The experiments were carried out at the Leibniz Institute of Plant Genetics and Crop Plant Research, Gatersleben, Germany (51° 49′ 23″ N, 11° 17′ 13″ E, altitude 112 m) under greenhouse and field conditions (Table S1). Thirty hexaploid winter wheat accessions (first experiment, Table S2) and 12 hexaploid spring wheat cultivars (second experiment, Table S3) were selected for phenotypic measurements. Control and tiller removal treatments were conducted in the field and greenhouse simultaneously. For detillering treatment, only the main shoot was maintained, and all other tillers were removed as soon as they became visible. In the subsequent experiment, the re-growth tillers were continuously removed two to three times per week until physiological maturity. For the control, no tillers were removed (Figure S1). Eighty plants per cultivar (40 plants for the control and 40 for tiller removal) were grown under field and greenhouse conditions.

For both the greenhouse and field experiments, seeds were sown in 96-well trays on the same date and germinated under greenhouse conditions (photoperiod, 16 h:8 h, light: dark; temperature, 20°C:16°C) for 14 days. Seedlings at the two- to three-leaf stage were transferred to 4°C and vernalized for 63 days. Vernalized seedlings were transferred to hardening conditions (photoperiod, 12 h:12 h, light:dark; temperature, 15°C) for 7 days to allow them to gradually acclimate. Finally, half of the plants were transplanted into 0.5 L pots (one plant per pot; 9 × 9 × 9 cm) under greenhouse conditions (photoperiod, 16 h:8 h, light:dark; temperature, 20°C:16°C). The plants were supplied with supplemental light (~250 mE m⁻²s⁻¹ PAR, photosynthetically active radiation) from low-intensity incandescent bulbs and were irrigated when required. The remaining plants were directly planted into a field with silty loam soil (20 plants per 2 m long row with 20 cm spacing between rows). All plants were manually irrigated as required.

The seven floral development stages were determined according to Kirby and Appleyard (1987) and Zadoks et al. (1974): the terminal spikelet (TS) stage, when spikelet initiation was complete; the white anther (WA) stage, when the stamens of florets 1 (F1) and 2 (F2) were completely enclosed by lemmas; the green anther (GA) stage, when all but the tips of florets were covered with glumes; the yellow anther (YA) stage, when the glumes were fully formed and the lemmas of the first three florets were visible; the tipping (TP) stage (Z49), when the first awns were visible; the heading (HD) stage (Z55), when 50% of the spikes were visible; and the anthesis (AN) stage (Z65), when 50% of the spikes had yellow anthers.

To detect the TS, WA, GA, and YA stages, each genotype was examined under a stereomicroscope every 2 days (Stemi 2000-c, Carl Zeiss Micro Imaging GmbH, Gottingen, Germany). For the TP, HD, and AN stages, the day of onset was recorded as the point at which 50% of the plants reached the respective stage. Thermal time was used to identify the duration of each stage, which was calculated as the sum of the daily average temperature [(Tmax+Tmin)/2; the base temperature was assumed to be 0°C].

During each stage, three plants per cultivar were randomly selected for all phenotypic measurements. Leaf area was measured immediately after dissection of fresh main culm leaf material using an area meter (LI-3100, LI-COR Ltd., Lincoln, NE, USA). Main culm shoots and spikes were dried separately in two cellophane bags at 60°C for 3–5 days for DW measurements. Stem DW refers to the DW of one main culm shoot, including leaves without spikes. Anther and ovary size measurements were only conducted on the main culm spikes and not on the spikes of tillers. From the GA to AN stages, spikelets from the center of the main culm spike were dissected and used to obtain digital images of anthers and ovaries for three plants per stage. When the anther or ovary was aborted, the size was not measured. Anther length (anther size) and ovary width (ovary size) were also measured under a stereomicroscope using the Carl Zeiss Imaging System AxioVision Rel. 4.8.2. For most plants, only the visible structures of the anthers and ovaries in the first four florets from the base were observed. In the detillering experiment, all traits were measured on the main shoot; tiller traits were not measured.

The methods used for data analysis and statistical analysis can be found in Chen et al. (2014).

