The tapetal tissue is essential for the maintenance of redox homeostasis during microgametogenesis in tomato

SUMMARY The tapetum is a specialized layer of cells within the anther, adjacent to the sporogenous tissue. During its short life, it provides nutrients, molecules and materials to the pollen mother cells and microsporocytes, being essential during callose degradation and pollen wall formation. The interaction between the tapetum and sporogenous cells in Solanum lycopersicum (tomato) plants, despite its importance for breeding purposes, is poorly understood. To investigate this process, gene editing was used to generate loss‐of‐function mutants that showed the complete and specific absence of tapetal cells. These plants were obtained targeting the previously uncharacterized Solyc03g097530 (SlTPD1) gene, essential for tapetum specification in tomato plants. In the absence of tapetum, sporogenous cells developed and callose deposition was observed. However, sporocytes failed to undergo the process of meiosis and finally degenerated, leading to male sterility. Transcriptomic analysis conducted in mutant anthers lacking tapetum revealed the downregulation of a set of genes related to redox homeostasis. Indeed, mutant anthers showed a reduction in the accumulation of reactive oxygen species (ROS) at early stages and altered activity of ROS‐scavenging enzymes. The results obtained highlight the importance of the tapetal tissue in maintaining redox homeostasis during male gametogenesis in tomato plants.


INTRODUCTION
Sexual reproduction in both animals and plants requires the formation of haploid gametes in a complex and highly regulated process. Quite unlike animals, in flowering plants the gametes are produced post-embryonically within specialized organs: the ovary and the anther. The female gametophyte (embryo sac) is produced from a germline originated in the ovules inside the ovary, whereas the male gametophytes (pollen) originate inside the anther. The formation of gametes in plants occurs late in development and does not depend on meristems but on cell-tocell communication or tissue interactions.
The anther has a relatively simple morphological structure, and the high accessibility of the organ has resulted in the anther being the object of numerous studies on the sexual reproduction of plants. Shortly after anther primordia initiation, several somatic and germinal cells originate. Typically, the primordia contain three cell layers (L1-L3) that will develop into the external epidermis, the archesporial cells, and the inner vascular and connective tissue ( Astrand et al., 2021;G omez et al., 2015). Archesporial cells further differentiate into three additional layers of somatic tissue: the endothecium, the middle layer and the tapetum, and a layer of microsporocytes (pollen mother cells, PMCs).
In Solanum lycopersicum (tomato), male sterility is a desirable trait to be used in hybrid seed production and cross-breeding programs. Over 50 male sterile mutants were isolated more than two decades ago (Gorman & McCormick, 1997), and yet only a limited number of genes involved in male gametogenesis have been identified. Mutations in the tomato SPOROCYTELESS/NOOZLE ortholog prevent the formation of both male and female sporocytes and the plants are fully sterile (Hao et al., 2017;Rojas-Gracia et al., 2017). Downstream of this gene, the Ms10 35 gene (DYT1 homolog) encodes a basic helix-loop-helix (bHLH) transcription factor specifically expressed in tapetal cells and meiocytes (Jeong et al., 2014). Another bHLH protein (Solyc01g081100) has been proposed as the best candidate to encode the tomato Ms32 gene (Liu et al., 2019). The Soly-c01g081100 gene is a homolog of the Arabidopsis bHLH10/ 89/90 gene that, together with DYT1-MYB35, forms a regulatory module during tapetum and pollen development (Cui et al., 2016). Recently, an ABORTED MICROSPORES (AMS) tomato homolog was identified that is essential for pollen development in this species (Bao et al., 2022).
Overlapping with the genetic network orchestrating anther development, additional factors and signaling molecules participate in the communication between the somatic and sporogenous tissues. These factors include hormones, secreted proteins, microRNAs (miRNAs) and cellular redox state (Dukowic-Schulze & van der Linde, 2021). Interestingly, although a high concentration of reactive oxygen species (ROS) causes irreversible DNA damage and cell death, at low levels ROS act as signaling molecules determining cell division and cell fate (Kelliher & Walbot, 2012;Yang et al., 2018). Several lines of evidence support the hypothesis that cellular redox state is an important morphogenetic factor controlling cell differentiation and proliferation during anther development in rice (Yu & Zhang, 2019). In tomato anthers, the modulation of ROS levels achieves the proper timing of programmed cell death (PCD) in tapetal cells (Yu et al., 2017), an essential process during pollen formation.
The tapetum layer, adjacent to the developing microsporocytes, has a central role during pollen development, and its premature or delayed degradation results in pollen abortion and male sterility (Bai et al., 2019;Liu et al., 2018). In this work, we studied the molecular and physiological changes associated with the absence of tapetum and its effect on pollen formation in tomato plants. In the absence of information concerning genes conferring tapetal cell identity, we looked for candidate genes based on gene homology. We identified the Solyc03g097530 gene as the closest homolog of the protein peptides TPD1, TDL1A and MAC1, and then mutant plants were produced using CRISPR/Cas9 technology. Mutant plants devoid of tapetum were used to gain insight into the genetic network and molecular signals involved in pollen formation in tomato plants. Our results provide evidence for a role of the tapetum in the modulation of redox homeostasis during early stages of pollen development.

