Among the world’s three most important staple food crops, wheat is grown on more than 215 million hectares (ha) worldwide, producing over 735 million tons (t) of grain (USDA, 2020). In South Asia, wheat is consumed by more than 1.8 billion people and is grown on about 47 million ha with an annual production of almost 145 million t (USDA, 2020). More specifically, India is the region’s largest wheat producer, with 30 million ha of wheat area accounting for 107 million t of grain (USDA, 2020).

The International Maize and Wheat Improvement Center (CIMMYT) annually develops and distributes improved wheat germplasm to hundreds of partners worldwide, free of charge in the form of targeted yield trials and observation nurseries. Recipients grow the trials and nurseries, keep promising lines for their breeding programs or for direct release and are asked to return performance data that are collectively interpreted and shared to guide further breeding. The data reflect significant variations in growing environments across years and locations and serve as an input for CIMMYT wheat breeding, resulting in lines that are diverse and specially selected for yield potential, disease resistance, end-use quality, and climate resilience. Research has shown that conditions at the CIMMYT wheat research station near Ciudad Obregón, an irrigated desert farm in northwestern Mexico, correlate with diverse wheat-growing environments worldwide (DeLacy et al., 1993; Lillemo et al., 2005) and, by controlling irrigation, create a simulation of six selection environments (SEs) that represent major CIMMYT target regions for wheat breeding and deployment in the developing world.

As part of CIMMYT’s efforts to describe, measure, and analyze genotype × environment (GE) interaction in the multi-environment testing of wheat lines, Cooper and DeLacy (1994) identified three areas: analysis of variance, indirect selection, and pattern analysis. Pattern analyses in the form of classification and ordination methods (Cooper and Hammer, 1996) involve developing biplots for the first two or three principal components for lines and environments to portray the relationships among environments, whereby the representation of discrimination among lines from the classification can be superimposed on the low-dimensional space defined by the ordination. Statistical applications by CIMMYT and partners to assess GE in extensive international wheat trials found that CIMMYT’s main testing location can be associated with various international sites (Crossa, 1990; Crossa et al., 1991; Cooper et al., 1993a, b, c, d) within certain mega-environments (MEs) that represent the world’s major wheat-producing areas (Rajaram et al., 1994; Braun et al., 1996).

A target population of environments (TPE) is a variable group of future production environments. A TPE is composed of many environments and future years or growing seasons. Naturally, GE may result from relatively static and predictable variation—for example, in soil or other conditions across the field—and dynamic, unpredictable, and often significant temporal variability—i.e., weather over different years.

The TPE delineations are based on climate, soil, and hydrological characteristics and can also include socioeconomic features, such as the resource levels of farm households. There are also different ways to group trials and environments into a TPE. Data from environmental sensors and satellites can be used to develop stratified hierarchical cluster analyses (Cooper and Hammer, 1996) of sites and thus identify homogeneous environments wherein line performances will be highly correlated.

SEs are where a breeder selects lines. The SE is identified as predicting the performance in a given TPE, but the SE may not itself belong to the TPE. To determine if lines tested in the SE are useful to predict the performance of those in the TPE, it is important to (1) compute the genetic correlations between the lines in the SE vs. the same (or related) lines in the TPE and show relatively high correlations between performances in both; (2) screen lines in the SE, with the repeatability (broad-sense heritability), where the SE is higher than the TPE; and (3) ensure that the SE allows a large number of lines to be screened at a low cost, such that the SE provides a high selection intensity.

In this study, we aimed to define TPEs for wheat breeding in India using historical meteorological and soil data and to estimate the genetic correlation, the direct and correlated response to selection between the TPEs and the SE of Ciudad Obregón, the prediction ability of the SE in Mexico, and the grain yield (GY) gains in each TPE over time.

The characterization of targeted breeding regions by defining TPEs is fundamental in designing strategies for any plant breeding program (Atlin et al., 2000) and paramount for CIMMYT, which seeks to develop wheat germplasm with high and stable yields for smallholder farmers worldwide. In South Asia—home to more than 300 million undernourished people and whose inhabitants consume over 100 million tons of wheat each year—92% of the varieties released contain CIMMYT breeding contributions and half of the spring bread wheat varieties are direct releases of CIMMYT breeding lines (Lantican et al., 2016).

