Identification of mega-environments in Europe and effect of allelic variation at maturity E loci on adaptation of European soybean

Plant material and locations

The genetic material underlying this study consisted of 75 European soybean cultivars from five different MGs, 000, 00, 0, I and II, with 15 genotypes per group (Supporting Information Table S1). Notably, this grouping is based on the classification of each cultivar in their country of origin. The experiment was planted at 33 locations, of which 22 produced data for the analyses (Fig. 1). Several reasons were responsible for the 11 sites that dropped out of the analysis, such as the weekly evaluation being too late so that there was no differentiation among the genotypes, field trials being eaten by rabbits or data being not reliable or not available at all.

The trial locations reached from southern France with a latitude of 43.7° to northern Germany with a latitude of 54.1°. The maximum difference in day length was about 2 h between the locations. Supporting Information Tables S1 and S2 provide additional information, including country of origin of the cultivars, participants, zip code and town of the locations and their latitude.

Experimental design and data collection

Experiments were conducted in 2014 in observation plots as randomized complete block design with two replications for each MG. Field trial management was performed following best local practices and the experience of the participants in cultivating soybean. Supporting Information Table S2 shows the different sowing dates at each location. The field trials were evaluated by a weekly survey of the growth stage beginning with calendar week 23 until at least calendar week 40. The phenological growth stages were recorded either with the German BBCH Scale (Munger et al. 1997) or with the American Code from Fehr and Caviness (1977). Full bloom (flowering), which corresponds to the BBCH Code 65 or R2 (Fehr and Caviness, 1977), and maturity, corresponding to the BBCH Code 89/99 or R8, were recorded in weeks from sowing. Furthermore, the reproductive phase was calculated from these data as weeks from flowering to maturity.

Statistical analyses

Variance components were estimated by the restricted maximum likelihood (REML) method in a full random model, and the significance of the variance component estimates was determined by likelihood ratio tests. Heritability (h²) was estimated based on fixed genotypic effects following the approach suggested by Piepho and Möhring (2007). Best linear unbiased estimates (BLUEs) were calculated for single locations, across locations and across locations within MEVs, with the aforementioned models, but assuming fixed genotypic effects, using the ASReml-R 3.0 package (Gilmour et al. 2009). Genotypic correlations among traits were estimated as Pearson's correlation coefficients based on the BLUEs of each MEV. The entire data analysis was conducted with RStudio version 3.2.0 (R Development Core Team 2015).

The phenotypic distance matrix of the locations was calculated by summing up the BLUEs of all genotypes at each location and subsequently calculating the absolute differences of these values among all pairs of locations. The smaller this distance value between two locations, the more similar they are with regard to the phenotypic values. Genotypes that did not mature at a specific location were set to the fictive value of 30 weeks after sowing. MEVs were identified based on the clustering of similar locations. The phenotypic distance between the genotypes was calculated in a similar way by summing up the BLUEs of one genotype at all locations and subsequently calculating the absolute differences of these values among all pairs of genotypes.

Summary statistics

Each MG was analysed separately across different numbers of locations, depending on how many locations the specific MG matured. The genotypes of the early MG 000 matured at each location, while genotypes of the very late MG II only matured at 10 of the 22 locations. Genotype, location and the genotype-by-location interaction term showed a significant contribution to the variation of flowering and maturity in each MG, while there was no significant genotypic variation for the reproductive phase for MGs 000 and I (Table 1). Location showed the strongest contribution to variation of flowering, maturity and reproductive phase, with a much stronger effect than the genotype or the genotype-by-location interaction term. Heritabilities for the MGs were high for flowering and maturity, ranging from 0.70 to 0.93, whereas the heritability for the reproductive phase was lower and more variable between MGs, with 0.12, 0.48, 0.54, 0.72 and 0.80 for MGs I, 000, II, 00 and 0, respectively (Table 1).

