Crop Science Journal of Natural Resources and Life Sciences Education
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

Published online 23 September 2005
Published in Crop Sci 45:2160-2171 (2005)
© 2005 Crop Science Society of America
677 S. Segoe Rd., Madison, WI 53711 USA
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via ISI Web of Science (2)
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Grüneberg, W. J.
Right arrow Articles by Hermann, M.
Right arrow Search for Related Content
PubMed
Right arrow Articles by Grüneberg, W. J.
Right arrow Articles by Hermann, M.
Agricola
Right arrow Articles by Grüneberg, W. J.
Right arrow Articles by Hermann, M.
Related Collections
Right arrow Other Tubers
Right arrow Biometrics

CROP BREEDING, GENETICS & CYTOLOGY

Genotype x Environment Interactions for a Diverse Set of Sweetpotato Clones Evaluated across Varying Ecogeographic Conditions in Peru

Wolfgang J. Grüneberga,*, Kurt Manriqueb, Dapeng Zhangb and Michael Hermannb

a Institute of Agronomy and Plant Breeding, Georg August Univ. of Göttingen, Von Siebold Str. 8, D-37075 Göttingen, Germany
b Dep. of Genetic Resources and Crop Improvement, International Potato Center, P.B. 1558, Lima 12, Peru

* Corresponding author (w.gruneberg{at}cgiar.org)


   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
Sweetpotato [Ipomoea batatas (L.) Lam.] is cultivated across a wide range of agrogeographical conditions. The objectives of this study were to analyze genotype x environment (G x E) interactions for sweetpotato yield (i.e., storage root yield, biomass, harvest index) and nutritional traits [i.e., root dry matter (RDM), starch (STA), and ß-carotene (BCR) content; and leaf carotene (BCL) and chlorophyll (CHL) content] in multienvironmental trials (MET) across ecogeographic regions. Nine clones of diverse origins were tested and compared with check clones at seven locations in Peru using two N treatments (N = 0 or 80 kg ha–1). The G x E analysis was conducted with regression, additive main effects and multiplicative interaction (AMMI), and cluster analyses. The G x E interactions were smaller than the genetic variation of nutritional traits. The G x E interactions were larger or nearly equal to the genetic variation of yield traits (except harvest index), and were mainly determined by subsets of genotypes and environments. The contribution of N input to G x E was often not significant. Genotypes were observed with wide adaptation and high yields (about 19 to 22 Mg ha–1) across all three environmental groups that were derived from the cluster analysis. However, a specifically adapted genotype was observed with considerable yield advantage over all widely-adapted genotypes in low-yielding environments (from 9 to 18 Mg ha–1). Locations differed in their selection ability for storage root yield. We concluded that it is possible to breed for high yield and wide adaptability in sweetpotato in Peru, and it can be ensured that low-yielding or marginal environments are not neglected in breeding efforts.

Abbreviations: AMMI, additive main effects and multiplicative interaction • BCR, root ß-carotene • BCL, leaf carotene • BIOM, biomass production • CHI, commercial harvest index • CHL, chlorophyll • CIP, International Potato Center • CYLD, commercial storage root yield • FM, fresh matter • G x E, genotype x environment • HI, harvest index • MET, multienvironmental trials • RDM, root dry matter • TYLD, total storage root yield


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
SWEETPOTATO is grown in tropical and subtropical regions under agrogeographic conditions that vary widely, but most of it is produced on marginal soils in low-input subsistence farming systems. With an annual planting area of approximately 9 million ha worldwide, it is the second most important tropical staple root crop, cassava (Manihot esculenta Crantz) being the most important. It is hypothesized that, because of its wide distribution and large genetic diversity, sweetpotato shows (i) large differences in genotypic expression in MET across regions, and (ii) G x E interactions are mainly explainable as subsets of genotypes and environments. However, little information is available about G x E variation in sweetpotato. A better understanding of G x E interactions in sweetpotato is required before agronomists and breeders can confidently decide (i) whether to develop widely or specifically adapted genotypes and (ii) how to allocate resources between tests for yield and nutritional traits. Such an understanding would also allow informed choices to be made regarding which locations and input system should be used in breeding efforts.

