Overexpression of the WAPO-A1 gene increases the number of spikelets per spike in bread wheat

Two homoeologous QTLs for number of spikelets per spike (SPS) were mapped on chromosomes 7AL and 7BL using two wheat MAGIC populations. Sets of lines contrasting for the QTL on 7AL were developed which allowed for the validation and fine mapping of the 7AL QTL and for the identification of a previously described candidate gene, WHEAT ORTHOLOG OF APO1 (WAPO1). Using transgenic overexpression in both a low and a high SPS line, we provide a functional validation for the role of this gene in determining SPS also in hexaploid wheat. We show that the expression levels of this gene positively correlate with SPS in multiple MAGIC founder lines under field conditions as well as in transgenic lines grown in the greenhouse. This work highlights the potential use of WAPO1 in hexaploid wheat for further yield increases. The impact of WAPO1 and SPS on yield depends on other genetic and environmental factors, hence, will require a finely balanced expression level to avoid the development of detrimental pleiotropic phenotypes.

More recently the WHEAT ORTHOLOGUE OF APO1 (WAPO1) has been proposed as the candidate gene for the SPS QTL on 7AL by multiple authors [24][25][26] , and validated in tetraploid wheat using a combination of mutants and transgenic lines 27 . This transgenic validation showed that WAPO-A1 from Triticum monococcum and Triticum turgidum increased SPS in the T. turgidum cultivar Kronos. It was also reported that overexpression of WAPO-D1 in the elite hexaploid variety KN199 had no effect on spike phenotypes 19 . However, transgenic overexpression of WAPO-A1 in hexaploid wheat or in cultivars that contrast in their WAPO-A1 haplotype has not yet been reported, and could be useful to build a better understanding of gene function, effect size, and pleiotropic effects for wider and diverse wheat breeding programs.
Multi-parental Advanced Genetic Inter-Crossed (MAGIC) populations are well suited for QTL mapping due to their high levels of allelic diversity and recombination [28][29][30] . Based on residual heterozygosity in the MAGIC populations, QTLs can be further validated and fine mapped through the development of Heterogeneous Inbred Families (HIFs). Selected HIFs will be isogenic at most loci in the genome, but still heterozygous for the region spanning the QTL of interest. The resulting segregating progeny will have properties similar to near-isogenic lines (hereafter named NILs) with and without the QTL of interest and are ideal genetic material to perform precise phenotyping and transcriptomic analysis, and to assess the allelic contributions to the trait of interest as well as for fine mapping and candidate causal gene identification 31 .
In this study, we describe the functional validation of WAPO1 in determining SPS in hexaploid wheat. We used two MAGIC populations across 14 site × year environments to identify two homoeologous SPS QTLs for 7AL and 7BL. Sets of contrasting NILs derived from HIFs segregating for the QTL on 7AL were used to validate and fine map the 7AL QTL, delivering a candidate gene WAPO-A1 and its homoeologue WAPO-B1. For WAPO-A1 we observed that increased expression in the spike in the NIAB MAGIC founder parents was positively correlated with SPS in the field. Therefore, our functional validation of the WAPO-A1 gene in hexaploid wheat used a transgenic approach in both low-and high-SPS genetic backgrounds. This transgenic overexpression approach provided functional validation in hexaploid wheat for the role of WAPO-A1 expression level in determining the number of SPS. Furthermore, we demonstrate that the increase in SPS in the transgenic lines is not always translated into an increased number of seeds per spike and can lead to pleiotropic or compensation effects under our growing conditions. The latter highlights the need to finely balance expression levels to achieve positive effects on yield when applied in breeding.

Results
The CSIRO and NIAB MAGIC populations have a wide phenotypic spread in SPS. Twelve field trials using an 4-parent MAGIC population 32 from CSIRO were performed between 2011 and 2014 across three field sites in New South Wales, Australia (Yanco, Narrabri, Wallendbeen) and in a Greenhouse in Canberra (ACT, Australia). Two field trials using an 8-parent MAGIC population 28 Table S1).
