Unequal redundancy among SEP4 paralogs determines tomato inflorescence architecture

Combining mutations in the tomato SEP4 paralogs J2, EJ2 and LIN results in synergistic increases in inflorescence branching and transforms j2 ej2 lin triple mutant inflorescences into highly branched, cauliflower-like structures [14] (Fig 1E–1G). To further dissect this example of genetic synergy into the individual contributions of J2, EJ2, and LIN, we analyzed single and double loss-of-function mutants in a cherry tomato cultivar (cv. Sweet-100) for changes in meristem development and inflorescence branching (S1B Fig). Confirming previous findings, the j2, ej2, and lin single mutants were not markedly affected in inflorescence branching but exhibited specific single mutant phenotypes (S1C–S1F Fig). Fruit abscission was lost in j2, sepals were elongated in ej2, and inflorescences developed additional flowers in lin [14]. In addition, j2 ej2 double mutants produced ectopic inflorescence meristems, which resulted in strongly branched inflorescences on primary and sympodial shoots bearing flowers with elongated sepals and missing abscission zones [14] (Fig 2A–2C). In comparison, lin j2 and lin ej2 double mutants initiated fewer ectopic inflorescence meristems, which led to weakly branched inflorescences at high penetrance on sympodial shoots and rarely on primary shoots (Fig 2C–2G). The lin j2 and lin ej2 double mutants also developed more flowers per inflorescence compared to WT, but were not markedly enhanced compared to lin single mutants (Fig 2H). In addition, lin j2 lacked fruit abscission zones while lin ej2 developed elongated sepals, reminiscent of the j2 and ej2 single mutants (S1G Fig). These observations indicate synergistic interactions between LIN and both J2 and EJ2 during inflorescence development; however, the effect was weaker when compared with synergy between J2 and EJ2 [14]. To account for genotype dependencies, we also revisited a previously published mutant collection in a plum tomato (cv. M82) [14], which confirmed weak inflorescence branching and more flowers per inflorescence in lin j2 and lin ej2 (Figs 2I, 2J, and S1G). In conclusion, we resolved synergistic interactions in lin j2 and lin ej2 double mutants that are weaker compared with interactions in j2 ej2. These results indicate unequal redundancy between SEP4 paralogs with a minor contribution of LIN to the suppression of inflorescence branching.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Unequal redundancy among the SEP4 paralogs J2, EJ2, and LIN suppresses inflorescence branching and flower production in tomato.

(A) Stereomicroscopy images of a j2 ej2 double mutant shoot apex after the floral transition with additional ectopic inflorescence meristems on the primary inflorescence. The orientation of de novo release of meristems on the primary inflorescence is indicated by white arrows in the zoomed image. Scale = 100 µm. (B) Detached inflorescence of the j2 ej2 double mutant (cv. S100) displaying strong inflorescence branching. Scale = 1 cm. (C) Quantification of inflorescence branching in the S100 cultivar shown as the percentage of branched inflorescences per genotype. cal, cauliflower inflorescence. (D, E) Stereoscope images of the shoot apex of lin j2 (D) and lin ej2 (E) double mutants. The orientation of the release of ectopic meristems on the inflorescence of the first sympodial shoot is indicated by white arrows in the zoomed image. Scale = 100 µm. (F, G) Images of detached inflorescences from the lin j2 (F) and lin ej2 (G) double mutants showing weak branching. Scale = 1 cm. (H) Quantification of flower number per inflorescence in the S100 cultivar. Note that the j2 ej2 mutant was excluded due to the unquantifiable (high) flower number. (I) Quantification of inflorescence branching in the M82 cultivar as in (C). (J) Quantification of flower number per inflorescence in the M82 cultivar as in (H). Letters in (H, J) represent the results from pairwise comparisons of means using one-way ANOVA and post hoc Tukey’s HSD test with 95% confidence level. The data underlying this figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003946.g002

SEP4 synergy acts through a limited set of genes after meristems transition to floral fate

Inflorescence architecture is determined by the rate at which meristems mature from vegetative to reproductive identity [7,15]. To gain molecular insights into how SEP4 synergy affects the rate of meristem maturation, we sequenced mRNA from micro-dissected meristems of the single and double sep4 mutants. We sampled meristems at the transition stage of meristem maturation (hereafter transition meristem samples), as well as a combined floral meristem/sympodial inflorescence meristem stage (hereafter floral meristem samples), during which J2, EJ2, and LIN are highly expressed (Figs 1A and S1A). We excluded the j2 ej2 lin triple mutant from this experiment due to the difficulty of obtaining staged meristem samples from a segregating triple mutant population. Multidimensional scaling (MDS) analyses of transcriptomes revealed that floral meristem samples clustered by the extent of inflorescence branching (Figs 3A, 3B, and S2) (see Methods). In contrast, transition meristem samples lacked clear clustering patterns, suggesting that floral meristem samples better captured relevant transcriptional changes associated with inflorescence branching. A differential expression analysis contrasting mutant genotypes with the WT yielded a total of 946 and 849 differentially expressed genes (DEGs; |log₂FC| > 0.58, FDR < 0.01) in the transition and floral meristem samples, respectively (Fig 3C and 3D; S1–S6 Tables). A rather small fraction of DEGs was shared between the three double mutants in the transition samples (68 DEGs, 7.2%) (Fig 3E and S7 Table), but this fraction increased in floral samples (104 DEGs, 12.3%) (Fig 3F and S8 Table). Furthermore, while gene ontology (GO) enrichment analysis of the 946 DEGs from the transition meristem samples did not yield any significantly enriched terms, the 849 DEGs in floral meristem samples showed significant enrichment for GO categories related to morphogenesis, flower development, and floral organ identity (Fig 3G and 3H). We concluded that SEP4 paralogs retained redundant functions to regulate a defined genetic module during inflorescence and floral meristem maturation stages, at which SEP4 genes are highly expressed (S1A Fig). Combined loss of SEP4 activity changes this genetic module and disturbs meristem maturation during the acquisition of floral identity.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. SEP4 synergy acts on a shared module of genes after the transition stage of meristem maturation.