Prerequisites for the analysis of G x E interactions are estimates of variance components relative to genotypes , G x E interactions , and the error term from trials conducted across locations. If multilocation trials are conducted that include further environmental factors, such as regions, years, or N input, then the G x E term can be further partitioned and corresponding variance components can be estimated. So far, there is no general theory for a detailed stability and adaptability analyses of partitioned G x E interaction variance components (i.e., {sigma}GxL2, {sigma}GxY2 and {sigma}GxLxY2, where L denotes location, and Y is year), so that MET with different environmental factors are usually analyzed by aggregating different environmental factors into one environmental factor and {sigma}GxE2, respectively. However, consideration of different environmental factors provides useful information about the contribution of a factor to G x E interactions. There are few reports of {sigma}G2, {sigma}GxE2, and {sigma}{epsilon}2 estimations for sweetpotato storage root yield in different regions of the world. Usually, only mean squares relative to genotypes (MSG), genotype–environment interactions (MSGxE), and the error (MS{epsilon}) are given, and variance component ratios for {sigma}G2/{sigma}GxE2/{sigma}{epsilon}2 have to be calculated from these MS values [i.e., yield variance component ratios of 1:1.27:1.93 for Cameroon can be calculated from the results of Ngeve and Bouwkamp (1993), and ratios of 1:0.69:0.55 for Peru can be calculated from the results of Manrique and Hermann (2002)]. However, occasionally stability parameters are reported for sweetpotato without giving variance components and MS values [i.e., the stability variance of sweetpotato clones in USA multilocation trials (Bacusmo et al., 1988)], which should be avoided. In sweetpotato, G x E analysis has been conducted using environmental variances (Bacusmo et al., 1988) or regression analysis (Janssens, 1985; Ngeve, 1993) within a region, but analysis of G x E interactions across regions using multivariate statistics has not previously been reported. Furthermore, there are many studies related to sweetpotato nutritional traits that show a considerable variability in sweetpotato germplasm for food quality characteristics (Collins, 1990; Woolfe, 1992; Ravindran et al., 1995), however, there is no information on the extent to which this genetic variation is associated with G x E interactions.

There were three objectives of this study. The first was to determine the magnitude of G x E variation in sweetpotato for yield and nutritional traits in trials conducted across ecogeographic regions, and the extent this G x E interaction caused by natural environmental factors (i.e., locations) and by artificial environmental factors (i.e., N input). The second objective was to determine the interpretable G x E variation using regression analysis and multivariate procedures, following the analytical protocol proposed by Fox et al. (1997). The third objective was to assess breeding options for wide or specific adaptation, as well as the suitability of locations and input systems for conducting selection of sweetpotato.


   MATERIALS AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
 REFERENCES
 
Nine sweetpotato clones were used for the G x E analysis. These included cultivars traditionally used by farmers from Peru (DLP2462, ARB535, and ARB-UNAP74) and East Africa (Wagabolige and Tanzania), the commercial cultivars Jewel (USA) and Xushu18 (China), and the advanced breeding lines JPKY16.005 (Japan) and SR92.499-23 (International Potato Center, CIP) (Table 1). Field experiments included one or two locally preferred sweetpotato cultivars as check clones (CC in Table 1). The plant material was a diverse set of sweetpotato cultivars in origin, cultivar type, storage root form, skin color, flesh color, and growth type. All apical cuttings were taken from virus-free mother plants grown in greenhouses at the CIP breeding station in San Ramon, Peru.


View this table:
[in this window]
[in a new window]
  		Table 1. Description of clones used for the G x E analysis and as local checks (CC = check clone at location OXA = Oxapampa, SRA = San Ramon, AQP = Arequipa, LMO = La Molina, TIM = Tingo Maria, TAC = Tacna, HCO = Huanuco).

 
Experiments were conducted at seven locations representing the wide ecogeographic variation present in Peruvian sweetpotato-growing areas, between the latitudes of 9 and 18° S. The locations represented the tropical midelevation valleys, the wet tropics, and the arid and irrigated Coastal Pacific lowlands (Table 2). Moreover, the experimental sites differed in soil type, previous crop, and N-fertilization to previous crop. At each location, two levels of mineral N fertilization were tested by either applying no mineral fertilizer or by applying urea at a rate of 80 kg N ha–1.


View this table:
[in this window]
[in a new window]
  		Table 2. Description of locations used for the genotype x environment interaction analysis.