As shown in Figure 1, we dissected the SEP into seven stages, including TS, WA, GA, YA, TP, HD, and AN, based on floral development and the abortion process. We quantified 80 plant growth traits in 30 winter wheat accessions to obtain overall information about plant growth during the SEP in both the field and greenhouse. The general range of values for most traits at the same stage was consistent between the greenhouse and field (Figure 2A). The trend in color switching was consistent for most traits, suggesting that the ranges of various plant growth traits at different stages are consistent and independent of the environment (Figure 2A).

Although the phenotypic ranges of most traits showed relatively high stability between field and greenhouse conditions, when investigating different traits at specific stages, we found that some traits significantly differed between the two environments (Figure 2B). Tiller number, leaf number, node number, leaf DW, tiller DW, and leaf area significantly differed between the greenhouse and field from WA to AN, TS to AN, TS to GA, YA to AN, WA to AN, and TS to HD, respectively. Node number did not change from GA to AN. However, leaf area was higher in the greenhouse, whereas leaf DW was higher in the field, implying that leaf thickness was higher in the field than in the greenhouse. Plant height was significantly higher at TS, YA, TP, HD, and AN in the field compared with the greenhouse, and spike DW was markedly higher under field conditions at GA, YA, TP, HD, and AN. Spike length was significantly different between greenhouse and field conditions only at the GA stage, as spike length increases rapidly at this stage and is sensitive to the environment; F1 (the first floret from the base of an individual spikelet) anthers and ovaries were significantly different between environments at AN.

The traits that could be used to distinguish between stages differed depending on the stage. In general, the distinguishing traits at different stages were consistent between the greenhouse and field, suggesting phenotypic stability, but the weights of their roles differed slightly, pointing to phenotypic plasticity (Figure 2C). Under greenhouse conditions, F1 ovary size and spike DW are two important traits for the AN stage, whereas F1 anther size plays an important role at GA, YA, TP, and HD, and node number can be used to distinguish the WA stage (Figure 2C). Under field conditions, F1 ovary size and tiller DW are two important traits for the AN stage, whereas F1 anther size is critical for GA, YA, and TP. Node number plays an important role at the TS stage, and tiller DW can be used to distinguish the HD stage. Finally, tiller DW has more influence in the field than in the greenhouse from YA to AN, and the influence of this trait increases from YA to AN under field conditions (Figure 2C).

As shown in Figures 1, 2, most of the traits examined are significantly influenced by the environment, and different traits play different roles in distinguishing the seven stages. The influence of genotype (σG2), environment (σE2), genotype-environment interaction (σG×E2), and residual (σe2) for different traits may explain the environmental effects and their different roles, as shown in Figures 1, 2. Therefore, we conducted variance component analysis according to the seven stages of the SEP. The results were as follows (as shown in Figure 3A): (1) the influence of genotype (σG2) on tiller number showed a declining trend from TS to AN, indicating decreasing stability during this phase; (2) the genotype and environment interaction (σE2) is important for determining leaf number at the WA stage, with the weight of σG×E2 decreasing from WA to AN, and genotypic effects (σG2) had the strongest influence on leaf number at TS; (3) residual (σe2) played an important role in determining tiller, leaf, and node number at all stages, pointing to phenotypic variation within a genotype for these three traits; (4) genotypic effect (σG2) had the strongest effect (explaining >50% of variance) in PCA of plant height and leaf DW at all stages, implying that plant height is a stable trait; (5) genotypic influence (σG2) played the largest role in the PCA of spike length from TP to AN; genotype-environment (σE2) had almost no effect, pointing to strong stability at all three stages; (6) genotypic effects (σG2) accounted for >50% of variance for all phenotypic components for leaf area from YA to AN, indicating strong stability during this phase; (7) similarly, genotypic effects (σG2) accounted for the highest amount of variance in the determination of spike DW from YA to AN; (8) for main shoot DW, genotypic effects (σG2) played the largest role at all stages, whereas environmental effects (σE2) were quite weak; (9) environmental effects (σE2) played the largest role in determining tiller DW from WA to AN, suggesting that tiller DW is highly sensitive to these effects during this phase; and (10) F1 anther and ovary size exhibited high stability at all stages examined, as revealed by the large proportion of genotypic effects (σG2) and the small proportion of environmental effects (σE2).