Both MAGIC populations showed transgressive segregation for SPS. In the 4-parent CSIRO population, recombinant inbred lines (RILs) had a mean SPS number ranging from 12.6 to 25.0 compared to 16.3 to 23.3 for the MAGIC parents. Yitpi and Chara were identified as having a reduced mean number of SPS when compared to Westonia and Baxter. On average the difference was at least one spikelet per spike between the CSIRO parents in the low and high SPS groups. The 8-parent NIAB population, RILs had mean SPS number ranging from 17.70 to 29.80 compared to 20.32 to 26.40 for the MAGIC parents. The NIAB MAGIC parents could also be broadly divided into high and low phenotype groups, with Soissons, Robigus and Brompton having a reduced SPS number compared to the other five MAGIC parents (Alchemy, Claire, Hereward, Rialto, Xi-19). QTL mapping identified ten major SPS QTLs. QTL mapping was conducted using MPWGAIM 33 with a separate univariate analysis for each trial. In total, 203 marker-trait associations (MTAs) for SPS were identified across 12 trials for the CSIRO MAGIC population and 18 MTAs for SPS were identified across 2 trials for the NIAB MAGIC population (QTL significance p < 0.05) (Supplementary Tables S2 and S3). The CSIRO MTAs cluster into ten major QTLs that are highly significant (max. -log10(p) > 5) and reproducible across at least two trials. For six of the identified QTL, we were able to assign candidate genes based on previously known allelic effects of flowering pathway genes, leaving four QTL that were initially not annotated. Of these, a SPS QTL on 7AL was observed most consistently at a high significance and was chosen for further downstream analysis (Table 1).
Despite substantially different environmental conditions, the major QTL for SPS on 7AL was observed in all 12 CSIRO trials and in two NIAB trials. This indicates that this QTL was environmentally and genetically very stable (Supplementary Table S4 Validation and fine mapping of the 7A SPS QTL using HIFs. To validate and fine-map the 7AL QTL, four Heterogenous Inbred Families (HIFs) were derived from selected F 7 RIL lines of the 4-parent CSIRO population containing heterozygosity in the QTL region. These RILs were utilized to develop sets of NILs with contrasting presence of the 7AL SPS QTL.
For two NIL pairs (F0748 and F1570) nearly all differences were statistically significant (p < 0.05). Because the selected NILs carried the 7AL QTL genotypes of all four founder parents, the allelic effects for each of the founders of the 4-parent CSIRO population could be determined. In addition, the SPS region was further delimited from 5.7 Mb to an interval of 2.2 Mb by NIL pairs F0035 and F1570 (Fig. 1B).
Additional phenotyping revealed the high SPS lines in each NIL pair also showed an increase in Grain number per spike in most cases, albeit mostly insignificant, (Supplementary Figure S1) and no differences in heading date exceeding 0.5 days were observed (Supplementary Figure S2).
OsHMA9 is a metal-efflux protein and OsAAH an allantoate deiminase. Neither have previously been linked to SPS phenotypes 37,38 .
OsAPO1 is an F-box protein that had previously been shown to improve rice yield, primary rachis branching and lodging resistance as the causal gene of qPBN6 39,40 and SCM2 QTL in rice 12 . The wheat orthologues of OsAPO1 on 7AL and 7BL (WAPO-A1 and WAPO-B1) were thus identified as good candidate genes for the   www.nature.com/scientificreports/ 7AL and 7BL SPS QTLs. We also confirmed that no SPS QTLs were detected in the WAPO-D1 genomic region. WAPO-A1 and WAPO-B1 genes and WAPO-A1 promoter sequences for Fielder, Chinese Spring, Robigus and Claire were obtained from public resources [41][42][43] . WAPO-A1 and 5 kb of upstream promoter sequence in Yitpi, Baxter, Chara and Westonia were obtained by Sanger Sequencing (GenBank ON210994-ON210998). Sequence alignments showed that the high SPS WAPO-A1 allele carries a 115 bp promoter deletion and two amino acid changes (C47F and D384N). Based on recent WAPO-A1 nomenclature 24 we reclassified our alleles accordingly ( Table 2). All the assigned WAPO-A1 alleles are consistent with the observed founder effects in both MAGIC populations for the 7AL SPS QTL.
