Introduction

With production in over 80 countries, cotton (Gossypium spp.) is the most economically important fiber crop, and the high-yielding, broadly adapted upland cotton species (Gossypium hirsutum L.) contributes to more than 90% of this global production. In the 2013/2014 season, cotton production was estimated at 120 million bales, with more than half of it attributable to China (32 million bales), India (30 million bales), and the U.S. (13 million bales) (USDA FAS 2015). In the United States, which is the largest exporter of cotton, total revenues from the cotton industry accounted for more than $27 billion USD in 2014 (NCC 2014). However, nearly 65% of cotton production in the U.S. is under rainfed (dryland) conditions (Cotton Inc. 2012). Competition for freshwater, increasing temperatures and sustained periods of minimal precipitation are projected to become more frequent and severe over the next decades (e.g., Chunlei and Richard 2013; Hansen et al. 2012; Lobell et al. 2011; Rahmstorf and Coumou 2011), threatening the sustainability of U.S. dryland cotton production (Reviewed in Dabbert and Gore 2014). In light of this, developing cotton germplasm with tolerance to combined heat and drought stress should be of central focus for climate-oriented cotton breeding programs.

Cotton breeding programs primarily select for improvement in both lint yield, of primary importance to producer, and fiber quality traits, which are critical for the textile industry. The expected response to selection for any trait can be estimated by the breeder’s equation (Lush 1937), with the response dependent on the selection differential (S) and trait heritability (h 2). When cotton is grown under standard agronomic practices, heritability estimates tend to be smaller for yield traits compared to fiber quality traits (Percy et al. 2006; Ulloa 2006), indicating a lower level of genotype-driven phenotypic variability for yield within breeding populations. This has important implications for breeding programs, because such traits are less responsive to selection and more likely to be affected by environmental conditions such as high temperature and water deficit.

Given that there is a need for simultaneous improvement in both lint yield and fiber quality, the genetic relationship between these traits merits examination. Correlations between traits can result in indirect selection, which can have unintended consequences on genetic gain and improvement of other economically important traits. Genetic correlations between traits primarily result from one of two phenomena: linkage disequilibrium (LD) or pleiotropy (Falconer and Mackay 1996; Lynch and Walsh 1998). In contrast to genetic correlations caused by LD, pleiotropy exists regardless of the population history and results in more permanent correlations that are unaffected by additional cycles of recombination (Bernardo 2010). Examples of previously identified relationships include favorable, positive correlations of micronaire with strength and lint percentage as well as unfavorable, negative correlations between lint percentage and length (Percy et al. 2006; Ulloa 2006). Such unfavorable correlations can present a difficult challenge when developing cotton cultivars with both improved yield and fiber quality.

Limited information exists on the extent of heritable variation or the magnitude and direction of genetic relationships between yield and fiber quality traits in upland cotton populations grown under the combination of high temperature and water deficit conditions. Therefore, studies focused on providing such information could help to facilitate cotton improvement by assisting the development of optimal selection and testing schemes to identify lines with high genetic potential under both stresses. In the present study, two upland cotton recombinant inbred line (RIL) populations were evaluated in multiple environments to estimate both (i) heritability and (ii) phenotypic and genotypic correlations between seven agronomic and fiber quality traits under managed well-watered (WW) and water-limited (WL) irrigation regimes in the presence of high temperature.

Materials and methods

Population development

In 2007, we crossed MSHT1B2R2, a heat tolerant inbred line, to DP 104 B2RF, a heat sensitive commercial inbred line to generate the MD population. In the same year, AZHT1B2R2, a heat tolerant inbred line, was crossed to TXHS1, a heat sensitive inbred line, to initiate construction of the AT population. Several F1 plants from each cross were self-pollinated in the greenhouse to generate F2 seed. Initially, 220 F2 individuals from each population were advanced to the F4 generation by single-seed descent. All F4 individuals were then tissue-sampled at first true leaf and assessed for the presence of both the RoundUp Ready Flex® trait (Monsanto Company, St. Louis, MO) and the Bollguard 2® trait (Monsanto Company). Bollguard 2® combines the proteins Cry1A and Cry1B from the bacterium Bacillus thuringiensis subsp. kurstaki, which help to protect against feeding damage of lepidopteron insect pests. In this study, the insect control provided by Bollguard 2® helped to ensure that most boll abscission was likely the result of abiotic stress. Seed of the selected F4-derived lines were advanced an additional two generations by self-pollinating and harvesting in bulk to produce F4:6 lines. Out of the initial 220 F2 plants, 113 and 117 F4 individuals from the MD and AT populations, respectively, were homozygous for both the RoundUp Ready Flex® and Bollguard 2® traits and produced sufficient seed quantities for field evaluation in multiple environments.