(A, B) Multidimensional scaling (MDS) plots of gene expression of 227 transition marker genes in transition meristem samples (A) and 241 floral marker genes in floral meristem samples (B). (C, D) Heatmap showing z-score normalised expression of differentially expressed (FC ≥ 1.5, FDR ≤ 0.01) genes in transition (n = 946) (C) and floral (n = 849) (D) meristem samples. (E, F) Overlap of genes differentially expressed in transition (E) and floral (F) meristem samples in j2 ej2, lin j2, and lin ej2 double mutants compared with the WT. The total number of DEGs per genotype is indicated in parenthesis. (G, H) The 10 most enriched gene ontology (GO) categories for all 849 (G) and common 104 (H) genes differentially expressed in j2 ej2, lin j2, and lin ej2 floral meristem samples. Dot size is proportionate to the count of genes per term. P values obtained by Benjamini–Hochberg (BH) method in clusterProfiler. The data underlying this figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003946.g003

SEP4 synergy is mirrored by a small group of genes with synergistic expression patterns

To pinpoint genetic modules that underlie synergistic changes in inflorescence branching, we focused on genes following synergistic expression patterns across floral meristem samples from single and double sep4 mutants. We adapted an established framework [39] to compare relative expression changes in single and respective double mutants, and isolate genes with non-additive expression patterns that deviate from expected additive effects (Fig 4A; S1–S6 Tables). By including directionality of non-additive expression changes relative to the predicted additive effects, we classified genes into three categories: (1) additive, in which the observed expression change in the double mutant did not significantly differ from the cumulative observed change in the two single mutants; (2) less-than-additive, in which the double mutant change was significantly smaller than the cumulative single mutant change; and (3) synergistic, in which the double mutant change was significantly greater than the cumulative single mutant change. Most genes followed non-additive (less-than-additive or synergistic) expression patterns (Fig 4B and 4C). For example, among the 351 DEGs in the j2 ej2 floral meristem samples, 220 (63%) followed non-additive expression patterns, including 139 less-than-additive and 81 synergistic genes. Among the non-additive expression changes, we observed more less-than-additive than synergistic changes. DEGs with less-than-additive expression are likely controlled by MADS-box heterocomplexes that contain multiple SEP4 paralogs. In such cases, loss of a single SEP4 paralog could already cause a reduction in heterocomplex activity that is not further enhanced in the double mutant. On this basis, less-than-additive DEGs are unlikely to explain the synergistic inflorescence phenotypes that are only observed in the double mutants. We concluded that synergistic inflorescence phenotypes are mirrored by widespread non-additive expression changes, but only a small set of genes with synergistic changes drive these complex expression changes.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Synergy among SEP4 paralogs acts on ANANTHA to drive the maturation of meristems towards floral fate.

(A) Schematic representation of additive and non-additive gene expression categories in double mutants when compared with single mutants. (B) Normalized expression (z-score) of differentially expressed genes in floral meristems j2 ej2 (top), lin j2 (middle), and lin ej2 (bottom) mutants compared with the WT, classified into additive, less-than-additive, and synergistic (more-than-additive) gene expression categories. Each category is split into genes that are down- and upregulated compared with the WT. Loess-smoothed trendlines are plotted in purple and yellow for down- and upregulated categories, respectively, with 95% confidence intervals shaded in red. n, number of genes per category. (C) Distribution of differentially expressed genes in j2 ej2 (top), lin j2 (middle), and lin ej2 (bottom) mutants into additive, less-than-additive, and synergistic (more-than-additive) gene expression categories. (D) Overlap of gene expression categories in 104 genes differentially expressed in j2 ej2, lin j2, and lin ej2. (E) Normalized gene expression (z-score) for genes in (D) in different tissues and developmental stages. Both expression mode and presence of J2 ChIP-Seq peak are indicated as in (D). Genes synergistically downregulated in j2 ej2 compared to the single mutants and WT are labeled with gene names. Magenta and blue boxes indicate gene clusters c2 (floral meristem) and c3 (flowers and fruit), respectively. (F) Macroscopic image of an an mutant inflorescence with overproliferating meristem tissue (cauliflower) instead of flowers. Scale = 1 cm. (G) Normalized gene expression of AN in floral meristem samples from single and double mutant combinations for j2 ej2, lin j2, and lin ej2. Data are represented as mean ± SE. (H) Browser view of the AN genomic region. Normalized coverage in counts per million (CPM) is shown for vegetative and floral meristem RNA-seq, meristem ATAC-Seq, and J2 ChIP-Seq (including IgG control). A significant J2 binding peak (P196) is indicated by a red square. Predicted MADS-box transcription factor binding sites (MADS TFBS) and conserved noncoding sequences (CNS) are indicated by magenta and cyan squares, respectively. Genomic positions on chromosome 2 (SL4.0) and the distance between the J2 binding site and the transcriptional start site (TSS) are indicated. (I) PCA of the 500 most differential genes in sep4t (j2 ej2 lin) and an inflorescence tissue, and staged WT meristems (EVM, early vegetative; MVM, middle vegetative; LVM, late vegetative; TM, transition; SIM, sympodial inflorescence; FM, floral). (J) AN expression levels in an and sep4t inflorescence tissue, and staged WT meristems. The data underlying this figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003946.g004

SEP4 synergy acts on the conserved floral meristem identity gene ANANTHA

To further investigate synergistic gene expression patterns that are associated with changes in inflorescence branching, we focused on the 104 DEGs that were shared between double mutants in floral meristem samples (Fig 4D and S8 Table). Among those shared DEGs, the j2 ej2 double mutant yielded the largest number of synergistic DEGs, consistent with the strong, synergistic inflorescence phenotype (Fig 2B). Most synergistic DEGs were downregulated in the j2 ej2 mutant but downregulated with additive or less-than-additive patterns in the lin j2 and lin ej2 combinations, in agreement with the milder inflorescence phenotypes (Fig 2F and 2G). When we investigated tissue-specific expression patterns of 104 shared DEGs using public expression data, we identified a cluster of genes with peak expression in floral meristems [7,40] (Fig 4E). This floral meristem cluster (c2) included AN, a known regulator of floral meristem identity and ortholog of Arabidopsis UFO [19,41]. Knockout of AN leads to cauliflower-like inflorescences reminiscent of the j2 ej2 lin triple mutants (Fig 4F) [19,42]. AN expression was synergistically downregulated in j2 ej2 but additively downregulated in lin j2 and lin ej2 genotypes, suggesting that SEP4 dosage-dependent changes in AN expression underlie gradual changes in inflorescence branching (Fig 4G). This model is further supported by a J2 binding site upstream of AN that we identified in published Chromatin Immunoprecipitation-sequencing (ChIP-seq) data [43]. Interestingly, the J2 binding site resides in an accessible chromatin region (ACR), which contains conserved noncoding sequences (CNS) and cis-regulatory elements (CREs) that were shown to impact AN transcription [44] (Fig 4H). Given that J2 and other SEP factors bind to a conserved MADS-box target motif (CArG-box) [43,45], we propose AN as a direct transcriptional target of SEP4 factors. To further test this hypothesis, we profiled gene expression of an and j2 ej2 lin triple mutant (sep4t) inflorescences by RNA-seq. Principal component analyses (PCA) of mutant transcriptomes with staged WT meristem profiles [7] revealed similar transcriptional states of an and sep4t meristems (Fig 4I). In addition, AN transcripts were barely detectable in sep4t meristems (Fig 4J), suggesting that sep4t cauliflower inflorescences can be largely attributed to the loss of AN expression. In conclusion, reductions in the dosage of SEP4 paralogs result in quantitative downregulation of AN, which explains the synergistic effect on inflorescence branching. However, other genes with peak expression in floral meristems and similar synergistic expression patterns likely act as additional drivers of SEP4 synergy.