 
In each experiment, a randomized complete-block design with three replications was used. The experimental plots consisted of four rows, each containing 25 plants. Planting distances were 1.0 m between rows and 0.3 m between plants. Harvests were conducted 150 d after planting. Total storage root yield (TYLD), commercial storage root yield (CYLD), and biomass production (BIOM) were recorded as Mg ha–1, whereas commercial harvest index (CHI = CYLD/TYLD) and harvest index (HI = TYLD/BIOM) were recorded as percentages. Five plants from the central two rows of every experimental plot were randomly selected at harvest time to generate subsamples of roots and leaves for the determination of RDM, STA, BCR, BCL, and CHL contents. Beta-carotene and CHL contents were determined with a spectrophotometer as described by Lichtenthaler and Wellburn (1983), and extractable STA and dry matter were determined following procedures described by Bainbridge et al. (1996) and Cunniff (1995), respectively. Storage RDM and STA contents were recorded as percentages on a fresh matter (FM) basis, whereas BCR, BCL, and CHL contents were recorded as mg 100 g–1 FM.

Statistical analyses were conducted using PLABSTAT (Utz, 1997), SAS6.12 (SAS Institute, 1988; 1997), and SAS-IML (SAS Institute, 1990) and our own SAS-IML-Macro STABSAS program that estimates stability and adaptability parameters, such as environmental variance, ecovalence, slopes and deviations of regression (Wricke and Weber, 1986), and the parameters derived from the AMMI analysis (Gauch and Zobel, 1988; Zobel, 1990). Data were classified relative to genotypes (G), nitrogen treatment (N), locations (L), and blocks or replications (R). In an initial ANOVA, each trait xi (namely, TYLD, BIOM, CHI, HI, RDM, STA, BCR, BCL, and CHL) was analyzed from each experimental site separately to determine outliers, experimental means, and coefficients of variation using the SAS procedure GLM and the model statement xi = G + N + G x N + R, which corresponds to the statistical model

where Yijkl is the plot value of the ith trait of the jth genotype, for the kth N treatment and lth block, µi is the trial mean of the ith trait; gij, nik, and gnijk are, respectively, the effects of genotypes, N treatment, and genotype–N interactions; blil is the effect of blocks; and {epsilon}ijkl is the plot error. In the next step, the variance components {sigma}G2, {sigma}N2, {sigma}L2, {sigma}GxL2, {sigma}GxN2, {sigma}GxNxL2 and {sigma}{epsilon}2 were estimated (excluding data for local checks), using PLABSTAT, with the model statement

which corresponds to the statistical model

where lin, glijn, nlikn, and gnlijkn are the effects of locations, genotype–location, N–location, and genotype–N–location interactions, respectively, and other effects as given in the above statistical model.

For the analysis of stability and adaptability, each combination of location N treatment was considered to be an environment (E). Variance components {sigma}G2, {sigma}E2, {sigma}GxE2, and {sigma}{epsilon}2 were estimated using PLABSTAT, with the model statement xi = G + E + EG + R:E + REG, which corresponds to the statistical model

where eik and geijk are the effects of environments and genotype–environment interactions, respectively, and other effects as designated above. For all traits for which {sigma}GxE2 was significantly and considerably larger than {sigma}G2 or {sigma}E2, the dynamic concept of stability (Becker and Leon, 1988) was applied by subdividing the interaction term into heterogeneity due to regression and residual deviations. This subdivision was performed with respect to genotypes and environments. For all remaining traits, the static concept of stability was applied (Becker and Leon, 1988) by calculating the variance of each genotype across environments and the variance of each environment across genotypes. Slopes of regression lines, deviations from regression, and environmental variances were recalculated by STABSAS for each genotype and environment. A multiple comparison of regression slopes was conducted using the least significant difference procedure (LSD with a 0.05 significance level). Heterogeneity of deviations from regression and differences of variances were determined using the Bartlett test and the Levene test, respectively (henceforth, B test and L test at the 0.05 significance level). For all traits where regression could not explain the majority of the G x E interaction, an AMMI analysis (Gauch and Zobel, 1988; Zobel, 1990) was conducted using STABSAS. From the complete AMMI model,

the additive main effects (gij and eik) and scores of the first principal component ({gamma}ij1 and {delta}ik1) were plotted in a biplot. In a final analysis for storage root yield, each genotype was considered in a multidimensional space of environments and each environment in a multidimensional space of genotypes and clustered using the SAS procedure CLUSTER that uses the residual sum of squares as a fusion strategy (Ward, 1963; Romesburg, 1990). Cluster summaries were plotted using the SAS macro DENDRO (Nicholson, 1995).


