1. Introduction
In potato cropping, farmers often abundantly apply nitrogen (N) fertilizer to ensure profits because potato plants are highly responsive to extra N [
1]. This practice reduces the nitrogen use efficiency (NUE) of the crop [
2,
3], which is already rather low because of a shallow root system and the common cultivation in sandy soils. Leaching of excess N causes eutrophication of ground and surface water and is, therefore, a serious threat to the environment in potato production areas. Governmental regulations limiting the N supply have been installed, and these make it necessary to improve NUE at lower levels of input. Moreover, the N fertilizer regulations are specified for maturity types, at least in the Netherlands, emphasizing the need to incorporate the effects of maturity type in NUE studies.
Effects of nitrogen availability on above-ground and below-ground crop development have been widely studied. High N input increases individual leaf size and leaf longevity [
4,
5] and promotes branching [
6], thus supporting a sustained leaf production, which enlarges the period of full soil cover (SC) [
7,
8]. Therefore, the crop intercepts more solar radiation and accumulates more dry matter with more N [
9], all resulting in higher yield, but lower NUE [
2,
3].
An increase in nitrate available for the plant was reported to lead not only to a reduction in the proportion of dry matter allocated to roots but also to an increase in the total root surface and root length [
10]. Differences in N uptake efficiency of two cultivars were attributed to the general differences in root morphology and to a particular N response of the cultivars [
10]. Moreover, high N availability tends to suppress or delay tuber bulking and affect dry matter partitioning between haulm and tubers [
5]. Additionally, N input also affects tuber size and quality parameters, including tuber dry matter content, tuber starch content, tuber protein content, tuber nitrate content, and processing quality [
4,
11]. With more N, the proportion of large tubers was shown to increase and the fry color was shown to become darker, while the effect on tuber dry matter content was ambiguous [
4,
12,
13].
Studies on canopy cover have shown a high correlation among the ability of genotypes to intercept photosynthetically active radiation, to change resource use efficiency, and to create tuber yield [
9,
14,
15]. Khan [
16] and Khan et al. [
17,
18] studied the canopy development (CDv) of potato using an eco-physiological model in which canopy growth is a function of thermal time, following the beta function as described by Yin et al. [
19]. This methodology allows the dissection of the complex trait of canopy growth into model parameters with biological meaning [
20,
21,
22]. The analysis of the curve parameters as new traits allowed capturing differences in N response among cultivars and maturity types, as well as among cultivars within the same maturity class, facilitating the understanding of the N effects on different stages of CDv [
3,
16,
17,
18,
23]. Furthermore, those canopy cover traits had high heritabilities [
16], and some of them showed high correlations with yield, maturity, and N content, allowing an interpretation of how the NUE of potato is affected and showing potential as selection criteria for NUE [
3,
11,
16]. In addition, these parameters were found to be related to genetic factors (quantitative trait loci; QTLs) that act during development of the canopy cover and are probably involved in the underlying physiological processes [
24]. The combination of this eco-physiological growth model and QTL analysis is a two-step approach, where the first step is to model the complex trait identifying biologically relevant parameters demonstrating genetic variation, and the second step is to use these parameters as new traits to find QTLs [
25,
26]. In potato, the two-step procedure was used to study the dynamics of senescence and the adaptation in potato under different day lengths [
27], and to identify QTLs related to canopy cover parameters [
16], as well as QTLs related to the N effects on the canopy cover parameters in a diploid mapping population [
24].
In recent years, association mapping approaches have become increasingly popular for genetic studies, offering a series of advantages that include higher mapping resolution and results that are applicable to a wider genetic background [
28]. Association mapping (AM) identifies QTLs by examining marker–trait associations resulting from linkage disequilibrium (LD) between markers and trait functional polymorphisms across a set of diverse germplasms [
28]. AM copes better with tetraploid, noninbred crops, such as potato [
29], than linkage analysis using segregating biparental tetraploid populations for which tetrasomic inheritance is complicated [
30]. AM can detect QTLs at the tetraploid level within a genetic background that is more representative of the breeding germplasm of the crop [
31]. Moreover, AM procedures can effectively compare a greater portion of the variation within a species while the traditional linkage analysis is limited to the variation in the two parents of the segregating population [
32]. However, in AM, it is important to consider the effect of population structure and/or kinship because any association may partially be caused by population admixture, leading to plausible but false marker–trait associations [
28,
32,
33]. The success of association mapping efforts depends on the possibilities of separating LD due to genetic linkage from LD resulting from other causes [
31].
