Diverse emmer wheat (T. dicoccon Schrank), including accessions identified as stress tolerant by Zaharieva et al. (2010), were crossed and backcrossed with hexaploid bread wheat to transfer A and B genome genetic variation into modern hexaploid wheat. Eight hexaploid and 11 tetraploid parents were recombined in 43 backcross combinations using the hexaploid as the recurrent parent. The pentaploid (AABBD) F₁ was backcrossed to the hexaploid parent and hexaploid progeny selected based on plant morphology. Doubled haploids were then produced from each hexaploid BC₁F₁ plant to produce an average of 10 homozygous lines per plant. A population of approximately 537 doubled haploid genotypes was developed (see Supplementary Table 1 for details) and subsequently evaluated for heat stress, along with their recurrent parents and commercial check cultivars. The parental materials and their pedigrees are reported in Table 1. The genotype Waxwing^∗2/Kiritati and its progeny were not included in the DNA study as DNA of the parent was not available.

Field experiments were sown at the IA Watson Grains Research Centre, The University of Sydney, Narrabri, NSW (30° 20′S 149° 45′E) during the cropping seasons 2014–2016. Experiments were arranged in randomized complete block designs with two replications. Genotypes were sown in plots comprising six rows spaced 33 cm apart and 6 m length, which was reduced to 4 m for harvest. Two experiments were sown adjacent to each other, each year. Experiment 1 (E1, Environment 1) was sown at the optimal sowing date for the region in mid-May. Experiment 2 (E2, Environment 2) was sown 8 weeks after the optimal sowing date.

A set of 200 wheat genotypes from the population were sown in 2014 and a different set of 196 wheat genotypes were sown in 2015. The commercial check cultivars were the same in both years. All genotypes from 2014 to 2015 combined were evaluated in 2016, representing 543 genotypes, including commercial checks, recurrent parents and additional materials. Experiments received supplementary irrigation as required to reduce the confounding effects of drought; particularly in late sown conditions. The experiments were fertilized and managed according to standard industry practices for the region.

A common set of traits including days to flowering and physiological maturity, percentage of screenings, thousand kernel weight (TKW) and grain yield was assessed each year following Pask et al. (2012). The grain yield/plot was measured in g/plot and later converted to tons ha^-1 for all further analyzes. For TKW, five hundred grains were counted using a CONTADOR seed counter (Pfeuffer GmbH, Flugplatzstraβe 70. D-97318 Kitizingen, Germany), then weighed (g) and the result multiplied by 2 to determine TKW. Care was taken to avoid broken grains. Percentage screenings were determined as the amount of split, small sized and shriveled grains in each plot sample (approximately 450–500 g) using an Agtator sieve shaker (Graintec Scientific, QLD, Australia) with 2.0 mm diameter sieves. Screenings were estimated after 40 shakes (standardized for wheat) using the formula; Screenings (%) = (weight of screenings)/(weight of sample) × 100.

DNA of all genotypes was extracted following the CTAB method proposed by Doyle and Doyle (1990). Five leaves per plot (about 0.5–0.6 g) were collected from random plants of the middle rows of plots during tillering and placed in a 15 mL centrifuge tube containing silica gel to extract the moisture. Tubes containing leaf samples were then kept at room temperature for 7 days to dry the samples for DNA extraction. The material was then genotyped by AgriBio, La Trobe University, VIC, Australia using the Infinium iSelect SNP 90K SNP Assay (Cavanagh et al., 2013; Wang et al., 2014) following the protocol prescribed by the manufacturer. In total 35,267 polymorphic SNP markers were used for further analysis (see Supplementary Data).

The best linear unbiased estimate (BLUE) of each wheat genotype was calculated using ASReml-R (Gilmour et al., 2009). The phenotypic data was combined across all 3 years within each sowing date and BLUEs calculated for each genotype in each environment. Genotypes and environments were considered fixed terms and ranges/rows within replicates within environments as random terms in the model. GGE bi-plots of the relationships between genotypes and environments were constructed on phenotypic data using the meta-analysis procedure from GenStat version 16 (Payne et al., 2011). GGE bi-plots were produced based on screenings, TKW and grain yield to estimate genotypic stability across both environments. Cluster analysis was performed using the multivariate analysis procedure from GenStat on 90K SNPs data to form similarity matrix and dendrograms were constructed on the basis of genetic distances.