   RESULTS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
DISCUSSION
 REFERENCES
 
Differences in the experimental mean among locations were large for storage root yield, CHI, and BIOM (Table 3). Differences among means were less pronounced for harvest index and nutritional traits. Low-yielding locations were found to exhibit a higher coefficient of variation (CV given as a percentage) than higher-yielding locations. This was true for all traits, with the exception of nutritional traits and harvest index. The storage root yield of commercial cultivars and breeding clones was far superior to that of both the farmer cultivars and the local controls (Table 4). Of the clones, Xushu18 most often gave the highest yield in medium- and high-yielding environments, whereas SR92.499-23 gave the highest yield in low-yielding environments (Fig. 1) . The yield advantage of commercial cultivars and breeding clones over farmer cultivars and local controls corresponds to superior harvest index values, often associated with reduced BIOM. However, clones JPY16.005 and SR92.499-23 are exceptions, as BIOM was still relatively high compared with farmer cultivars and local controls (Tables 4 and 5).


View this table:
[in this window]
[in a new window]
  		Table 3. Experimental means () and coefficient of variation for observed traits at seven locations.

 

View this table:
[in this window]
[in a new window]
  		Table 4. Clone means for observed traits for each N treatment (N = nitrogen application, in kg ha–1) across locations.

 


View larger version (30K):
[in this window]
[in a new window]
  		Fig. 1. Storage root yield of local control (check) clones and another nine clones of sweetpotato used for analysis of genotype x environment interactions across 14 environments: OXA = Oxapampa, SRA = San Ramon, AQP = Arequipa, LMO = La Molina, TIM = Tingo Maria, TAC = Tacna, HCO = Huanuco. Nitrogen fertilization: N0 = 0 and N80 = 80 kg ha–1.

 

View this table:
[in this window]
[in a new window]
  		Table 5. Estimates obtained using the dynamic concept of genotype x environment interaction for tuber yield, commercial harvest index, and biomass.

 
All clones exhibited higher storage root and biomass yields at nitrogen level N80 compared with N0, but the response to N application was not large (Table 4). Only in low-yielding environments was a more sizeable response to N observed in most commercial cultivars and breeding clones (Fig. 1). Considerably higher storage RDM contents (27.5 to 29.5%) and STA contents (16.8 to 17.6%) were observed in local control clones and farmer cultivars than in advanced breeding clones and commercial cultivars; however, JPKY16.005 was an exception, as its RDM and STA content were relatively high (Tables 4 and 6). Moreover, BCR contents of the storage roots of local control clones and farmer cultivars were greater than those of the other clones, except for clones SR92.499-23 and Jewel. Nitrogen fertilization resulted in decreased harvest indices and increased STA, CHL content, and BCR contents of storage roots, but this effect was often not significant. The responses of storage RDM and BCL contents to N fertilization were inconclusive.


View this table:
[in this window]
[in a new window]
  		Table 6. Estimates obtained using the static concept of genotype x environment interaction for harvest index and quality traits.{dagger}

 
The {sigma}G2 component was significant for all traits except CHI (Table 7). This was associated with a significant {sigma}GxL2 variance component for all traits except BIOM. The effect of N and its contribution to G x E interaction was small, and was only significant for the {sigma}GxNxL2 variance component for storage root yield, biomass, and storage RDM. The magnitude of the variance components {sigma}G2, {sigma}GxL2, {sigma}GxN2, and {sigma}GxNxL2, and the large number of environments resulted in high operational broad-sense heritabilities (h2), which were greater than 0.80 for most traits (Table 7).


