Development of Commercial Thermo-sensitive Genic Male Sterile Rice Accelerates Hybrid Rice Breeding Using the CRISPR/Cas9-mediated TMS5 Editing System

Hybrid rice breeding offers an important strategy to improve rice production, in which the cultivation of a male sterile line is the key to the success of cross-breeding. CRISPR/Cas9 systems have been widely used in target-site genome editing, whereas their application for crop genetic improvement has been rarely reported. Here, using the CRISPR/Cas9 system, we induced specific mutations in TMS5, which is the most widely applied thermo-sensitive genic male sterility (TGMS) gene in China, and developed new “transgene clean” TGMS lines. We designed 10 target sites in the coding region of TMS5 for targeted mutagenesis using the CRISPR/Cas9 system and assessed the potential rates of on- and off-target effects. Finally, we established the most efficient construct, the TMS5ab construct, for breeding potentially applicable “transgene clean” TGMS lines. We also discussed factors that affect the editing efficiency according to the characteristics of different target sequences. Notably, using the TMS5ab construct, we developed 11 new “transgene clean” TGMS lines with potential applications in hybrid breeding within only one year in both rice subspecies. The application of our system not only significantly accelerates the breeding of sterile lines but also facilitates the exploitation of heterosis.


Results
Development of TGMS lines using CRISPR/cas9 editing of TMS5. The CRISPR/Cas9 editing system mediates targeted genome editing through the complex of Cas9 endonuclease and guide RNA, which provides a simple and efficient technique compared with other genome engineering technologies, such as zinc-finger nucleases (ZFNs) and transcription activator-like effector nucleases (TALENs) 39,40 . CRISPR/Cas9-induced editing events take place mainly in the T-DNA transformed callus cells before regeneration in rice 26 . Therefore, homozygous mutants with high mutation frequencies can be obtained in the T 0 generation, and mutants without exogenous T-DNA can be isolated after segregation of the T 1 generation. These mutants are also named "transgene clean" plants 41 . We used the CRISPR/cas9 editing system, which is able to target multiple sites via a single construct 26 , to create TMS5-targeted mutations to develop commercial TGMS lines. To obtain target sequences with high editing and low off-target efficiencies, 10 target sites (Supplementary Figure 1) screened from the TMS5 coding region were randomly divided into five groups, and the two target sequences in one group were cloned into the same vector to transform the ZH11 callus (see Methods). Transgenic plants were obtained based on five vectors. To evaluate the editing efficiencies of each target sequence, the target sites in the leaves of T 0 plants were sequenced. Mutations were detected in 9 of 10 target sites. Among these nine mutation sites, the single mutation frequencies of the target sequence ranged from 46.2% to 88.2% and from 69.2% to 94.1% T 0 plants contained mutations in at least one of the two target sites (Fig. 1A). Notably, the homozygotic mutation frequency for single target sequences reached 29.4% and the plants carrying homozygous mutations for at least one of two target sequences reached 32.4% (Fig. 1A). Furthermore, individuals with pollen sterility were obtained for all T 0 plants under high-temperature (HT) treatment. The percentages of pollen-sterile individuals among the T 0 plants ranged from 30.0% to 85.3% (Fig. 1A). TMS5b showed the highest editing, homozygous mutation, and pollen sterile frequencies (Fig. 1A). The above results indicated that the binary construct for targeting the TMS5a and TMS5b sites (named the TMS5ab construct) had the highest targeted editing efficiency.
To obtain TGMS plants, T 0 sterile plants were treated at a low temperature (~22 °C, daily average temperature [DAT]). Plants with restored fertility were obtained from tranformants of each construct. Furthermore, we obtained "transgene clean" TGMS lines in T 1 plants with each construct (Figs 2 and 3A and Supplementary Figure 2A). Taken together, we obtained "transgene clean" TGMS lines through the CRISPR/ Cas9-mediated TMS5 site-directed mutagenesis system.