Several papers reported on association mapping studies in tetraploid potato. Gebhardt et al. [
34] and Simko [
35] reported markers associated with resistance to diseases using a form of
t-test. Malosetti et al. [
31] proposed an AM approach based on mixed models with attention for the incorporation of the relationships between genotypes, whether induced by pedigree, population substructure, or otherwise. D’hoop et al. [
36] applied a simple regression-based AM approach for quality traits in potato with promising results for these traits in a large set of tetraploid cultivars. In this paper, we combine the model for canopy development and association analysis to study the genetic basis of developmental physiological and agronomic traits in relation to contrasting N levels. We performed genome-wide AM for canopy development parameters and agronomic traits in a set of 169 tetraploid potato cultivars. Our cultivar set was phenotyped and studied for canopy development under contrasting N levels; in addition to effects of environmental factors, we observed genetic variation in the model traits and in agronomic traits [
3], as required for a genome-wide association analysis. Moreover, we analyzed N-dependence of the detected QTLs to show the genetic response to such an important factor and to show the usefulness of the canopy development analysis in combination with genetics studies.
In summary, our objectives were to carry out a genetic analysis on model-based phenotypic variables associated with above-ground and below-ground development. Since these variables are known to be sensitive to nitrogen supply, the analysis was carried out using data of phenotyping experiments with two contrasting nitrogen input levels. Similarly, these variables vary among maturity types and show strong interactions between maturity type and nitrogen supply; we were, therefore, interested in the genetic background of the cultivar × nitrogen interaction and whether marker–trait associations were consistent across maturity types and nitrogen supplies.
3. Discussion
In this study, we combined canopy development modeling with an association mapping analysis to reveal the genetic basis of developmental, physiological, and agronomic traits with varying N availability. We applied established methodologies as used by D’hoop [
36,
42]. The association analysis was done after correction for relatedness, which is the accepted standard because it decreases the probability of false positives [
31,
43]. In potato, D’hoop et al. [
42] showed an increased level of LD within specific cultivar groups demonstrating the importance of correcting for relatedness.
Canopy cover of potato shows a very large genetic variation [
3,
17,
18,
44,
45,
46] and a large genotype × environment interaction [
47,
48,
49]. It is controlled by multiple interacting genes, each having only a relatively small effect [
27], like in many other crops [
50,
51,
52,
53]. Moreover, the genotype × environment interaction is strongly influenced by the maturity type of the cultivar [
17,
18]. However, the impact of the genotype × environment interaction on the various components of canopy cover over time is variable [
3,
17]. These effects come about through the impact of environmental factors on stem development (stem number, stem branching, and sympodial growth), leaf appearance, leaf expansion, and leaf senescence [
54]. In the specific case of nitrogen supply, the interaction between the cultivar and nitrogen supply affects the number of branches (either the lower lateral branches or the sympodial branches at the top of the stem), the number of leaves on the main stem and on the different types of branches, the rate and duration of leaf expansion (resulting in the final size of the individual leaves), the duration of the life span of the leaf, and the rate of leaf senescence [
55]. In close interaction with the maturity type [
3], nitrogen supply supports a rapid development of the canopy early in the growing season and a long duration of canopy cover throughout the remainder of the season, thus enabling an advanced, enhanced, and prolonged light interception, allowing high tuber yields [
3,
17,
56].
Our results showed effects of N levels on the relationship between traits based on the genetic correlation (
Figure 2), similar to the results based on phenotypic correlation for both N levels reported by Ospina et al. [
3]. We demonstrated the effect of N input on canopy development and yield traits (
Figure 3), as well as the strong contribution of maturity type, which is the major factor determining development, to the genetic variation. The genetic variation resulted in QTLs consistently detected in both years at both N levels for 20 of the 24 traits included in this study. Many of these QTLs accumulated in a single region on Chromosome 5 that is known to be linked to maturity type as shown by Kloosterman et al. [
40], who identified an allelic variation of the CDF1 (cycling DOF factor) gene at this locus which strongly influences phenology, plant maturity, and onset of tuberization, reflecting the importance of this region for quantitative developmental traits.