View this table:
[in this window]
[in a new window]
  		Table 7. Variance components and operational broad-sense heritabilities of observed traits.{dagger}

 
In an ANOVA in which the factors of N and location were aggregated into the factor environment, the estimated ratios of {sigma}G2/{sigma}GxE2/{sigma}{epsilon}2 were 1:0.78:0.21, 1:3.41:4.58, and 1:4.31:9.76 for storage root yield, CHI, and BIOM, respectively. The ratios of variance components were 1:0.10:0.20, 1:0.75:0.58, 1:0.29:0.44, 1:0.12:0.07, 1:0.93:3.34, and 1:1.11:2.95 for harvest index, storage RDM, STA content, BCR content of storage roots, BCL content, and CHL, respectively. The {sigma}GxE2 variance component was significant for all traits (results not presented) and graphical illustrations reveal that all G x E interactions led to crossovers—that is, changes in the ranking of genotypes (results presented only for storage root yield). The crossover interactions observed for storage root yield show the considerable yield advantage of clone SR92.499-23 in low-yielding environments over local controls, and over most other clones, too (Fig. 1).

The subdivision of G x E sum of squares (Table 8) into heterogeneity of regressions and deviations from regressions for traits that had a {sigma}GxE2 considerably larger than {sigma}G2 or {sigma}{epsilon}2, showed that the variance component relative to regression was significant for storage root yield for genotypes and environments, and for CHI with respect to genotypes. The heterogeneity of regression with respect to genotypes explained about 1/4 of the total G x E interactions for storage root yield, and about 1/5 for the CHI. For storage root yield, high regression slopes (bi > 1) associated with low MS deviations were observed for the clones Jewel, JPKY16.005, and Tanzania, whereas the clone Xushu18 exhibited a high value of bi, associated with high MS deviations (Table 5). For CHI, a similar association between regression slopes and MS deviations was observed for these clones (results not presented). Differences in biomass means, slope of regressions, and MS deviations among clones were less pronounced. Environments for which steep regression slopes for storage root yield and low MS deviations were found were San Ramon N80, La Molina N0, La Molina N80, Tacna N0, Tacna N80, Huanuco N0, and Huanuco N80. For BIOM, striking bi differences were observed among environments; however, the least significant difference was relatively large (LSD = 2.25). Only the BCR content of storage roots exhibited significant differences for environmental variances of genotypes (Table 6), but those for harvest index did exhibit differences at the P = 0.10 level. However, in the case of carotene content, the environmental variance and mean were highly correlated (r = 0.839, P = 0.005, where r is the Pearson product-moment correlation, and P is the significance probability of getting a larger value of |r| under the null hypothesis that r is truly equal to zero), whereas, for harvest index, no such correlation was found (r = –0.190, P = 0.625). With regard to locations, no significant differences were observed for variances across genotypes for harvest index and nutritional traits (results not presented).


View this table:
[in this window]
[in a new window]
  		Table 8. An ANOVA for genotype (G) by environment (E) interaction (G x E) with subdivision (SUB) of G x E interactions using regression analysis for tuber yield, commercial harvest index, and biomass (Het. R. = heterogeneity due to regression, Dev. R. = deviations from regression lines).

 
For storage root yield, the first and second principal components of the AMMI analysis explained 69.5 and 13.6% of G x E interaction, respectively. The AMMI biplot (Fig. 2A) displays large differences in the additive main effects for genotypes and environments and a pattern of G x E interaction. Low-yielding environments (Oxapampa N0, Oxapampa N80, Arequipa N0, Arequipa N80, and San Ramon N0) exhibited negative values for the first principal component axis (PCA1). Medium- and high-yielding environments (San Ramon N80, Tacna N0, Tacna N80, Tingo Maria N0, Tingo Maria N80, Huanuco N0, and Huanuco N80) exhibited positive values or values close to zero for PCA1. Low-yielding genotypes (Wagabolige, ARB535, ARB-UNAP74, and DLP2462) clearly showed negative values for PCA1. The medium- to high-yielding genotypes exhibited negative (SR92499-23) and positive (Xushu18) PCA1 values, and also some that were close to zero (Jewel, JPKY16.005, and Tanzania). The pattern observed from the AMMI analysis of CHI corresponds to that observed for storage root yield; furthermore, scores of PCA1 and PCA2 were correlated (results not presented). For the case of biomass, the first and second principal components of the AMMI analysis explained 40.1 and 13.6% of G x E interaction, respectively (Fig. 2B). Differences in the additive main effect with respect to environments were large, whereas those for genotypes were much smaller. No clear pattern could be observed, but some genotypes exhibited PCA1 values close to zero (Wagabolige, Jewel, and SR92499-23), although there were large differences in the additive main effects among these three clones (from 10 to 22 Mg ha–1).