SEP4 synergy is relayed to SEP1/2/3 paralog expression levels

In addition to the floral meristem cluster (c2), we identified a cluster (c3) of genes that were primarily expressed in samples derived from developing flowers and fruits, and that included two additional SEP paralogs, TOMATO MADS-BOX5 (TM5) and TM29 (Fig 4E). Previous studies had demonstrated by RNA in-situ hybridization that TM5 and TM29 are expressed in floral meristems at the onset of organogenesis and that individual knock-down by antisense technology transforms carpels, the female reproductive floral organs that give rise to the tomato fruit, into ectopic inflorescences [46,47]. TM5 and TM29 are lowly expressed during the floral meristem stage of meristem maturation (Fig 4E) and differentially expressed in floral meristem samples across all tomato sep4 double mutants (Fig 5A), which suggested an additional layer of SEP paralog redundancy in floral meristems. Therefore, we revisited the phylogenetic relationship of tomato SEP proteins using hierarchical orthologous groups (HOGs) [48] (Figs 5B and S3A; S9 and S10 Tables). As previously reported [14], we detected an expansion of the SEP family in tomato to six members compared with only four in Arabidopsis (Fig 5B and 5C) [11]. A similar expansion had been observed in the SEP family of the related Solanaceae species petunia [34], and we observed more than four SEP proteins in all other Solanaceae species in our dataset (Fig 5C). This family expansion in Solanaceae is driven by duplication events in the SEP4 clade, with two duplication events that resulted in the paralog pairs J2-EJ2 and LIN-RIN in tomato (S3A and S3B Fig). In contrast, the SEP1/2 clade, which comprises SEP1 and SEP2 in Arabidopsis, is reduced to TM29 in tomato, and this contraction to a single SEP1/2 paralog is conserved throughout the Solanoideae subfamily of Solanaceae (Fig 5B and 5C). Paralog copy number is conserved in the SEP3 clade, which is the ancestral SEP subclade when using the AGAMOUS LIKE6 (AGL6) sister clade as an outgroup, and comprises tomato TM5 and the Arabidopsis ortholog SEP3.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. The SEP1/2 and SEP3 orthologs TM29 and TM5 function redundantly to regulate floral organ identity in tomato.

(A) Normalized gene expression of TM5 (top) and TM29 (bottom) in floral meristem samples from single and double mutant combinations for j2 ej2, lin j2, and lin ej2. Data are represented as mean ± SE. (B) Maximum-likelihood tree of SEP proteins in Arabidopsis and Solanaceae. The AGL6 sister clade was used as outgroup. SEP1/2, SEP3, SEP4 subclades are highlighted in different shades of purple. Tomato, physalis, petunia and Arabidopsis genes are labeled in red, yellow, magenta, and green font, respectively. Numbers at nodes represent ultrafast bootstrap support values from 1,000 replicates. Branch lengths are proportional to the number of amino acid substitutions, branches in green show duplication events, and the scale bar indicates the average number of substitutions per site. (C) SEP gene counts across distinct subclades in Arabidopsis and Solanaceae species shown in the phylogenetic tree in (B). Note that gene content of tetraploid N. tabacum is halved to allow comparisons with the other (diploid) genomes. (D) Images of a WT inflorescence with a close-up of a single flower (left) and a dissected flower (right), showing the pedicel with calyx (sepals), petals, stamens, and carpels (right). Note the presence of abscission zones on the pedicel of the flower. (E–G) Images of the tm29 (E), tm5 (F), and tm5 tm29 (G) mutants as in (D). Note petal fusions in (E–G), bending anthers in (F), and changes in both petal, stamen, and carpel identity in (G) highlighted in magenta font and asterisks. (H) Macroscopic image of a tm5 tm29 double mutant inflorescence with close-up on flowers that produce pseudofruits. (I) Macroscopic images of tm5 tm29 pseudofruits sectioned along the longitudinal axis (top) and with partially removed pericarp (bottom). The pericarp and locules are labeled. Note that locules of pseudofruits are filled with leaf-like organs. Scale bars in (D–I) indicate 1 cm. The data underlying this figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003946.g005

Redundancy between tomato SEP1/2 and SEP3 subclades confers the conserved SEP function to specify floral organ identity

Since TM5 and TM29 were expressed in floral meristems (S4A Fig) and synergistically downregulated in j2 ej2 floral meristem samples (Fig 5A), we reasoned that TM5 and TM29 may play roles during inflorescence development. Thus, we generated tm5 and tm29 mutants and isolated loss-of-function alleles that disrupt the conserved MADS-box domain (S4B Fig). Single tm5 and tm29 mutants did not show obvious changes in inflorescence architecture (S4C Fig). Moreover, we observed only very subtle defects in floral organ development: petals on tm5 and tm29 mutants were partly fused, anthers in tm5 flowers were bending outwards, and tm5 fruits produced fewer seeds (Figs 5D–5F and S4D–S4F). We could not confirm the strong floral organ defects of tm5 and tm29 single mutants from previous antisense knock-down studies [46,47]. However, the mild flower defects observed in tm5 hinted towards unequal redundancy between TM5 and TM29 with a more prominent role for the SEP3 ortholog TM5 during floral organ specification. Indeed, organ defects were dramatically enhanced in tm5 tm29 double mutants, which developed flowers with petals and stamen transformed to leaf-like organs (Figs 5G and S4G). In addition, tm5 tm29 flowers gave rise to parthenocarpic (seedless) fruits that contained leaf-like organs instead of seeds (Figs 5H and 5I and S4H–S4I), similar to petunia sep123 (fbp2 fbp5 pmads12) mutants, which develop ovaries filled ovules converted to leaf-like organs. In contrast to tomato tm5 tm29, petunia fbp2 fbp5 pmads12 flowers retain partial stamen identity [49]. This transformation of floral organs into leaf-like organs in all floral whorls in tm5 tm29 is reminiscent of the sep123 triple mutant in Arabidopsis, after which the SEPALLATA gene family was named [26]. However, in contrast to Arabidopsis, loss of SEP1/2/3 activity in tomato still permits pericarp development in tm5 tm29 mutants. We concluded that TM5 and TM29 act redundantly to confer the conserved SEP1/2/3 function in specifying floral organ identity across all four whorls. Unequal redundancy between TM5/TM29 during floral organ specification indicates that they are an ancient pair of paralogs that diverged due to compensatory drift [3].

