Identification of bread wheat genotypes with superior grain yield and agronomic traits through evaluation under rust epiphytotic conditions in Kenya

Bread wheat (Triticum aestivum L.) cultivars adapted to specific environments and resistant to prevalent pathogens are preferred for obtaining high yield. This study aimed to identify wheat genotypes with superior grain yield (GY) and yield associated traits from 168 genotypes of International Maize and Wheat Improvement Center’s 13th Stem Rust Resistance Screening Nursery evaluated over two seasons during 2019 and 2020 under high disease pressure of both stem rust (SR) and yellow rust (YR) in a 21 × 8 α-lattice design with 3 replications in Kenya. Effects due to seasons were significant for YRAud, SRAud, 1000-kernel weight (TKW), days to heading (DH), plant height (PH) and number of spikelets spike−1 (SS), while genotypes and genotypes × season interaction effects were significant for all traits except number of kernels spike−1. Respectively, heritability values of 0.95, 0.93, 0.87, 0.86, 0.77 and 0.75 were observed for area under disease progress curve for SR (SRAud), YR (YRAud), TKW, DH, biomass (BM) and GY. Path analysis showed positive direct effects on GY via PH, SS, BM, and TKW. Biplot analysis identified 16 genotypes with superior desirable traits GY, BM and harvest index. The SR contributed the highest reduction in GY and TKW while YR contributed the most reduction in BM. These identified genotypes with superior GY combined with adequate resistance to both SR and YR are potentially valuable resources for improvement of locally adapted wheat cultivars.


Results
Temperature and rain fall. Respectively, mean minimum temperature of 9.6 °C and 10.6 °C and maximum temperature of 22.8 °C and 23.8 °C were noted during main season 2019 (MS2019) and off season 2020 (OS2020). Mean monthly rainfall of 657.5 mm was recorded during MS2019 and 681.9 mm during OS2020. The summary of the mean temperature and rainfall observed during both seasons is given in Table 1.
Analysis of variance and genotypes' performance across seasons. Effects due to seasons were significant area under disease progress curve for stem rust (SR Aud ) and yellow rust (YR Aud ) at p < 0.001, 1000-kernel weight (TKW), grain yield (GY), days to heading (DH), plant height (PH) and number of spikelets spike −1 (SS) at p < 0.05, however, they were not significant (p > 0.05) for spike length (SL), number of kernels spike −1 (KS) and biomass (BM). Effects due to both genotypes and genotype × season interaction were significant for SR Aud and YR Aud (p < 0.01). Genotype effects were significant (p < 0.01) for all yield-related traits except for KS whereas season × genotype interaction was significant (p < 0.001) for all yield-related traits ( Table 2).
During MS2019, SR infection resulted in 23.12% higher mean SR Aud than observed in OS2020. However, low YR infection was observed during MS2019 with YR Aud means that were 35.32% less than those of OS2020. Generally, wheat lines in MS2019 performed better agronomically with plants that were 3.53% earlier in heading and 2.38% shorter than performance in OS2020. The MS2019 also produced lines with higher SS, GY, TKW, Table 1. Mean temperature (minimum and maximum), total rainfall, mean monthly rainfall and total number of rainy days recorded during seasons MS2019 and OS2020 at Njoro. SE standard error.  (Table 3).