View larger version (27K):
[in this window]
[in a new window]
  		Fig. 2. The AMMI biplot of nine sweetpotato clones evaluated for (A) storage root yield and (B) biomass production in 14 environments in Peru.

 
Cluster analysis of genotypes in a multidimensional space of environments using the variable storage root yield (Fig. 2) showed two main groups formed by low-yielding clones (Wagabolige, ARB535, ARB-UNAP-74, and DLP1462) and medium- to high-yielding clones (Jewel, JPKJ1600, Tanzania, SR924992, and Xushu18). Medium- to high-yielding clones were grouped according to high bi values, low MS deviations (Table 5), and PCA1 scores close to zero (Jewel, JPKJ1600, Tanzania) (Fig. 2A) and high bi values, large MS deviations, and low or high PCA1 scores (SR924992 and Xushu18). The cluster analysis of environments in a multidimensional space of genotypes, using the variable storage root yield (Fig. 3) , showed that the N environments at each location were usually clustered at the first fusion steps, except for San Ramon. Environments were clearly grouped according to yield and G x E contribution: (i) low-yielding with a medium contribution to G x E (Oxapampa, San Ramon N0, and Arequipa); (ii) medium- to high-yielding with a high contribution to G x E (Tingo Maria); and (iii) medium- to high-yielding with a low contribution to G x E (San Ramon N80, La Molina, Tacna, and Huanuco).



View larger version (18K):
[in this window]
[in a new window]
  		Fig. 3. Cluster analysis of 14 environments in Peru in multidimensional space of storage root yield for seven clones of sweetpotato. OXA = Oxapampa, SRA = San Ramon, AQP = Arequipa, LMO = La Molina, TIM = Tingo Maria, TAC = Tacna, HCO = Huanuco. Nitrogen fertilization: N0 = 0 and N80 = 80 kg ha–1.

 

   DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
REFERENCES
 
A considerable storage root yield advantage over locally preferred cultivars may be achieved by recommending new cultivars (Fig. 1), but advanced breeding materials tend to be inferior to local control clones and farmer cultivars relative to nutritional traits (Table 4). In some areas, sweetpotato is the major source of nourishment, and quality aspects are likely to determine whether or not a new clone is accepted by farmers. Owing to the genetic variance and G x E interaction (Table 7), nutritional traits, such as dry matter, STA, and carotene content, may be improved with high selection efficiency in the early stages of a sweetpotato breeding program.

Medium to high values of nutritional traits were observed in advanced clones, such as JPKY16.005, SR92.499-23, and Jewel, though this was often true for only one nutritional trait. This might be due to the choice of the multitrait selection procedure in sweetpotato breeding. Index selection is clearly superior to selection in successive stages using tandem or independent culling selection (Wricke and Weber, 1986). Therefore, further studies aiming at informed choices of multitrait selection in a sweetpotato breeding program are needed if its goal is the production of material with high yield and important nutritional traits. We want to highlight that the present study clearly shows small G x E interactions in MET across regions for sweetpotato nutritional traits, which means that selection may be conducted centrally, in one or two environments, without loss of efficiency for all nutritional traits, even if the breeder is aiming at the development of material for different ecogeographic regions.

The genetic variance for yield-related traits was significant (except CHI), which means that a considerable improvement in yield can be made, but for CHI, only limited improvement can be expected. The G x E interactions for yield-related traits, except harvest index, were large in trials across ecogeographic regions (Table 7), and were mainly determined by subsets of genotypes and environments. The heterogeneity of regression with respect to genotypes only explained 23 and 0% of total G x E variation for storage root yield and BIOM, respectively (Table 8). Hence, the use of regression analysis to estimate yield stability of advanced sweetpotato clones in MET is questionable. In recent years, AMMI analysis has often been used for G x E studies in MET and pancontinental trials, for example those conducted by CIMMYT (Crossa et al., 1991). The present study also clearly shows that results from sweetpotato MET should be evaluated using this multivariate tool.