We also classified WAPO-B1 alleles based on recent nomenclature 22 (Table 2). All assigned WAPO-B1 alleles are also consistent with the observed founder effects except for Soissons, which has a high SPS founder effect on 7BL, while carrying the low SPS WAPO-B1.hap1 allele. Westonia carries the WAPO-B1.hap3 allele, previously reported to carry a frameshift mutation/premature stop codon 22 . Based on our alignments and the protein sequence of previous WAPO-B1 gene model TRIAE_CS42_7BL_TGACv1_578478_AA1895640.1 we believe that this gene model is annotated incorrectly (Supplementary Figure S3) and that as a result, this InDel is unlikely to cause a premature stop as it falls into the 5' UTR of WAPO-B1. This observation would be consistent with the fact that WAPO-B1.hap3 only has an intermediate SPS phenotype across four trials.

WAPO-A1 expression in the field is correlated with SPS. To investigate if the expression of WAPO-
A1 in wheat is correlated with SPS in the field we collected samples from tiller dissections of the eight NIAB MAGIC parents from a 2017 field nursery at growth stage GS32 44 , because wheat-expression.com 45 data suggests that WAPO1 homoeologous are expressed in the spike at this developmental stage. Expression analysis using qRT-PCR confirmed that higher WAPO-A1 expression in wheat inflorescences is correlated (R 2 = 0.77) with increased SPS (Fig. 3). The low expression of WAPO-A1 in Soissons may be a result of the presence of the Ppd-D1a allele in this variety, which accelerates flowering time, so we possibly missed the expression peak of WAPO-A1.
Transgenic complementation of the WAPO-A1a allele with WAPO-A1b leads to increased SPS and spike length. To functionally validate the WAPO-A1 candidate, Yitpi, carrying the low SPS WAPO-A1a allele, was complemented with the native high SPS WAPO-A1b allele and promoter from Westonia (~ 8 Kb www.nature.com/scientificreports/ genomic fragment, see methods) using Agrobacterium mediated transformation. Six sets of T 2 single-insertion homozygous lines next to their respective null-segregant control line were grown in the glasshouse and phenotyped (Fig. 4A,B and Supplementary Figure S4). Four out of six transgenic lines showed an increase in SPS compared to their null segregant. Homozygous T 2 lines #5 and #3 displayed the strongest phenotype reaching more than 27 and 25 SPS, respectively, as compared to 23 in their null segregants. Homozygous plants from lines #9 and #12 displayed a gain of about one SPS, while lines #4 and #7 did not show a significant difference in SPS compared to their null segregants. The number of SPS in a non-transformed Yitpi control was similar to the null-segregant lines (Fig. 4B).
Samples of developing spikes at GS32 were collected from these lines for gene expression analysis using qRT-PCR (Fig. 4C), which revealed a strong correlation between the level of WAPO-A1 expression and the number of SPS and spike length (Fig. 4D,E). These results are consistent with a causal, positive correlation between WAPO-A1 expression and the number of SPS and spike length.   mine if SPS and seeds per spikelet could be increased beyond the levels of the high SPS WAPO-A1b allele, we also transformed Fielder, which already contains a native high-SPS WAPO-A1b allele, with the native high-SPS WAPO-A1b allele and promoter from Westonia. We generated five independent primary transgenics in the Fielder background and isolated three single-insertion lines, for which we identified homozygous T 2 and nullsegregants for phenotypic analysis of SPS, empty basal spikelets and seeds per spike. All three transgenic lines showed an increased number of SPS in comparison to the nulls and the number of SPS was strongly correlated with the level of transgene expression (Fig. 5, Supplementary Figure S5). These results further validated the conclusion that SPS number is closely correlated with WAPO-A1 expression levels. It also demonstrates that it is possible to increase SPS number beyond the levels determined by the native high SPS allele alone.