RESULTS
Identification of the tomato Solyc03g097530 gene as a candidate gene to specify tapetum identity Previous studies in Arabidopsis, maize and rice established that tapetum identity in these species is determined by the combined function of an LRR receptor kinase and a small protein ligand encoded by the gene homologs TPD1, TDL1A and MAC1 Yang et al., 2003;Zhao et al., 2008). Following a gene homolog strategy, we looked for a tomato gene candidate to be involved in the specification of tapetum identity. The TPD1, the second option sequence was used as a bait in the Plant Comparative platform Phytozome (https://phytozome-next.jgi.doe. gov) (Goodstein et al., 2012) against the tomato genome, and two homologous sequences were identified: Soly-c03g097530 and Solyc11g012650. The expression of these genes was analyzed in vegetative tissues (leaves) and flower buds using quantitative real-time polymerase chain reaction (qRT-PCR). The results showed that Soly-c11g012650 was preferentially expressed in leaves, whereas Solyc03g097530 was expressed in developing flowers, reaching the highest level in flowers at anthesis ( Figure S1). Phylogenetic analyses were performed using a list of homolog genes from different plant species obtained in a BLAST search using the TPD1 gene (At4g24972) as a bait. These sequences also included the Arabidopsis closest homolog At1g32583 and the rice ortholog OsTDL1A (Zhao et al., 2008). In the phylogenetic tree, Soly-c03g097530 grouped with TPD1 and related TPD1-like homologs from Solanaceae ( Figure S1).
The protein sequence of Solyc03g097530 (176aa) was aligned with Arabidopsis TPD1 and two protein homologs that have been functionally characterized: TDL1A from rice (Zhao et al., 2008) and MAC1 from maize . These proteins showed high amino acid identity, mainly in the C-terminal region, with six highly conserved cysteine residues and a putative dibasic cleavage site (Figure 1a). In addition, the Solyc03g097530 protein and its homologs contain a predicted signal peptide at their N-terminal regions (Figure 1a, underlined). The subcellular location of the protein was determined by fusing the yellow fluorescent protein (YFP) to the C-terminal end of Solyc03g097530, transiently expressed in Nicotiana benthamiana leaves. The control protein (35S:GFP) exhibited both cytoplasmic and nuclear localization ( Figure S2), whereas the Solyc03g097530-YFP protein was localized in the proximity of the plasma membrane, where it formed small dots, and in the cytosol, as large aggregates (Figure S2).
The expression of Solyc03g097530 was analyzed in different plant tissues, including seedlings (apical and basal regions), leaves and developing flowers, using qRT-PCR. The gene was expressed in all the tissues analyzed, reaching the highest level in flowers at anthesis (Figure 1b). The spatial and temporal expression pattern of Soly-c03g097530 was evaluated during flower development using in situ hybridization (Figure 1c-h). Solyc03g097530 RNA was not detectable in inflorescence meristems and flower buds before differentiation of the anther primordium ( Figure 1c). Expression was first detected at floral stage 6 at the internal layers of the developing anther that will generate the sporogenous tissue (Figure 1d,e). Later, at the tetrad stage, Solyc03g097530 transcript was localized at the tapetum and the microspores, which were still surrounded by the callose wall ( Figure 1f). The expression of the gene continues during the following floral stages in the tapetal cells that gradually disintegrated and in the pollen grains (Figure 1g,h). On the ovary, we detected transient expression in the ovule primordia of flowers at stage 8 ( Figure S3).
To evaluate the conservation of Solyc03g097530 function during the development of the anther, we designed an experiment to complement a loss-of-function tpd1 mutant. We used a mutant line (N843482, tpd1 mutant) that contains a T-DNA insertion in the TPD1 gene. Mutant plants were indistinguishable from the wild type except that the anthers did not produce pollen grains ( Figure S4). To complement the mutant phenotype, we generated a genetic construct by fusing 2.7 kb of the promoter region of Arabidopsis TPD1 and the coding sequence of the Solyc03g097530 gene that was used to genetically transform heterozygous tpd1 plants. We obtained 32 independent transformants and four of them were homozygous for the mutation. These four plants produced viable pollen ( Figure S4) and seeds, and hence recovered fertility, demonstrating the ability of the Solyc03g097530 protein to replace TPD1 function.
Altogether, these results suggest that Solyc03g097530 (from now on SlTPD1) is the ortholog of TPD1, TDL1A and MAC1 genes, and therefore is a good target to generate tomato plants devoid of tapetum.