Previous reports have shown that CRISPR/Cas9 technology has a relatively high potential for off-target activity, which affects its application. The specificity of the seed sequence, including the 10-12 nucleotides close to the protospacer adjacent motif (PAM), has a significant impact on the off-target activity 42 . The TMS5ab construct has the highest efficiency among the constructs used for CRISPR/Cas9-mediated TMS5 editing; therefore, we analysed the off-target efficiency of TMS5ab. We predicted five and nine potential off-target sites for TMS5a and TMS5b, respectively. After sequencing those potential off-target sites, no genome editing was detected in any examined samples (Table 1). These results indicated that the TMS5ab construct had the highest editing efficiency and an undetectable off-target effect. The target sequences with high editing and low off-target efficiencies will improve the breeding efficiency and reduce the breeding cost associated with the development of commercially Scientific RepoRts | 6:37395 | DOI: 10.1038/srep37395 applied TGMS lines. Therefore, we exploited the TMS5ab construct for the transformation of other rice varieties for commercially applied TGMS lines.
Editing characteristic of TMS5 and factors influencing editing efficiency. Next, we analysed the mutation types and frequencies of the nine effective target sequences (Fig. 1B). Among all types of induced mutations, single-nucleotide insertions were most frequently detected, with a frequency up to 51.96%, consistent with previous reports 25,26 . The second highest frequency of mutation was nucleotide deletion, in which single-nucleotide deletions showed the highest mutation rate of 10.29%. Together with the increased number of deleted nucleotides, the mutation frequency of the target site became gradually reduced. The longest deletion fragment was 253 bp in length. Nucleotide substitutions were also observed, accounting for only 2.94% of all mutations. The TMS5ab construct gave rise to a higher percentage of plants with pollen sterility compared to the other constructs. The number preceding ", " is the total number or frequency of mutations, whereas the number following ", " is the number or ratio of homozygote plants. (B) CRISPR/Cas9-induced mutation types and frequencies. In all types of induced mutations, single-nucleotide insertions were most frequently detected. The largest deletion was 253 base pairs long. (C) TMS5ab construct-induced mutation frequencies in 11 cultivars of two different rice subspecies. Excluding YNSM, the mutation frequencies of TMS5a in the other ten cultivars were all higher than 72.72%.
Next, we analysed the factors that affect editing efficiency. First, we analysed the GC content of the target sequences because targeted genome editing relies on the binding of the single-guide RNA (sgRNA) containing the target sequence to the target sites. By comparing the GC contents in the target sequences and the corresponding editing efficiency, we found that a high GC content is correlated with a high target sequence editing efficiency ( Fig. 1A and Supplementary Figure 3). We then proceeded to test the sgRNA and Cas9 expression levels from the leaves of T 0 plants carrying different target sequences (Fig. 4). The Cas9 expression levels were lowest in TMS5abS-4 plants, which displayed modified TMS5b target sites and no modified TMS5a sites. In contrast, the Cas9 expression levels in TMS5efS-2, TMS5ghS-2 and TMS5ijS-1 plants were 14.1, 15.3 and 58.1 times larger than that in TMS5abS-4, respectively. However, no genome editing occurred in any of the target sites in these three lines. A similar phenomenon was also observed for the relationship between the target and the sgRNA and the editing efficiency. This finding suggested that once the sgRNA and Cas9 expression levels reach a certain threshold value, there is no obvious correlation between their expression levels and the editing efficiencies. Furthermore, because PAM is essential for identification of the target site by the target sequence-sgRNA, we analysed the effects of different PAMs on the editing efficiencies of the target sites ( Fig. 1A and Supplementary Table 1). The average editing efficiencies of the target site PAMs with AGG (TMS5c, d and f), TGG (TMS5b, g, h and j), and CGG (TMS5a and i) were 80.6%, 68.8%, and 70%, respectively. Although AGG displayed a higher average editing efficiency compared with TGG and CGG, TMS5b, which contained TGG as the PAM, displayed the highest editing efficiency among all nine target sequences. These findings indicated that the type of target sequence PAMs had no significant effect on the editing efficiency. We also found that the TMS5e target sequence was highly complementary to the sgRNA, forming a secondary structure that was likely to inhibit binding of the target sequence to the genome (Supplementary Figure 4). This was the plausible cause of the genome editing failure and is supported by previous studies 25,26 .