We performed three-way PCA to further dissect phenotypic variance components and to uncover their contributions to different traits at all seven stages. In three-way PCA, if the influence on a trait at a specific stage is great, this influence causes large differences between cultivars. Our results were as follows: (1) Tiller number is sensitive to the environment, with the most evident differences in tiller number between genotypes observed at stages WA and GA in the field (Figure S2). (2) The clearest differences in leaf number among genotypes were detected at YA and AN in the greenhouse, as supported by the wider range (i.e., substantial variation) in leaf number in the greenhouse vs. the field (Figure S3). (3) Genotypes could be distinguished based on node number most easily in the field at stages TS, WA, and GA (Figure S4), whereas little variation in node number was detected in the remaining stages in both the field and greenhouse. (4) The differences between genotypes based on plant height were not as obvious as those for the other traits in the field and greenhouse at all seven stages. Genotypes could be distinguished based on plant height most easily at stages TP and HD in the greenhouse and at YA and HD in the field (Figure S5), which is consistent with the finding that plant height increased a bit more rapidly in the field than in the greenhouse (Figure 2B). (5) Stages TP, HD, and AN in the greenhouse and GA, YA, and WA in the field contributed greatly to the differences in spike length detected at different stages (Figure S6). (6) For leaf area, the WA, GA and YA stages in the greenhouse and AN in the field played different roles in differentiating genotypes (Figure S7). (7) The differences in spike DW among genotypes were most obviously influenced at HD and AN in both the greenhouse and the field (Figure S8). (8) The differences in leaf DW among genotypes were much more obvious in the field environment than in the greenhouse. The greatest differences in leaf DW among genotypes were detected at the TS, YA, HD, and AN stages in the field (Figure S9). (9) The differences in main shoot DW among genotypes were most obvious at the AN stage under both greenhouse and field conditions. In addition, the differences in main shoot DW among genotypes differed at TP and HD in plants grown in the greenhouse (Figure 3B). (10) The differences in tiller DW among genotypes were much more evident in the field than in the greenhouse at all seven stages, which is in agreement with the relatively wide range of tiller DW (Figure 2B). Obvious effects on tiller DW were detected under two conditions, F-HD (field-heading) and F-AN (field-anthesis), with significant differences in tiller DW detected among the 30 wheat accessions (Figure 3C). (11) F1 anther size exhibited the most significant differences between genotypes at YA and HD in the greenhouse and at GA and TP in the field (Figure S10). (12) F1 ovary size markedly differed among genotypes at AN under both greenhouse and field conditions (Figure S11). Finally, marked differences in F1 ovary size were detected at the HD stage in the greenhouse and at the GA stage in the field and greenhouse.

Based on the seven stages shown in Figures 1, 2, we further investigated the five stages of anther and ovary size, as we found that anther and ovary size increase from GA (maximum number of floret primordia, grain yield potential) to AN (Guo et al., 2015). We previously found that ovary size at AN is closely associated with grain number per spike and that floral abortion occurs after GA (mainly from GA to AN) likely due to the competition for assimilates between florets within an individual spikelet (Guo et al., 2016). To further dissect the phenotypic components of anther/ovary size and to examine the roles of anther and ovary size in determining grain yield traits (including grain number per spike, grain weight per spike, individual grain area, individual grain width, individual grain length, and thousand kernel weight (TKW), we measured anther and ovary size across the four floret positions (Figures 4A–C, F1 [the first floret from the bottom of an individual spikelet], F2, F3, and F4) and five stages (GA, YA, TP, HD, and AN) under control and detillering treatment in the field and greenhouse to observe the dynamic growth of florets and the effects of detillering on floret growth.

As shown in Figure 5A, favorable (greenhouse) conditions significantly increased F4 anther size at all five stages (GA, YA, TP, HD, and AN) under both control and detillering treatment. F4 ovary size at all five stages was also significantly higher under greenhouse conditions vs. the field under both control and detillering treatment, except for AN under detillering treatment. Similarly, greenhouse conditions markedly increased anther and ovary size at positions F2 and F3 at all five stages under control and detillering treatment (Figures S12–S14). However, the effects of greenhouse conditions on anther and ovary size at position F1 were not as significant as those at F2, F3, and F4 (Figures S12–S14).