Sltpd1 mutants specifically lack tapetal cells and are male sterile
To demonstrate the function of SlTPD1 in the control of tapetum specification, tomato lines with mutations targeted to the third exon of the gene were generated using CRISPR/ Cas9 ( Figure S5). Among the T 0 generation, we selected six diploid plants that showed percentages of edition over 80% and that mostly contained biallelic mutations ( Figure S5). All the plants showed complete male sterility and developed seedless (parthenocarpic) fruits. Histological sections of the mature anthers revealed collapsed locules containing dense debris, but they did not contain viable pollen (Figure S5). In these plants, we observed a strong correlation between male sterility and the development of parthenocarpic fruits ( Figure S5). F 2 plants were obtained after pollination with wild-type pollen, and stable single-mutation lines were obtained. Two mutant lines (Sltpd1 Del5 and Slt-pd1 Del2 ) containing deletions of five and two nucleotides, respectively, were chosen for further analysis. During the design of the gRNA guide, the uncharacterized Soly-c09g089880 gene was identified as a potential off-target, although the designed guide showed four mismatches. To discard the edition of this gene in the selected lines, we amplified by PCR a fragment of the Solyc9g089880 gene using genomic DNA from these plants (Sltpd1 del2 and Sltpd1 del5 mutants), flanking the putative target site. We found no changes in the sequence of the putative off-target gene.
To establish the role of the SlTPD1 gene during male gametogenesis, we compared the development of anthers from the wild type and the Sltpd1 mutants. In wild-type tomato anthers, cells from the L2 layer differentiate into archesporial cells that undergo periclinal divisions (parallel with the epidermis) ( Figure 2a). In the Sltpd1 mutant, anther development was slightly different to that in the wild type, showing cells with squared rather than rectangular shape and a reduced number of periclinal divisions (Figure 2d). From stage 8, we observed clear differences between the two genotypes. Whereas epidermis, endothecium and middle cell layers were formed, tapetum was not present in the mutant, and sporogenous cells seemed more abundant and disorganized compared with the wild type (Figure 2e). At stage 10, wild-type microsporocytes completed meiosis and formed tetrads surrounded by callose and separated from the adjacent cell layers ( Figure 2c). Eventually, callose was degraded, releasing the microspores that continued to develop into mature pollen grains during floral stages 12-16 (Figure 2g-i). Simultaneously, the tapetum started to degrade and was not visible by stage 16 (Figure 2i). In the mutant anthers, microsporogenous cells continued to divide and enlarge in size ( Figure 2f). After extra rounds of divisions, the cells occupied the complete cavity of the locule ( Figure 2j). Cell counting showed that by floral stage 8, the number of sporogenous cells in Sltpd1 anther locules was roughly double that of the wild-type anthers (24.0 AE 5.8 versus 49.0 AE 4.7 cells per locule section; P < 0.001, Student's t-test). At stage 10, sporogenous cell number further increased (66.1 AE 11.3 cells per locule section; P < 0.001, Student's t-test), and cells seemed to have initiated meiosis but failed to complete it ( Figure 2j). Finally, cells degenerated causing the collapse of the anther locules and the deposition of dense cell debris ( Figure 2l).
The presence of tapetal tissue and meiocytes was tested by performing in situ hybridization assays using probes for the tapetum-specific TomA5B (Solyc01g086830) gene (Aguirre & Smith, 1993) and the tomato homolog (Solyc04g008070) of the meiosis marker SOLO DANCER (SDS) gene (Azumi et al., 2002). In the wild-type anther, the TomA5B probe is first visible at stage 7 and is strongly hybridized with the tapetal cells at floral stages 8 and 10 ( Figure 3a). The signal decreased dramatically by stage 12, when tapetum degeneration starts ( Figure 3a). In the Slt-pd1 mutant anthers no signal was obtained in any of the floral stages analyzed (Figure 3a Callose deposition occurs around the sporogenous cells prior to meiosis initiation and later, between meiotic products (Jaffri & MacAlister, 2021). After the completion of meiosis, callose is quickly degraded after the release of callases (b-1,3-glucanases) by the tapetum. Using aniline blue staining, the pattern of callose deposition and degradation was analyzed in the mutant plants. In wild-type anthers, the deposition of callose appears as an intense fluorescence signal around the tetrads that quickly disappears at the termination of meiosis ( Figure 3c). In the mutant plants, the accumulation of callose was observed as a diffuse signal surrounding the sporogenous cells, and the fluorescence signal persisted in time until the collapse of the anther locule ( Figure 3c).