Field trial management

Multi-environment, on-farm field trials were conducted at Casa Grande, AZ (32°53′N, 111°49′W; elevation of 426 m) in 2011–2013, Hale Center, TX (34°5′N, 101°48′W; elevation of 1042 m) in 2011–2012, and Unadilla, GA (32°13′N, 83°46′W; elevation of 130 m) in 2011-12. The soils in the farm fields of AZ, TX, and GA are classified as Denure sandy loam, Pullman clay loam, and Dothan loamy sand, respectively (http://websoilsurvey.sc.egov.usda.gov). Precipitation (mm), minimum and maximum air temperatures (°C), and maximum vapor pressure deficit (VPD, kPa) data were collected at each location from before first flower to harvest with a Davis Vantage Pro 2 weather station (Davis Instruments, Hayward, CA, USA) located a maximum of 50 m from each field site (Fig. 1). A threshold ambient air temperature of 32 °C was used to declare the presence of heat stress based on a robust data set of U.S. county cotton crop yields from 1950 to 2005, in which cotton yields increased up to 32 °C but sharply decreased above 32 °C (Schlenker and Roberts 2009). In addition, 32 °C was shown to be a critical temperature threshold for boll retention (Reddy et al. 1999). The number of days above 32 °C during the first flower to harvest period for each location is provided in Table 1.

Fig. 1
figure 1figure 1figure 1

Plot of daily precipitation (mm), air temperature (maximum and minimum, °C), and vapor pressure deficit (maximum, kPa) across the growing seasons in which the MSHT1B2R2 × DP 104 B2RF (MD) and AZHT1B2R2 × TXHS1 (AT) populations were evaluated in a–c AZ (2011-13), d–e GA (2011–2012), and f–g TX (2011–2012) under well-watered (WW) and water-limited (WL) irrigation regimes. The x-axis denotes the day of year (Julian calendar) from before planting to harvest. The horizontal black, dashed line at 32 °C represents the air temperature threshold above which lint yields are sharply decreased per Schlenker and Roberts (2009). The blue, green, and red vertical lines demarcate the day of year for planting, first flower, and harvest, respectively. (Color figure online)

Full size image
Table 1 Number of days with a maximum air temperature warmer than 32 °C, average daily air temperature (maximum and minimum, °C), average daily maximum vapor pressure deficit (VPD, kPa), and average daily precipitation (mm) from first flower to harvest for each environment in which the MSHT1B2R2 × DP 104 B2RF (MD) and AZHT1B2R2 × TXHS1 (AT) populations were evaluated
Full size table

We evaluated two sets of experimental lines, with each set consisting of a RIL population (MD or AT), respective parental lines, and eight heat stress check lines (four heat-tolerant and four heat-sensitive lines); remaining empty plots in the experimental design were planted with the commercial cotton cultivar DP 1044 B2RF. Table 1 provides the day of year (DOY; Julian calendar) on which the experimental trial was planted in each environment. For each set of experimental lines, a well-watered (WW) and water-limited (WL) experiment was arranged in a 7 × 21 (0, 1) alpha lattice design with two replications per irrigation regime, for a total of 588 plots. Each experimental unit consisted of a two-row plot that was 12.19 m in length, with a 3.05 m alley at the end of each plot in all environments. Inter-plant spacing on the planter was set to a standard agronomic practice of 0.08 m in all three locations. Additionally, inter-row spacing was set to the standard agronomic practices of each location with 0.97 m in AZ, and 1.02 m in both GA and TX. At the AZ location in 2011, 0.61 m of plants were removed from both ends of each plot prior to defoliation. This is because the plants at the ends of plots were taller and more productive because of their likely ability to extract additional moisture from the unplanted alleys relative to plants within the middle of plots. As a result, the 2011 AZ plots were 10.97 m in length when data were collected. This phenomenon was not observed for plots in any of the other environments.

The on-farm field experimental trials conducted at Hale Center, TX, and Unadilla, GA, used overhead sprinkle irrigation, while those conducted at Casa Grande, AZ, used subsurface drip irrigation in 2011 and furrow irrigation in 2012 and 2013. Within each environment, the scheduling of irrigations was performed in consideration of cotton grower practices for the location, within season precipitation data, soil types within a field, and visible diagnosis of the physiological status of heat stress check lines. For the WW treatment, irrigations were applied to refill the root zone water content to the upper limit of soil water holding capacity (i.e., field capacity). For each environment, when at least 50% of the plots from both the MD and AT populations were at first flower (i.e., a plot had at least one plant with a visible flower), the WL plots immediately started to receive 50% of the irrigation amounts applied to the WW plots. As a result, the first flower DOY was synchronized with initiation of the WL treatment (Table 1), having the intent to minimize the interaction between phenology and soil moisture deficit.

All plots were harvested using a mechanical two-row plot picker with an on-picker weigh system (Master Scales, Greenwood, MS, USA). For each environment, Table 1 provides the DOY on which the experimental trial was harvested. Prior to mechanical harvest, 50 boll samples were randomly harvested by hand from each two-row plot. Boll samples were ginned on a 10-saw experimental gin (Continental Eagle Corporation, Prattville, AL, USA). Each population was phenotyped for two agronomic and five fiber quality traits: lint yield (kg ha−1), lint percentage (%), micronaire (unit), length (upper half mean, mm), strength (kN m kg−1), uniformity (%), and elongation (%). All fiber quality traits were assessed using high-volume instrumentation (HVI) (HVI1000 M, Uster Technologies, Charlotte, NC, USA).

