities that may reflect specific breeding histories. For example, the cultivar ‘Merian’ (TP394) from Belgium clustered with the Swiss landraces, indicating the latter being the initial source of breeding germplasm used ( Figure 3 ). According to information from the breeder, the cultivar ‘Merian’ orginates to 75% from ILVO breeding material from before 1970, which may also have included some Swiss landraces (T. Vleugels, personal communication). An interesting cluster of three accessions was observed at the PC1, PC2 coordinates -5, -0.5 ( Figure 3 ): The Swiss landrace ‘Ueberstorf’ (TP074), as well as the cultivars ‘Grasslands Sensation’ from New Zealand (TP036) and ‘AberClaret’ from the United Kingdom formed a distinct group and the Pearson correlations of their allelic profiles ranged from 0.89 (TP036/TP074) to 0.94 (TP006/TP074; data not shown). ‘Grasslands Sensation’, also known as ‘G40’ or ‘Swiss’ was bred based on four Swiss red clover cultivars, i.e. ‘Renova’, ‘Mont-Calme’, ‘Leisi’ and ‘Changins’, all tracing back to Swiss landraces (Kölliker et al., 2003; Claydon et al., 2010). ‘AberClaret’ and the cultivar ‘AberChianti’ (TP007; Figure 3 ), are advertised as high-yielding up to their fourth and fifth harvest year (Marshall et al., 2017), a typical trait of Swiss ‘Mattenklee’ cultivars and landraces, to which they show considerable similarity in the PCA ( Figure 3 ). The Canadian cultivar ‘Altaswede’ (TP105) clustered within the Northern European accessions. This is in agreement with it being described as ‘single cut’ clover (Valle, 1958), the predominant phenotype of red clover in Northern Europe. Indeed, ‘Altaswede’ was selected by the University of Alberta, Canada, based on Swedish seed stock around 1919 (Aasen and Bjorge, 2009). These examples are clear evidence that red clover breeders sometimes make use of the ‘breeder’s exemption’, i.e., the right to use registered cultivars from the market in their own breeding programs and integrate desired traits into their own breeding pools. Nevertheless, the distinct grouping of the accessions according to their region of origin suggests that this is, so far, not frequently done and breeders mostly rely on their own, local genetic resources. This is in contrast with many other breeding programs of crops such as wheat or rice, where the incorporation of foreign material is far more widespread (Garrett et al., 2017; Luttringhaus et al., 2020).

While the characterisation of genetic diversity is an important first step in the efficient utilization of genetic resources, targeted incorporation of germplasm into the breeding process requires detailed phenotypic characterisation. Conducting field trials with perennial forage legumes is challenging. First, the establishment phase is critical to ensure reliable data in the following years. Second, pest, disease or weed occurrence can substantially reduce the plant density after each cut or during harsh winters (Boller et al., 1998; Öhberg et al., 2008; Zanotto et al., 2021), having a strong impact on the data assessed in subsequent cuts and growing seasons. Although the multi-site trials in this study were all managed by experienced red clover researchers and field teams, there were some setbacks during the trial period. The trial in CZE had to be abandoned after year one, as the damage caused by mice and Fusarium spp. was too severe to get reliable data in subsequent years. Several other trials experienced disease pressures that resulted in the loss of several accessions and missing data in the following cuts. Despite the difficult conditions encountered at some trial locations, heritability values of field trials were, with a few exceptions, at a high level (h² > 0.6, Figure 6 ) and similar to heritability values reported in other perennial forage crops (Annicchiarico et al., 1999; Han et al., 2006; Li et al., 2015).