Identification of global transcriptional changes associated with the absence of tapetum
To identify the transcriptional changes associated with the absence of tapetum, RNAseq analyses were performed using Sltpd1 mutant anthers from floral stage 8 (meiotic stage). Differentially expressed genes (DEGs) were selected  Table S1).
At early floral stage 8, Gene Onthology (GO) analyses revealed enrichment in genes related to pollen and tapetum development (five out 13 categories that correspond to 49 genes; Figure 4b; Table S1). Among these genes, homologs of AMS, bHLH91, DYT1, and MYB35 showed strong downregulation (Table 1). Accordingly, in Arabidopsis, DYT1 is required to activate the expression of bHLH010/bHLH089/ bHLH091 genes, which in turn facilitate DYT1 nuclear localization and promote MYB35 expression (Cui et al., 2016). In addition, we detected the strong downregulation of genes required during late stages of pollen development, including a polygalacturonase homolog of the Arabidopsis QRT3 gene involved in microspore separation (Rhee et al., 2003) and a fatty acid-CoA reductase (Table 1).
When looking at signaling pathways, an important group of redox-related genes was observed grouped under 'cellular response to hydrogen peroxide' and 'defense response'. A specific expression heat-map analysis of redox-related genes revealed the differential expression of 70 genes, 53 of which were downregulated and 17 of which were upregulated (Figure 4c; Table S1). Among the downregulated genes (Table 1), we detected two Respiratory burst oxidase homolog (Rboh) genes (also known as NADPH oxidases), key enzymes that catalyze the formation of ROS in plants and a glutaredoxin (GRX) that shows homology with the MIL1 gene from rice involved in microspore development (Hong, Tang, Zhu, et al., 2012). Moreover, nine peroxidases are downregulated in the mutant anthers, including homologs of the previously characterized PRX9 and PRX40 involved in pollen development in Arabidopsis (Jacobowitz et al., 2019). Peroxidases are multifunctional proteins that catalyze the oxidation of a variety of substrates by H 2 O 2 and act as efficient components of the antioxidative system controlling ROS.
We analyzed the contribution of the genes involved in redox homeostasis to the development of the tomato anther. From the list of DEGs, a subset of key ROS-related genes was selected, and expression levels were checked in anthers from different developmental stages (St6-St20). The expression of two tomato RBOH genes (SlRbohA/Soly-c01g099620 and SlRbohE/Solyc06g075570) was analyzed by qRT-PCR. Moreover, we analyzed the expression of SlRBOH1/SlRbohG, recently identified as a brassinosteroid (BR)-regulated gene involved in tapetal cell degeneration and pollen development (Yan et al., 2020). In Sltpd1 mutant anthers, we detected an important reduction in the expression level of SlRbohA and SlRbohE at early stages of anther development (Figure 5a, b). The expression levels of SlRBOH1/SlRbohG did not significantly change during the floral stages analyzed (Figure 5c). The expression of the GRX-C9-like gene SlGRXC9/ Solyc08g036570 was analyzed, and high levels of expression were detected in the wild type at the earliest stages analyzed (floral stages 6 and 8), whereas in the mutant samples, the expression level was greatly reduced (Figure 5a). The expression of two TGA-like transcription factors (Solyc06g074320/SlTGA9 and Solyc10g078670/ SlTGA10), downregulated in the RNA-seq, were also analyzed. Quantitative RT-PCR experiments indicated that although SlTGA9 showed reduced expression in the mutant at floral stages 8, 14 and 15, SlTGA10 expression was strongly reduced in the mutant anthers from floral stage 8, and this low level persisted until floral stage 16 (Figure 5b, c). In Arabidopsis, ROXY1/ROXY2 glutaredoxins interact with TGA9/TGA10 transcription factors during another development (Murmu et al., 2010). Thus, the interaction between SlGRXC9 and SlTGA9/SlTGA10 was tested in planta through bimolecular fluorescence complementation (BiFC). SlGRXC9 and SlTGA9 interacted in nuclei of N. benthamiana leaves, but no interaction could be observed between SlGRXC9 and SlTGA10 ( Figure 6).
Globally, the expression analyses suggested that the absence of tapetum is associated with the disruption of a genetic network involved in the modulation of ROS levels, especially during early stages of anther development. The presence of ROS was tested in the anthers of wildtype and Sltpd1 mutant plants. We analyzed and quantified the presence of the superoxide anion (O 2 .-) and H 2 O 2 , considered the major ROS forms in plant cells (Huang et al., 2019), using 3,3 0 -diaminobenzidine (DAB) and nitroblue tetrazolium (NBT) staining, respectively. The quantification of NBT staining of the anthers, a proxy for superoxide anion presence, detected the highest levels at floral stages 8 and 10, but no differences were observed between wild-type and mutant anthers (Figure 7a). DAB staining assays showed that in both wild-type and mutant anthers the level of H 2 O 2 is higher at floral stage 8 and then decreases progressively. Interestingly, at early stages (St8 and St10), the level of H 2 O 2 was significantly lower in Slt-pd1 than in the wild type (Figure 7b, c). These results suggest that a critical H 2 O 2 threshold should be reached during the early stages of anther development, concurring with the meiotic stage. In plants, the maintenance of ROS levels also relies on the action of non-enzymatic and enzymatic scavenging mechanisms. This last mechanism includes enzymes such as catalase (CAT), peroxidases (PRXs) and superoxide dismutase (SOD) (Huang et al., 2019). To study the functionality of this enzymatic scavenging mechanism in the flowers of the mutant plants, we measured SOD and PRX activities. Compared with the wild type, SOD activity showed a significant reduction in the mutant plants at floral stages 6 (pre-meiotic), 12 and 20 (anthesis) (Figure 7d). Remarkably, PRX activity was much reduced in Sltpd1 mutant anthers at many of the floral stages analyzed (Figure 7e). The downregulation of PRX activity at floral stage 8 was in agreement with the global downregulation of PRXs shown in the RNA-seq experiment (Table 1).