Development of commercial TGMS lines in rice. The TMS5ab construct established in our study
showed promising utilization potential. To develop the applied TGMS lines, we selected 11 fertile elite cultivars of two different rice subspecies (including seven maintainer lines of the three-line system, three indica and one japonica conventional variety) as the hosts for the targeted genome editing using the TMS5ab construct over the course of only one year (Fig. 1C). We obtained various numbers of T 0 transgenic plants in all 11 cultivars. Excluding the identification of only one YNSMS (Yuenongsimiao (YNSM) transformed with TMS5ab) plant carrying the edited TMS5a, the editing efficiencies among the other cultivars ranged from 72.72% to 100%. While no homozygous TMS5a mutation was found in YNSMS, the homozygous mutation rates of the other cultivars ranged from 11.1% to 54.55%. These results indicated that we had produced TMS5-targeted mutation plants in all 11 cultivars and that the editing efficiency of the CRISPR/Cas9 system may vary depending on the genetic backgrounds of the different cultivars. After LT treatment of the T 0 sterile individuals, plants with restored fertility were selected.
To obtain TGMS plants without exogenous T-DNA, plants with restored fertility from 11 cultivars were planted under high temperature (HT) conditions. Additionally, sterile plants without hygromycin phosphotransferase (HPT) and Cas9 were obtained by PCR and Southern blot detection (Supplementary Figure 2B,C  and Supplementary Table 2). These "transgene clean" sterile plants were grown under LT conditions, and T 2 seeds were obtained. The "transgene clean" TGMS lines segregated in the T 1 generation. While exhibiting similar   Figure 5B). The phenotypes of these TGMS lines as a consequence of genome editing were stable in the T 2 and T 3 generations. To obtain TGMS lines that were applicable to two-line hybrid breeding, we next treated these TGMS lines with different temperature gradients and found that fertility was gradually reduced with increasing temperature (Fig. 5). TGMS lines of ZZBS and YJSMS showed relatively low critical sterility-inducing temperature (CSIT), i.e., below 24 °C (DAT), while in the ReBS, TFBS and WSSMS TGMS lines, the CSIT were all about 24 °C (DAT). The CSIT of the GAZS TGMS lines was approximately 26 °C (DAT), and the CSIT of ZS97BS exceeded 26 °C (DAT). ZZBS, ReBS, TFBS, WSSMS and YJSMS exhibited a pollenless sterile type, whereas ZS97BS and GAZS displayed completely abortive pollen at 28 °C (DAT). These results demonstrated that CSIT was regulated by the rice genetic background. The populations of "transgene clean" YJSMS were grown in the experimental plots of the China National Hybrid Rice R&D Center (Changsha, China) in the summer of 2015 and they displayed pollenless male sterility (Supplementary Figure 5). To confirm that TMS5 editing would not lead to female organ abortion, GAZS and YJSMS were pollinated with normal pollen. The sterile plants of GAZS and YJSMS yielded seed settings via artificial pollination (Supplementary Figure 6), indicating normal female organ development and fertilization functions. These results indicated that these lines could be used as male sterile lines, which are almost indistinguishable from traditional male sterile lines. Taken together, the findings suggest that practical TGMS lines with good agronomic characteristics can be developed by targeted editing of TMS5 using the CRISPR/Cas9 system.
To test the combinatorial capacity of the obtained TGMS lines, we crossed HNBS with Huahang1179 (HH1179), YJSMS with R94, YJSMS with R340, YJSMS with R297, YJSMS with R173, WSSMS with R364, WSSMS with R173, WSSMS with R340, and WSSMS with R297. Subsequently, we obtained the F 1 generation of HNBS and HH1179 and found that the offspring were stronger and provided a greater yield than those of the control variety (Supplementary Figure 7). In addition, the hybrid demonstrated a 31.4% increase in the main panicle length, a 47.9% increase in grain number per main spike, a 19.8% increase in thousand seed weight, a 14.4% increase in single plant yield, and a 15.9% increase in plot yield (Supplementary Table 3). Taken together, these results indicated that TGMS lines with TMS5 targeted editing using the CRISPR/Cas9 system could be applied to improve production in hybrid rice breeding.