Effects of nitrogen availability on potato development were reported by many authors [
2,
3,
4,
5,
6,
9,
11,
12,
13,
14,
16]. As stated above, in general, more available nitrogen advances, enhances, and prolongs canopy cover as a result of improved haulm growth [
57], as well as the initiation of more leaves with a longer life span [
4]. Our canopy development model describes this elegantly and in a quantitative way with biologically meaningful parameters, thus illustrating the genotype × environment interaction in an analytical way. The three phases of canopy development (see
Section 5.6) responded to N with cultivars having a faster buildup phase of the canopy, resulting in a shorter time to reach maximum coverage (
t1) and a higher maximum cover (
Vx) for a longer period (
t2–t1) at high N input, all resulting in higher photosynthetic potential [
3,
14].
Most traits included in this study had relatively high genetic correlations between high and low nitrogen conditions, except for
AP3, te–t2 (Phase 3 of CDv), and
t1. These high correlations reflect the consistency of the genotypic behavior under varying N availability, at least for canopy development parameters associated with the period of maximum canopy cover. Phase 3 of CDv was difficult to phenotype precisely due to the senescence process itself, which starts with yellow leaves until an uncertain point when the canopy collapses. The yellowness could start early if conditions are not favorable but the crop continues to take up nitrogen. On the other hand, wind and rain can accelerate the collapse and those factors are difficult to predict. Therefore, Phase 3 parameters showed the largest random error, explaining the low heritabilities of
AP3 and
te–t2 [
3]. The relationships between some of the traits based on their genetic correlation coefficients were slightly different between N input levels. For instance, yield has an absolute correlation with
AUC higher at low N than at high N, as a result of changes in the relationship with other traits. Under high N input, there are no nutrient constraints for canopy development leading to an expansion of the duration of the potato growth phases [
3,
14]. However, it is known that, with high N input, important traits determining yield such as leaf area index (LAI) and radiation use efficiency (RUE) are positively affected [
24]. LAI continues to increase even when the soil coverage is 100% [
9,
58]; the maximum coverage is also reached faster and sustained longer at high nitrogen input [
3]. Therefore, although yield and
AUC are highly correlated under both N conditions, the contribution of LAI and RUE under high N may not be fully captured by the
AUC. This could also be reflected in the QTLs detected; QTLs common to both N levels for yield and
AUC were found on Chromosome 5 (
Table 2) and colocalized with QTLs for maturity (mt_as), while QTLs exclusively detected at high N for yield were located on Chromosomes 9 and 12. A possible explanation is that the latter two might be associated with the contribution of RUE and/or LAI to yield.
Genomic regions with possible pleiotropic effects were detected on Chromosomes 2, 5, and 6 (
Figure 4). The QTL hotspot on Chromosome 5 was the most noticeable, accumulating QTLs for 50% of the traits on this maturity-related region, as similarly shown by Hurtado-Lopez et al. [
39] with developmental traits related to senescence and flowering and with plant height. Most of the traits with QTLs in this region were highly correlated with maturity assessed in our trials (mt_as), emphasizing the importance of maturity and the genomic region on Chromosome 5 for crop development. Moreover, as a general remark, the colocalization of QTLs was mostly determined by the correlation between traits. Furthermore, there was an N dependency of some QTLs for several traits. The region on Chromosome 2 accumulated QTLs for six traits (
AP1,
t1, TbnA, TbwA, TbwMX, and SCYi) at high N input, while the region on Chromosome 6 was related to four traits (with QTLs for TbnB, TbnA, TbwB, and
tm1) at low N input. This shows the strong effect of available N on the genetic response, as well as its complexity.
Regarding the N-dependent QTLs, at high N input, more QTLs involving more traits were detected than at low N input, along with a higher percentage of marker–trait associations on Chromosome 5. Gallais and Hirel [
59] found in maize more QTLs for some traits at high N input than at low N input (vegetative development, N uptake, and yield components), while, for other traits, it was the opposite (N utilization efficiency and protein content). This is a reflection of the difference in the expression of the genetic variability between high and low N input that may be trait-dependent. The trait × nitrogen interaction was translated into a QTL × nitrogen interaction in those studies, as well as in our study.