SlTPD1 is required to specify tapetum identity in tomato
The stamens are the male reproductive organs of angiosperms and the place where the pollen is produced within the flower. The different tissues that compose anthers sequentially develop from the anther primordia, suggesting that cell-to-cell communication is critical to coordinate growth and development (van der Linde & Walbot, 2019). The tapetum is the limiting tissue between the somatic and germinal cells, and it is in a dynamic state during its short life period, facilitating the passage of nutrients and molecules to the sporogenous cells and microspores (Pacini & Cresti, 1978).
In this study, we evince the pivotal role of the tapetal tissue during pollen development throughout the isolation and characterization of the SlTPD1 gene. SlTPD1 shows homology with the TPD1 protein from Arabidopsis and, when transformed into the tpd1 mutant, it was sufficient to complement the fertility defects. In addition, protein sequence alignments also showed strong domain conservation with the monocot proteins TDL1A/MIL2 and MAC1 from rice and maize. Therefore, SlTPD1 is the ortholog of the TPD1, MAC1 and TDL1A/MIL2 genes. On the other hand, although the putative receptor for the SlTPD1 protein has not yet been identified in tomato, our data also suggest the conservation of the receptor/ligand module in tomato plants.
In Arabidopsis, mutant plants in either EMS1 (TPD1 receptor) or TPD1 genes share a phenotype, the lack of tapetum and the production of extra sporocytes (Yang et al., 2003;Zhao et al., 2002). We detected the expression of SlTPD1 by in situ hybridization on the anther wall early in development until late stages, where it appeared associated with the tapetum and microsporocytes. In this aspect SlTPD1 slightly differs from TPD1 that is preferentially expressed in microsporocytes, whereas EMS is predominantly expressed in the tapetum (Yang et al., 2003;Zhao et al., 2002). In maize, MAC1 is expressed early in anther ontogeny, where it suppresses archesporial cell proliferation, suggesting that cell position rather than lineage  . This hypothesis is in agreement with the phenotype of Sltpd1 mutants that showed defects in the shape and pattern of division of the archesporial cells. It has been shown that the ectopic expression of TPD1 activates cell division, possibly by altering the expression of cell-cycle genes (Huang, Wijeratne, et al., 2016). Taken together, we propose a dual role for SlTPD1 in the control of archesporial cell divisions and the determination of tapetal cell identity in tomato plants.
Most TPD1 homologs are expressed in different tissues outside the anther, including leaves, roots, seedlings (Hong, Tang, Shen, et al., 2012;Wang et al., 2012;Yang et al., 2003) and ovules Yang et al., 2005). At present, a possible role of these proteins during vegetative development remains elusive. However, in monocots, TPD1 orthologs have been reported to control megaspore mother cell proliferation during ovule development (Sheridan et al., 1996;Zhao et al., 2008). Using in situ hybridization, the expression of SlTPD1 was detected in anthers and the developing ovules. Sltpd1 mutant plants did not show obvious defects in ovule development, and flowers formed normal seeded fruits when pollinated with wild-type pollen. A peculiar and distinctive phenotype of the tomato Sltpd1 mutants is the formation of seedless fruits (parthenocarpic). Parthenocarpy, the formation of fruits in the absence of pollination and fertilization, is often the consequence of the precocious activation of molecular events normally triggered by these processes (Molesini et al., 2020). Also, it could be achieved by external applications of different hormones or growth regulators (Vivian-Smith & Koltunow, 1999). In tomato plants, several reports suggest a role for developing stamens or male gametophytes in the repression of ovary growth (Hao et al., 2017;Medina et al., 2013;Okabe et al., 2019;Rojas-Gracia et al., 2017). Mutations in SlTPD1 caused complete male sterility and the production of small parthenocarpic fruits. This phenotype could support this repressive effect exerted by male gametogenesis progression. Alternatively, the abnormal progression of male gametogenesis could result in the production of signaling molecules that indirectly activate premature ovary growth. In this regard, antisense plants targeting SlRBOHB/SlWfi1, a tomato gene involved in the generation of ROS, show several developmental defects including parthenocarpic fruit development (Sagi et al., 2004). However, SlRBOHB/SlWfi1 was not differentially expressed in the mutant anthers of Sltpd1 mutants, suggesting that the specific alteration of SlRBOH genes could lead to similar phenotypes during reproductive development.

Tapetal tissue is essential to maintain redox homeostasis during pollen formation
The results presented in this study show that the absence of tapetal tissue in Sltpd1 mutants has a huge impact on the transcription of genes involved in redox homeostasis in the anther at early stages. Moreover, a reduction in ROS levels seems to be associated with the failure of pollen mother cells to progress into meiosis. In agreement with this observation, pioneering work in maize showed that hypoxia triggers meiotic fate acquisition, acting as a positional cue for germ cell production (Kelliher & Walbot, 2012).
Besides the production of ROS as a product of several metabolic processes, specific enzymatic machinery is in charge of maintaining redox homeostasis in plants. ROS production in extracellular spaces partly relies on RBOH genes, also known as NADPH oxidases, which catalyze the generation of superoxide radicals. Enzymatic scavenging mechanisms involve CAT, PRXs and SOD, although PRXs can act as both ROS-generating and ROS-processing components (Mittler, 2017). Cellular changes of ROS levels can act as a signal to drive differentiation and morphogenesis during reproductive development. In Arabidopsis, RBOHE is specifically expressed in the tapetum, and the genetic interference with the temporal ROS pattern resulted in altered tapetal PCD and male sterility (Xie et al., 2014). In addition, PRX9 and PRX40 are extensin PRXs specifically expressed in the tapetum that act as scavenging molecules contributing to tapetal cell wall integrity (Jacobowitz et al., 2019). ROS signaling includes GRXs that act as sensors of redox status, altering signal transduction pathways that result in biological responses (Song et al., 2002). Studies in Arabidopsis, maize and rice have highlighted the importance of GRXs in the formation of the anther and the differentiation of microsporocytes (Hong, Tang, Zhu, et al., 2012;Kelliher & Walbot, 2012;Xing & Zachgo, 2008). In rice, a mutation in the anther-specific GRX MICRO-SPORELESS 1 (MIL1) prevents the completion of meiosis during male gametogenesis. MIL1 encodes a CC-type GRX that specifically interacts with TGA transcription factors (Hong, Tang, Zhu, et al., 2012). The rice glutaredoxin MALE STERILE CONVERTED ANTHER 1 (MSCA1) is required for the creation of a hypoxic niche critical for the specification of anther germinal cells (Kelliher & Walbot, 2012). In Arabidopsis, ROXY1 and ROXY2 also encode CC-type GRXs and are required for the formation of the adaxial anther lobe, possibly with other GRXs or redox regulators. ROXY1 and ROXY2 proteins can interact with the TGA transcription factors TGA9 and TGA10 in Nicotiana tabacum (tobacco) leaves (Xing & Zachgo, 2008). These authors suggest that this interaction results in the modification of TGA9/10 and its activation as a transcriptional factor (Murmu et al., 2010). A similar genetic network to the ones described in rice and Arabidopsis should operate in tomato anthers, where ROS produced in the tapetal cells orchestrate anther wall development and the progression of pollen mother cells into meiosis. Using a tomato mutant lacking the tapetum, we identified several elements of this network that were included in the proposed working model  Figure 8). ROS produced by RBOHs (SlRbohB and SlRbohE) and PRXs results in the accumulation of H 2 O 2 in tapetal tissue. GRXs, including SlGRXC9, could target TGA transcription factors (SlTGA9 and SlTGA10) for activation, thus regulating the expression of a set of genes required for further stages of pollen and anther development. This genetic network is severely affected by the absence of SlTPD1 and the concomitant loss of the tapetum, especially during early stages of anther formation. The identification of this network of genes will facilitate dissecting this regulatory mechanism controlling anther and pollen development in tomato plants.