Table 1. Summary statistics across all test locations

	Maturity group
	II	I	0	00	000
No. locations	10	13	18	20	22
Flowering (R2)
	0.053***	0.068***	0.178***	0.157***	0.083***
	1.943***	2.020***	1.915***	2.400***	2.545***
	0.1***	0.092***	0.144***	0.184***	0.115***
	0.104	0.186	0.196	0.129	0.105
h2	0.76	0.83	0.93	0.93	0.92
Maturity (R8)
	0.087***	0.098***	0.134***	0.222***	0.062***
	3.843***	4.501***	4.943***	4.725***	3.699***
	0.299***	0.124***	0.273***	0.211***	0.110***
	0.137	0.291	0.222	0.182	0.184
h2	0.70	0.83	0.84	0.93	0.89
Repro (R2–R8)
	0.062*	0.005	0.121***	0.06***	0.01
	3.239***	3.033***	2.375***	1.756***	1.529***
	0.389***	0.259***	0.211***	0.281***	0.145***
	0.217	0.349	0.48	0.281	0.275
h2	0.54	0.12	0.80	0.72	0.48

• Variance components of genotype ( ), location ( ) and genotype-by-location interaction ( ), as well as heritability (h²) for each maturity group and the three traits flowering, maturity and reproductive phase (repro).
• * Significantly different from zero at the 0.05 level of probability.
• ** Significantly different from zero at the 0.01 level of probability.
• *** Significantly different from zero at the 0.001 level of probability.

Identification of European mega-environments

The grouping of locations (Fig. 1) into MEVs was performed based on maturity because this is the essential criterion to determine if a genotype fits to a certain region and optimally exploits the available growth period. Based on the phenotypic distances between the locations, the heatmap with cluster plot revealed six MEVs with locations that were most similar to each other with regard to the time of maturity across all cultivars (Fig. 2). The first MEV (MEV1) comprises the four northernmost locations GER3, NED1, NED2 and BEL1. MEV2 groups the German and Austrian locations GER4, GER5, AUT1 and AUT3 together. MEV3 consists of SUI1, ROU1, AUT2, FRA1 and AUT4. MEV4 comprises CRO1, CRO3 and SUI2; MEV5 groups FRA2, ITA3 and SRB1; and MEV6 consists of the three southernmost locations FRA3, ITA2 and ROU4. The northern MEVs MEV1 and MEV2 were clearly distinct from the most southern MEVs MEV5 and MEV6. MEV4 clustered closer to the northern locations than MEV3, even though MEV3 is the more northern of the two MEVs. In general, however, locations that are on a similar latitude clustered together.

Characterization of maturity groups within mega-environments

Locations within one MEV were also similar with regard to the number of genotypes reaching maturity. In MEV1, the complete set of MG 000 matured, but also some genotypes of MG 00 reached maturity at the two more southern locations BEL1 and NED1. Three MGs reached maturity in MEV2 and even all MGs in MEV3 (Figs 2a & 3). Only three MGs matured in MEV4, whereas all cultivars matured in MEV5 and MEV6. The MGs were not strictly separated but nevertheless showed the expected trend, with genotypes of MG 000 always being the earliest in flowering and maturity. There were two outliers within the cultivars originally classified as belonging to MG 000, Diamant and Perla, that were always later maturing. Two general trends can be observed in these data: one for the different MGs within MEVs and another for each MG between MEVs (Fig. 3, Table 2). First, we observed that MGs in one MEV differ by around 1 week to maturity, with a slightly larger difference between MGs 000 and 00. The differences between the MGs within the MEVs become smaller when moving north to south. In general, MG 00 showed the largest variation for flowering and maturity compared to the other MGs. The second trend was a north–south gradient between MEVs, as each MG tended to become earlier in the more southern MEVs.

Allelic variation at the E1, E2, E3 and E4 maturity loci

Adjusted entry means (BLUEs) were calculated across 10 locations where almost all genotypes reached maturity, which enabled the comparison of all 75 cultivars from the different MGs (Fig. 4). The earliest cultivars were Gallec, Capnor and Abelina from MG 000, which needed 17.4 weeks from sowing to maturity, and the latest cultivars were Mitsuko and Blancas from MG II with 20.8 weeks from sowing to maturity. This difference of about 3.5 weeks observed across the 10 locations would probably be much greater across all 22 locations but cannot be assessed as the late genotypes did not reach maturity at the northern locations. Nevertheless, these data provide a robust basis to assess the effects of the E1, E2, E3 and E4 maturity loci.