A trait showing low heritability across all trial locations was CP. The reason for this is not entirely clear, but may be associated with the low overall variability for this trait (coefficient of genetic variation (CV_g) = 0.07-0.18, Supplementary Table S2 ). Only the heritability of CP of the second cut (CP_Y1.C2) for trial site NOR was satisfactory with h² = 0.72. It was highly correlated with DOF (r = 0.76, Supplementary Figure S3 ) and negatively correlated to DMY_Y1.C2 (r = -0.28). Hence, accessions that had a high CP in C2 generally flowered later, yielded more in C1 and less in C2 and may have concentrated protein levels in C2 (data not shown). Despite several publications reporting red clover forage quality values (Schubiger et al., 1998; Hoekstra et al., 2018), to our knowledge no heritability values have been reported so far. Thus, it remains unclear whether the heritability is truly low or, whether there was not enough variation in the accessions studied. The latter has previously been reported for alfalfa, where limited variability led to low narrow sense heritability of CP (Guines et al., 2002). In addition, the CP levels of red clover are already high compared to forage grasses (Schubiger et al., 1998) and therefore not a primary breeding goal. In alfalfa, leaf-stem ratio influences forage quality and is an important breeding aim (Annicchiarico, 2015).

During the experiment, the plant density of several red clover accessions declined at specific locations. In NOR, the plant density in autumn of Y1 was on average 39% for Eastern European accessions compared to 63% for native Nordic accessions (data not shown). In comparison, in SRB the plant density was on average 83% for Nordic accessions compared to over 90% for accessions from other regions. The overall high plant density in SRB after Y1 ( Figure 9 ) is likely caused by the late sowing (autumn compared to spring for other trials) and the mild winter following sowing. In the Swiss trial, which experienced high southern anthracnose disease pressure, average plant densities after Y1 were highest for local, American and Eastern European accessions (62%, 57% and 59%, respectively) and lowest for Northern European accessions (21%), which is in agreement with the level of southern anthracnose resistance observed for these accessions in artificial inoculation experiments (Frey et al., 2022). The low plant density after Y1 resulted in non-satisfactory data quality in the second year for many accessions. Often, accessions that failed to persist were of foreign origin and not adapted to the conditions of the trial location, which indicates a strong effect of geographical origin on performance. Indeed, the clustering by geographical origin of accessions was also apparent when phenotypic data per trial was analysed ( Figure 8 ). The genetic distinctness of these populations is, thus, manifested in their phenotypic performance. It also indicates a degree of local adaptation among accessions originating from similar environments.

Almost 80 years ago, Pieters and Hollowell (1937) reported that “claims of superiority are made for all of these regional strains, and the evidence of comparative trials shows that in most cases each such regional variety is superior to others in the environment where it was developed”. The present work provides strong support for this hypothesis with trial locations covering a wide range of eco-environmental conditions relevant for European red clover cultivation areas. Breeding material and cultivars from Northern Europe showed best performance at the NOR location regarding DMY and vigor ( Figure 9 ). This trend of local adaptation with accessions performing well at the location where they were bred was also observed for other trial locations and is in agreement with previous findings (Valle, 1958; Rosso and Pagano, 2005). In the CHE trial, Swiss breeding materials and cultivars showed best performance regarding DMY_Y1 and VIG_Y1, while in the GBR trial Central European accessions performed best. In CZE and SRB, Eastern European breeding materials performed best ( Figure 9 ). As highlighted above, performance of red clover is often driven by persistence, and the plant density of poorly adapted accessions declines throughout the growing season and/or winter. While red clover has a natural senescence, its persistence can be improved by improving resistance for relevant diseases that occur in the growing location (Taylor, 2008). Depending on the region, different traits are relevant for persistence and hence have resulted in accessions adapted to the location of origin.

The five trial locations cover a wide range of latitudes and climate zones, and large accession x location interactions were observed. From correlation analysis among locations ( Figure 7 ), we could show that especially the Northern European trial location (NOR) differed from the remaining sites. Negative correlation coefficients indicated the presence of cross-over interactions, i.e. that ranking of accessions is reversed in NOR compared to other locations. This observation clearly highlights the need for locally adapted breeding programs. We observed a strong adaptation of Northern European accessions to the prevailing conditions and/or management. According to Boller et al. (2010), European cultivars of red clover can be grouped by flowering time: early-flowering types adapted to southern latitudes are capable of rapid regrowth and continuous flowering, while late-flowering types adapted to northern latitudes remain vegetative after the first cut. This may explain the poor performance of Northern European accessions of red clover in the more southern locations, where the cutting regime is optimized for fast regrowing types. At these locations, the first cut is most likely too early for Northern European accessions, which show lower yield as they are not able to resume generative growth afterwards. In their native habitat of NOR, the first cut happened 1-1.5 months later than in the remaining trial locations and allowed the Northern European accessions to reach their full yield potential.