Variance components and broad sense heritability. Genetic variance (σ G ) for all traits except PH, KS, SL and SS surpassed variance for season (σ E ), genotype × season (σ G×E ) and error (σ e ) though the genetic, season, genotype × season and error variance for MKW and HI were negligible. The highest proportion of genetic vari- Table 2. Mean squares of the wheat genotypes evaluated for stem rust (SR), yellow rust (YR), grain yield (GY) and yield related traits during MS2019 and OS2020 at KALRO, Njoro. YR Aud yellow rust area under disease progress curve (AUDPC), SR Aud stem rust AUDPC, TKW thousand kernel weight, GY grain yield, SS number of spikelets spike −1 , DH days to 50% heading, PH plant height, SL spike length, KS number of kernels spike −1 , BM biomass, HI harvest index, CV coefficient of variation, R 2 coefficient of determination. The expected mean squares determined the random error as a test for the blocks and the genotype × season effects, the replicates as an error term to test the effects due to seasons, the genotype × season as an error term to test the effect due to genotypes and blocks as an error term to test the effect due to replicates; ***, ** and * = significance at p < 0.001, p < 0.01 and p < 0.05, respectively.  www.nature.com/scientificreports/  Although the correlation between GY and DH (r = − 0.30) and between GY and SL (r = − 0.16) were equally negative, the association between GY and PH (r = 0.14) was positive (Table 7).

Source of variation df
Genotype × Trait biplot and trait relationship analyses. GY as well as yield related traits were plot-  www.nature.com/scientificreports/ showed high values for TKW, MKW, KS and BM whereas the low yielding genotypes were tall and late with long spikes and more SS (Fig. 1). Stepwise regression analysis revealed that SR Aud played significant role in influencing the TKW with model R 2 of 0.33 and C p statistic of 18.29. However, a partial value of R 2 = 0.06 that resulted into model R 2 of 0.39 was observed for a model with both SR Aud and YR Aud with a C p value of 3.00. GY was negatively influenced by SR Aud with model and partial R 2 of 0.22 and C p of 42.33. However, YR Aud was the second variable with a model R 2 of 0.38 and C p of 3.00. Lastly, contrary to GY and TKW, the results showed that YR Aud contributed to the reduction of BM with R 2 of 0.44 and C p statistic of 28.51 while a model with both SR Aud and YR Aud resulted in a partial R 2 of 0.08, model R 2 of 0.52 and a C p of 3.00 (Table 8).

Discussion
In the present study, significant effects due to seasons indicated that seasonal conditions that prevailed in MS2019 were entirely different from OS2020 and this influenced the performance of wheat genotypes over growing seasons in different years. The mean monthly rainfall and temperature was higher in OS2020 than in MS2019, thus creating conducive relative humidity for rust infection that resulted in yield reduction. Differential performance among different CIMMYT's High Rainfall Wheat Yield Nurseries for yield across low and high rainfall seasons have been reported 31 . Effects due to genotypes were significantly different for various traits like DH, PH, SL, SS, TKW, GY, BM, HI, SR Aud and YR Aud indicating presence of high level of genetic diversity among genotypes hence genotypes with desirable traits can be selected for use in local breeding. Based on the findings from this study, the MS2019 could be suitable for selection of genotypes for traits including earliness, shortness, higher SS, GY, TKW, MKW and BM since despite higher SR infections the lines had higher means for these traits compared to OS2020. Of the evaluated genotypes, 51 had multiple sibs sharing common parents in different genetic backgrounds. Phenotypic variation of these lines for above said traits suggests these lines were subjected to selection procedures across diverse environments. These variations could have been combined through the shuttle breeding strategy adopted by CIMMYT through which early generation segregating populations are evaluated in contrasting environments to identify appropriate genetic variation for wide adaptation, durable rust resistance and to enhance yield gains 32 . www.nature.com/scientificreports/ Significant genotype × season interaction observed for studied traits indicated differential response of genotypes to environments resulting in non-uniform phenotypic response of lines to SR, YR, yield and yield-related traits. This variation could be either due to interaction of the genetic and non-genetic factors during plant growth or possibly presence of diverse genes or their combination in these genotypes with differential efficiency to control growth and response to rust infections. Previous studies reported significant genotype × environment interactions for yield and yield components in wheat genotype 33,34 . Therefore, plant breeders should take climatic factors into consideration when selecting genotypes that are stable across environments to avoid abandoning good breeding lines and/or carrying over poor genetic stocks.