None of the cultivars carried the wild-type, photoperiod-sensitive E1 allele, but two photoperiod-insensitive alleles were identified at the E1 locus, e1-nl and e1-as. At the E2 locus, the photoperiod-sensitive E2 allele and the photoperiod-insensitive e2-ns allele were found, with e2-ns and e2-dl being indistinguishable as they are in perfect linkage disequilibrium in this set of material. Four allelic variants of the E3 locus were identified, comprising two photoperiod-sensitive alleles (E3-Ha or E3-Mi), as well as two photoperiod-insensitive alleles e3-tr and e3-fs. Furthermore, two alleles at the E4 locus, the wild-type E4 allele and the photoperiod-insensitive e4-SORE-1 (e4), were found in these cultivars. Thus, a total of 10 E haplotypes for these four genes were identified across the five MGs, with each E haplotype comprising cultivars from different MGs (Fig. 5). Four haplotypes that contain the e1-nl allele were associated with the early MGs 000 and 00, whereas e1-as was found in cultivars from all MGs, dependent on the allelic state at the other E loci. For the latest MG included in this study, MG II, seven cultivars each carried e1-as and the photoperiod-insensitive allele at E2 (e1-as e2 E3-Ha E4) or at E3 (e1-as E2 e3-tr E4). Surprisingly, we found two cultivars being heterozygous at one of the four E genes, namely the cultivar Ruzica carrying e3-tr/e3-fs and Carla_TD being heterozygous at the E4 locus.

However, also within-haplotypes variation for flowering and maturity was observed (Fig. 4). In general, cultivars from E haplotypes containing the e1-nl allele were earliest and reached maturity at almost all 22 locations. Furthermore, we also observed variation among the cultivars with regard to the relative duration of the vegetative and reproductive growth phases. The most extreme genotypes were Lucija (e1-nl E2 E3 E4) with 40.5% vegetative growth and 59.5% reproductive growth and at the other end Ecudor, Castetis (both e1-as E2 e3-tr E4) and Buga (e1-as e2 e3-fs E4) with 47.2% vegetative growth and 52.8% reproductive growth. E haplotypes containing the e1-nl allele had an on average shorter relative proportion of vegetative growth phase compared to haplotypes carrying the e1-as allele.

Clustering of cultivars

The clustering of the cultivars based on their phenotypic distance from all 22 locations revealed at least seven clades (Fig. 5). At first glance, two major groups can be identified that can roughly be described as early material consisting of MGs 000 and 00 and late material consisting mainly of MGs 0, I and II. However, cultivars from the original MGs did not strictly cluster together but rather were often assigned to new groups of cultivars with higher phenotypic similarity. The first group from the left comprises all cultivars of MG 000 except for Perla and Diamant, which were more similar to cultivars of MG 00. MG 00 was divided into two groups with one half of the cultivars being more similar to the cultivars of MG 000, which could therefore be classified as early 00. These cultivars have in common with the MG 000 that they carry the e1-nl allele, except for Proteix and Sigalia. Furthermore, seven MG 00 cultivars and nine MG 0 cultivars clustered together, which might be classified as MG 00/0. Within this cluster, the cultivars of the two MGs 00 and 0 each grouped together. In contrast to the early MG 00 group, these cultivars carried the e1-as allele in either the e1-as e2 e3-tr E4 or e1-as e2 E3 E4 haplotype. Within the second large cluster containing the later-maturing MGs, four subsets of cultivars could be distinguished by the cluster plot but not so much by the heatmap. Two of these clades contain cultivars of MGs 0 and I, another clade comprises cultivars of MGs I and II and a small clade clusters six cultivars of MG II.

Phenotypic performance of E haplotypes dependent on MEV

Similar to the MGs, the different E haplotypes each reached earlier full bloom and maturity in a gradient from the northernmost MEV1 to the southernmost MEV6 (Fig. 6). The pattern of the E haplotypes was very similar between the MEVs, whereas the variation among and within them became smaller at the lower-latitude MEVs. MEV1 provided the best differentiation for the haplotypes with the e1-nl allele, whereas for other haplotypes, MEV4 revealed the most variation. The E haplotypes performed very similarly for flowering at MEV5 and MEV6 but reached maturity much earlier at MEV6, also resulting in a shorter reproductive phase.