Discussion
In recent years, RNA interference (RNAi) and antisense approaches have become more commonly used in reverse genetics systems 40 . In our previous work, we knocked down TMS5 expression using RNAi. Although homozygous  15 . Therefore, sterile heterozygous TGMS plants cannot be used for hybrid rice breeding. We then knocked down TMS5 expression using an antisense RNA technique and obtained very low rates of TGMS plants with a relatively high CSIT (Supplementary Figure 8), potentially due to incomplete suppression of TMS5. This result suggested that both RNAi and antisense RNA technology provide sequence-dependent post-transcriptional inhibition of gene expression, resulting in incomplete depletion of target genes 40 . Nevertheless, TGMS lines with low CSIT guarantee seed production in two-line hybrid rice breeding 43 . Importantly, using a modified sgRNA to cleave a target gene exon via DSBs and nonhomologous end-joining (NHEJ) repair in the CRISPR/Cas9 system, knockout genotypes can be deliberately generated in plants. CRISPR/Cas9-mediated site-directed mutagenesis produced a loss of function of TMS5 and generated TGMS rice with low and stable CSIT (Fig. 5). Moreover, regardless of the presence of the CRISPR/Cas9 transgene, the target gene mutations could be stably transmitted to future generations 41 . Therefore, "transgene clean" mutants could be obtained in the T 1 generation (Figs 2, 3 and 5 and Supplementary Figure 2). In summary, the mutations generated using this strategy are essentially similar to the spontaneous or induced mutations that occur in conventional breeding, in addition to the advantage that the objective trait of CRISPR/Cas9 editing in plants can be deliberately improved 39 .
Currently, hybrid breeding is an important approach for the improvement of rice yield in China 6 . Several years or decades were required to obtain and select sterile lines under specific temperature and light conditions in the two-line breeding system [21][22][23][24] . In contrast, only one year was required to produce "transgene clean" tms5 TGMS plants in the T 1 generation using the CRISPR/Cas9-mediated site-directed mutagenesis system. Commercial sterile lines are mainly produced using the available sterile lines as the sterile gene donor (female parent) in traditional cross and backcross breeding programs 21 . However, there are limited genetic resources for the intensive use  of a single source of male-sterile cytoplasm, and variety collapse of a source of sterile cytoplasm is considered to be disastrous in developing hybrids 44 . Therefore, by developing TGMS lines in different genetic backgrounds, the genetic diversity-especially the cytoplasm diversity-can be increased, improving the efficiency of the exploitation of heterosis. The methodology applied in our work to produce TGMS lines will mitigate limitations related to traditional hybrid breeding.
The production of hybrid rice using the three-line system involves CMS, maintainer and restorer lines. Commercial CMS lines can be mainly categorized into three groups: wild-abortive (WA), Honglian (HL) and Baotai (BT) 45 . The utilization of different types of three-line CMS lines is restricted because special CMS-restorer genes are required from the male parents, and the germplasms of maintainers and restorers are not abundant. For example, approximately 5% of the rice germplasms can serve as restorer lines in China, Southeast Asia and America 10,46,47 . The CMS and maintainer lines possess the same nucleus but different cytoplasms 8 . In the present study, we used CRISPR/Cas9-mediated editing of nuclear genes encoding TMS5 in maintainer lines to obtain TGMS lines with extremely similar characteristics to their corresponding CMS lines. Almost all of the normal varieties with or without specific CMS-restorer genes could restore the fertility of TGMS lines 12,14 . This method retains all of the merits of CMS lines, avoids the problems caused by limited genetic resources and improves the heterosis utilization efficiency. Therefore, we termed this direct and efficient TGMS breeding method as "turning three-line into two-line".
The breeding of japonica hybrid rice has not been widely applied due to a lack of male-sterile lines with high agronomic value. The japonica hybrid rice occupies approximately only 3% of the japonica rice planting area in China 10,48 . The promotion of japonica hybrid rice breeding will be a breakthrough in rice production in China. The sterility gene in japonica P/TGMS lines has been derived mainly from Nongken58S by employing a single genetic background, and it possesses the characteristics of high CSIT 43 . Nevertheless, it will be very convenient to use CRISPR/Cas9-mediated TMS5 editing to cultivate sterile lines of japonica or wide-compatibility sterile line breeding to improve japonica or japonica-indica hybrid rice breeding 48 .
In two-line hybrid breeding, the cultivation of sterile lines with low CSIT is the key for ensuring the purity of hybrid seeds 43 . We tested CSITs of 7 TGMS lines generated using the TMS5ab editing system (Fig. 5) and found that varieties with different genetic backgrounds possessed different CSITs and that most of them met the requirements of breeding programs. These results suggested that the CSIT of tms5 TGMS lines was determined according to their genetic backgrounds but not tms5. Furthermore, TGMS lines with higher CSITs could be crossed with lower CSIT lines to select new TGMS lines with lower CSITs.