Plant material and growth conditions
Tomato (Solanum lycopersicum L.) seeds from cultivar Moneymaker were grown in a glasshouse in pots with a 2:1:1 mixture of peat:vermiculite:perlite with a temperature regime of 25-30°C day and 18-20°C night. Natural light was supplemented with lamps to obtain a 16-h light/8-h night photoperiod. Flower samples were collected at different developmental stages according to bud size (Brukhin et al., 2003). In this study, eight floral stages were analyzed that correspond to the following sizes: St6, 0.3 cm; St8, 0.4 cm, meiotic; St10, 0.5 cm, tetrad of microspores; St12, 0.6 cm; St14, 0.7 cm; St16, 0.8 cm, pollen mitosis; St18, 0.9 cm; and St20, 1 cm, anthesis. For the characterization of tomato fruits, the size and weight of at least 30 fruits from the different genotypes were analyzed when fully ripe. Nicotiana benthamiana plants were grown under the same conditions as the tomato plants. Four-week-old plants were used for transient expression experiments.
Arabidopsis thaliana Columbia (Col) plants were used as the wild-type genotype. The tpd1 mutant corresponds to the T-DNA insertion line N843482 (SAIL_1174_B09) obtained from the Nottingham Arabidopsis Stock Centre (NASC, https://arabidopsis.info). The line was genotyped using the T-DNA-specific primer LBb1 and the gene-specific primer pair TPD1-LP1/TPD1-LP2 that amplifies 427 bp from the TPD1 gene (Table S2). Arabidopsis plants were grown in seed pots in a growth chamber with a 2:1:1 mixture of peat:vermiculite:peat at 21°C under long-day conditions (16-h light/8-h dark) and irrigated with Hoagland's solution.

Phylogenetic analysis
The phylogenetic tree was inferred by the neighbor-joining method using Poisson-corrected amino acid distances. A total of 1000 bootstrap pseudo-replicates were used to estimate the reliability of internal nodes. Tree inference was performed using MEGA 6 (Tamura et al., 2013). The data set comprised 18 TPD1-like genes from different plant species, obtained from the GenBank database (Table S1).

Plasmid assembly
Construct for the complementation of the Arabidopsis tpd1-2 mutant A fragment of 2.7 kb from the TPD1 promoter (5 0 region of the gene) was fused to the coding sequence of the tomato SlTPD1 gene. First, both fragments were independently amplified by PCR and cloned into the intermediate vectors pENTRY 5 0 TOPO and pCR8 (Invitrogen, now ThermoFisher Scientific, https://www.thermofisher.com), respectively. Second, a multisite gateway recombination reaction was performed to introduce both sequences into the binary destination vector pK7m24GW,3 (https://gatewayvectors.vib.be/) to obtain the final construct pAtTPD1::SlTPD1.
Design of gRNA and CRISPR/Cas9 construct for SlTPD1 gene editing For the design of optimal gRNAs, the target site was selected using the Breaking-Cas design tool (Oliveros et al., 2016). Among the proposed guides, we chose the one with the highest score and with the lowest number of off-target genes. This tool is freely available online (https://bioinfogp.cnb.csic.es/tools/breakingcas). CRISPR/Cas9 plasmid assembly was performed using the Golden Braid (GB) modular framework and tools (https://gbcloning.upv. es/). First, a single gRNA sequence was obtained by the annealing of complementary primers and then assembled with GB1001 (U626 promoter) and GB0645 (scaffold RNA) parts into the destination vector pDGB3a1. In successive multipartite GB reactions, this first module was assembled with the GB0639 and GB0226 parts (containing hCas9 and nptII transcriptional units, respectively) into the final destination vector. The final construct was then transformed into Agrobacterium tumefaciens strain LBA4404. The primers used are listed in the Table S2.