The rather large difference between the two photoperiod-insensitive alleles at the E1 locus can be observed in the comparison of the e1-nl E2 E3 E4 and e1-as E2 E3 E4 haplotypes, which are identical except for their allelic state at E1. The haplotype with the e1-nl allele resulted in an average of 2–3 weeks earlier flowering compared to the haplotype carrying the e1-as allele. However, the effect on maturity was much smaller, especially in the southern MEVs, which led to a very long reproductive phase for the haplotype e1-nl E2 E3 E4. Genotypes with the e1-nl allele were in general the earliest E haplotypes regarding both flowering and maturity, except for the combination with the wild-type E2 allele. Particularly, in MEV1, only genotypes with e1-nl and the e2 allele matured, whereas the cultivars with e1-nl and the wild-type E2 did not reach maturity. Furthermore, MEV1 showed a differentiation among the e1-nl genotypes insofar that the combination with the e4 allele was the earliest maturing group followed by the E3 E4 and then the e3-tr E4 allele combinations. By contrast, these three combinations were very similar regarding flowering.

Comparing the effect of the E2 alleles within genotypes carrying the e1-nl allele, the photoperiod-sensitive E2 allele clearly delayed flowering at MEV1, whereas almost no difference was observed at all other MEVs. Nevertheless, the same comparison revealed a large effect on maturity, as the wild-type E2 allele delayed maturity by around 1 week even in the southern MEVs. The effect of the E2 locus on flowering was greater in the background of the e1-as allele, either combined with the photoperiod-sensitive E3 or photoperiod-insensitive e3-tr allele, as the difference between E2 and e2 was up to 2 weeks for the time to flowering (MEV2) and only slightly less pronounced for maturity.

The difference between genotypes carrying different E3 alleles was not as clear. A single comparison could be made between e3-tr and E3 within the same e1-as e2 E4 background, where the photoperiod-insensitive e3-tr allele promoted flowering by on average 1 week and had much less effect on maturity. Another comparison could be made between the alleles e3-tr and e3-fs in an e1-as e2 E4 background. Here, the e3-tr allele also promoted earlier flowering compared to e3-fs, while for maturity there was no difference. Very long periods of the reproductive phase compared to the other haplotypes were observed for e1-as e2 E3 E4 and e1-nl E2 E3 E4 in MEV6. The comparison of the two haplotypes e1-as e2 E3 E4 and e1-as e2 E3 e4, differing only in their allelic state at E4, showed that the photoperiod-insensitive e4 allele promoted early flowering and maturity. This effect became less pronounced in the southern MEVs.

Correlation of flowering, maturity and reproductive phase among MEVs

The correlations among flowering, maturity and reproductive phase were evaluated separately within MGs and MEVs. Generally, there was a moderate positive correlation between flowering and maturity that was only significant at few MEVs (Supporting Information Table S4). This correlation increased at MEVs where MGs showed a greater than average variation. In general, flowering and reproductive phase tended to be negatively correlated, whereas maturity and reproductive phase were positively correlated.

Definition of mega-environments in Europe

In accordance with the results of Cavassim et al. (2013), we observed that the location effect showed by far the highest contribution to the variation of maturity, which illustrates the strong effect of environmental factors on soybean phenology. Moreover, the highly significant and large genotype-by-location interaction variance component observed for each MG and for all three traits clearly indicated the need to define MEVs within Europe, just as in other soybean-growing regions. For MEV identification, the trait maturity was used, because the highest priority for farmers and breeders is to know if a cultivar reaches maturity in a certain environment and when. In our approach, MEVs were defined as clusters of locations providing similar growth conditions, that is, locations that were most similar with regard to the phenotypic performance of the cultivars. This resulted in a strong reduction of the variation explained by location within each MEV as compared to the analysis across all locations (Table 1, Supporting Information Table S5). By contrast, the genotype-by-location interaction variance was still significant for most MGs in most MEVs (Supporting Information Table S5), illustrating that the employed approach is different to classical MEV delineation studies. However, it appears to be a good method for a rough classification based on a moderate number of locations (three to six within MEVs) and genotypes (15 per MG). The results provide information for farmers and breeders about how long a cultivar of a certain MG needs to mature in a specific MEV. This approach revealed six MEVs that follow a latitudinal gradient, which is in line with the adaptation zones defined in the USA (Scott and Aldrich 1983; Zhang et al. 2007).