SEP4 paralogs retained a function in specifying floral organ identity

The fact that neither single nor double tm5 tm29 mutants developed branched inflorescences contradicted the hypothesis that synergistic reduction of TM5 and TM29 expression is responsible for inflorescence branching in j2 ej2 mutants. We reasoned that the effect of TM5 and TM29 on meristem activity only becomes visible when SEP4 activity is reduced, so we combined tm5 and tm29 with j2 and ej2 mutations. Indeed, loss of J2 or EJ2 activity in tm5 tm29 j2 and tm5 tm29 ej2 triple mutants enhanced the vegetative identity of sepals, petals, and stamen (Fig 6A and 6B). Yet, the most dramatic changes were again observed in the fourth floral whorl. While tm5 tm29 mutants developed parthenocarpic fruits with leaf-like organs, the additional loss of one SEP4 paralog in tm5 tm29 j2 and tm5 tm29 ej2 triple mutants caused the development of ectopic inflorescences in the center of each flower (Figs 6A, 6B, S4J, and S4K). This process is reiterated in each new flower, indicating a reversion from floral to inflorescence meristem identity (Fig 6C–6E). These results show that when SEP4 function is partially lost in the absence of SEP1/2/3 (in tm5 tm29 j2 or tm5 tm29 ej2), floral meristems produce three whorls of vegetative organs. Mutant floral meristems fail to terminate in the fourth whorl and instead revert to inflorescence meristem identity, producing an ectopic inflorescence with flowers that reiterate the process. This shows that tomato SEP4 (J2 and EJ2) can confer the SEP1/2/3 function in the fourth whorl in the absence of TM5 and TM29, consistent with partial redundancy of SEP1/2/3/4 during carpel development (S4L Fig).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 6. Redundant SEP1/2/3/4 activity acts on FALSIFLORA during the specification of floral organ identities.

(A, B) Images of tm5 tm29 j2 (A) and tm5 tm29 ej2 (B) triple mutants with a close-up of single (left) and dissected (right) flowers, showing the pedicel with calyx (sepals), petals, stamens, and carpels (right). Note the changes in petal, stamen, and carpel identity in (A, B) highlighted in magenta font with asterisks. (C, D) Images of dissected tm5 tm29 j2 (C) and tm5 tm29 ej2 (D) carpels showing the formation of ectopic inflorescences. (E) Images of tm5 tm29 j2 (left) and tm5 tm29 ej2 (right) inflorescences showing the reiterated formation of ectopic inflorescences in flowers. (F, G) Macroscopic (left and middle) and stereomicroscopic (right) images of a sep1234q quadruple (F) and falsiflora (G) mutant inflorescences with overproliferation of meristem tissue (cauliflower) and bracts. (H) Macroscopic images of detached flowers from different sectors (sec) of a fa^CR F1 individual. Note the gradual changes in floral organ identity from sec-2 to sec-7, and the cauliflower inflorescence in sec-8. (I) Schematic representation of FA with guide RNAs (gRNA) target positions indicated by arrowheads. Purple boxes represent exons. The Cas9 cleavage sites for gRNAs are indicated with arrowheads. (J) Quantification of FA haplotype frequencies in different fa sectors and a WT control by amplicon deep sequencing. The total number of reads, percentage of wild-type (%WT) and modified (%mod.) reads are shown. Note that haplotype-1 (h01) represents the WT haplotype at both target regions. (K) Browser view of the FA genomic region. Normalized coverage (CPM) is shown for vegetative and floral meristem RNA-seq, meristem ATAC-Seq, and J2 ChIP-Seq (including IgG control). A significant J2 binding peak (P366) is indicated in red. Predicted MADS-box binding sites (MADS TFBS) and conserved noncoding sequences (CNS) are indicted in magenta and cyan, respectively. The distance between the J2 binding site and the transcriptional start site (TSS) are indicated. (L) PCA of the 500 most differential genes in fa and sep1234q meristems and staged WT meristems (EVM, early vegetative; MVM, middle vegetative; LVM, late vegetative; TM, transition; SIM, sympodial inflorescence; FM, floral). (M) Developmental stage prediction for fa and sep1234q meristem tissue from transcriptome data using RAPToR. Age estimates for replicate samples are shown. Diamonds and error bars indicate mean values and standard deviation. (N) Expression (normalized read counts) of FA in staged WT meristems and fa and sep1234q meristem tissue. (O) Proposed regulatory feedback between SEP genes and AN/FA. Stage-specific expression patterns of SEP1-4 and AN/FA (left) support a sequential, positive feedback between SEP genes and the AN/FA module (right): (1) At the transition from vegetative to floral meristem identity, SEP4 genes redundantly activate AN, supported by sep4-triple and an cauliflower-like inflorescences, low AN expression in sep4 single and combinatorial mutants, and J2 binding upstream of AN. (2) AN/FA positively feedbacks on SEP4 to maintain floral meristem identity, supported by downregulation of SEP4 genes in an meristems. (3) AN/FA activates SEP1-3 genes in floral meristems to specify floral organ identity, supported by SEP1-3 downregulation in an and fa meristems. (4) SEP1-4 positively feedbacks on FA to maintain floral meristem identity, supported by low FA expression in sep4t and sep1234q meristems. Solid and dashed arrows mark regulation with support for direct (genetic analyses, RNA-seq, ChIP-seq) and indirect regulation (genetic analyses, RNA-seq), respectively. Scale bars indicate 1 cm (A–E, F–G (left), H), 0.5 cm (F–G middle), 500 µm (F–G right). The data underlying this figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003946.g006

SEP1/2/3 paralogs confer a minor function in driving meristem transitions to floral identity

We further reduced SEP4 activity in the absence of SEP1/2 or SEP3 paralogs by isolating tm29 j2 ej2 (sep1/2 sep4^partial) and tm5 j2 ej2 (sep3 sep4^partial) triple mutants. The triple mutants developed inflorescence branching patterns reminiscent of j2 ej2 (Figs 2B and S5A). In addition, loss of TM5 (SEP3) or TM29 (SEP1/2) increased the occurrence of vegetative organs (bracts) in the inflorescence and defects in floral organ identity (S5A and S5B Fig). As a result, tm5 j2 ej2 mutants developed petals instead of stamens, indicating that loss of J2 and EJ2 further enhances effects in tm5 on stamen identity. We also isolated the tm5 tm29 j2 ej2 quadruple (sep1234q) mutant, which developed highly branched inflorescences that failed to differentiate flowers and instead developed modified leaves (bracts) and overproliferative meristematic tissue (Fig 6F). This severe loss of floral identity from the combined mutation of TM29 (SEP1/2), TM5 (SEP3), and J2/EJ2 (SEP4) shows that SEP1/2/3 paralogs confer floral meristem identity when SEP4 dosage is reduced. In conclusion, SEP1/2/3 paralogs retained redundant functions with SEP4 paralogs in promoting meristem maturation towards floral identity.