Plant transformation
Arabidopsis transgenic plants were obtained using the floral-dip method (Clough & Bent, 1998). Briefly, plants were grown under long-day conditions until flower transition occurs and then the main stem was removed to allow the growing of secondary meristems. Agrobacterium inoculation (C58C1 strain carrying the construct of interest) was performed by immersion of the shoots (2-5 cm length) in a suspension containing 5% sucrose and 0.05% Silwet L-77. Transformant plants were selected in the presence of kanamycin and transferred to soil for further analyses.
Tomato transformants were obtained by in vitro cocultivation of the Agrobacterium strain LBA4404 (carrying the binary vector of interest) and cotyledon explants (Ellul et al., 2003). Transformants were selected in the presence of kanamycin and after rooting, transferred to the glasshouse.

Genotyping of CRISPR/Cas9 edited plants
Genomic DNA was extracted from young leaves or unopened flower buds. A 530-bp fragment from the SlTPD1 genomic region flanking the targeted region was amplified using oligos SlTPD1G For and SlTPD1G Rev, purified and sequenced. T 0 plants with percentages of edition over 80% were selected using the online tool TiDE (http://shinyapps.datacurators.nl/tide/) (Brinkman et al., 2014). We then used the online software ICE 2 CRISPR analysis tool (https://ice.synthego.com) to identify the number and type of edition for each plant.
For the genotyping of stable and Cas9-free edited plants, PCRbased molecular markers were designed. We used cleaved amplified polymorphic sequences (CAPS) markers (Konieczny & Ausubel, 1993) that detect polymorphisms that occur in restriction sites. The deletions present in the Sltpd1 del2 and Sltpd1 del5 alleles generated new restriction sites for BseGI and NcoI enzymes, respectively. Using SlTPD1G For and SlTPD1G Rev oligos (Table S2), a 530-bp fragment was obtained from genomic DNA. BseGI generated two fragments of 308 bp and 220 bp in the Sltpd1 del2 allele and NcoI generated two fragments of 299 bp and 226 bp in the Sltpd1 del5 allele. Neither of the enzymes cut the wild-type fragment.

Subcellular localization of SlTPD1
The coding sequence of SlTPD1 was cloned via Gateway LR reaction into the pEarleyGate101 vector (containing the YFP fluorescent tag) (Earley et al., 2006) to generate the expression vector SlTPD1-YFP. The vector was transformed into Agrobacterium tumefaciens strain C58 and used to agroinfiltrate 4-weeks-old N. benthamiana leaves. After 48 h of infiltration, the localization of the fluorescence fusion protein was determined on N. benthamiana leaf disks by confocal scanning microscopy (LSM 780; Zeiss, https://www.zeiss.com/). A 35S:GFP construct was used as the control.

Expression analyses by qRT-PCR
Total RNA was extracted from frozen tissue using the E.Z.N.A. Plant RNA Kit (Omega Bio-tek, https://www.omegabiotek.com). RNA was treated with DNAseI (ThermoFisher Scientific) to remove genomic DNA and quantified in a NanoDrop ND-1000 Spectophotometer (ThermoFisher Scientific). For first-strand cDNA synthesis, 1 lg of DNase-treated RNA was used for reverse transcription using a PrimerScript RT reagent kit (TaKaRa, https://www. takarabio.com) and a mix of oligo poli-dT and random hexamers. The resulting cDNA was used for qRT-PCR with the MasterMix qPCR ROX PyroTaq EvaGreen 59 (CmB, https://www.cultek.com/) and the reaction was run on a QuantStudio 3 (Applied Biosystems, now ThermoFisher Scientific). Relative expression levels were calculated by normalizing to the reference genes ACT (Arabidopsis experiments) or SlActin8 (tomato experiments) and using the DD Ct method. All primers showed amplification efficiencies between 90% and 110%. The primers used are listed in the Table S2.

Histological techniques
For histological studies, tissue was fixed in FAE overnight at 4°C and stored in 70% ethanol. Samples were embedded in acrylic resin (Technovit 7100; Kulzer, https://www.kulzer.com) according to the manufacturer's instructions. For histological analysis, resin sections were stained with 0.05% toluidine blue in 0.1 M, pH 6.8, phosphate buffer (O'Brien et al., 1964) and visualized with a Leica DM 5000B microscope (Leica Microsystems) under bright field.

Aniline blue staining in cryosections
For assays in which fresh tissue was needed, samples were fixed in NEG-50 (Richard Alan Scientific), rapidly frozen in liquid nitrogen, and cut into 16 lm sections using a cryostat (Microm HM 520). Cryosections were stained for 10 min in the darkness with 0.5% aniline blue in 0.07 mM sodium phosphate buffer and visualized in a Leica DM 5000B microscope (Leica Microsystems, https:// www.leica-microsystems.com).

Pollen viability assay
Alexander's staining was carried out as previously described (Peterson et al., 2010), with 2 min of incubation at 50°C on a hot Histochemical localization and quantification of hydrogen peroxide (H 2 O 2 ) and superoxide radical (O 2 Á-) Hydrogen peroxide localization was performed in anthers obtained from flowers in different developmental stages. Immediately after dissection, anthers were submerged in a 1 mg ml À1 DAB-HCl (pH 3.8) solution for 16 h under light conditions (Unger et al., 2005), then cleared in 80% ethanol for 20 min and observed in a binocular microscope (Leica Microsystems). Hydrogen peroxide levels were quantified following a similar method. After staining in DAB-HCl and clearing with ethanol, anthers were pulverized in liquid nitrogen, dissolved in 0.2 M HClO 4 and centrifuged at 12 000 g for 10 min. The absorbance of the supernatant was quantified at 450 nm. H 2 O 2 concentrations were obtained through a standard curve for known hydrogen peroxide concentrations diluted with 0.2 M HClO 4 -DAB (Kotchoni et al., 2006). Superoxide radical was measured as formazan formation over time from tetrazolium blue. Flowers from different developmental stages were weighed, submerged in 50 mM potassium phosphate buffer (pH 7.8) containing 0.1% NBT and 10 mM sodium azide, and left to stain for 2 h before clearing in 70% ethanol. After staining, tissue was rapidly frozen in liquid nitrogen and ground. Formazan was selectively extracted using 200 ll of DMSO and absorbance was measured at 550 nm.