Statistical analysis

As part of a data quality control measure, a mixed linear model was fitted separately for each trait to detect and remove significant outliers using Studentized deleted residuals (Neter et al. 1996) using PROC MIXED in SAS version 9.4 (SAS Institute 2013). The model used separately for each RIL population was as follows:

$$\begin{aligned} {\text{Y}}_{ijklmn} &= \mu + {\text{env}}_{i} + {\text{irg}}_{j} + ({\text{env}} \times {\text{irg}})_{ij} + {\text{rep}}({\text{irg}} \times {\text{env}})_{ijk} \\ & \quad + {\text{column(rep}} \times {\text{irg}} \times {\text{env)}}_{ijkl} + {\text{block(rep}} \times {\text{irg}} \times {\text{env)}}_{ijkm} \\ & \quad + \,{\text{genotype}}_{n\,} \, + ({\text{genotype}} \times {\text{env}})_{in} + ({\text{genotype}} \times {\text{irg}})_{jn} \\ & \quad + ({\text{genotype}} \times {\text{env}} \times {\text{irg}})_{ijn} + \varepsilon_{ijklmn} \\ \end{aligned}$$
(1)

where Y ijklmn is an individual phenotypic observation; µ is the grand mean; env i is the effect of the ith environment; irg j is the effect of the jth irrigation regime (WL or WW); (env × irg) ij is the interaction between the ith environment and the jth irrigation regime; rep(irg × env) ijk is the effect of the kth replication within the jth irrigation regime within the ith environment; column(rep × irg × env) ijkl is the effect of the lth column within the kth replication within the jth irrigation regime within the ith environment; block(rep × irg × env) ijkm is the effect of the mth block within the kth replication within the jth irrigation regime within the ith environment; genotype n is the effect of the nth genotype; (genotype × env) in is the interaction between the nth genotype and ith environment; (genotype × irg) jn is the interaction between the nth genotype and jth irrigation regime; (genotype × irg × env) ijn is the three-way interaction of the nth genotype, jth irrigation regime, and ith environment; and ε ijklmn is the random error term. The model terms rep(irg × env) ijk , column(rep × irg × env) ijkl , and block(rep × irg × env) ijkm were fitted as random effects.

Once outliers had been removed for all traits, an iterative mixed linear model fitting procedure was conducted separately for each trait in ASReml-R version 3.0 (Gilmour et al. 2009) using Eq. 1 with model terms defined previously. Compared to the model used to detect and remove significant outliers (Eq. 1), the terms env i , irg j , and (env × irg) ij were treated as fixed effects with all other terms fitted as random effects in this implementation. Likelihood ratio tests were conducted to remove terms from the model which were not significant at α = 0.05 (Littell et al. 2006). With the final model for each trait, the predict function in ASReml-R version 3.0 (Gilmour et al. 2009) was used to calculate an adjusted mean for each line accounting for the fixed effects in the model; this was done with respect to each irrigation regime (WW or WL) across environments as well as within each environment. Sequential tests of fixed effects were conducted with degrees of freedom being calculated with the Kenward and Rogers approximation (Kenward and Roger 1997) in ASReml-R version 3.0 (Gilmour et al. 2009). For each population, Spearman’s correlation coefficients (r s ) were used to assess the strength of association (non-parametric) between ranked WW and WL trait adjusted means using PROC CORR in SAS version 9.4 (SAS Institute 2013).

For each trait, broad-sense heritability on an entry-mean basis (Ĥ 2) was estimated for the separate irrigation regimes using a mixed linear model. Equation 1 was reformulated to remove the irrigation regime term. Next, all terms were fitted as random effects in order to obtain variance component estimates. The variance component estimates from each model were used to estimate Ĥ 2 (Holland et al. 2003) as follows:

$$\hat{H}^{2} = \frac{{\hat{\sigma }_{G}^{2} }}{{\hat{\sigma }_{G}^{2} + \frac{{\hat{\sigma }_{GE}^{2} }}{{n_{env} }} + \frac{{\hat{\sigma }_{E}^{2} }}{{n_{plot} }}}} = \frac{{\hat{\sigma }_{G}^{2} }}{{\hat{\sigma }_{P}^{2} }}$$
(2)

where \(\hat{\sigma }_{G}^{2}\) is the genotypic variance, \(\hat{\sigma }_{GE}^{2}\) is the genotype by environment interaction variance, \(\hat{\sigma }_{E}^{2}\) is the residual error variance, n env is the harmonic mean of the number of environments in which each line was observed, n plot is the harmonic mean of the total number of plots in which each line was observed, and \(\hat{\sigma }_{P}^{2}\) is the phenotypic variance. Standard errors for the heritabilities were approximated using the delta method (Holland et al. 2003).