SEP1/2/3/4 redundancy acts on the conserved floral identity gene FALSIFLORA

The inflorescences of sep1234q mutants closely resembled those of the historical mutant falsiflora (fa) [17], which lacks the activity of the tomato ortholog of Arabidopsis LFY, a floral identity gene that is deeply conserved across flowering plants [16,18,50]. Hence, we generated fa mutants in cherry tomato (cv. Sweet-100) using CRISPR–Cas and observed the expected fa phenotypes in multiple first-generation (T0) plants (Figs 6G and S5D–S5F). Due to high Cas9 editing activity, we recovered only one out of 11 fa^CR T0 individuals that developed a few fertile flowers with pollen for backcrossing to the WT. Cas9-free fa-null/ + F1 heterozygotes did not develop obvious floral organ or inflorescence defects (S5G Fig), confirming the recessive nature of the fa mutant [17]. However, we identified an F1 individual that retained Cas9 activity and developed distinct sectors with abnormal phenotypes ranging from floral organ defects to fa-null inflorescences that overproliferated bracts and meristematic tissue (Figs 6H and S5H). The distinct phenotypic sectors suggested that Cas9 had induced additional fa mutations, leading to gradual reductions in FA dosage and a spectrum of floral organ and inflorescence defects. We quantified the frequency of FA mutations in individual flowers from sector 1 (sec-1) to sec-7 or inflorescence tissue (sec-8) by amplicon deep sequencing (Fig 6I). Remarkably, the frequency of the functional FA (WT) haplotype indeed decreased with phenotypic severity, from sectors with mild floral organ defects to sectors with strong organ defects and branched inflorescences, which contained only out-of-frame fa mutations (Figs 6J and S5I; S11 Table). These results show that gradual reductions in FA gene dosage can lead to both floral organ defects and inflorescence branching. Given the phenotypic similarities between fa and sep1234q mutants, we reasoned that SEP1/2/3/4 activity is required for maintaining FA expression in reproductive meristems, which guides meristem maturation towards floral fate. This model is supported by a J2 binding site upstream of FA with two CarG-box elements (Fig 6K) [43] and a recent report that deletion of this site leads to floral organ defects and mild inflorescence branching [51].

To investigate the functional relationship between SEP1/2/3/4 and FA, we profiled gene expression in fa and sep1234q inflorescences by RNA-seq and compared mutant transcriptomes with staged WT meristem profiles [7]. Principal component analyses of the 500 most variable genes resolved subsequent stages of meristem maturation on the first component (PC1) (Fig 6L). To estimate the developmental stage of fa and sep1234q meristem tissue, we applied an established framework [52] and staged mutant transcriptomes on the WT meristem maturation trajectory as a reference (Fig S6 and Methods). This analysis predicted fa and sep1234q near late vegetative (LVM) and transition (TM) meristem stages, suggesting that fa and sep1234q meristems revert to a maturation stage that precedes the transition to reproductive development (Fig 6M). In addition, FA transcript levels were strongly reduced in sep1234q meristems, consistent with a model in which SEP1/2/3/4 activity is required to maintain FA expression (Fig 6N). To further investigate this potential positive maintenance relationship, we examined expression of FA, AN and SEP1-4 genes across the sep4t, sep1234q, an, and fa RNA-seq datasets (S7A and S7B Fig). In addition to low AN transcript levels in sep4t (j2 ej2 lin) mutants, we found that J2 and EJ2 were downregulated in an, supporting a positive feedback loop between SEP4 genes and AN (Fig 6O). Further, TM5 and TM29 were downregulated in both an and fa, and AN and FA were also lowly expressed in sep4t and sep1234q, indicating a second layer of feedback maintenance between SEP1-3 and AN/FA during floral organ specification. Interestingly, LIN showed high expression in an, fa, and sep1234q while TM29 maintained elevated transcript levels in sep4t and sep1234q, suggesting potential compensatory mechanisms that warrant further investigation (S7A and S7B Fig). Together, these results indicate that incomplete functional divergence of SEP genes ensures redundancy on the dose-sensitive FA/AN module, which is conserved across flowering plants to drive meristem maturation during inflorescence and flower development [50].

SEP gene synergy emerged from redundancy on the conserved FA/AN module

In this study, we dissected synergistic interactions between closely related SEP4-clade MADS-box genes in the model crop tomato. Previous studies showed synergy between the SEP4-clade paralogs J2, EJ2, and LIN on the suppression of inflorescence branching [14,53]. Here, we found that SEP4 synergy on gene expression programs underlying inflorescence branching is more pronounced between the direct paralogs J2 and EJ2 than with the distant paralog LIN. This pattern is consistent with paralog redundancy generally declining over evolutionary time, likely driven by divergence in coding sequences and gene expression patterns [54]. Transcriptome analyses show that SEP4 synergy acts through a defined set of genes in floral meristems including AN, a conserved floral identity gene and ortholog of Arabidopsis UFO [19,50]. By integrating data from previous studies [43,44], we propose a model (Fig 6O) in which SEP4 paralogs directly activate AN expression by binding an enhancer element ~3.5 kbp upstream of AN when meristems transition to floral identity, a critical window in which inflorescence architecture is determined [7,14,15]. Further, we found that synergy between four SEP1/2/3/4 paralogs acts through the expression of FA, a second deeply conserved floral identity gene and ortholog of Arabidopsis LFY [16,22,50]. Remarkably, loss of SEP1/2/3/4 activity leads to lower FA transcript levels, and lowering FA gene dosage recapitulates the phenotypic spectrum arising from gradual reduction of SEP1/2/3/4 dosage. These results indicate that SEP1/2/3/4 synergy converges to maintain FA dosage. We found evidence that SEP1/2/3/4 factors directly regulate FA by binding to an upstream cis-regulatory element, a model supported by published ChIP-Seq data [43] and CRISPR-Cas deletion of this element, which leads to inflorescence branching and homeotic floral organ conversions [51]. Thus, residual SEP1/2/3/4 redundancy acts on the FA/AN module during subsequent stages of meristem development to drive identity transitions to reproductive development, a function that remained conserved when SEP genes functionally diverged. Together, these findings show that SEP gene synergy emerged as a relic of functional redundancy.