In the present study, we selected 10 target sequences in the TMS5 coding region to examine the efficiency of inducing target site editing and found that the efficiency of TMS5b was highest (up to 88.2%). Among the five constructs, the editing efficiency of TMS5ab was as high as 94.1%, and the homozygous rate of one of the target sequences was as high as 32.4%. These findings showed that the production of "transgene clean" TGMS plants was very convenient using the CRISPR/Cas9 system in the T 1 generation and that the breeding efficiency could be markedly improved. Furthermore, the editing efficiency of the CRISPR/Cas9 system was affected by several factors. As previously reported 25,26 , targets with higher GC contents demonstrated relatively higher editing efficiencies ( Fig. 1A and Supplementary Figure 3). In addition, as previously described 26,49 , a more stable secondary structure resulted in imperfect complementary base pairing of the target sequence and the target site ( Fig. 1A and Supplementary Figure 4). As a result, the target sequence with a less stable secondary structure with sgRNAs tends to be selected in this system 49 . We also noted that the editing efficiency of the same target sequence was altered in different rice varieties (Fig. 1C). This observation may be due to the reduced activity of the CRISPR/ Cas9 editing system in some varieties or to the higher percentage of faultless repairs after DSB 25 . The sgRNA expression level might be the limiting factor for target site editing in Arabidopsis 26 . However, when the sgRNA and Cas9 expression levels reached certain thresholds, their expression may not be relevant to the target site editing efficiency in rice.
In the present study, we designed 10 target sequences in the TMS5 gene and selected the most optimal TMS5ab construct for the development of commercial TGMS lines by comparing the editing efficiency and analysing the off-target efficiency of the target sequences. Using the TMS5ab construct, the TMS5 gene was edited in 11 elite ice lines to develop the "transgene clean" TGMS lines within only one year. This method can shorten the breeding period, reduce the working complexity and cost, and facilitate the exploitation of heterosis. Our work will greatly promote the breeding process of TGMS lines and will maximize the number of cross combinations of three-line sterile lines, which will facilitate the development of hybrid rice breeding. This work can be applied for not only hybrid rice breeding but also potentially breeding of other hybrid crops.

Methods
Plant materials, growth conditions and generation of transgenic plants. TMS5as  Guangzhou (23_N, 113_E), during normal rice growing seasons, or in a growth chamber at DATs of 22,24,26,28 and 30 °C with three replicates as previously described 15 . Vector constructs. The TMS5as vector was constructed as previously described 50 . An antisense fragment from TMS5 was amplified and cloned into pYLox vector 15 . CRISPR/Cas9 was constructed as previously described 26 . Briefly, according to the design principles of the target sequences in the CRISPR/Cas9 system, 19 to 20 bases upstream of the PAM motif were selected as candidate target sequences, and target sequences in TMS5 were determined for TMS5a, TMS5b, TMS5c, TMS5d, TMS5e, TMS5f, TMS5g, TMS5h, TMS5i and TMS5j (Supplemental Table 4) by Blast analysis of the rice genome (http://blast.ncbi.nlm.nih.gov/Blast.cgi) to ensure that no other genes were targeted. Equivalent amounts of forward and reverse primers (0.05-0.1 μ M) for one target were mixed and incubated at 90 °C for 1 min, followed by a gradual cool-down to room temperature to form the target adaptor. Approximately 1 μ g each of the pYLgRNA-OsU6a and pYLgRNA-OsU3 vectors was digested using 10 U Bsa I (NEB), respectively. The ligation reaction (10 μ l) contained 1 μ 1 of 10 x T4 DNA ligase buffer, 20 ng of pYLgRNA-OsU6a or pYLgRNA-OsU3 vector, 1 μ 1 of the target sequence adaptor, and 35 U of T4 DNA ligase (Takara, Dalian, China). The target sequence adaptors TMS5a, TMS5c, TMS5e, TMS5g and TMS5i, respectively, were ligated into the linearized pYLgRNA-OsU6a vector (digested with 10 U of Bsa I), and TMS5b, TMS5d, TMS5f, TMS5h, and TMS5j, respectively, were cloned into pYLgRNA-OsU3 to generate sgRNA expression cassettes with target sequences. The sgRNA expression cassettes with the target sequences TMS5a, TMS5c, TMS5e, TMS5g and TMS5i were amplified from pYLgRNA-OsU6a using primers Pps-GGL and Pgs-GG2 and the sgRNA expression cassettes with target sequences TMS5b, TMS5d, TMS5f, TMS5h, and TMS5j were amplified from pYLgRNA-OsU3 using primers Pps-GG2 and Pgs-GGR with High- Fidelity

Mutation detection and analysis of transgenic plant lines.