Peroxidase (PRX) and superoxide dismutase (SOD) activity
Flowers at different developmental stages were collected and frozen in liquid nitrogen. Frozen tissue was ground and homogenized in extraction buffer (0.1 M Tris, pH 7.0, 0.1% ascorbic acid, 0.1% L-cysteine, 0.5 M sucrose and 10 mg ml À1 PVP) and centrifuged at 4°C for 15 min, saving the supernatant. Total protein was quantified using the Bradford method (Bradford, 1976). Briefly, 10 ll of crude extract was added to a tube containing 1 ml of Bradford solution (0.01% Coomasie Brilliant Blue G-250, 4.7% ethanol, 8.5% phosphoric acid) and mixed. After 2 min, the absorbance was measured at 595 nm. A standard curve was generated using known concentrations of BSA.
For SOD activity, 25 mg of protein from the crude extract was added to 1 ml of SOD buffer (50 mM PBS, pH 7.6, 0.01 mM EDTA, 50 mM sodium carbonate, 12 mM L-methionine, 10 lM riboflavin, 50 lM NBT) and incubated at room temperature (22-25°C) under light conditions for 10 min. Absorbance was measured at 550 nm and SOD buffer without extract was used as a negative control. SOD activity was quantified as the quantity of enzyme required to inhibit 50% of the photoreduction of NBT.
For PRX activity, 25-75 mg of protein from the crude extract was added to 1 ml of PRX buffer (0.85 mM hydrogen peroxide in HEPES, pH 7.0, 0.125 M 4-aminoantipyrene, 8.1 mg ml À1 phenol) and the change in absorbance was measured for 2 min at 510 nm. A standard curve was generated using known concentrations of horseradish peroxidase.

RNA-Seq analyses
Total RNA was extracted from stage-8 stamens from frozen tissue of wild-type and Sltpd1 plants using a NucleoSpin RNA Plant kit (Macherey-Nagel, https://www.mn-net.com/) and measured in a NanoDrop ND-1000 Spectophotometer (ThermoFisher Scientific). The RNA quality was assessed based on the RNA integrity number (RIN) using a Bioanalyzer 2100 (Agilent, https://www.agilent. com/) and samples with RIN > 8 were selected for the experiment. RNA sequencing was performed using the BGISEQ Technology platform at BGI (https://www.bgi.com/global). A total of three biological replicates were used for each sample set. GO enrichment, KEGG enrichment and statistical analysis were performed through the Dr Tom platform (BGI).

Statistical analysis
SPSS STATISTICS 27 (IBM, https://www.ibm.com) was used for statistical analysis. For each data set, a Shapiro-Wilk normality test was run. For normally distributed data, a Student's t-test was used for pairwise comparison. Non-normally distributed data were analyzed with a Mann-Whitney test.

Bimolecular fluorescence complementation (BiFC)
The coding sequences of SlGRXC9/Solyc08g036570, SlTGA9/Soly-c06g074320 and SlTGA10/Solyc10g078670 were amplified by PCR and cloned into pCR8 using the pCR TM 8/GW/TOPO TM TA Cloning Kit (Invitrogen, now ThermoFisher Scientific). Then cDNAs were fused in frame at the N terminus with the N-terminus half (nYFP) or C-terminus half (cYFP) of the fluorescent YFP protein via Gateway. The vectors were transformed into Agrobacterium tumefaciens strain C58 and used to agroinfiltrate 4-week-old N. benthamiana leaves to a total OD 600 of 0.1. After 72 h of infiltration, protein-protein interactions were analyzed on leaf disks by confocal scanning microscopy (LSM 780; Zeiss). Intensity was adjusted using the negative control Del2GAI_cYFP (Gallego- Bartolomé et al., 2012), which does not interact with SlTGA9, SlTGA10 or SlGRXC9.

SUPPORTING INFORMATION
Additional Supporting Information may be found in the online version of this article. Figure S1. Identification of TPD1-like genes in tomato. Figure S2. Subcellular localization of Solyc03g097530 (SlTPD1) protein in Nicotiana benthamiana leaves. Figure S3. Complementation of the male-sterile floral phenotype of the Arabidopsis tpd1 mutant using the Solyc03g097530 (SlTPD1) gene. Figure S4. Expression of Solyc03g097530 (SlTPD1) in the ovary detected using in situ hybridization. Figure S5. Characterization of CRISP/Cas9-mediated Soly-c03g097530 (SlTPD1) edited tomato plants. Table S1. Accession numbers of TPD1-like gene sequences from different plant species used for the phylogenetic analysis. Table S2. Oligonucleotides used in this study. Table S3. List of differentially expressed genes (DEGs) between wild-type and Sltpd1 mutant anthers from floral stage 8.