The strong influence of photoperiod on the short-day plant soybean is recognized as the primary reason for this pattern of adaptation from south to north (Cober 1996). Scott and Aldrich (1983) defined 10 adaptation zones for the USA with narrow belts of around 200 km in latitudinal direction, which was confirmed by recent results of Zhang et al. (2007). Zhang et al. (2007) also stated that the adaptation zones for early MGs are more clearly defined as narrow belts compared to later MGs, as the earlier MGs 0 to III are each best adapted to regions within 2° of latitude, the equivalent to a range of 220 km from north to south. Applying the classification of adaptation zones of the USA to Europe, there should be at least five zones, assuming a zone or belt every 200 km or 2° of latitude. Even though six MEVs were identified in our study, these belts are not equal to the zones of the USA, and because of the higher latitude of the northern part of Europe as compared to the USA, a higher density of different adaptation zones might exist, each with a more narrow range of latitude.

It must be noted that this study can only provide a first assessment of potential MEVs within Europe, as the number of test locations was high with 22, but there were still larger gaps in the direction of latitude not covered by any location. For example, MEV1 defined by four test locations covers a latitudinal range of about 330 km and should be divided into at least two distinct zones. An additional border might be between the two Dutch locations NED2 and NED1, because at NED1 and BEL1 several cultivars of MG 00 reached maturity, whereas at NED2 and GER3 only very early cultivars of MG 000 reached maturity. However, because of the too low number of locations in the northern part of Europe, this conclusion requires further research with additional locations. Likewise, it is most likely that there is one or more additional MEVs between MEV1 and MEV2, as the distance between the southernmost location of MEV1 (BEL1) and the northernmost location of MEV2 (GER5) is rather large (about 200 km). Even though this study was the largest study of its kind ever realized in Europe, our results clearly illustrate the need for further research on adaptation zones in Europe, with a finer grid of test locations. As it is difficult to get even more participants for such a study, this might be based on national or regional trials, which are performed anyhow in most countries and which could be performed with overlapping checks to facilitate a joint analysis.

At first glance, MEV3 and MEV4 do not fit into the classification of MEVs according to latitude, as even cultivars of later MGs reached maturity at MEV3 but not at the lower-latitude MEV4. According to the cluster tree, the more northern MEV3 actually showed a higher similarity to the two southernmost MEVs, MEV5 and MEV6, than MEV4. Several reasons are conceivable, such as differences in planting date, altitude, temperature, moisture, irradiation or year effects. Notably, two of the three locations from MEV4 are located in Croatia, where the weather was very special in 2014 insofar that it was one of the rainiest years with lower temperatures compared to the average of the last decades, so that MGs I and II did not reach maturity, while they normally do. Thus, regarding the phenotypic performance, the two MEVs are rather similar (Fig. 3), except that MGs I and II matured in MEV3, but not in MEV4, a pattern that likely explains the clustering of these two MEVs with the southern or northern MEVs, respectively. Furthermore, MG 000 matured slightly earlier in MEV3 than in MEV4, while conversely, MGs I and II appeared to be more delayed in flowering in MEV3 than in MEV4. This indicates that the similarity of MEV3 and MEV4 with either the southern or northern MEVs is likely due to particular year effects and can be expected to be reversed in other years, such that MEV3 might then cluster with the northern and MEV4 with the southern MEVs.

In general, the alpine climate is strongly influenced by mountains and therefore rather complex and not uniform, likely explaining the difficulty in grouping the Swiss locations in MEVs. For instance, SUI1 is located near the Léman Lake near Geneva and clustered with the more northern locations of MEV3. SUI2, by contrast, is geographically close, situated near the lake of Neuchâtel, but was more similar to the southern locations of MEV4. Notably, SUI1 is known for growing vine and orchards, whereas at SUI2 field crops are grown. Latitude is the likely reason why the MGs took a longer period from sowing to maturity at the Swiss locations than at locations from lower latitudes. Interestingly, however, almost all MGs reached maturity at these two locations, which is likely due to factors other than latitude but made them more similar to the more southern locations. This example clearly illustrates that it is not only latitude and thus photoperiodic response that determines soybean adaptation and maturity. Consequently, a more fine-scaled classification of soybean-growing regions will have to take into account also other factors such as for example temperature, as implemented already in the Canadian system using crop heat units (Brown & Bootsma, 1993).