To determine the mutation of target sites, genomic DNA from the leaves of transgenic rice plants was extracted using the sodium dodecyl sulphate method 51 . PCR amplifications were performed using primer pairs (Supplemental Table 4) surrounding the designed target sites. The PCR products were directly sequenced or cloned into the pEASY-Blunt (TransGen Biotech, Beijing, China) vector and sequenced using the Sanger method. Mutations were identified by comparing the sequences of transgenic and WT plants. The sequencing chromatograms from mutations were analysed, and mutations containing normal sequencing chromatograms were considered homozygotes. Mutations containing superimposed sequencing chromatograms were considered heterozygous or biallelic mutations. Heterozygous sequences from direct sequencing were decoded using the degenerate sequence decoding method 52 . The relevant PCR primers are listed in Supplementary Table 4.
The identification of "transgene clean" plants was conducted using T 1 progeny of T 0 homozygous mutations. The transgenic lines were analysed by PCR using HPT and Cas9 primers and agarose gel electrophoresis. The pYLCRISPR/Cas9Pubi-H plasmids and ZH11 DNA were selected as positive and negative controls, respectively. HPT-and Cas9-negative plants were considered "transgene clean" plants. Some "transgene clean" plants identified by PCR were further confirmed by Southern blotting. Southern blots were performed using a DIG-High Prime DNA Labeling and Detection Starter Kit I (Roche Diagnostics, Mannheim, Germany) according to a standard program supplied by the manufacturer. In brief, PCR products of HPT were labelled with GIG-dUTP by random priming (Roche Diagnostics) and used as hybridization probes. Genomic DNA from the leaves of rice plants was extracted using the Cetyltrimethylammonium bromide (CTAB) method. Genomic DNA was digested overnight using EcoRI (Takara, Dalian, China) and then separated by 0.8% agarose electrophoresis and transferred onto Hybond N + nylon membranes (Amersham Bioscience, Bucks, UK) using the capillary transfer method. Hybridization was performed at 65 °C in DIG Easy Hyb buffer. The membranes were washed twice with wash buffer (2 x SSC, and 0.1% SDS) at 68 °C and then blocked with blocking solution and incubated with anti-digoxigenin-AP solution for 30 min. Subsequently, the membranes were washed twice with wash buffer (0.1 M maleic acid, 0.15 M NaCl, 0.3% (v/v) Tween 20, pH 7.5) at room temperature for 15 min each, followed by equilibration with detection buffer (100 mM Tris-Cl, 100 mM NaCl, pH 9.5) for 2-5 min. One millilitre of CSPD ready-to-use solution was added to the membranes to stimulate chemiluminescence. The chemiluminescence signals were detected and analysed using X-ray films.
Expression analyses. Total RNA was isolated from the leaves or panicles of T 0 rice plants using TRIzol (Invitrogen, Carlsbad, CA, USA). DNase I-treated total RNA (2.0 μ g) was used for reverse transcription using an M-MLV-RT kit (Takara, Dalian, China) with a mixture of oligo(dT) and sgRNA RT primers (Supplemental Table 4). Real-time quantitative PCR was performed with gene-and target-specific primers (Supplemental Table 3) using the SsoFast ™ EvaGreen Supermix kit (Bio-Rad, USA) and CFX96 Real-Time PCR Detection System (Bio-Rad) as previously described 15 . Three replicates were evaluated for each gene. The internal standard gene OsActin1 was used to normalize the cDNA levels of the target genes. Total protein was extracted from rice panicles, and immunoblot analysis was performed using a previously described method 15  RNA secondary structure analysis. Secondary structure analysis of target-sgRNA sequences was performed with the RNA Folding Form (http://unafold.rna.albany.edu/?q= mfold/RNA-Folding-Form).
Characterization of phenotypes. Rice florets were photographed using an OLYMPUS DP70 digital camera under an OLYMPUS SZX10 dissecting microscope. Mature pollen grains were stained using 1% I 2 -KI solution and photographed using an OLYMPUS BX51 or Leica DNRXA microscope.