For each irrigation regime, phenotypic (\(\hat{r}_{pij}\)) and genotypic correlations (\(\hat{r}_{gij}\)) between traits and their standard errors were estimated using multivariate REML in PROC MIXED of SAS version 9.4 (SAS Institute 2013) as previously described (Holland 2006; Holland et al. 2001). To eliminate model convergence issues arising from differences in the scales of measurement between traits (e.g. lint yield and fiber length), a data standardization procedure was implemented using PROC STANDARD in SAS version 9.4 (SAS Institute 2013). For each RIL population, raw trait data with outliers removed were standardized to have a mean of zero and standard deviation of one within each environment prior to model fitting. The model used separately for each RIL population was as follows:

$$\begin{aligned} {\text{Y}}_{ijklmno} & = \, \mu \, + {\text{ env}}\left( {\text{trait}} \right)_{ijk} + {\text{ rep}}\left( {{\text{env }} \times {\text{ trait}}} \right)_{ijkl} + {\text{ column}}\left( {{\text{rep }} \times {\text{ env }} \times {\text{ trait}}} \right)_{ijklm} \hfill \\ & \quad + {\text{ block}}\left( {{\text{rep }} \times {\text{ env }} \times {\text{ trait}}} \right)_{ijkln } + {\text{ genotype}}_{o} + \, \left( {{\text{genotype }} \times {\text{ env}}} \right)_{ko} \hfill \\ & \quad + {\text{ rep }} \times {\text{ genotype}}\left( {\text{env}} \right)_{klo} + \, \varepsilon_{ijklmno} \hfill \\ \end{aligned}$$
(3)

where Y ijklmno are the paired phenotypic observations for the ith and jth traits; µ is the multivariate grand mean; env(trait)ijk is the effect of the kth environment on the combined ith and jth traits; rep(env × trait) ijkl is the effect of lth replication within the kth environment on the combined ith and jth traits; column(rep × env × trait) ijklm is the effect of the mth column within the lth replication within the kth environment on the combined ith and jth traits; block(rep × env × trait) ijkln is the effect of the nth block within the lth replication within the kth environment on the combined ith and jth traits; genotype o is the effect of the oth genotype; (genotype × env) ko is the effect of the interaction between the oth genotype and kth environment; rep × genotype(env) klo is the effect of the interaction between the lth replication and oth genotype within the kth environment; and ε ijklmno is the random error term. The model terms genotype o , (genotype × env) ko , and rep × genotype(env) klo were fitted as random effects with all other terms fitted as fixed effects. The REPEATED statement was used to estimate the covariance of the error associated with the ith and jth traits measured in the same plot.

The formula for estimating phenotypic correlations was as follows:

$$\hat{r}_{pij} = \frac{{\hat{\sigma }_{Pij} }}{{\hat{\sigma }_{Pi} \hat{\sigma }_{Pj} }}$$
(4)

where \(\hat{\sigma }_{Pij}\) is the estimated phenotypic covariance between traits i and j, \(\hat{\sigma }_{Pi}\) is the estimated phenotypic standard deviation of trait i and \(\hat{\sigma }_{Pj}\) is the estimated phenotypic standard deviation of trait j.

The formula for estimating genotypic correlations was as follows:

$$\hat{r}_{gij} = \frac{{\hat{\sigma }_{Gij} }}{{\hat{\sigma }_{Gi} \hat{\sigma }_{Gj} }}$$
(5)

where \(\hat{\sigma }_{Gij}\) is the estimated genotypic covariance between traits i and j, \(\hat{\sigma }_{Gi}\) is the estimated genotypic standard deviation of trait i and \(\hat{\sigma }_{Gj}\) is the estimated genotypic standard deviation of trait j.

When either type of correlation was zero or non-significant (P > 0.05), the two traits were considered independent (Falconer and Mackay 1996).

Results

Phenotypic variability of agronomic and fiber quality traits

We evaluated two RIL populations that segregated for tolerance to heat and drought stress under two irrigation regimes (WW and WL) at three experimental field locations over multiple years for a total of seven environments (Fig. 1). The maximum air temperature and vapor pressure deficit (VPD), which influence cotton leaf temperature and evaporative cooling (Carmo-Silva et al. 2012), were highly significantly different (P < 0.0001) among the three locations from first flower to harvest (Supplementary electronic material Table 1). The AZ location (2011–2013) had the highest average number of days above 32 °C (90.66 days) and the highest average VPD (1.94 kPa), whereas the GA and TX locations (2011–2012) averaged 31.50 days above 32 °C (Table 1). The average VPD for the GA and TX locations (2011–2012) were 0.18 and 0.91 kPa, respectively. Taken together, these three locations provided favorable weather conditions for assessing the influence of heat and drought stress on the extent of phenotypic variation for two agronomic (lint yield and lint percentage) and five fiber quality (micronaire, length, strength, uniformity, and elongation) traits in the MD and AT populations.