The Fielder control lines have fewer SPS (SPS = 17.02) than the Yitpi control (SPS = 23.44), despite carrying the favourable WAPO-A1b allele. This difference in SPS between the two cultivars could be explained by different photoperiod responses. Fielder carries the photoperiod insensitive PPD-D1a allele while Yitpi carries the photoperiod sensitive PPD-D1d allele, known to increase SPS 4 .
Importantly, all the homozygous transgenic lines of Fielder, but not Yitpi containing the native high SPS WAPO-A1b transgene had spikes displaying a "compact head" phenotype (Supplementary Figure S6), which was also previously described 27 . This phenotype was observed in all three Fielder families in the homozygote T 2 lines but was not seen in the null-segregants.
Increases in WAPO-A1 expression may cause pleiotropic effects. In hexaploid wheat an increase in infertile spikelets in WAPO-A1b lines compared to WAPO-A1a has previously been described 22,26 . In Kronos, a tetraploid wheat, a reduction in spike fertility has also been reported 27 , in lines containing additional WAPO-A1 transgene copies. In our studies the transgenic lines with higher levels of WAPO-A1 expression show an increased number of empty basal spikelets (Supplementary Figure S7). This was especially pronounced in the highest expressing lines, both in Yitpi (T 2 line #5) and Fielder (T 2 line #1). Similar to phenotypes previously described in tetraploid wheat (Kronos) WAPO-A1 overexpression lines 27 we observed one or two malformed florets at the base of some spikes of the transgenic lines. The phenotypic alterations are indicative for developmental disorders like homeotic transformations or failed organ development (Supplementary Figure S8).
In rice, lines with increased OsAPO1 expression reveal a suppression of tiller outgrowth 11 . Similarly, we observed a negative association between tiller number and WAPO-A1 expression across our transgenic Yitpi lines in the greenhouse pot experiment (Supplementary Figure S7). However, the null segregants also showed variability in this trait meaning the decrease was only significant in one line (T 2 line #5).

Increased SPS in transgenic lines does not translate into equivalent increases in seed yield.
Under our greenhouse conditions we could not measure a clear benefit in yield (as seeds per spike or per plant) associated with transgene expression (Supplementary Figure S4), either due to the pleiotropic phenotypes described above or potential source limitations under these growth conditions.
The transgenic Yitpi line with the highest transgene expression and number of SPS (T 2 line #5, T 2 line #3) showed a clear penalty in both seeds per spike and seeds per plant. We also observed an extreme branching phenotype and floral defects in a small proportion of the spikes developed by Yitpi T 2 line #5 (Supplementary Figure S8). In fact, the spikes were often so long that they did not fully emerge from the underlying leaf sheath, and the florets inside were often infertile (Fig. 4A). Taken together, our data indicate, that a too high expression of WAPO-A1 reveals a rather negative effect on seed production. www.nature.com/scientificreports/ However, lines with a lower increase of the transgene and a resulting smaller increase in SPS such as Yitpi T 2 Line #12 (WAPO-A1 expression increase 37%; Seeds per spike increase 11%) and Fielder T 2 line #3 (WAPO-A1 expression increase 50%; Seeds per spike increase 21%) indicated that increases in seeds per spike could be observed, suggesting that one can select an optimal range of WAPO-A1 expression levels to achieve the desired phenotypic and yield effects.