In the present research, we used available climate and soil data to define three main TPEs in India, the largest wheat producer in South Asia (USDA, 2020) and one of the main breeding targets for CIMMYT. These TPEs coincided with the three main geographical zones in India: TPE 1, representing the NWPZ; TPE 2, the NEPZ; and TPE 3, the CPZ. This characterization corresponds to data provided by the Government of India and represents more than 28 million ha, which is equivalent to over 97% of India’s total wheat (Singh et al., 2019).

Based on the GY data (Figure 4) and the recorded minimum temperature (Table 2), which is a critical parameter for wheat production (Asseng et al., 2015), TPE 1 is associated with optimum wheat-growing conditions, whereas wheat grown in TPE 2 and TPE 3 is subject to stress caused by higher temperatures or drought. Higher temperatures combined with higher rates of relative humidity, as occurs in TPE 2, can cause the development of leaf diseases other than rusts. Indeed, one of the most important leaf diseases in NEPZ (TPE 2) is spot blotch caused by the fungi Bipolaris sorokiniana, which can cause significant GY losses (Devi et al., 2018). Similarly, given the lower average temperatures in the NWPZ (TPE 1), yellow rust caused by Puccinia striiformis f. sp. tritici is currently the most relevant disease for that region, even though all three types of wheat rust occur throughout India’s wheat-growing areas (Joshi et al., 2007).

Regarding soil characteristics, TPE 1 can be classified as a loam type of soil, TPE 2 as clay loam, and TPE 3 as clay type of soil (Soil Science Division Staff, 2017). In general, the soil of the three TPEs can be considered of medium texture. The cation exchange capacity of TPE 3 indicates a higher availability of essential minerals for the plants than in TPEs 1 and 2 (Table 2). The available soil water capacity was lower in TPE 3, which may be due in part to the greater proportion of coarse fragments in its soils (Table 2), meaning that wheat crops in this environment may suffer drought stress more frequently. This can be aggravated by the suboptimal irrigation common in TPE 3 and occasionally in TPE 2 due to lack of water (Joshi et al., 2007).

CIMMYT target environment characterizations date back to its precursor organization, the joint Mexico-Rockefeller Foundation Office of Special Studies, which operated in the 1940s, and were initially restricted to Mexico but became global with the Center’s formal establishment in the 1960s. Target environment definitions at that time were based on required traits, the need for yield stability, and the diversification of production systems. In the 1970s, 15 agroecological zones were defined. In the 1980s, CIMMYT redefined those agroecological zones and identified instead 12 wheat MEs, defined as broad, not necessarily contiguous but frequently transcontinental, areas with similar biotic and abiotic stresses, cropping system requirements, and consumer preferences (Rajaram et al., 1994); six each for spring and winter wheat growth habits. DeLacy et al. (1993) showed that the major discrimination factors among MEs were latitude and the presence/absence of stresses. Hodson and White (2007) used geospatial criteria to refine ME definitions and found that, in India, ME1 (high yield potential, irrigated) and ME5 (lower yield potential, irrigated, and temperature stressed) were predominant. From our results, and given the geographical projection reported by Hodson and White (2007), TPE 1 overlaps with ME1 and TPE 2 with ME5, whereas TPE 3 relates to ME4 (lower yield potential, drought stressed). The MEs can be broad and transcontinental, making latitude a fundamental factor, but we defined the TPEs in India at the regional level. In fact, when ME1 was first defined in India, it encompassed the NWPZ (TPE 1) and part of the NEPZ (TPE 2), whereas in our analysis, these two regions appear to be different in terms of climate and some soil characteristics (Table 2). Similarly, ME5 included also the CPZ, whereas in our analysis, this appeared to be classified as TPE 3.

The average H² in each TPE in all cases was lower than in the SEs. Given that each ESWYT cycle varies from year to year, estimations of GE variance that include not only the SE but also the year effect are not possible due to the lack of connectivity between cycles. Nonetheless, we have somewhat addressed this constraint by analyzing various years of data consisting of “good” and “bad” years in the SEs and TPEs; the year effect is also implied in the pedigree-based prediction models. The average H² across sites in India considering TPE subdivision was similar to that of the H²across SEs owing to the fact that many more environments are present across the TPEs than in the SE, and the response of the genotype within the TPEs becomes G + G^∗TPE, thus passing this term to the numerator of the H² formula and so contributing to the response to selection by separating G^∗TPE from G^∗E (Atlin et al., 2000).