For both RIL populations, there was a highly significant effect of environment and the interaction between environment and water regime for all traits (P ≤0.001 to ≤0.0001), as well as of water regime for the two agronomic traits (P ≤ 0.0001) across the seven environments (Table 2). The effect of water regime was highly significant (P ≤ 0.0001) for the five fiber quality traits in the AT population, but ranged from non-significant (fiber strength) to highly significant (P ≤0.05 to ≤0.0001) in the MD population. The mean yield of lint for the MD and AT populations under the WL treatment was only ~50% of that observed under the WW treatment (Tables 3, 4). There was a minor increase in the mean of lint percentage and micronaire along with a minimal decrease in the mean of fiber length and uniformity in both populations under WL conditions. In the AT population, the mean of fiber strength and elongation had nominally decreased and increased, respectively, in response to the WL treatment, while the opposite nominal response patterns were observed for these two traits in the MD population.

Table 2 F values and significance levels of fixed sources of variation in the analysis of variance (ANOVA) for seven agronomic and fiber quality traits for the MSHT1B2R2 × DP 104 B2RF (MD) and AZHT1B2R2 × TXHS1 (AT) populations, their parents, and commercial check varieties evaluated in seven environments under well-watered (WW) and water-limited (WL) irrigation regimes
Full size table
Table 3 Mean, range, and standard deviation (SD) of adjusted means for agronomic and fiber quality traits for the MSHT1B2R2 × DP 104 B2RF (MD) population, midparent values, and estimated broad-sense heritability (Ĥ 2) on an entry-mean basis with their standard error (SE) from an evaluation in seven environments under well-watered (WW) and water-limited (WL) irrigation regimes
Full size table
Table 4 Mean, range, and standard deviation (SD) of adjusted means for agronomic and fiber quality traits for the AZHT1B2R2 × TXHS1 (AT) population, midparent values, and estimated broad-sense heritability (Ĥ 2) on an entry-mean basis with their standard error (SE) from an evaluation in seven environments under well-watered (WW) and water-limited (WL) irrigation regimes
Full size table

Heritability of agronomic and fiber quality traits

Broad-sense heritability (Ĥ 2) on an entry-mean basis relates to the extent of phenotypic variability among related lines within a population that is attributable to genetic effects (Holland et al. 2003). In general, estimates of broad-sense heritability on an entry-mean basis for lint percentage and fiber quality traits were high, while the heritability of lint yield was relatively lower (Tables 3, 4). When considering both populations, lint yield heritabilities were low to moderately high, ranging from 0.32 to 0.63 under WL and WW conditions (Tables 3, 4). Heritabilities for lint percentage and fiber quality traits were high to very high, ranging from 0.72 to 0.93 in both populations under WL conditions, with a similar range of heritabilities (Ĥ 2 = 0.80 to 0.95) under WW conditions. The highest heritability estimated in the MD population was for lint percentage (Ĥ 2 = 0.95) under WW conditions, while in the AT population the highest estimated heritability was for fiber length (Ĥ 2 = 0.94) under WW conditions.

Phenotypic and genotypic correlations between agronomic and fiber quality traits

Phenotypic correlations (\(\hat{r}_{pij}\)) between traits are influenced by both genetic and environmental factors, while genotypic correlations (\(\hat{r}_{gij}\)) are affected by genetic factors (Lynch and Walsh 1998). Overall, we found genotypic correlations (\(\hat{r}_{gij}\) = 0.64–0.70) between lint yield and percentage to be among the strongest estimated, whereas phenotypic correlations (\(\hat{r}_{pij}\) = 0.22–0.37) between these two agronomic traits were relatively weaker by nearly two-fold. Additionally, there was a wide range of non-significant to moderately strong phenotypic (\(\hat{r}_{pij}\) = −0.43 to 0.40) and genotypic (\(\hat{r}_{gij}\) = −0.59 to 0.43) correlations between lint percentage—the most heritable of the two agronomic traits—and the five fiber quality traits (Table 5; Supplementary electronic material Tables 2, 3). Furthermore, the mean of the absolute value of genotypic correlations for these six highly heritable traits was very similar between water regimes for both the MD (mean \(\left| {\hat{r}_{gij} } \right|\) = 0.34, WW; 0.33, WL) and AT (mean \(\left| {\hat{r}_{gij} } \right|\) = 0.30, WW; 0.32, WL) populations. Lastly, the across environment correlation coefficient values (Table 5; Supplementary electronic material Tables 2, 3) were also largely consistent with those generated from an analysis of the seven traits within each environment (results not shown).