Discussion
MAGIC populations are well suited for QTL dissection. The use of two separate MAGIC populations across 14 site × year environments enabled us to capture a wide range of genetic and phenotypic variation and to identify homoeologous SPS QTL on 7AL and 7BL. The additional recombination present in MAGIC populations allowed for improved mapping resolution which aided the direct identification of candidate genes, as was the case with the NIAB MAGIC population, or via the generation of HIF lines 31 . This pipeline for candidate gene discovery via HIF lines derived from MAGIC populations is repeatable using other yield-component traits.
MAGIC populations can also reveal the effects of stacking of favourable alleles through transgressive segregation 48 as well as epistatic effects 49 . Within the 4-parent CSIRO population alleles at the Ppd-D1, PPD-B1, VRN-A1, VRN-B1 and VRN-D1 loci are segregating in addition to WAPO-A1 and WAPO-B1 32,35 . In addition to WAPO-A1 and WAPO-B1, within the 8-parent NIAB population Ppd-D1, ELF3-B1 and ELF3-D1 are also segregating 4,7,28,50 . All of these loci were identified as significant SPS QTL in the respective populations with the exception of ELF3-B1 and ELF3-D1, which may be due to their minor effect on SPS or due to more complex environmental interactions of ELF3 51 .

Overexpression of WAPO-A1 in hexaploid wheat increases SPS. The WAPO-A1 gene has been
proposed as the candidate gene for a major QTL for SPS on 7AL mapped in several populations [22][23][24][25][26]  Here, we provide functional evidence in hexaploid wheat for a similar role of the WAPO-A1 gene in determining SPS in a dose dependent manner. We have generated two sets of transgenic lines to test the effect on supplementing cultivars carrying either the low-SPS WAPO-A1a or the high-SPS WAPO-A1b alleles with a WAPO-A1b allele from a high-SPS cultivar (Westonia). In both sets of transgenic lines, we observed a positive correlation between WAPO-A1 expression and the number of SPS, which is consistent with variation in WAPO-A1 being responsible for the observed QTL. Kuzay et al., 2022 27 provided strong evidence for a role of WAPO-B1 in determining SPS, by a reduced number of SPS present in wapo-b1 mutants. However, it was previously reported that overexpression of WAPO-D1 in the elite hexaploid variety KN199 had no effect on spike phenotypes 19 . Different WAPO1 homoeologues may therefore have different effects on SPS.
At the molecular level, WAPO-A1 increases the expression of B-Class and C-Class MADS box genes, which affect lodicules, stamen and carpel development 52 . This is a possible cause behind the floral abnormalities and reduced fertility observed in both the transgenic Yitpi and Fielder lines. In the Fielder transgenics we also observed a compact spike phenotype, which indicates that the overexpression of WAPO-A1 in these lines alters spike development in several ways. Similar phenotypes in the florets and spikes have been reported in transgenic lines overexpressing the WAPO-A1 gene in the tetraploid wheat cv Kronos 27 .
In transgenic lines displaying a low level of expression of the WAPO-A1 transgene, and a small increase in SPS, we observed an increase in the number of seeds per spike (Yitpi line #12 and Fielder line #3). This is in line with the idea that a subtle rather than a strong increase of WAPO-A1 expression levels could be associated with increased grain number. This is similar to the situation in rice in which SCM2, a weaker allele of OsAPO1 compared to Ur1, has been reported to increase spikelet numbers without reducing panicle number 12 . For commercial application of transgenic lines, further validation of the positive yield component phenotypes observed in transgenic lines with modest increases in WAPO-A1 expression in a replicated field trial environment will be required. The long timelines, high costs and current lack of commercial transgenic traits in wheat also represent significant challenges for product development.

WAPO-A1 activity is finetuned via interconnected, developmental pathways. QTL mapping in
the two MAGIC populations revealed large phenotypic effects of PPD-D1a and VRN-A1a (and likely VRN-B3 as well) on the number of SPS. More minor but significant effects on SPS from PPD-B1a, VRN-B1a and VRN-D1a were also observed. The MAGIC populations studied here thus form an excellent platform to identify the genetic backgrounds that optimise the SPS, and grain number potential associated with WAPO-A1b for specific environments.