As shown in Figure 5B, according to variance component analysis of anther and ovary size at F1, F2, F3, and F4, the genotypic effects (G), environmental effects (E), interaction between genotype, environment, and treatments (GET), and residual (ERR) explained a relatively high proportion of the variance compared to the other components, including interaction between genotype and environment (GE), genotype and treatments (GT), and environment and treatments (ET).

Analysis of phenotypic similarity based on anther size, ovary size, and grain size and number revealed that anther and ovary size at the same stage were close to each other, and grain number and size were close to ovary size at AN (Figure S15).

To construct a plant phenotypic map across the 12 spring wheat genotypes, we performed clustering analyses on the comprehensive phenome-wide data set across treatments (control and detillering) and environments (field and greenhouse) (Figures 6A,B). This map provides an overall view of the phenotypic similarity of plant and floret growth and facilitates further evaluation of floret growth and grain yield traits. Hierarchical cluster analysis (HCA) revealed that most traits under detillering treatment were separated from the same traits under control conditions irrespective of genotype, but the traits in the field could not be clearly differentiated from the traits in the greenhouse. To clearly visualize the phenotypic similarity among lines, as revealed by genotype similarity (the distance between genotypes in cluster analysis), we performed self-organizing map clustering analysis of the data set (Figure 6B). In the self-organizing map plot, the genotypes under detillering treatment could be distinguished from the same genotypes under control conditions, but the genotypes in the field and greenhouse could not be clearly separated. The results of HCA and self-organizing map clustering analysis support each other and are consistent with the finding that greenhouse conditions have strong effects on anther and ovary size compared to field conditions, while the differences in anther and ovary size between detillering and control treatments were not as significant as they were between the field and greenhouse.

We then constructed a neighbor-joining tree based on the 31 anther and ovary growth-related traits to investigate the phenotypic similarity of the genotypes in different release years. We constructed phenotypic similarity trees for the genotypes under greenhouse and field conditions, respectively (Figure S16). We found that genotypes with the same release years tended to cluster closely together, suggesting that anther and ovary size across different floret positions and stages are influenced by the release years of the cultivars, but the effects are not very strong. In addition, the two phenotypic similarity trees in the field and greenhouse are not very similar, but the relative relationships of some cultivars within the same groups are similar. Finally, we determined the phenotypic distance matrices of the two trees and found that they are positively associated (Pearson's coefficient r = 0.407, P < 2e-04, Mantel test; Figure S17).

We performed principal component analysis (PCA) to capture phenotypic variation in the entire population and to extract specific phenotypic traits regarding the discrimination of groups (Figure 7, Figures S18–S22). The top six principal components (PCs) explain 99.13, 98.61, 98.41, 98.35, and 97.93% of the total phenotypic variation at stages GA, YA, TP, HD, and AN, respectively, indicating that the plants showed the most phenotypic differences at the early growth stages. Although there were no significant differences in phenotypic variation explained by the top six PCs between the five stages, the cumulative variance explained by the top six PCs decreased with floret growth. Also, the cumulative variance explained by the first PC decreased with floret growth, with the highest explained cumulative variance detected at GA (88.8%) and the lowest at AN (50.09%).

Overall, the first PC can separate field- and greenhouse-grown plants, but it cannot distinguish between plants under control vs. detillering treatment or among the three genotype groups, whereas the other PCs cannot separate plants based on environment, treatment, or genotype. These results are in agreement with the results of clustering analysis and the finding that greenhouse conditions have strong effects on anther and ovary size compared to the field, whereas the differences in anther and ovary size between detillering and control treatments were not as significant as they were between the field and greenhouse.