N-independent associations were mostly located at the maturity locus on Chromosome 5 (data uncorrected for maturity). Khan [
16] used a similar phenotyping approach to study potato canopy development and reported a major QTL hotspot on Chromosome 5 in a diploid biparental population (SH × RH) affecting all parameters of the canopy cover curve in several environments. Ospina [
24], using the same diploid biparental mapping population, reported the same QTL region on Chromosome 5 at both high and low nitrogen levels. In addition, QTLs for growth and yield traits in this region were found in drought tolerance QTL mapping in the greenhouse of the C × E diploid mapping population [
60] and in multiple environments for the same population [
27,
39]. Therefore, the overall and predominant effect of maturity on canopy development and on yield appears to be stable across different environments, nitrogen conditions, and populations.
N-independent QTLs different from those of the maturity locus on Chromosome 5 were found only for SCYi and TbnB (
Figure 4). These two traits were not correlated with the maturity assessment (mt_as) (
Figure 1). For SCYi, there were N-independent QTLs on Chromosomes 1, 2, 5, and 11, while, in the diploid mapping population, SH × RH [
24], N-independent QTLs were found on Chromosomes 5 and 10 (referred to as linkage groups V and X in the multitrait QTL analysis [
24]). This might suggest that the genetic background, as well as the population type, influences the genomic regions related to a trait. For TbnB, we found N-independent QTLs, as well as a QTL at low N level, on Chromosome 3. Schönhals [
61] also found associations for tuber number on this chromosome (as well as on Chromosomes 1, 5, and 6), using markers for candidate genes that were functionally related to tuber yield and starch. A comparison of the results with previous reports is difficult because different markers were used in different populations. For the markers used in the detection of QTLs with the SH × RH diploid biparental population by Ospina [
24], there were no physical positions available (these makers were not used in this association analysis), while, for the SNPs used here in the association mapping, there are no genetic positions known on the SH × RH genetic map.
After the maturity correction, the number of N-independent marker–trait associations was drastically reduced. Since most of the traits were maturity-related, the maturity correction was expected to have a strong impact on the detection of QTLs. Only the N-independent QTLs for SCYi and TbnB remained after the correction (these were not linked to maturity, and the traits did not correlate with maturity). Similarly, D’hoop et al. [
62] showed the impact of maturity. In their phenotypic analysis, the presence or absence of maturity as a term in the model influenced the genotypic effects for two traits studied, underwater weight and maturity trait (both traits are physiologically correlated), but not for the majority of quality traits, which were not correlated with maturity. Their association analysis using maturity-corrected values in a model with a correction for relatedness showed a reduction in the number of marker–trait associations detected for these two traits (underwater weight and maturity), while, for other quality traits, there was no clear trend [
63].
The maturity correction of our data using predefined information of the cultivars (the maturity grouping factor) was effective in removing the maturity effect, although it might have affected the detection of association, most probably reducing the power since the overall variation was reduced. However, it allowed detection of new QTLs (
Table 3) that did not colocalize in the main region related to maturity on Chromosome 5. For instance, a new N-independent QTL for DM% on Chromosome 3 was detected. This QTL was expected to be N-independent since the DM% trait did not have a significant nitrogen effect (
Appendix B). The results with the diploid population SH × RH [
24] also showed a QTL for DM% on Chromosome 3. In general, the maturity dependency of some QTLs resulted from the physiological relationship of canopy development traits and agronomic traits with maturity (these relationships were discussed by [
3,
24]). Kloosterman et al. [
40] identified the causal gene within the Chromosome 5 maturity locus. Allelic variation of the CDF1 (cycling DOF factor) gene at this locus strongly influences phenology, plant maturity, and onset of tuberization. The CDF1 gene has a great effect on the plant life-cycle length by acting as a mediator between photoperiod and the tuberization signal. This major effect acts on several processes of the plant, resulting in a strong linkage between maturity and traits related to CDv and yield. Khan [
16] also mentioned the dependency of tuber yield on its components (especially tuber bulking parameters and CDv traits), as these are physiologically and surely genetically related; genotypes with higher tuber bulking rates show limited haulm growth and canopy duration, leading to an early maturity type [
16].