With regard to storage root yield, we observed three interesting categories of high-yielding genotypes (Fig. 2A, Fig. 4) : (i) high-yielding with wide adaptation, such as Jewel, JPKY16005, and Tanzania; (ii) high-yielding with specific adaptation to medium- and high-yielding environments, such as Xushu18; and (iii) high-yielding with specific adaptation to low-yielding environments, such as SR92.499-23. However, a clone may change category if BIOM is considered; for example, clone SR92.499-23 appeared to be widely adapted for BIOM in this study (Fig. 2B). The present results suggest that breeding sweetpotato for high yield and wide adaptation is possible. However, the target region low-yielding or marginal environments is important for sweetpotato cultivar development, and aiming only at the genotype category high-yielding with wide adaptation is risky in sweetpotato improvement programs because these widely adapted clones may not deliver improved sweetpotato production in the low-yielding marginal environments (Fig. 1).



View larger version (14K):
[in this window]
[in a new window]
  		Fig. 4. Cluster analysis of nine clones of sweetpotato in multidimensional space of storage root yield across 14 environments in Peru.

 
The need to conduct separate breeding programs for low- and high-yielding environments is debatable (Atlin and Frey, 1989). Selection is often performed in high-yielding environments because differences between genotypes are larger in high-yielding than in low-yielding environments. This was also clearly observed in the present study (Fig. 1; bi and MS dev. in Table 5), but differences between genotypes were greater toward both extremes—the rich and the poor environments (Fig. 1)—which is consistent with observations of Ceccarelli (1996) in the case of selecting barley (Hordeum vulgare L.) under low- and high-yielding environments. The benefit of conducting direct selection under conditions similar to the target environment has been demonstrated in the case of barley (Ceccarelli, 1994), and in a wheat (Triticum aestivum L.) selection study, it has been shown that there is an advantage to select in both low- and high-yielding environments by an index (Ud-Din et al., 1992). However, Fox et al. (1997) tend to consider both the advantage of a selection in the poor environments as well as the advantage of selection in the poor and the rich environments, as an exception in grain crop breeding. The present study clearly shows that, for sweetpotato, attention needs to be paid to testing in low-yielding marginal environments if farmers working in such environments are to be the main beneficiaries of the new cultivars. Hence, yield testing in early stages of a sweetpotato breeding program should use at least one favorable and one less-favorable environment. Otherwise, it is not possible to select and increase favorable alleles in sweetpotato breeding material for all three categories of high-yielding genotypes observed in this study. The procedure of using less-favorable environments in sweetpotato selection (i.e., simultaneously by index selection or subsequently in successive stages) merits further investigation.

The choice of test environments—input systems and locations—is of major importance to yield improvement. We show that, for sweetpotato, N input is usually not an environment differentiating factor at experimental stations of research institutes (Fig. 3). The N effect may be more pronounced under no-input on-farm conditions. That N had a very small effect here may be a consequence of the fact that, usually, considerably more fertilizer is used on the test sites of research institutes than is used on farms, and for sweetpotato the residual N from fertilization of previous crops (Table 2) was most likely sufficient to show only a very small N response in this study. An indication for this is that the effect of adding N was greater for most commercial cultivars and breeding lines in low-yielding environments than in high-yielding environments (Fig. 1). Hence, sweetpotato selection for low-input conditions may require no-input growing areas at experimental stations of research institutes. In contrast to the choice of input system, the choice of location is clearly of significant importance in sweetpotato selection. We observed three environmental groups of sweetpotato test environments in Peru (Fig. 2A, 3): (i) low-yielding with a medium contribution to G x E (Oxapampa, Arequipa, and San Ramon N0); (ii) medium- to high-yielding with a high contribution to G x E (Tingo Maria); and (iii) medium- to high-yielding with a low contribution to G x E (San Ramon N80, La Molina, Tacna, and Huanuco). This suggests that at least one environment from each environmental group should be used as a test site in later stages of sweetpotato testing in Peru if adaptability of genotypes—wide or specific—is an aim of a sweetpotato breeding program. In that case, Arequipa, San Ramon N80, and Tingo Maria should all be used as test sites. Although the three environments clearly differ in their G x E contribution, they all showed medium to high selection ability for genotypes (medium to high bi in Table 5), and among these, San Ramon N80 (high bi and low MS dev. in Table 5) seems to be an appropriate environment to select in early and late stages of a breeding program.