We assessed the connections between floret size (anther and ovary size) and grain yield traits (grain number per spike, grain weight per spike, grain area, grain width, grain length, TKW), as shown in Figure 8. At HD, under all four conditions (greenhouse-control, greenhouse-detillering, field-control, field-detillering), F4 ovary size was relatively closely connected with grain number per spike, suggesting that F4 ovary size at HD plays an important role in determining grain number per spike. F1 anther size was closely associated with grain size traits (TKW, individual grain area, width, length), implying that F1 anther size at HD plays a critical role in determining grain size. In addition, F1 anther size at TP was strongly correlated with individual grain width under all four conditions, indicating that F1 anther size at TP plays an important role in determining grain width. Moreover, association analysis between anther/ovary size and grain number/size showed that anther/ovary size shares relatively close connections with grain number/size traits at the late floral development stages compared with early floral development under both control and detillering treatments in both the field and greenhouse. Finally, there were no significant differences between different treatments or environments, suggesting that the connections between anther/ovary size and grain number/size are relatively stable.

As shown in Tables 1, 2, we divided the 12 spring wheat accessions into two group according to their release years (Table S3). The release years of the first six accessions range from 1931 to 1953, whereas those of the six remaining accessions range from 1959 to 1997. In the greenhouse, ovary size at anthesis at positions F1, F2, F3, and F4 in accessions with early release years was consistently larger than that of accessions with later release years. However, in the field, ovary size at anthesis in accessions with early release years was not consistently larger than that of accessions with later release years, which may be attributed to environmental influences (Tables S4, S5).

Finally, detillering treatment markedly increased ovary and anther size at anthesis under both greenhouse and field conditions (Tables 1, 2; Tables S4–S9). Therefore, the increased assimilate partitioning to the spike leads to improved anther and ovary size. In summary, the effects of release year and detillering treatment on anther size at anthesis were not as significant as their effects on ovary size (Tables 1, 2; Tables S4–S9).

Tiller and leaf growth are closely coordinated in wheat plants under favorable conditions (Bos and Neuteboom, 1998; Miralles and Richards, 2000). The low-tillering wheat genotypes first described by Atsmon and Jacobs (1977) are ideal materials for investigating tiller and leaf growth. Richards (1988) identified a single recessive gene (tin) located on chromosome 1A in a wheat uniculm line. Kuraparthy et al. (2007) mapped another tiller inhibition gene (tin3) on chromosome 3A in wheat. Physiological studies showed that tin lines have a greater harvest index and grain weight than the wild type, making them quite suitable for growth under terminal drought stress due to their reduced non-productive tiller number and soil water use prior to anthesis, greater biomass at anthesis, and increased water-soluble carbohydrate levels in stems (Duggan et al., 2005a,b; Mitchell et al., 2012; Hendriks et al., 2016). The flag leaves of tin lines are larger than those of the wild type, with longer leaf blades at higher phytomers compared to free-tillering lines, as well as reduced total leaf area (Moeller et al., 2014). The tin gene is positively related to root length and biomass (Hendriks et al., 2016). Precocious internode development is affected by tiller bud outgrowth in the tin mutant (Kebrom et al., 2012). Taken together, these findings point to the importance of tiller and leaf growth in wheat.

In the current study, we described tiller and leaf growth at a fine scale to facilitate the manipulation of the growth of these plant parts in a narrow time window to improve grain yield. We found that tiller and leaf growth were most strongly influenced by the environment (field and greenhouse) (Figure 2). PCA showed that tiller and leaf number exhibited large phenotypic variation within genotypes (high error, Figure 3A); tiller DW was sensitive to the environment (high environmental effects, Figure 3A) at all stages except TS, which is consistent with the finding that tiller DW significantly increased under field conditions at all stages except TS. Leaf area at AN and leaf DW at TS, WA, and GA were relatively stable (weak environmental effects, Figure 3A), which is supported by the observation that the environment had no significant effect on leaf area at AN or leaf DW at TS, WA, and GA.