PPD1, VRN1 and VRN3 all influence the timing of transition from vegetative to inflorescence meristem (IM) and from IM to terminal spikelet [3][4][5]53 . Later transitions are associated with increased SPS 53 , either linked to the duration or resources available for development. This would explain the low SPS observed for Soissons, carrying the early PPD-D1a allele, despite having the high-SPS WAPO-A1b allele. Similarly, in our greenhouse experiments Fielder (WAPO-A1b, PPD-D1a) also had fewer SPS than Yitpi (WAPO-A1a, PPD-D1d). WAPO-A1 native alleles on the other hand have not been reported to cause large changes in flowering time in field trials 25,53 and while transgenic expression delayed flowering, CRISPR and EMS mutants had no significant effects 27 .
Our results show that the WAPO-A1b allele has a larger effect on the number of SPS in the winter wheat NIAB MAGIC population (+ 2.1 SPS) compared to the spring wheat CSIRO MAGIC population (+ 1.1 SPS; VRN-A1a and/or VRN-B1a and/or VRN-D1a present, see Table 1). An epistatic interaction between WAPO-A1, VRN-D3, PPD-B1 and Qsn.csu-6B has recently been described 53 . These data together support the conclusion that the  56 , which interact with the WAPO1 orthologues OsAPO1 or HvUFO, respectively. OsAPO1 and OsAPO2 form a regulatory module together with LARGE2 32 . OsAPO2 has also been previously shown to be transcriptionally regulated by SP3 57 in rice. Together the WAPO1 pathway genes are expressed in the IM and lateral meristems and extend IM maturation and control spikelet meristem identity acquisition 58 . This may also explain the extreme branching phenotype observed in some spikes of the transgenic Yitpi T 2 line #5.
To further elucidate the complexity of these regulatory networks, further work should focus on determining epistatic relationships between the identified QTLs at the phenotypic level as well as relating this to more comprehensive studies of expression changes and molecular interactions in different genetic backgrounds. WAPO1 breeding applications. Increasing SPS is a potential path to increasing seeds per spike, grain number and ultimately yields. PPD1 and VRN1 have already been deployed to locally adapt cultivars to geographic adaptation zones and due to their importance for local adaptation it can be assumed that there would likely be extensive trade-offs in their use in improving SPS. WAPO-A1b on the other hand is potentially valuable for wheat breeders as it can increase SPS in locally adapted varieties without having strong impacts on flowering time.
However, the increased number of SPS found in most of the transgenic lines is not always translated in an increased number of seeds per spike. Perhaps, under our glasshouse growing conditions, the plants did not have sufficient resources to fill the extra spikelets, which might be especially relevant in pot experiments, where a large number of tillers and spikes per plant is formed as compared to the situation in dense canopies under field conditions. Additionally, strong overexpression of WAPO-A1 produces some defects during floret or seed development.
These results align with those of other publications that report on the pleiotropic effects of WAPO-A1b, including reduced spike fertility 26 and increased seeds per square meter but reduced TGW 15 . As previously proposed 24 , our results imply that WAPO-A1b promotes an increase in sink capacity which may require introgression into high-biomass productive varieties and cultivation under conditions that support high yield potential. The observed trade-offs are not unique to WAPO1 and have been shown in other genes that increase sink capacity such as FT2 59 and FUL2 3 . In rice it has also been reported that OsAPO1 increases yield 13 , but that the effect on yield is source dependent 60 and dependent on gene expression levels as described above.
It might be possible to overcome these trade-offs by stacking OsAPO1/WAPO-A1 with genes that have complementary phenotypic effects, and some promising examples have been reported in both rice and wheat. For example, stacking WAPO-A1 with GNI-A1 and qGSNP-A-5A to further increase SPS without reducing TGW was possible in a Japanese dihaploid wheat population 14 .