The extent of the genotype-by-environment [G^∗E(TPE)] interaction variance was more than five times greater than that of G across TPEs, whereas across SEs, the G^∗SE variance was lower than the G variance. Similarly, the average genotypic variance was more than three times greater across SEs than across TPEs. Furthermore, the size of G^∗E in each TPE was substantially lower when compared to the one across all sites in India [G^∗E(TPE)], which indicates that the subdivision in TPEs is meaningful by reducing the extent of the genotype-by-environment.

Additionally, the fact that the variance of G increases substantially within each TPE when compared with pooled variance across all sites in India indicates that there is a part of G that can be further exploited in each TPE. Altogether, these results suggest that greater gains can be achieved by targeting selection in the TPE than without TPE subdivision because of a higher precision given that H² disregarding the TPEs is substantially lower than the one including the TPEs in the mixed model. Although the size of the G^∗TPE variance is smaller than that of the G^∗E(TPE), the size of G^∗E(TPE) relative to G is substantial (Atlin et al., 2000).

The observed phenotypic and genetic correlations between TPEs suggest that line performance information from one TPE can be used to make inferences about performance in others. This result is useful in the potential design of testing strategies for TPEs, although our results from the variance components still suggest that response to selection can be greater if TPE-targeted decisions are made from the data derived across SEs.

Regarding the CR, the results indicated that with TPE subdivision, the efficiency can be higher in TPE 1 and TPE 2 than across India. The CR for TPE 3 was comparable to that of the analysis across sites, but not higher. Ideally, for the CR to be at its maximum, the H² in the SE needs to be higher than in the TPEs, and the genetic correlation needs to be close to unity as possible; however, these conditions are not necessarily met (Searle, 1965). Genetic correlations are highly influential in the CR because its changes are directly related to the opportunities of achieving an indirect response by selecting in the SE for the TPEs (Cooper and DeLacy, 1994). In our results, the H² in the SE is higher than in TPEs, and the genetic correlations are, in average, of medium magnitude, which altogether has permitted to make genetic progress in the TPEs. It is important to note that, when calculating the correlated response to selection (indirect selection efficiency), it is assumed that the selection intensity in the SE is the same as in the TPEs (Falconer, 1952; Searle, 1965) for which the observed values may vary, considering that there is higher selection intensity in the SE, going from an initial 9,000 lines to the 49 lines finally included in each ESWYT.

The results derived from the DR indicate that, in terms of selection intensity (i), to achieve a selection response across TPEs and in each of the TPEs similar to the one in the SEs, i would need to be proportionally higher in each TPE in relation to the SE. This is largely due to the fact that H² in individual TPEs is lower than H²across SEs, so gains in TPEs can be raised by improving TPE testing conditions, thus increasing H². One of the aspects that contribute to maintain high-medium H² in the SE and that are standard practices are: the mechanized operations (planting, crop management, and harvest), adequate management of the irrigation water including drip irrigation systems, and a crop rotation strategy to uniformize soil moisture during May–September prior to the wheat-planting season in November.

In applying pedigree-based prediction models to assess the predictive ability of SEs, a large portion of total variance was accounted for by the main environmental effects (Table 5), which in this case considers previous information of SEs and TPEs and the TPE/SE–year combinations. In most cases, environmental variance fell significantly with the inclusion of MVs, which indicates this information contains, as expected, an explainable proportion of the environmental variance and the year effect. The SEs in which the MVs did not contribute to total phenotypic variance were BLHT and B2IR, indicating that the major source of variation within these SEs is given by the spatial features of the environment rather than by year-to-year MV variation. In agreement with previous findings, the inclusion of the GE term in the prediction models and environmental covariables generally tends to increase the prediction accuracy of the models (Jarquín et al., 2014; Pérez-Rodríguez et al., 2015; Cuevas et al., 2017).