Table 5 Phenotypic (\(\hat{r}_{pij}\); below the diagonal) and genotypic (\(\hat{r}_{gij}\); above the diagonal) correlations between seven agronomic and fiber quality traits for the MSHT1B2R2 × DP 104 B2RF (MD) and AZHT1B2R2 × TXHS1 (AT) populations that were evaluated in seven environments under well-watered (WW) and water-limited (WL) irrigation regimes
Full size table

Within each RIL population, the strength and direction of phenotypic and genotypic correlations between all seven traits under WW conditions were highly concordant (i.e., high pattern similarity) with their counterparts estimated for WL conditions (α = 0.05) as determined by linear regression (r 2 = 0.92–0.95; Fig. 2a). The majority of the points above the line of equality for the comparison of genotypic correlations between WW and WL irrigation regimes in the MD population indicated that the WW treatment tended to have stronger genotypic correlations. The opposite pattern was observed for the AT population, with a trend for the WL treatment to have stronger genotypic correlations. Additionally, both the MD and AT populations tended to have stronger phenotypic correlations under WL conditions. Phenotypic and genotypic correlations were only slightly less concordant between populations (r 2 = 0.71–0.82; Fig. 2b), with an inclination for the AT population to have stronger phenotypic and genotypic correlations than the MD population.

Fig. 2
figure 2figure 2

The degree of relationship (r 2) between all well-watered (WW) and water-limited (WL) pairwise trait combinations (n = 21) of phenotypic (\(\hat{r}_{pij}\)) or genotypic (\(\hat{r}_{gij}\)) correlation coefficients within (a) or between (b) the MSHT1B2R2 × DP 104 B2RF (MD) and AZHT1B2R2 × TXHS1 (AT) populations. The red diagonal line represents equality in the value of correlations. The points that are not on the diagonal line represent correlations that are not identical. (Color figure online)

Full size image

With potential implications for genetic improvement in cotton breeding programs, lint yield had a moderately strong, positive correlation with lint percentage at the genotypic level (\(\hat{r}_{gij}\) = 0.64–0.70) in both populations under WW and WL conditions. In addition, lint percentage had a moderately strong genotypic correlation with micronaire (\(\hat{r}_{gij}\) = 0.39–0.43) and length (\(\hat{r}_{gij}\) = −0.59 to −0.53) across populations and irrigation regimes. Interestingly, on average, the negative genotypic correlation of lint percentage with strength and uniformity in the MD population (\(\hat{r}_{gij}\) = −0.48 to −0.37) was 2.57-fold stronger relative to the AT population (\(\hat{r}_{gij}\) = −0.35 to −0.09) across both irrigation regimes. Overall, we found fiber length to have a moderately strong, positive genotypic correlation with strength (\(\hat{r}_{gij}\) = 0.36–0.58) and uniformity (\(\hat{r}_{gij}\) = 0.18–0.46), but a weak to moderately strong negative genotypic correlation with elongation (\(\hat{r}_{gij}\) = −0.21 to −0.05) and micronaire (\(\hat{r}_{gij}\) = −0.58 to −0.42). The strongest estimated genotypic correlation between fiber quality traits was for uniformity with strength in the MD population, where it was 0.71 under both WW and WL conditions.

Stability of agronomic and fiber quality traits

The AT and MD populations had nearly equivalent lint yield under WW conditions (AT: 1652.77 kg ha−1; MD 1646.39 kg ha−1). However, the AT population yielded 8.54% higher lint (846.36 kg ha−1) than the MD population (779.75 kg ha−1) under WL conditions. As shown in Table 6, Spearman correlation coefficients (α = 0.05) were strongest between WW and WL adjusted means of lint yield in the MD population (r s  = 0.95) relative to the AT population (r s  = 0.79). This pattern is further evident when considering the performance of the top 10 highest yielding RILs for lint yield under WW relative to WL conditions—the rank order of the RILs were less variable between irrigation regimes for the MD population (Table 7). Taken together, these results suggest that the MD population had slightly higher lint yield stability, but lower lint yield than the AT population under both irrigation regimes over multiple environments. In contrast and indicative of higher stability, there were minimal differences among the strength of correlations (r s  = 0.95–1.00) for each of the more heritable traits (lint percentage and five fiber quality traits) between irrigation regimes across both RIL populations (Table 6).

Table 6 Spearman’s correlation coefficients (r s ) between ranked adjusted means of agronomic and fiber quality traits for the MSHT1B2R2 × DP 104 B2RF (MD) and AZHT1B2R2 × TXHS1 (AT) populations evaluated in seven environments under well-watered (WW) and water-limited (WL) irrigation regimes
Full size table
Table 7 The top 10 recombinant inbred lines (RILs) of the MSHT1B2R2 × DP 104 B2RF (MD) and AZHT1B2R2 × TXHS1 (AT) populations evaluated in seven environments based on the rank order of adjusted means for lint yield under a well-watered (WW) compared to a water-limited (WL) irrigation regime
Full size table