Maturity group classification in Europe

Our results showed that the cultivars from all MGs tended to reach full maturity earlier when moving towards lower latitudes where also the differences between the MGs became smaller. Dependent on the photoperiod sensitivity of the cultivars, they matured later or even failed in reaching full maturity at the northern locations with their longer day lengths. Similar results were reported by Lima et al. (2000), where cultivars when grown at a lower latitude flowered and matured earlier. Nevertheless, there are some locations where the genotypes responded differently than expected from the locations' latitude. As mentioned earlier, other factors also play a major role in soybean development, for instance temperature or some more complex effects due to geographical and regional climate such as irradiation. Surprisingly, genotypes from later MGs (I and II) reached full maturity at the locations GER4 and GER5 from MEV2 located in southern Germany and at the Austrian locations AUT2 and AUT4 from MEV3. The recommendations for these regions in Germany and Austria are cultivars of MGs 000 and 00, according to Hahn and Miedaner (2013) based on results from value for cultivation and use (VCU) trials. It must be noted that the results of our study are based on field trials of only a single year and must therefore be treated with some caution due to potential effects of the particular year on phenotypic performance. Furthermore, we only considered the physiological maturity, which however might differ from the actual possibility to harvest the mature soybean. Moreover, from a practical point of view, reaching maturity is not necessarily equivalent with good adaptation, which might be best assessed by yield, as this integrates a whole range of additional parameters. Future field trials should consequently concentrate on several years to clarify whether the findings are reproducible over years and whether genotypes of later MGs could be considered for growing in a certain region. If genotypes of later MGs could indeed be used by breeders to expand or better exploit the growing period, this could be a means to increase soybean yields in these regions.

Clustering of cultivars reveals groups across maturity groups

In general, the clustering of the cultivars based on their phenotypic performance was in accordance with their provided MG classification. However, more than five groups were identified, which means that cultivars from some MGs are grouped together differently (000, 00, 00/0, 0/I, I/II, II and I/0). MG 000 is characterized by early maturity at all locations and the ability to reach maturity at the northernmost locations included in this study. However, two cultivars of this MG, Diamant and Perla, responded differently than the other cultivars and also clustered with the MG 00 cultivars, suggesting that they might better be classified as belonging to the latter group. These two genotypes are very early in Romania where they originate, but planted at northern locations they matured too late, probably caused by a higher temperature requirement. Notably, these two cultivars also differ from the other MG 000 cultivars as they carry the e1-as and not the e1-nl allele. MG 00 always showed the largest phenotypic variation and can be clearly separated into two distinct groups with an early and a later set of cultivars. These early MG 00 cultivars even matured at the high-latitude locations BEL1 and NED1. Except for two of these cultivars, Sigalia and Proteix, they are also similar to those of MG 000 in that they carry the same E haplotypes of e1-nl.

Within the second big cluster comprising the later-maturing MGs, four subsets can be identified according to the cluster plot even though they are not as distinctly separated by their phenotypic distance as the earlier material. Cultivars of MG 0 and MG I clustered together as did cultivars of MGs I and MG II. The method used for clustering the cultivars is probably more accurate for early material, as less information is provided for later material because this only matured at 10 locations, and within these locations, the differentiation was not as pronounced as for the early material. Taken together, the classification of the soybean cultivars, which was performed in each country of origin without reference to each other, appears to be generally correct. Nevertheless, our analysis also indicated that in a harmonized classification system some cultivars may better be grouped into an adjacent MG.

Allelic variation at E maturity genes and their effects in European soybean

Photoperiod sensitivity can be measured as the difference in days to flowering and maturity under long-day and short-day conditions (Xu et al. 2013). The greater the difference between long-day and short-day conditions, the greater the photoperiod sensitivity, independent from the actual time required to reach either flowering or maturity. Thus, a genotype can be photoperiod insensitive, although it flowers relatively late at high latitudes, if the difference between long day and short day is small. Conversely, a cultivar that is very early at southern locations is not necessarily photoperiod insensitive, as may be the case for the two cultivars Perla and Diamant, for which the flowering may be shortened due to the impact of high heat accumulation in the southern locations. This illustrates the necessity to test a cultivar for its maturity in the target environment, as it is for example not possible to classify cultivars into early MGs by their phenotype at the lower-latitude MEV5 or MEV6. Furthermore, knowing the allelic state at the maturity E loci might help to assign a cultivar to a MG. In our study, the number of locations where a cultivar did not mature can be taken as a proxy for the extent of photoperiod insensitivity, as cultivars that matured at all locations are likely more photoperiod insensitive. Three different E haplotypes might therefore be classified as being sufficiently photoperiod insensitive to be grown at the northernmost locations tested in this study: e1-nl e2 e3-tr E4, e1-nl e2 E3 E4 and e1-nl e2 E3 e4. Notably, all three haplotypes have in common that they carry the photoperiod-insensitive allele at E1 and E2 and two of them additionally carry the insensitive allele at either E3 or E4.