An alternative approach may be to exploit genes known to interact at the molecular level. In rice it was recently shown that that OsAPO1 physically interacts with OsAPO2 and a HECT-domain E3 ubiquitin ligase LARGE2/ OsUPL2. LARGE2 modulates the protein stability of OsAPO1 and OsAPO2 61 . Extrapolating from these results in rice, further finetuning of WAPO1 activity and resulting SPS and seed number might be possible in wheat through posttranscriptional mechanisms and may help to overcome the constraints due to the limited number of WAPO-A1 alleles that have currently been identified.

Conclusion
This study has provided functional validation in hexaploid wheat for the role of WAPO-A1 in determining SPS. We demonstrated that the expression level of this gene in MAGIC lines under field conditions as well as in transgenic lines in the greenhouse is positively correlated with the number of spikelets per spike. We also identified potential pleiotropic effects of excessive WAPO-A1 expression. Future work can focus on identifying additional existing or novel WAPO1 alleles that have an optimized level of expression and protein activity or identifying pathways and genes that can be used to optimize the expression of WAPO1 or that have complementary phenotypic effects. The optimal activity of WAPO-A1 resulting into yield gains can be expected to be dependent on the yield potential of the environment and agronomic practice.

Methods
Plant materials. The founder parents for the NIAB and CSIRO wheat MAGIC populations were selected based on wide genotypic and phenotypic diversity.
Six of the NIAB MAGIC population founder parents are winter wheat cultivars from the United Kingdom (UK) (Alchemy, Brompton, Claire, Hereward, Rialto, Robigus) and one from France (Soissons) and one is a facultative spring variety (Xi-19) 28 .
All plant material used in this study was derived from commercially released wheat varieties that were obtained and used in accordance with applicable institutional, national and/or international regulations. QTL mapping of SPS. The 4-parent CSIRO MAGIC population had previously been genotyped using both the Infinium 9 k and 90 k SNP chips as well multiple DArTs and microsatellites for key genes 32,33 and the 8-parent NIAB MAGIC population using the Infinium 90 k SNP chip 28,34 . QTL analyses were conducted using whole genome average interval mapping using MPWGAIM 33 . For the NIAB MAGIC population a genetic map adapted from Gardner et al., 2016 34 was used. For the CSIRO MAGIC population a previously published map was used 33 . Skimmed genetic maps containing only unique mapping locations were used to calculate MAGIC RIL haplotypes and founder probabilities, and which were then used for QTL mapping using MPWGAIM. Assigned candidate genes are based on previously published allelic information 35 . Development of HIF lines and QTL fine mapping. Following the process described by Barrero et al., 2015 31 , four founders of heterogeneous inbred families (HIFs) were identified from the F 7 MAGIC RILs through genotyping of the 7AL SPS QTL region using a selection of KASP markers designed based on SNPs from the Infinium iSelect 90 K array 63 . After selfing of each (heterozygous) HIF founder, homozygous progeny with contrasting SPS QTL genotypes were selected and self-pollinated to generate sets of contrasting NILs. Plants from four families were then phenotyped across six field trials as described above.
Genotyping of the F 12 lines with the Infinium iSelect 90 K array 63 was used to capture any additional recombination in the SPS QTL region and to assign founder genotypes. All NILs were checked and found to contain < 1% residual heterozygosity.

7AL and 7BL SPS QTL Candidate gene sequencing. Publicly available WAPO-A1 and WAPO-B1
sequences were obtained from the reference wheat line Chinese Spring 41 as well as from Brompton, Robigus and Cadenza (parent of Xi-19) 43 . Additionally, full length sequences for WAPO-A1 including 5 kb upstream promoter sequences in Yitpi, Baxter, Chara and Westonia were obtained via cloning and Sanger Sequencing. Exome sequences for WAPO1 homoeologues in Alchemy, Claire, Hereward, Rialto, Soissons and Xi-19 were obtained via Exome Capture sequencing.