An open question is which other MADS-box factors interact with SEP proteins to form heterotetrameric complexes that bind to the CArG-box target motifs upstream of AN and FA. We speculate that SEP proteins interact with AP1/FRUITFULL (FUL)-clade MADS-box proteins that are co-expressed in transition and floral meristems [7,11,55]. This model is supported by published yeast two-hybrid assays showing interactions between the AP1/FUL-like proteins MACROCALYX (MC) and FUL2, and the SEP proteins TM5, J2, EJ2, and RIN [55,56]. A recent study [55] reported that heterotetramers of J2 and either MC or FUL2 bind target DNA in electrophoretic mobility shift assays (EMSAs). We revisited affinity purification-sequencing (DAP-seq) experiments with in vitro-reconstituted heterotetramers from this study but did not find evidence for J2/MC or J2/FUL2 binding at the J2 target sites upstream of AN and FA identified in in vivo meristem ChIP-seq data [43]. However, genetic evidence supports a role of MC, FUL2, and the FUL-like gene MBP20 in meristem maturation and floral meristem identity. Combined loss of FUL2 and MBP20 function results in a flowering delay and moderately branched inflorescences that resemble those of weak j2 ej2 mutants [55]. The mc mutant also flowers late but produces flowers with elongated sepals similar to ej2 mutants [55,57]. However, in contrast to ej2, mc inflorescences revert to vegetative growth after the formation of few flowers and this vegetative reversion is strongly enhanced in mc ful2 mbp20, indicating redundant roles of AP1/FUL genes in conferring floral identity [55,57]. Together, these findings show that AP1/FUL-genes are involved in both guiding meristem transitions and specifying floral organ identity, paralleling the roles of FA and SEP genes described in this study. This dual function supports a model in which the floral transition and floral organogenesis represent a developmental continuum rather than distinct phases [58].

Feedback regulation among SEP paralogs calibrates their gene dosage

We propose that transcriptional feedback regulation among SEP paralogs coordinates subsequent stages of meristem maturation by maintaining a critical dosage of the FA/AN module. SEP4 synergy influences the expression of two closely related SEP1/2/3 genes, leading to a dose-dependent downregulation of SEP1/2/3 paralogs in floral meristems when SEP4 activity is reduced. Whether SEP4 proteins directly activate SEP1/2/3 paralogs remains to be determined since we could not identify J2 binding sites near TM5 and TM29, suggesting either an indirect relationship or a direct control over longer genomic distances. More likely appears that transcriptional feedback between SEP4 and SEP1/2/3 paralogs is bridged by the FA/AN module to guide meristems during the acquisition of floral identity (Fig 6O). The Arabidopsis orthologs of the transcription factor FA (LFY) and the F-box protein AN (UFO) form a heteromeric transcriptional complex that regulates targets which LFY homomers cannot bind [41]. UFO forms a ring-like expression domain in vegetative and inflorescence meristems, which has been proposed to establish a pattern that in combination with local activation of LFY facilitates floral meristem initiation and bract suppression [58]. In contrast, tomato AN is specifically expressed during the floral stage of meristem maturation while FA is expressed throughout meristem maturation including stages that precede the floral transition. FA homomers likely confer the FA function in promoting the transition to flowering, consistent with a flowering delay observed in fa but not in an mutants [16,22]. The activation of SEP4 paralogs at the floral transition precedes AN expression in floral meristems (Fig 6O), and AN expression is diminished when SEP4 activity is reduced. SEP4 function is thus required for AN expression and the formation of putative FA/AN heterocomplexes that drive meristem maturation post-transition towards floral identity. SEP1/2/3 paralog expression is instead activated in floral meristems during floral organ initiation [11], and a critical SEP dosage is essential during these maturation stages to maintain FA expression. Progressive loss of SEP activity in tm5 tm29 (sep123) and tm5 tm29 j2 ej2 (sep1234q) mutants causes homeotic organ conversions and loss of floral meristem determinacy. Similarly, quantitative reductions in FA dosage cause floral organ conversions and loss of floral meristem identity [51], and FA expression is substantially reduced in sep1234q mutant meristems. Thus, transcriptional feedback regulation between SEP1/2/3/4 paralogs and AN/FA (Fig 6O) ensures a critical SEP dosage to maintain FA and AN expression after the floral transition for accurate floral meristem patterning and floral organ differentiation. This model warrants further investigation using spatially resolved approaches to delineate feedback maintenance between SEP and AN/FA genes. While the spatiotemporal expression domains of TM5 [46], TM29 [47], J2/SLMBP21 [59], AN [19], and FA [16] in floral meristems have been characterized by in situ hybridization experiments at tissue level individually, examining how these patterns are modified in sep and an/fa mutant backgrounds at cellular resolution will provide additional mechanistic insights.

Gene dosage effects maintain SEP redundancy but constrain functional divergence

Multiple duplication events led to the expansion of the SEP gene family and allowed SEP paralogs to functionally diverge and acquire new functions [25,34]. We show that SEP redundancy is conserved during the transition of meristems to floral identity, seen in reversion of floral meristems to vegetative identity in sep1234q mutants. Redundant SEP roles during floral meristem development have been described in other species [34,35], which indicates that the SEP function in guiding meristem maturation towards floral identity is evolutionary conserved. SEP3, which belongs to a conserved paralog subclade that is maintained as one copy per diploid genome in most dicot species, thus evolved the derived and more specialized function in floral organ identity. Nevertheless, the dosage sensitivity of SEP genes during inflorescence and flower development imposed a constraint to the functional divergence of the SEP family. The molecular basis that underlies incomplete divergence of SEP paralogs remains to be determined. Divergent expression patterns from cis-regulatory changes may have led to functional divergence without unbalancing SEP dosage, and expression atlas data for SEP genes [7,11] hints towards regulatory sequence evolution although more detailed analyses of cell type-specific expression patterns are required. Future analyses of non-coding sequences may reveal cis-regulatory regions and transcriptional enhancers that control spatiotemporal SEP expression patterns underlying both divergent and redundant functions. In summary, our work demonstrates that dosage sensitivity of paralogous genes can impose an evolutionary constraint to the capacity of gene families to fully diverge in function.