The high proportion of genetic variance relative to environmental variance exhibited by SR Aud , YR Aud , TKW, PH, DH and GY indicates the expression of these traits was under minimal environmental influence. On the other hand, higher estimates of environmental variance relative to genotypic variance for KS, SS, PH, SL and BM suggested that genotypes exhibited variable response for these traits across the seasons with greater influence by environmental conditions, therefore, selection based on phenotypic value of observed traits is unreliable. Knowledge of heritability is appropriate for predicting response to selection of a particular trait under certain environmental conditions. Moreover, it helps determine whether or not a particular trait can be improved by selection, by improvement of management practices, or both. Theoretically, genotypes with broad genetic background selected in contrasting environments would be expected to have low broad sense heritability for target traits due to occurrence of high G × E interaction variance which results in unreliable ranking of genotypes across environments 35 . High heritability observed for both SR Aud and YR Aud , indicates a large proportion of observed variance is heritable and selection for these traits is potentially effective, although this is dependent on the magnitude of dominance and epistatic effects which constitutes a proportion of genetic variance that is non heritable. Comparatively higher heritability of 83.09% for GY were observed in F 3 segregating populations evaluated in Pakistan 36 .
A positive correlation of GY was observed with PH, KS, TKW and BM, however, the direct effects of PH on GY were negative. On the other hand, negative correlation of GY with DH, SS and SL were observed but the direct effects of SS on GY were positive indicating that the undesirable effect of SS on GY was influenced by other traits. Other studies have reported correlation and path analysis on agronomic traits in wheat 37,38 . Although late maturity is normally associated with more accumulation of dry matter which translates into high GY, the negative relationship between earliness and GY observed in this study is desirable as an escape strategy in case heat and drought stresses prevail during growth. Despite the fact that PH had a positive though not significant correlation with GY, its direct effects on yield were negative which could imply that the indirect effects of other traits on PH significantly impacted yield. The high direct effects on GY via KS, TKW and BM indicate that these traits can be used indirectly as a selection criterion to improve GY.
The biplot analysis enabled a visual comparison of the traits, genotypes, and their interrelationships. It displayed the patterns of variability of the traits on the first and second principal components as accounting for 57.07% of the total variations present. This proportion can be considered low, with a reflection on the complexity Table 8. Regression coefficients for SR Aud , YR Aud , 1000-kernel weight, biomass, and yield of the test genotypes. GY grain yield, TKW thousand kernel weight, BM biomass, SR Aud stem rust area under disease progress curve (AUDPC), YR Aud yellow rust AUDPC, R 2 coefficient of determination, Partial R 2 coefficient of partial determination-indicates the proportion of variation explained by the full model that could not be explained by the regressor in the reduced model, Model R 2 adjusted coefficient of determination-the best model is the one with the largest model R 2 , C p Mallows C p statistic: indicates amount of bias in estimating the regression coefficient and therefore predicting the response. www.nature.com/scientificreports/ of the relationships among the evaluated traits, in accordance with the findings of comparatively lower variability of 68% in a study by Mohammadi et al. 39 who evaluated Iranian durum wheat varieties for plant height, grain yield, days to maturity and a 1000-kernel weight across rain fed and irrigated environments. In a related study, Bhatta et al. 25 noted 54% total variation on the first 2 principal components associated with agronomic and quality traits of synthetic and bread wheat accessions in Western Siberia. Traits TKW, MKW, BM, PH and HI had longer vectors and were more responsive in discriminating between lines that performed well for these traits against the ones that relatively performed low. Positive correlation between TKW, MKW, BM, KS, GY and HI as indicated by acute angles in biplot suggested that genotypes plotted along these vectors potentially possess multiple desirable traits that can be selected simultaneously for development of high yielding genotypes with good agronomic attributes. Applying a genotype by trait biplot in a study on soybean (Glycine max L.) cultivars in Ontario, Yan and Rajcan 29 were also able to visually compare and select promising cultivars for multiple traits including seed yield, oil content, protein content, plant height and days to maturity. From regression analyses, reduction in GY and TKW was better explained by SR Aud whereas YR Aud better explained the reduction in BM. Effects of SR on GY in wheats evaluated across different environments in Kenya have been previously reported 40 . Stem rust normally infects stem sheaths, leaves and occasionally glumes of wheat plant, and both mesophyll and palisade layers are ruptured during rust establishment 41 . Estimated gains in dry weight of spring wheat kernel result from temporary accumulation of photosynthates in plant stems near the time of anthesis and if SR infection occurs around this critical stage, it is expected to exert a negative influence on GY 42 .