Discussion

The exploitation of standing genetic diversity in the cotton germplasm pool through modern breeding approaches represents a sustainable strategy for the continued improvement of agronomic and fiber quality traits for high temperature, water-limited production environments. Within cotton breeding populations, the range of heritable variation for agronomic and fiber quality traits as well as the direction and strength of correlations between them influences the rate of genetic gain. Although such findings from populations grown under abiotic stress conditions would be immensely valuable to help establish and shape climate-oriented cotton breeding programs, there is an overwhelming lack of information to support breeding efforts due to a paucity of studies investigating the effect of combined heat and drought stress on field-grown cotton populations (Reviewed in Dabbert and Gore 2014). To address this issue, we assessed the heritability of and the phenotypic and genotypic correlation between two agronomic and five fiber quality traits for two cotton RIL populations in the presence of high temperature under managed WW and WL irrigation regimes across seven environments. This study provides the most extensive multi-location (AZ, GA, and TX), multi-year (2011–2013) evaluation to date of cotton RIL populations under heat and drought stress.

We first assessed the MD and AT populations in seven environments having optimal weather conditions for evaluating the consequences of combined heat and drought stress on levels of phenotypic variation for agronomic and fiber quality traits. As expected, we observed an approximate 50% reduction in lint yield for WL relative to WW plots in both RIL populations, which considerably negated the beneficial average 2.7% increase in lint percentage across the two RIL populations under WL conditions. Although we did not directly measure physiological traits on cotton plants in this study, it is highly probable that the applied drought stress resulted in a stomatal inhibition of photosynthesis from the closing of stomata (Carmo-Silva et al. 2012), thereby negatively affecting yield gains through a concomitant reduction in carbon assimilation and photosynthetic capacity (Reviewed in Chaves et al. 2003). Furthermore, this drought-induced stomatal closure plausibly led to a substantial increase in leaf temperature under high daytime temperatures that constrained the capacity for evaporative leaf cooling via transpiration, and thus caused a metabolic limitation to photosynthesis (Carmo-Silva et al. 2012). It is likely such an increase in canopy temperature under WL conditions exacerbated heat stress, contributing to higher yield losses (i.e., lower boll load) from increases in flower sterility and boll abortion (Brown and Zeiher 1997; Burke et al. 2004; Hodges et al. 1993; Oosterhuis and Snider 2011; Reddy et al. 1992; Snider et al. 2009, 2011).

In contrast to the 50% reduction in lint yield, there was neither a consistent decrease nor a change of this magnitude for any of the fiber quality traits (micronaire, length, strength, uniformity, and elongation) when comparing WL to WW plots, although the effect of water regime was significant for most of these five traits in both RIL populations. As expected in light of previous cotton drought studies (Pettigrew 2004; Wen et al. 2013), fiber length and uniformity declined under the WL treatment in both populations, but these undesirable effects on fiber quality varied minimally from 1.17 to 6.09%. Given that high turgor is critical for the expansion and elongation of each single-cell fiber strand (Dhindsa et al. 1975), it is plausible that an imposed water deficit disrupted the needed level of turgor and, as such, decreased overall fiber length and reduced uniformity in how the individual fibers elongated relative to each another (Aboukheir et al. 2012; Guthrie et al. 1993; Ruan et al. 2001). Micronaire increased under WL relative to WW conditions in the MD and AT populations, in agreement with the general findings from a previous drought study of eight cotton genotypes (Pettigrew 2004). However, the average fold increase was only 1.03 across the two RIL populations. The elevated micronaire values for the WL plots indicate that the fibers were more coarse, which could have resulted from increased fiber maturity (Pettigrew 2008) and/or higher cellulose deposition during fiber development due to excess carbohydrate supply (Guthrie et al. 1993). Such excess supply would be expected if the carbohydrate demands of boll load were lowered due to the combination of heat and drought stress (Guthrie et al. 1993). Contrary to the other fiber traits and still in need of explanation through further research, strength and elongation nominally increased (1.32%) and decreased (−0.13%), respectively, in the MD population under WL conditions, whereas these two traits showed inverse directional responses at slightly larger magnitudes (−3.27% and 7.94%) to the WL treatment in the AT population.

Estimates of broad-sense heritability on an entry-mean basis were high to very high for lint percentage and the five fiber quality traits in the MD and AT populations under WW (0.80–0.95) and WL (0.72–0.93) conditions. On average, broad-sense heritability estimates for these six traits were only slightly (4.23%) larger under WW relative to WL conditions across both RIL populations. Similarly, these five fiber quality traits were also highly heritable in the TM-1 × NM24016 upland cotton RIL population when evaluated under WW (Ĥ 2 = 0.79–0.87) and WL (Ĥ 2 = 0.81–0.88) irrigation regimes over three years in the hot arid environment of central Arizona (Pauli et al. 2016). Taken together, these results highlight the lack of responsiveness exhibited by lint percentage and fiber quality traits to a water deficit treatment, suggesting that variation for these traits is predominantly under the control of genetic effects. Additionally, these high estimates of heritability imply that these six traits should respond very favorably to selection under either WW or WL conditions in both RIL populations when using the experimental field design implemented in this study (Holland et al. 2003).