Based on the pedigree-based predictions (Figure 3), TPE 1 is mainly associated with SEs B5IR and B2IR, and FDRT. The SEs that displayed higher association with TPE 2 were B2IR, F5IR, and FDRT, while B5IR, B2IR, and BLHT displayed higher association with TPE 3. The patterns we found of correlated, and direct, response to selection are being used to select germplasm included each year in ESWYT, which is targeted to TPE 1 and requires lines of normal maturity, rather than the early maturing lines desirable for TPEs 2 and 3.

Previous findings have indicated that simulated SEs at CIMMYT-Ciudad Obregón correlate with international sites and that this has resulted in GY genetic gains in target regions (Trethowan et al., 2002; Lillemo et al., 2004; Tadesse et al., 2010; Manès et al., 2012; Crespo-Herrera et al., 2017). The possibility of evaluating large numbers of lines for GY over different SEs has created an opportunity to develop and adapt germplasm for regions beyond the breeding site. This study suggests expanding the scope of such opportunities. For instance, thanks to CIMMYT’s collaboration with the Government of India and financial support from the USAID Feed the Future Innovation Lab on Applied Wheat Genomics at Kansas State University, more than 500 wheat lines that undergo Stage 2 evaluation are now routinely tested at three sites, each in a different India TPE, by the Borlaug Institute for South Asia (BISA sites; Table 1).

Earlier testing of lines in TPEs, coupled with SE data, could raise genetic gains in ESWYT germplasm (Sharma et al., 2012; Crespo-Herrera et al., 2017), since more information from the TPEs could be available to make crossing and selection decision of parental lines.

In general, the current breeding strategy justifies the indirect selection in the SE for the TPEs, given the magnitude of H², the relative size of the G and G^∗E variances, the magnitude of DR and its implications on the selection intensity, and the fact that a centralized breeding operation that has for a long term delivered genetic gains to India is more resource efficient. Nonetheless, current advances in statistical methods and selection tools may permit a fine-tuning of the testing strategy, for instance, one possible option is to evaluate Stage 1 lines in all the SEs to make selections for Stage 2 evaluations, which also can be conducted in both the SE and TPE. This change would require a modification in trial designs and allocation of resources, as these are limited in both the SE and TPE. Currently, the ESWYT germplasm is tested in the TPEs after 2 or 3 years of testing at CIMMYT-Ciudad Obregón. Hence, earlier information from the TPE can improve the selection for the final ESWYT germplasm and consequently the potential varieties that can be grown in the TPEs.

Based on our results, greater gains in GY were achieved in TPE 1 (118 kg/ha/year) and TPE 3 (123 kg/ha/year) than in TPE 2 (46 kg/ha/year) over the period we studied. ESWYT lines are bred for optimal environments and show better adaptation in TPE 1 and TPE 3. In TPE 2, yield gains are possibly lower due to a shorter growing season with continuous and terminal heat stress, where the germplasm requires some degree of earliness to cope with heat stress (Mondal et al., 2013). These germplasm requirements limit the adaptation of ESWYT lines intended for optimal conditions where earliness is not required due the crop cycle duration. Another annually distributed CIMMYT trial, the International Heat Tolerant Wheat Yield Trial (HTWYT), should feature lines better adapted to TPE 2. The HTWYT lines are selected based on their performance under heat stress with the additional restriction that their performance is comparable to the main checks for the optimal SE.

Wheat breeding must tackle the yield uncertainties that farmers face due to environmental variations. Golling (2006) estimated an annual benefit in excess of $140 million due to more stable wheat yields—independent of gains in genetic yield potential—as a result of breeding research. Yield stability is paramount to smallholder farmers and becoming more important as the earth warms and rainfall grows more erratic. According to Cooper and DeLacy (1994), to develop an understanding of the GE interactions in TPEs, it is important to characterize such TPEs in terms of the key environmental challenges that may lead to the definition of strategies for exploiting the GE interactions. In this study, we defined three TPEs in India based on meteorological and soil characteristics that overlap with the three main wheat production zones in India. Furthermore, the TPEs are associated with a set of SEs at CIMMYT-Ciudad Obregón that have allowed GY gains in the germplasm delivered to the TPEs and across India. The results described herein contribute to an integral strategy to the efficiency of CIMMYT’s wheat breeding program and further ensure the delivery of genetic gains to target environments.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: http://orderseed.cimmyt.org/iwin-results.php.

All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.