None of the cultivars carried the photoperiod-sensitive wild-type E1 allele, which was rather surprising because the cultivars used in this study cover five MGs and even the late MG II cultivars carried the photoperiod-insensitive e1-as allele. Therefore, the E1 locus, either through the dysfunctional e1-nl or the partially functional e1-as allele, plays a major role for early flowering as well as early maturity, and photoperiod insensitivity at this locus is probably a requirement for soybean adapted to Central or Northern Europe. This conclusion is corroborated by a recent study by Langewisch et al. (2014), who did not observe the e1-nl allele in later-maturing US American lines of MGs II, III and IV but found the partially functional e1-as allele to predominate, suggesting that this allele was essential for the development of productive lines, adapted to the North American growing areas. Xu et al. (2013) reported three major allelic combinations conditioning photoperiod insensitivity in soybean accessions originating mainly from northern Japan and China: the combination e3 e4, photoperiod-insensitive e1 (e1-nl or e1-fs) alleles and e1-as e3. In our collection of European cultivars, no combination with e3 e4 was present, but novel extremely early material of MG 0000 exists also in Europe, and it will be interesting to determine their allelic state at the four E loci.

An interesting cultivar is Lucija of MG 0 with a unique haplotype in our material (e1-nl E2 E3 E4), as it flowered very early, almost in a similar range as the other genotypes carrying e1-nl in combination with additional photoperiod-insensitive E alleles. Jia et al. (2014) identified the same E haplotype in the Russian cultivar Sunset that was classified as MG 0000, substantiating the major contribution of the e1-nl allele to pre-flowering photoperiod insensitivity. However, maturity was much more delayed and at MEV1 Lucija did not even reach maturity. This indicates that for an adaptation to higher latitudes e1-nl is not sufficient but requires additional photoperiod insensitivity alleles that also act on the reproductive development, that is, the time between flowering and maturity. Our data do not allow us to draw clear-cut conclusions on the effect of the E2 locus in European soybean material, but as Xu et al. (2013) concluded that the dysfunctional allele at the E2 locus is required for photoperiod insensitivity, this warrants further research.

For the E3 locus, two groups were identified, carrying either one of two photoperiod-insensitive e3 alleles (e3-tr or e3-fs) or a wild-type allele (E3-Ha or E3-Mi). Because e3-tr and e3-fs are both photoperiod-insensitive alleles, the cultivar Ruzica scored as heterozygous probably appeared phenotypically homogenous for maturity in the field, even though it seems to be a mixture of two haplotypes. The E3 and E4 loci code for phytochrome A genes and were shown to respond to pre-flowering and post-flowering photoperiod (Xu et al. 2013; Jiang et al. 2014), but our results do not allow us to evaluate the effect of the E3 locus on either pre-flowering or post-flowering in European soybean. Both groups showed a similar range for flowering and maturity, likely because of the rather diverse allelic combinations at the other E loci in both groups. Notably, the role of the photoperiod-sensitive and photoperiod-insensitive E3 alleles on flowering and maturity is not clear, as Jia et al. (2014) found that the e3-tr allele promoted flowering but delayed maturation. This might also contribute to the observed variation, as well as the observation by Jia et al. (2014) who pointed out that a possible interaction with the low average temperatures at higher-latitude regions might be the reason for the converse roles of alleles at E3.

In contrast, the photoperiod-insensitive e4 allele, which was found in combination with the partially functional e1-as allele and the E3 allele in the cultivar Proteix, clearly led to an earlier flowering and maturity. This confirms the effect of e4 on adaptation to higher latitudes because Proteix also successfully matured at NED1 and BEL1. The same haplotype but carrying the e1-nl allele instead was the earliest maturing haplotype, especially at the most northern MEV1, suggesting an effect of the E4 locus also on post-flowering photoperiod or thermal response. These results indicate a greater importance of the E4 than E3 locus for adaptation to very-high-latitude environments, which is in accordance with the results of Tsubokura et al. (2013). As for the cultivar Ruzica, segregating the two e3 alleles, it appears that the cultivar Carla_TD segregates the wild-type E4 and the photoperiod-insensitive e4 alleles. At first glance, this is surprising, as one would expect this to cause the segregation of two phenotypes in the field. A possible explanation lies in the origin of this cultivar. Carla_TD was bred in Romania so that it is possible that the two homozygous allelic classes, E4/E4 and e4/e4, did not differentiate at this lower-latitude site and the cultivar therefore appeared homogenous for flowering and maturity.