Plant material and growth conditions

Wild-type seed for the S. lycopersicum cv. M82 (LA3475) and S. lycopersicum cv. Sweet-100 double-determinate [42] were from our own stocks. Mutant seed for j2, ej2 and lin single and combinatorial mutant combinations in the M82 background were from our own stocks [14]. For phenotypic analyses, seeds were germinated on soil in 96-well plastic trays and grown under long-day conditions (16-h light/8-h dark) in a greenhouse supplemented with artificial light from LED panels (~250 μmol m⁻² s⁻¹), oscillating day-night temperature (25 °C day/20 °C night), and relative humidity of 50%–60%. Plants were grown in 5 L pots (2 per pot) under standard fertilizer regime and drip irrigation. For meristem analyses, seeds were pre-germinated on moist filter paper at 28 °C in the dark for 72 hours. Germinated seedlings with similar radicle length were transferred to soil in 96-well plastic flats and grown in the greenhouse on flooding tables.

Plant phenotyping

To quantify flowering time, the number of leaves before the emergence of a floral meristem was counted. A minimum of nine plants was included for each of the genotypes. Additionally, the number of seedlings in vegetative, transition and floral meristem stage was assessed for each of the genotypes 20 days after sowing in a minimum of 22 plants per genotype. To quantify the number of flowers per inflorescence, flowers and floral buds were counted on five inflorescences per plant (excluding the first inflorescences) and 8 replicate plants per genotype.

Inflorescence branching was quantified by counting the number of branch points in five inflorescences per plant and eight replicate plants per genotype. Whenever that number exceeded 22 branching points, the inflorescence was considered branched but “uncountable”. All phenotyping experiments were performed for all the genotypes in parallel on a single day.

Stereoscopy and photography

Seedlings grown in 96-well trays were harvested individually and all primordia except for the last two were removed by hand or with a needle. Z-stack images of developing meristems were acquired in different maturation stages spanning from vegetative to inflorescence using a Leica M205 FCA stereomicroscope with a Leica 10450028 objective connected to a Leica DFC7000T lamp (Leica Microsystems, Wetzlar, Germany). Photographs of inflorescences were taken using either an Olympus OM-D camera with an ED 14–42 mm f/3.5–5.6 EZ objective or an iPhone11 equipped with a 12 MP wide-angle camera.

CRISPR-Cas

CRISPR–Cas9 mutagenesis in tomato was performed as previously described [60–62]. Briefly, guide RNAs (gRNAs) were designed using CRISPOR [63] and the Sweet-100v.2.0 reference genome [42]. Binary vectors for CRISPR-Cas mutagenesis were assembled using the Golden Gate cloning system as previously described [60,64,65]. The gRNA target sequences (gRNAs) were amplified using KOD Hot Start DNA Polymerase (Merck Millipore) using target-specific forward primer, a universal reverse primer and the pICH86966::AtU6p::sgRNA vector (Addgene 46966) as template. The gRNAs were combined with the AtU6p promoter in Level1 vectors, and assembled with Nos::NptII::ocs (addgene #51144) and SlUbi::SpCas9-P2A-GFP::nos [60] in Level2 binary vectors. Plasmids were verified by Sanger sequencing and transformed into Agrobacterium tumefaciens (AGL-1) by electroporation using a Bio-Rad GenePulser II (1 mm cuvettes, 1.8 kV, 25 µF, 200 Ω, ~5 ms pulse). Constructs were transformed into the double-determinate cherry tomato cultivar Sweet-100 or M82 by Agrobacterium tumefaciens-mediated transformation as previously described. CRISPR–Cas9 editing was verified in the T0 generation by genotyping, Sanger sequencing or amplicon deep sequencing as described [60], and transgene-free mutant plants were isolated in the T1. J2, EJ2 and LIN were targeted simultaneously with a multiplex construct containing two gRNAs per gene. TM5 and TM29 were targeted with two independent constructs with two gRNA per gene, and mutations were combined by crossing T1 or T2 individuals. FA was targeted with three gRNA sequences. All gRNAs used in this study are listed in S12 Table.

Genotyping

Frozen leaf or cotyledon tissue was homogenized with two metal beads at a frequency of 18–20 Hz for 1 min in a mix mill. Genomic DNA was extracted using extraction buffer (0.1 M Tris-HCl pH 9.5, 0.25 M KCl, 0.01 M EDTA) by vigorous vortexing. Cell extracts were transferred to new multi-well plate, incubated at 95 °C for 10 min, and subsequently cooled at 4 °C for 5 min. Equal volume of 3% (w/v) BSA was added to the lysate, mixed vigorously by vortexing, and centrifuged at 3,700 rpm for 15 min. Target regions were amplified with gene-specific primers and products were separated on agarose gels. For CAPS analyses, products were digested using restriction enzymes (NEB). For Sanger sequencing, PCR products were purified with ExoSap (Thermo). All primer sequences and genotyping information are listed in S13 Table.

Amplicon deep sequencing

Genetic sector analyses were performed using amplicon deep sequencing as previously described [60,66]. Briefly, gDNA was extracted from individual flowers and gRNA target regions were amplified using gene-specific primers (S13 Table) with common adapter sequences. Gene-specific amplicons were diluted 10-fold in H₂O and used as template for the addition of sample-specific indexes using adapter primers with unique barcodes as described before [42,66]. Equal volumes of each indexed amplicon reaction were pooled and subsequently purified using SPRIselect beads (Beckman Coulter). A single Illumina sequencing library was prepared using the xGen DNA Library Prep MC Kit (Integrated DNA Technologies) and sequenced on the MiSeq System (Illumina) at the Genome Technologies Facility (GTF) at UNIL. Raw reads were analyzed using the S100 reference genome [42,60] and a previously published pipeline [60] based on Trimmomatic (v0.39) [67], FASTX-Toolkit (v0.0.14) [68], PEAR (v0.9.6), HISAT2 (v2.2.1) [69], SAMtools (v.1.17) [70], SMAP [71,72], and MAFFT (v7.505) [73], retaining haplotypes with a minimum frequency of 5%. Haplotypes were categorized in “WT” (no insertions or deletions), “in-frame” (total InDel size a multiple of three or zero), “out-of-frame” (total InDel size not a multiple of three nor zero). Haplotype frequencies and sequences were visualized in R [74,75] and Benchling [76], respectively.