The reduction in yield and grain quality due to rust infection could be due to reduction of photosynthetic area and destruction of phloem tissue that are responsible for mobilization and remobilization of photosynthates. Severe infections due to compatible reaction between rust and host genotype result into altered phloem transport to divert nutrients to actively growing urediniospores at the expense of developing spikes of wheat resulting in shriveled kernels and hence poor yield 41 . Though YR does not destroy tissues as done by SR, it severely infects and kills leaves at vegetative and reproductive stages through destruction of photosynthetic functions of leaves and affecting movement of assimilates, consequently reduces biomass that is often observed at physiological maturity 43 . This phenomenon was clearly observed in the step wise regression analysis, since high YR infections contributed to a greater reduction of the BM compared to SR infection which contributed the largest reduction in GY and TKW.
This study demonstrates that small plot yield tests along with data on yield related traits, YR and SR are useful for preliminary screening of a large set of lines to identify promising lines for large scale yield testing. Additionally, small plot yield trials help in strategic use of resources in terms of field space, labor and time that are increasingly required to conduct large size yield plots for large number of entries. However, according to a study by Fischer and Kertesz 44 , small plot wheat yield and harvest index estimation, respectively, explained 46% and 53% of the variations that could occur in large plot testing. This scenario was also evident in this study where the harvest index, which is a predictor of the yielding ability of genotypes showed comparatively lower range than those reported in most experiments performed in large plot testing, which reflects on the influence of interplant and interplot competitions that occurs in small plots. We found that most of the wheat genotypes possessing adult plant resistance (APR) to both rusts performed better than genotypes carrying either a race specific resistance gene or combination of race specific resistance genes. We identified 16 genotypes that possess adequate level of resistance to both SR and YR, and superior GY and yield-related traits. These genotypes would serve as a valuable resource for the selection or further improvement of locally adapted wheat cultivars. Field preparation, trial design and crop management. A well-drained plain field that was previously under a cover crop of canola (Brassica napus) was used for this study. Land was disc ploughed once and harrowed twice to achieve a fine tilled seedbed suitable for planting wheat. Each line was planted in a 2-row plot measuring 0. www.nature.com/scientificreports/ During both MS2019 and OS2020, the field was immediately irrigated using sprinklers after planting to supply adequate moisture to initiate germination and seedling growth. Supplemental irrigation was done when the rainfall was inadequate. After planting, a pre-emergence herbicide Stomp 455C was applied to supply pendimethalin at an equivalent rate of 1.37 kg ha −1 while Buctril MC a post emergence herbicide, was applied at GS13 46 to supply Bromoxynil octanoate at an equivalent rate of 0.28 kg ha −1 + MCPA ethyl-hexyl ester at 0.28 kg ha −1 , both mixed at a rate of 1.25 kg ha −1 to selectively control annual broad leaf weeds. Calcium ammonium nitrate (CAN) was top dressed when plants attained GS30 46 at an equivalent rate of 100 kg ha −1 to supply 33 kg N ha −1 . Systemic insecticide Thunder OD 145 was applied at a rate of 0.25 kg ha −1 , to supply imidacloprid at 0.03 kg ha −1 and betacyfluthrin at 0.01 kg ha −1 at tillering (GS 25) and ear emergence (GS 55) stages 46 to control Russian wheat aphid.