Anthesis occurred during the first 2 weeks of September in E1, whereas anthesis was delayed by an average of 4 weeks in E2. The maximum temperature between anthesis and physiological maturity in 2014 was 35°C in E1 and 40.8°C in E2. The respective E1/E2 maximum temperatures in 2015 and 2016 were 35.4/35.6°C and 29.9/36.7°C. High temperature was more pronounced in the later stages of grain filling in all environments (Supplementary Figures 1–3).

The GGE bi-plot analysis accounted for 100% of the total variation of all traits examined in both environments. There was a significant environmental effect and three typical examples of the responses of the recurrent parent (circled in red) and its derivatives for screenings, TKW and grain yield are presented in Figures 1–3. The responses of all remaining cross combinations are presented in Supplementary Figures 4–8. A dendrogram constructed using DNA diversity (part D in each figure) indicated that many genotypes were closely related to the recurrent parent. This is likely an artifact of the line development process as material was first selected visually for hexaploid type before under-going haploidization and subsequent chromosome doubling. This process skewed the population toward the more stable recurrent parent genotype. The impact of emmer introgression on each trait is presented below.

The derived materials generally had higher screenings than the recurrent parent with the exception of the parents T.dicocconP194625/Ae.squarrosa (372)/2/3^∗Pastor and 2-49/Cunningham//Kennedy (Figures 2A, 3A). Nevertheless, while the majority of progeny from other crosses had high screenings, there were some with lower screenings than the parents in crosses to PBW 502, DBW-16, and PBW 550 (Table 2). Interestingly, DBW-17 progeny had higher screenings in E1, but reduced screenings when sown late indicating possible tolerance to higher temperatures among some progeny (Figure 1A).

The derived progeny significantly varied for TKW on the basis of recurrent parent. Crosses to DBW-17 (Figure 1A), 2-49/Cunningham//Kennedy (Figure 2B), T.dicocconP194625/Ae.squarrosa (372)/2/3^∗Pastor (Figure 3B) and Sokoll produced many progeny with higher TKW than the recurrent parent (Table 2). Some crosses, such as those made to PBW 502, DBW-16, and PBW 550, produced progeny with generally similar TKW, although some with significantly lower TKW were observed. All progeny derived from Berkut had lower TKW than the recurrent parent. The recurrent parent 2-49/Cunningham//Kennedy derived line #9 and #13 had greater TKW and lower screening percentages under normal and stressed environments (Table 2).

The grain yield of derived progeny varied significantly with the recurrent parent used. Higher yielding progeny were generally observed in E1 compared with E2. High yielding progeny compared to the recurrent hexaploid parent in E1 were observed for crosses to 2-49/Cunningham//Kennedy (Figure 2C), T.dicoc-conP194625/Ae.squarrosa (372)/2/3^∗PASTOR (Figure 3C), PBW 502 and Sokoll. Some progeny were high yielding in both environments, including Sokoll and T.dicocconP194625/Ae.squarrosa (372)/2/3^∗Pastor. Progeny higher yielding than the recurrent parent in E2 only were observed in crosses to DBW-16 and DBW-17 (Figure 1C and Table 2), indicating possible tolerance to higher temperatures. Berkut progeny #26 produced greater yield (5.29 t ha^-1) under optimal sowing than recurrent hexaploid parent and check cultivars, whereas the PBW 550 derived line #37 produced greater yield (3.24 t ha^-1) under heat stress.

The range of genetic variation contributed by emmer wheat, determined as the difference between individual progeny and the recurrent parent, varied from 0.01 to 0.44 (Table 2). The greatest emmer wheat contribution was observed for the Berkut progeny #45.

The emmer derived lines in the top and bottom group for each trait in each environment and the emmer parent genetic contribution are given in Table 3. Emmer derived lines in the top 10% based on yield under optimal sowing had an average 11% contribution from the emmer parent compared with only 6% contributed in the bottom 10% (Table 3). Similar trends were observed under heat stress (delayed sowing) where the top 10% of progeny had an emmer contribution of 14% compared with 5% in the bottom group. Grain weight in the top 10% of progeny in both optimal and delayed sowing conditions also had a higher emmer contribution (up to 11%) than the bottom group (up to 6%) (Table 3). However, screenings showed a different emmer contribution depending on sowing date. Under optimal sowing, progeny with lower screenings had a smaller contribution (4%) from the emmer parent compared with 10% in the bottom grouping. This was reversed in late sowing where those with lower screenings had a higher comparative emmer contribution (11 vs. 4%) in the bottom group.