Based on the Chinese Spring reference sequence primers were designed and WAPO-A1, B1 and D1 sequences were cloned from Chinese Spring, Chara, Baxter, Westonia and Yitpi using a Zero Blunt TOPO Cloning kit (Invitrogen). Subsequently the 5 kb promoter sequences for WAPO-A1 were cloned from overlapping PCRs using the same methodology. Clones were selected on LB + Kanamycin medium and sent for Sanger sequencing after two nights of incubation at 37 °C. Sanger Sequencing was performed by LGC (UK). Sequences are available under GenBank accession numbers ON210994 to ON210998. PCR conditions and primers are available in Supplementary Table S6.
Wheat Exome Capture sequencing was conducted using a set of internal exome probes based on the IWGSC RefSeqv1 annotations (pre-release) including CDS + UTR sequences. Probe design and balancing was carried out by Roche. Genomic DNA was extracted using a Qiagen DNeasy Mini Kit (Qiagen, US). Library preparation, sequence capture and sequencing were performed at Fasteris. Sequencing was performed on an Illumina HiSeq3000 to obtain an average of 8.2 Gb of raw sequence data (2 × 150 bp paired end reads) per sample. Reads for the eight NIAB founders were mapped to the repeat-masked Chinese Spring reference genome (IWGSCv1.0) including mitochondrial (EMBL accession: AP008982) and chloroplast (EMBL accession: AB042240) genomes with BWA 65 (Li, 2013). This includes an additional remapping step to fine-map InDels. Variant calling and filtering were conducted using GATK 66  WAPO-A1 sequence including 5.4 kb of the native promoter was synthetically design at GeneArt (Thermo Fischer Scientific, US) and received in the standard cloning vector pUC57. This cassette was subcloned into the binary vector VecBarII (using the XbaI and HindIII sites), which has been extensively used at CSIRO for wheat transformation 70 . CSIRO's Transformation Unit performed the transformation of the Yitpi cultivar or Fielder, using Agrobacterium 71 . Yitpi was chosen over Chara for the complementation experiment because of previous work done at CSIRO showing higher transformation rates in Yitpi (19-27% for Yitpi vs. 5-9% for Chara, data not shown). Thirteen Yitpi and five Fielder primary transformants were generated and grown in pots in the glasshouse. To assess the copy number in each primary transgenic, 25 T 1 seeds from each T 0 plant were germinated in petri dishes and a Basta-resistance 'leaf assay' was used to determine transgene segregation 72 . After a week, the resistance to Basta was scored to determine segregation ratios (data not shown). Based on the segregation ratios six T 0 Yitpi and three T 0 Fielder transformants were selected for further work. From each of them eight T 1 positive lines and two null-segregants were transplanted to soil and grown in glasshouse conditions. For each T 1 plant, about 25 T 2 seeds were used in leaf-tipping assays to identify homozygote lines. A T 2 homozygous and a nullsegregant lines from each family were chosen for final analysis. Ten plants of each were transplanted to pots in the glasshouse and they were phenotyped at maturity. Samples were also collected for gene expression analysis in developing spikes.
Gene expression of WAPO-A1 in transgenic and control lines was analyzed by RT-qPCR in developing spikes at GS32. Expression levels were calculated relative to the housekeeping gene TaActin1 (CJ961169). Primer sequences are in Supplementary Table S7.
All transgenic work was conducted at CSIRO in Canberra (ACT, Australia) in certified Physical Containment Level 2 (PC2) facilities, including laboratories and glasshouses, following the standards set by the Australian Office of the Gene Technology Regulator (https:// www. ogtr. gov. au/).

Data availability
DNA sequences generated in this study are available in NCBI GenBank (accession numbers ON210994 to ON210998). Other datasets not compatible with Springer Nature mandated databases are available for academic partners for non-commercial purposes upon request sent to the corresponding author, provided that bilateral terms-of-use agreements can be concluded. License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.