The importance of spike DW and both anther and ovary size in determining grain yield traits has been demonstrated in previous studies (Abbate et al., 1997; González et al., 2003a; Gonzalez et al., 2011; Fischer, 2007; Xie et al., 2015). Extending the SEP is an important way to increase spike DW at anthesis (Fischer and Stockman, 1980; González et al., 2003a, 2005a; Demotes-Mainard and Jeuffroy, 2004), which would guarantee an adequate supply of assimilate for grain filling after anthesis (Schnyder, 1993; Calderini and Reynolds, 2000; Borras et al., 2004; Ghiglione et al., 2008; Reynolds and Tuberosa, 2008). Increasing ovary size at anthesis would also improve the chances of grain setting in wheat (Guo et al., 2016). In the current study, we found that spike DW showed variable stability over the seven stages of the SEP, suggesting that the reliability of manipulating spike DW varies at different stages. For example, genotypic effects play a large role in determining spike DW at tipping, heading, and anthesis, indicating that manipulating DW would be highly reliable at these three stages. In addition, leaf area is critical for the production of photosynthetic products, which influence spike DW. Therefore, spike DW could be increased by manipulating leaf area. The high proportions of genotypic effects at the yellow anther stage and anthesis imply that it would be highly reliable to manipulate leaf area at these two stages.

Favorable conditions (greenhouse) had a significant effect on anther and ovary size, especially for anthers and ovaries at distal positions (e.g., F3, F4) within individual spikelets. This finding is in agreement with the results of variance component analysis, i.e., that environmental effects explain a large proportion of the variance in this trait (Figure 5). Moreover, this result is supported by the finding that the first PC can distinguish traits (anther and ovary size) measured in the field and greenhouse (Figure 7). Although the environment strongly affected anther and ovary size, anther and ovary size showed relatively high phenotypic similarity, since anther and ovary size at the same stages were clustered together, indicating relatively high stability. Furthermore, the moderate degree (Mantel test, r = 0.407, P = 2e-04) of correlation of phenotypic distance between genotypes under both field and greenhouse conditions and the relatively high similarity of cluster trees in the field and greenhouse indicate that both genotype and environment have relatively strong effects on anther and ovary size. These results suggest the possibility of genetically manipulating anther and ovary size and the potential risk of exposing plants to stressful environmental conditions during the floret growth phase.

Manipulating the SEP by altering the expression of known flowering-time genes (e.g., Ppd, Vrn) is an important way to regulate floret fertility/grain number in wheat (González et al., 2005b; Serrago et al., 2008; Ursula et al., 2017). In addition, sugar content in spikes is closely associated with floret fertility (Ghiglione et al., 2008; Dreccer et al., 2014). Although many strategies have been designed to increase grain yield in wheat, little is known about the role of anther and ovary size in determining grain yield (Guo et al., 2016). The regulation of anther and ovary growth during the pre-anthesis phase is poorly understood. It would be worth investigating the sensitivity of anther and ovary growth to different environments (e.g., heat and drought stress) in wheat.

In this study, we dissected floret growth (anther/ovary size) during the SEP to assess the key stages of floret growth to improve grain yield. Unexpectedly, we found that ovary size (ovary width in this study) at the heading stage is closely associated with grain number, while anther size (anther length in this study) at the tipping stage is closely associated with grain width. These findings indicate that ovary size at heading plays an important role in floret survival/grain setting and that anther size at the tipping stage is crucial for ovary/grain swelling. Phenotypic association analysis revealed the relatively high importance of anther and ovary size at the heading stage in determining grain number and size. We previously found that ovary size at anthesis plays an important role in determining grain number per spike (Guo et al., 2016), whereas in the current study, we found that ovary size is important during the heading stage. These different results may be attributed to environmental effects on anther and ovary size detected in this study. On the other hand, the results of this and the previous study imply that ovary size at late stages (i.e., heading and anthesis) is quite crucial for determining grain number and size.

As expected, we found that detillering clearly improved anther and ovary size, which may be attributed to the increased distribution of assimilates to grains. Unexpectedly, wheat accessions with early release years (1931–1953) had greater ovary sizes than those with later release years (1959–1997). These findings suggest that the increase in grain number that has occurred over the course of wheat breeding in the past decades may have led to a reduction in ovary size, which might further decrease grain size.

In summary, there are two important differences between the current and previous studies: (1) We studied the SEP at different developmental stages and (2) we obtained an overall view of plant and floret growth during the SEP. The analyses of phenotypes during the SEP in narrow time windows performed in this study suggest that plant and floret growth are not only controlled by phenology genes, but they are also strongly influenced by the environment. Therefore, future studies should focus on the stages of SEP in greater detail, particularly via physiological and genetic analysis.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The reviewer MRT and handling Editor declared their shared affiliation.