Broad-sense heritability estimates for lint yield were dramatically different from those of lint percentage and the five fiber quality traits. On average, lint yield heritabilities were 1.6-fold smaller for the WL (MD, Ĥ 2 = 0.32; AT, Ĥ 2 = 0.46) compared to the WW plots (MD, Ĥ 2 = 0.60; AT Ĥ 2 = 0.63) across the RIL populations. Comparatively, estimates of broad-sense heritability on an entry mean basis for lint yield in the TM-1 × NM24016 upland cotton RIL population were moderately higher under both the WW (Ĥ 2 = 0.68) and WL (Ĥ 2 = 0.71) treatments, with a negligible difference in the estimate of heritability between the two irrigation regimes (Pauli et al. 2016). However, Pauli et al. (2016) evaluated the RIL population at a single location over three years under highly managed drip irrigation regimes, potentially allowing for higher estimates of lint yield under WL conditions. In contrast to lint percentage and fiber quality traits, the lower estimates of heritabilities observed for lint yield in our study suggest that selection will be less effective in the WL conditions given the greater influence of the environment. In this case, lint yield can be expected to be more responsive to selection in high-yielding environments (i.e. optimal irrigation), with the need for an increased number of environments and/or replications if evaluation and selection under combined heat and drought stress are important to cotton breeding programs.

The highest absolute value for a phenotypic correlation observed between lint percentage and fiber quality traits never exceeded 0.43 in either RIL population irrespective of water regime. Comparatively, the maximum absolute value of phenotypic correlations between the same five fiber traits in the TM-1 × NM24016 upland cotton RIL population evaluated under WW and WL conditions was 0.57 (Pauli et al. 2016). When considering the indirect selection of lint yield (low heritability) through its strong genotypic correlation with lint percentage (high heritability), unfavorable correlated responses to different degrees in micronaire (\(+ \hat{r}_{gij}\)), length (\(- \hat{r}_{gij}\)), strength (\(- \hat{r}_{gij}\)), and uniformity (\(- \hat{r}_{gij}\)) in both RIL populations under either irrigation regime would likely result. Such an outcome would lead to price discounts at the gin due to undesirable fiber characteristics for the textile industry. Additionally, direct selection on lint yield alone would partly present a similar challenge given that it has a moderately strong, negative genotypic correlation with strength and length in the MD and AT populations, respectively. It appears that the most favorable genotypic correlations existed for length with strength, as the magnitude and direction of this association suggests that these traits are amenable to simultaneous genetic improvement.

The yield of lint for the AT population was nominally 1.09-fold larger than the MD population under WL conditions, and both populations had essentially the same lint yield under WW conditions. Interestingly, Spearman’s rank correlation coefficients between WW and WL RIL adjusted means of lint yield were strongest for the MD population, suggesting that this population had higher stability for lint yield than the AT population. Therefore, these results imply that the selection of high-yielding RILs for environments prone to high temperature and water deficit should occur when the AT population is evaluated under imposed heat and drought stress conditions. Conversely, selection of high-yielding RILs from the MD population for favorable or adverse environments would be seemingly successful under either WW or WL conditions in the presence of high temperature, but the WW treatment did result in higher heritabilities for lint yield that would be more amenable for selection. Lint percentage and the five fiber quality traits, however, were highly stable with minimal fluctuations in the rank order of RILs between WW and WL irrigation regimes regardless of RIL population. Importantly, future experiments are needed to explore the extent to which canalization, plasticity, and developmental stability—major processes involved in the regulation of phenotypic variation (Debat and David 2001)—are responsible for changes or lack thereof in the rank order of RILs for lint yield and other fiber-related traits between irrigation regimes in both the MD and AT populations.

Conclusions

When considering the projected changes to the global climate (e.g., Chunlei and Richard 2013; Hansen et al. 2012; Lobell et al. 2011; Rahmstorf and Coumou 2011), there is likely to be an increase in the prevalence of the co-occurrence of high temperature and low rainfall in cotton production environments (Reviewed in Dabbert and Gore 2014). Therefore, it has never been more urgent for cotton breeding programs to focus on the development of germplasm with improved tolerance to both heat and drought stress. To lay the foundation for these breeding efforts, we investigated the degree to which the heritability of and phenotypic and genotypic correlation between seven agronomic and fiber quality traits were affected by the coincidence of imposed heat and drought stress in two cotton RIL populations across seven environments. We showed that heritabilities and correlations among these traits within populations were influenced by combined heat and drought stress, and we were able to identify notable differences in the magnitude of trait correlations between RIL populations. Furthermore, the differential stability of lint yield between RIL populations provided insight into the variable importance that contrasting WW and WL irrigation regimes may have in the testing and selection of stable, high-yielding genotypes. In summary, our findings from the study of these two RIL populations in multi-environment evaluation trials under managed WW and WL conditions could be used by breeders to help develop robust, cost-efficient methods for progeny testing and maximization of trait heritability and selection progress in the development of high-yielding, stress-tolerant cotton cultivars.