Sun et al. (2011) proposed that within photoperiod-insensitive lines temperature plays a major role for the regulation of the expression of the photoperiod-responsive FLOWERING LOCUS T (FT) homolog GmFTa. Interestingly, we found that even cultivars carrying the same E1–E4 haplotype differed in flowering and maturity. This suggests either the contribution of other E loci to this variation or may substantiate a more important role of temperature in the pre-flowering and post-flowering response of photoperiod-insensitive lines, mediated by allelic variation in temperature response genes.

Taken together, our results revealed the well-known E maturity loci as key components that facilitate the adaptation of soybean to the higher latitudes in Europe. It must be noted here that we have only evaluated the presence and the effects of mutations in these genes established in other soybean material, mainly in Asia. Other mutations may be present in European soybean, which warrants further research at the molecular level. In addition, it will be interesting to study the effects of each of the different alleles in more detail, both as single or in combination with other alleles at other E loci, which could be performed through phenotyping of segregating populations.

Duration of developmental phases and implications for yield improvement

While a cultivar must certainly reach maturity within the given growing season of a certain environment, a well-adapted cultivar is also defined by best exploiting the yield potential of that environment (Zhang et al. 2007). Previous reports indicated that yield is also affected by the relative duration of the vegetative and reproductive growth phases. In particular, the length of the reproductive phase, which is mainly characterized by the seed-filling period, was shown to be positively correlated with yield (Egli 1994; Gay et al. 1980; Smith & Nelson 1986). We found that within the same E haplotype the relative proportions of the vegetative and reproductive phases varied considerably. For example, the two cultivars Sigalia and Fortuna showed a similar length from sowing to maturity but differed clearly in the relative duration of the two developmental phases. Interestingly, Sigalia is known to be a high-yielding cultivar, but whether this is due to the longer reproductive phase of this cultivar requires further research.

For a better comparison, the correlations among the three traits were calculated for each MG within each MEV, but the results have to be treated with caution because the analysis was based on only 15 genotypes. The reason for the moderately positive correlations between flowering and maturity observed here and also by Jia et al. (2014), as compared to the high correlations reported by Alliprandini et al. (2009) and Liu et al. (2016), is mainly due to the low variation within each MG. Especially flowering is less variable among very early material because of its photoperiod insensitivity, which was also observed by Zhai et al. (2014). Although Cooper (2003) suggested to select for earlier flowering in order to lengthen the seed-filling period and to better exploit long days and irradiation, this does not appear suitable for the very early material. First, as mentioned earlier, we observed only little variation for flowering among the very early material, and second, this would probably hamper a sufficient vegetative growth, leading to a reduced yield potential.

The generally high positive correlation between maturity and reproductive phase indicated that the reproductive phase is mainly determined by the duration until maturity and is rather independent from flowering in very early soybean material. As we observed similar vegetative periods among the early material, variable reproductive periods seem to be a characteristic of the photoperiod-insensitive cultivars. The higher diversity for maturity observed also by Jia et al. (2014) and Xu et al. (2013) suggests that post-flowering response is probably more important for maturity and MG classification and consequently, post-flowering response to thermal time should be further examined in this material. A better understanding of the genetic mechanisms underlying the variation in the length of the reproductive phase might facilitate the optimization of this phase to tailor cultivars to specific target environments and to fully exploit the yield potential.

Assuming that a cultivar is best adapted to a certain environment when it exploits the whole growing season, the latest MG that reaches maturity would be the best adapted, for example MG 000 at MEV1, MG 00 at MEV2 or MG 0 at MEV3. However, these MGs are not necessarily the highest-yielding MG at a given MEV, so that for the proper designation of a well-adapted MG yield data should be considered also, as potentially the proportion of vegetative and reproductive growth. The greater delay of flowering of MGs I and II at MEV3 and the ensuing shortening of the reproductive phase may result in lower yields, which might be the reason why these MGs are not recommended and grown in these areas of Austria and Germany.