A limitation of this study is the fact that it was conducted in only 1 yr. This is usually not a good thing for G x E studies; however, temperature and rainfall in the Andes are mainly determined by altitude and distance to the Amazon basin—only minor changes occur from year to year—such that we expected that the major differences and similarities apparent among genotypes and environments will be consistent across time. The large contribution this genotype–location interaction has made to G x E interaction in the Andes has already been shown by Collins et al. (1987). Nevertheless, any final decision concerning test sites for testing advanced breeding clones in the later stages of a breeding program should be based on information from subsequent years. Moreover, the ability to use a location to represent two environmental groups in the early or late stages of a breeding program (e.g., San Ramon in this study) would require data generated across at least 2 yr.

Moreover, the present study shows that the environmental groups determined by analysis of G x E interactions are not reflected by easily determined environmental attributes such as altitude, rainfall, or our initial classification into wet tropics, tropical midelevation valleys, and Coastal Pacific lowlands. This has already been observed for other crops, for example, adaptation of a wheat genotype to Ecuadorian highlands and Western Australia (cited by Fox et al., 1997). Further studies are needed (i) to confirm the environmental groups observed here, (ii) to assess to what extent confirmed environmental groups determined through analysis of G x E interactions can be described by ecogeographic attributes, and (iii) to link environmental groups to other environmental groups in other regions of the world.

Finally, we found that clones with high yields and high yield stability tend to have a high harvest index and high harvest index stability (Tables 4 and 6). The number of genotypes used in this study is not large enough to conclusively show that harvest index is a major determinant of yield stability in sweetpotato. However, in the case of wheat and maize (Zea mays L.), yield stability has been shown to be related to reproductive behavior and photo-insensitivity (Bolaños and Edmeades, 1993; Evans, 1993). In the case of sweetpotato, wide yield stability is determinated by harvest index stability or insensitivity and photo-insensitivity. This merits further investigation because the possibility of selecting for high yield and wide geographic yield stability in sweetpotato by selecting for the highly heritable harvest index would considerably increase the efficiency of a breeding program aimed at wide geographic stability.

In conclusion, the magnitude of G x E interactions for sweetpotato nutritional traits was small, and selection for nutritional traits may be conducted centrally in one environment (or very few environments), even if the breeder is aiming at the development of material for different ecogeographic regions. For sweetpotato storage root yield, however, a clear pattern of G x E interactions exists. Therefore, breeding for high-yielding, widely adapted clones should be possible, but yield performance in low-yielding or marginal environments merits specific consideration. The three major environmental groups observed in Peru lend themselves to be used to test for high-yielding widely adapted sweetpotato clones and to ensure that low-yielding or marginal environments are not neglected in breeding efforts.


   ACKNOWLEDGMENTS
 
The authors gratefully acknowledge the financial support provided by the German Federal Ministry for Economic Cooperation and Development (BMZ), which enabled the research reported in this paper to be conducted (GTZ Project No. 98.7860.4-001.00, Contract 81026661).

Received for publication October 20, 2003.


   REFERENCES
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 




This article has been cited by other articles:


Home page
 Crop Sci.Home page
W. H. Pfeiffer and B. McClafferty
HarvestPlus: Breeding Crops for Better Nutrition
Crop Sci., December 18, 2007; 47(Supplement_3): S-88 - S-105.
[Abstract] [Full Text] [PDF]

Home page
 Crop Sci.Home page
A. S. Zanklan, S. Ahouangonou, H. C. Becker, E. Pawelzik, and W. J. Gruneberg
Evaluation of the Storage Root-Forming Legume Yam Bean (Pachyrhizus spp.) under West African Conditions
Crop Sci., September 1, 2007; 47(5): 1934 - 1946.
[Abstract] [Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via ISI Web of Science (2)
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Grüneberg, W. J.
Right arrow Articles by Hermann, M.
Right arrow Search for Related Content
PubMed
Right arrow Articles by Grüneberg, W. J.
Right arrow Articles by Hermann, M.
Agricola
Right arrow Articles by Grüneberg, W. J.
Right arrow Articles by Hermann, M.
Related Collections
Right arrow Other Tubers
Right arrow Biometrics

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
The SCI Journals Agronomy Journal Vadose Zone Journal
Journal of Plant Registrations Soil Science Society of America Journal
Journal of Natural Resources
and Life Sciences Education		 Journal of
Environmental Quality