Due to the natural expansion across Eurasia and Northern Africa and the recent domestication at several locations (Kjærgaard, 2003), red clover has evolved as a very diverse crop. The outcrossing nature of red clover has, in addition to the large between-accession diversity, resulted in an even higher within-accession diversity (Kölliker et al., 2003; Dias et al., 2008; Pagnotta et al., 2010; Collins et al., 2012). This allows red clover accessions to rapidly adapt to new challenges. Being a perennial species, adaptation in red clover has even been detected during continuous cultivation without allowing reproduction. Collins et al. (2012) found considerable genetic shifts within accessions grown at locations different than their origin after three to five years of continuous cultivation. Especially the Swedish environment was very selective and composite accessions became more similar to the Swedish accessions than to their original accession (Collins et al., 2012). Not only different environments but also changing environmental conditions between years may contribute to allelic diversity, as was shown in natural populations of perennial ryegrass (Keep et al., 2021).

A peculiarity of forage crops is the co-existence of wild relatives and cultivated forms in many landscapes which carries the potential for intercrossing. Diversity of wild red clover accessions has been shown not only to be influenced by the geographic location, but also by the proximity to sites where landraces have been grown for decades (Herrmann et al., 2005). This indicates potential geneflow between wild and cultivated forms. It also demonstrates the potential of utilizing wild germplasm for improving forage crop cultivars. Potentially useful adaptation to environmental conditions has also been demonstrated for wild alfalfa and other Medicago ssp., where candidate genes associated with adaptation to environmental gradients have been identified (Blanco-Pastor et al., 2021). Thus, variability in environmental conditions seems to drive genetic diversity in forage crop accessions and cultivars, which is a prerequisite for adaptation to changing climatic conditions.

As indicated by the large accession x location interactions observed in our study, breeding a ‘one-fits-all’ cultivar seems to be very unrealistic, due to the very variable environmental conditions demanding different adaptive traits from the genetic material as discussed above. However, there are trial locations that were more similar to each other based on assessed trait values and cultivars being able to perform well in such regions are more likely to be found. Growing a reduced set of diverse accessions at more locations throughout Europe might help to define such macro-environments representing the different red clover growing conditions. Knowledge of available genetic resources and their characteristics is key for breeding red clover well adapted to environmental conditions. This will become even more important in view of changing climatic conditions. For example, with increasing temperatures diseases adapted to warmer climates like southern anthracnose, which has already extended its local distribution (Jacob et al., 2016; Hartmann et al., 2022), will most likely continue to spread to Northern latitudes. Materials resistant to this disease might be a valuable source to complement the genetics from other regions not yet affected. The present work will serve as a valuable basis to identify interesting materials expressing desired characteristics. In conclusion, the detailed description of the EUCLEG red clover diversity panel provides an important source for further targeted improvement of red clover. The knowledge gained about the importance of local adaptation and genetic differentiation will help to design novel breeding strategies that consider broader scale adaptation of future cultivars. Additionally, the large amount of SNP markers provided for the EUCLEG red clover diversity panel, together with phenotypic data assessed in the presented field trials or other complementary phenotypic studies, may enable the implementation of genome wide association studies to unravel the genetic architecture of agronomically important traits.

The datasets presented in this study can be found at https://zenodo.org/deposit/7034131.

Study design and curation of the accession collection: CG, RK, and AP. Generation of phenotypic data: CG, HA, JR, LJ, and LS. Generation of genotypic data: LS, TR, and LF. Data analysis and interpretation: MN, RK, CG, and LF. Drafting of the manuscript: MN, RK, and CG. All authors contributed to the article and approved the submitted version.