We showed how genetic factors determining canopy development and yield traits in potato cultivars interact with N levels. The different QTL regions detected for a trait under contrasting N conditions may imply that the phenotypes are the result of a tradeoff between these QTL regions. The detection of N-dependent QTLs emphasizes the importance of direct selection under limiting N conditions only if the QTLs contribute to the traits of interest. The contribution of genetic factors to growth and yield is affected by N input, with different interactions between the traits under low N than under high N and, therefore, different contributions of the traits to the observed phenotype. Ospina et al. [
3] mentioned that, to breed for NUE under low input, the strategy should be to select for high yield under low N and combine this with a high responsiveness to more N input. This allows for selection of better adapted genotypes in N-limiting conditions. In addition, to bypass the strong linkage with maturity that is observed for developmental traits, NUE, and yield, the selections should be done within each maturity group. Thus, the phenotyping should be made more discriminative to exploit the variation in a narrow maturity category. Additionally, the strong correlation of most of the traits mentioned with maturity can mask useful genetic variation for these traits, as exemplified in this paper. An early selection is required to increase the number of individuals in the target maturity class. This might be achieved by developing marker selection for maturity. Small differences in the existing trait will then be detectable, and selectors and breeders may be able to identify new traits and be more discriminant in their assessments for these other traits.
In general, the reports in other crops on QTL detection under contrasting N conditions have shown the great influence of environmental conditions. For example, in barley, the detection of QTLs was reported to show an extensive G × E/QTL × E interaction, with QTLs changing between years irrespective of N levels [
64]. In maize, contrasting results have been reported when looking for QTLs under high and low N input. QTLs for grain composition and NUE-related traits detected at low N corresponded to QTLs detected at high N input [
65], while other results showed that QTLs for NUE were only detected at low N input [
66]. Hirel et al. [
67] suggested that, depending on the recombinant inbred line population, the response of yield to various levels of N fertilization could be different and, thus, controlled by a different set of genes. We only reported QTLs consistently detected in both years for each N level. Therefore, our research does not directly address the G × E interaction, for which more experiments would be needed.
However, our research also demonstrated extensive G × E/QTL × E interactions, as illustrated in
Table 1; the number of marker–trait associations was significantly higher across all datasets (950 marker–trait associations detected irrespective of the year) than the number of associations detected in both years (only 166 marker–trait associations detected in both years). In the context of our experimental setup, nitrogen level was the major control factor driving the differences within this very constant physical environment. The total number of marker–trait associations (
n = 166) detected in both years could be broken down in groups that were detected at both N levels (
n = 50), only at high N (
n = 69), or only at low N (
n = 47) (
Table 1). Moreover, there was a strong effect of maturity type on the detection of these marker–trait associations. When data were corrected for maturity type, only 86 marker–trait associations were found, which could be broken down into three groups; eight were detected at both N levels, 48 were only detected at high N, and 30 were only detected at low N (
Table 1). Therefore, the nitrogen dependency of some QTLs could be interpreted as a QTL × N interaction.
Our approach focused on contrasting N input levels using a single N application, and this is a first step to understanding the genetic factors involved in the response of potato to N. It is important to mention that fertilization practices such as split application might have an additional effect on the plant response to nitrogen, especially in relation to the different maturity types. Additionally, soil mineral N supply during the growing season is difficult to control and understand [
7] since it is a dynamic factor. Goffart et al. [
68] mentioned that soil mineral N supply is influenced by several predictable and unpredictable factors, such as weather conditions, chemical and physical soil properties, type and evolution of organic matter previously incorporated in the soil, cultural practices, maturity type of the cultivar, and crop duration. This N dynamic in the soil could result in different levels of available N. This difference will affect the crop development response of the cultivar and, thus, the variation of the traits, thereby affecting the consistency in the detection of QTLs or marker–trait associations.
Lastly, the understanding of the influence of an intrinsic major genotypic factor such as maturity type is valuable to refine breeding strategies, as well as to develop cultivars suitable to low N input or otherwise limiting conditions. Furthermore, the results presented here suggest that breeding schemes should be done within maturity groups, with the main idea of improving characteristics that are highly influenced by maturity, such as DM%, N content, and NUE, within a maturity group.