Methods
Data collection. The first round of natural infections of YR disease severity evaluation was done when about 50% of the test genotypes headed and the susceptible checks showed 50% disease severity. Evaluation of genotypes for YR was done over at least 2 occasions. Later, genotypes were evaluated for SR when susceptible checks showed 50% SR severity and notes were taken over 3 occasions. In all instances, YR and SR severities were estimated at 7-day intervals based on the modified Cobb's Scale (0-100%) 47 . Phenological traits viz. DH, PH, SL, KS, SS, BM, HI, TKW and MKW along with GY were measured for all wheat genotypes. For each genotype, plants were considered to have headed when 50% part of spike emerged from the boot. PH was measured at physiological maturity from the base of the plant at the soil level to the tip of the spikes excluding awns from a random sample of 5 plants per genotype. The SL was measured from a random sample of 5 spikes from base to tip excluding the awns. Both KS and SS were determined from a sample of 5 random spikes per plot. At physiological maturity, plots were harvested by cutting at the base for estimating GY and BM. Both GY and BM were recorded in grams per plot area (g m −2 ) then converted into t ha −1 as: The HI was computed by determining the ratio of GY to the total BM of plants upon harvesting from samples obtained from each plot. One thousand kernels were counted from threshed grains using a Contador seed counter (brand Pfeuffer, Serial number: 14176107) and weighed to estimate TKW. The MKW was estimated by dividing the TKW by 1000 seeds for each genotype.
Data analyses. Both SR and YR rust severity notes were converted into area under disease progress curve 48 . where Y ijklm is the observation of experimental units, µ is the overall mean, S i is the effect due to ith season, R j (i) is the effect due to jth replicate in the ith season, B k (ij) = effect due to kth block in the jth replicate in the ith season, G l is the effect due to lth genotype in the kth block in the jth replicate, SG il is the effect due to interaction between ith season and lth genotype in the ith season in the jth replicate and ε ijklm is the random error component. Effects due to genotypes were considered fixed while replicates, blocks, season and genotype × season effects were treated as representatives hence considered random. To test all pairwise comparisons among means of seasons, the Tukey's test was calculated for each trait whenever season effects were significant 50 .
Estimates of genetic, genotype × environment (genotype × season) and error variance components were computed using PROC MIXED procedure in SAS software using restricted maximum likelihood (REML) method with genotype as a random factor. These components were used to estimate broad sense heritability (H 2 ) on genotype mean basis, given by: where σ 2 P is the total phenotypic variance, σ 2 G is the genotypic variance, σ 2 GE is the genotype × year variance, σ 2 e is the error, R is the number of replications and E is the number of seasons 51 . Pearson correlation analysis was conducted using the PROC CORR procedure in SAS software to establish the relationship among GY and yield related traits. For Path analysis 26 , multicollinearity test was first checked among all independent variables using variance inflation factor statistic. To partition the correlation coefficients for yield related traits into direct and indirect effects contributing to GY, the PROC CALIS procedure of SAS 49 was used. To depict the proportion of variance explained by each yield component and associations among parameters, a biplot of genotypes and trait analysis was performed in XLSTAT Microsoft excel-2013 add-in software 52 .
A regression analysis was performed to determine the contribution of SR and YR infections on the GY losses of rust infected wheat genotypes. This was conducted using the PROC REG forward selection method in SAS software where phenotypic value of a given trait was modeled as: where Y i is the expected value of dependent variable for a set of independent variables X 1 , and X 2 ; β 0 is the expected value of dependent variable at X 1 or X 2 , = 0; β 1 , β 2 , is the partial regression coefficients for every unit increase or decrease in independent variables X 1 and X 2 , respectively and ε i is the random error. Y ijklm = µ + S i + R j (i) + B k (ij) + G l + SG il + ε ijklm , www.nature.com/scientificreports/ Use of plant material. Experimental research and field studies on common wheat plants used in this study, including the collection of plant material, were conducted following relevant institutional, national, and international guidelines and legislation. Permission from Kenya Plant Health Inspectorate Service (KEPHIS) was obtained for import of wheat seeds in Kenya.

Data availability
The data used to present the findings reported in this study is available upon request through the corresponding author.