Cis-regulatory sequence analyses

Accessible chromatin regions (ACRs) were identified using published ATAC-seq data [77]. Binding site for J2 were identified using published J2 ChIP-seq data [43]. Conserved non-coding sequences (CNS) were retrieved from the conservatory database [78]. MADS-box transcription factor binding sites were predicted on the SL4.0 reference genome [79] by FIMO/MEME Suite [80] using default parameters and the non-redundant set of profiles for MADS box factors from the JASPAR CORE database (release 2020) [45] (S14 Table). Data was visualized in jbrowse2 [81].

Meristem dissection, RNA extraction, library preparation and sequencing

Thirteen days after sowing, three representative seedlings per 96-well flat were dissected to determine the average developmental stage of the seedling population. When representative seedlings had reached the transition or floral meristem stage of meristem maturation [7], seedings with similar number of leaves were rapidly harvested starting at 13:00 hours for a maximum of 2 hours to minimize circadian effects. After removal of leaves, shoot apices were fixed in ice-cold acetone and vacuum infiltrated for 45 min at 4 °C. Meristems were micro-dissected using a sharp syringe needle under a Leica MS5 stereomicroscope connected to a Leica CLS 100× lamp (Leica Microsystems, Wetzlar, Germany). Dissected meristems of the same genotype and morphological stage were pooled in 2 ml round-bottom RNAse-free Eppendorf tubes (Eppendorf, Hamburg, Germany) filled with ice-cold acetone, and stored at −70 °C. RNA was extracted from pools of 18–38 meristems and three replicate pools per genotype-stage combination using the Arcturus PicoPure RNA isolation kit (Applied Biosystems) according to the manufacturer protocol with minor modifications. Volumes of extraction buffer and 70% EtOH were increased 2.2-fold and RNA was treated with DNAse to remove contaminating gDNA using the RNAse-Free DNase kit (Qiagen). Sequencing libraries were prepared from 300 ng total RNA as starting material at the Genomic Technologies Facility (GTF) of the University of Lausanne. Indexed libraries were prepared using the TruSeq Stranded mRNA Library Prep kit from Illumina according to the manufacturer’s instructions. Fragment size and concentration were assessed with a Bioanalyzer. Libraries were sequenced on two Illumina NovaSeq6000 lanes to produce between 15,819,728 and 22,075,610 SR100 reads per sample (S15 Table).

Read quality control and alignment

Read quality of raw reads was assessed using the FastQC (v0.11.7) [82]. Illumina TruSeq adaptors were trimmed with Trimmomatic (v0.36; with parameters SE -phred33 ILLUMINACLIP:TruSeq3-SE.fa:2:30:10 LEADING:3 TRAILING:3 SLIDINGWINDOW:4:15 MINLEN:36) [67]. Trimmed reads were aligned to the SL4.0 reference genome [79] using STAR (v2.7.8a; with parameters--runMode alignReads --outFilterType BySJout --outFilterMultimapNmax 20 --outMultimapperOrder Random --alignSJoverhangMin 8 --alignSJDBoverhangMin 1 --alignIntronMin 20 --alignIntronMax 1000000 --alignMatesGapMax 1000000 --outSAMtype BAM Unsorted) [83]. Alignments were sorted and indexed using SAMtools (v.1.17) [70] and gene expression was quantified as unique reads aligned to the SL4.0 gene annotation (ITAG4.0) [79] using HTSeq-count (v.0.11.2; with parameters --format = bam --order = pos --stranded = no --type = exon --idattr = Parent) [84].

Differential expression (DE) analyses and identification of non-additive expression pattern

Transcriptomic data were processed using the edgeR (v4.4.2) [85] and limma (v3.62.2) [86] packages for differential expression analysis. Synergistic expression patterns were identified following the analysis framework described by Schrode and colleagues [39]. In brief, lowly expressed genes were filtered retaining only those with ≥10 reads in at least 2 samples. Counts were normalized using the trimmed means of M-values (TMM) method and transformed into log2 counts per million (logCPM) with the voom function() which also estimates mean-variance relationships and assigns precision weights for downstream linear modelling [39,86]. For each genotype (e.g., WT, j2, ej2, j2ej2), a design matrix was constructed, and gene-wise linear models were fitted using the lmFit() function. For each set of singles and corresponding double mutant contrasts were defined as follows:

• Additive expectation = (j2 − WT) + (ej2 − WT)
• Combinatorial effect = (j2ej2 − WT)
• Synergy (interaction) = (j2ej2 − WT) − [(j2 − WT) + (ej2 − WT)]

For each gene, moderated t-statistics were calculated using the empirical Bayes procedure implemented in eBayes(), shrinking gene-wise variance estimates toward a common prior, stabilizing the standard errors (SE). Statistical significance was defined as genes with 95% confidence intervals excluding zero after Benjamini–Hochberg (BH) multiple-testing correction (FDR < 0.05). Each logFC estimate was tested as LogFC ± 1.96 × SE. Based on the calculated significance, genes were then classified as:

• Non-additive: when the synergy contrast was significant (FDR < 0.05). It can be:
  • Synergistic: when the double mutant effect exceeded the additive expectation in the same direction (i.e., more up or more down than predicted).
  • Less-than-additive: when the double mutant effect was weaker than expected.
• Additive when the double mutant was significantly different from control (FDR < 0.05) but the synergy contrast was not significant.

To further evaluate the robustness of synergy detection, the proportion of non-null hypotheses (π₁) across all synergy tests was estimated using the Storey–Tibshirani qvalue method [87]. Median standard error (SE) of contrasts was used to estimate the probability of detecting deviations from additivity at varying effect sizes and sample numbers [39]. Differentially expressed genes (DEGs) per genotype (single versus control or double versus control) were determined with a threshold of |log2FC| ≥ 0.58 (~1.5-fold change) and FDR < 0.01.

Multidimensional scaling (MDS) analyses were conducted using CPM values on: all expressed genes, 2,583 genes dynamically expressed during meristem maturation [15] and in 227 transition marker and 241 floral meristem marker genes, in the transition and floral stage datasets, respectively [14]. The samples linej2_TM_3 and M82_FM_1 were excluded from the differential gene expression analyses due to outlier behavior in the transition marker and floral marker MDS analyses (S2 Fig). Gene-normalized z-scores were calculated with the formula z = (x − μ)/σ, where x is the CPM value for a specific gene, μ is the mean of CPM values across samples, and σ is the standard deviation. All data visualization and statistical analyses were conducted in R [74,75]. Gene-normalized z-scores were plotted as heatmaps using pheatmap (v.1.10.12) [88] and dotplots using ggplot2 package (v3.5.1) [89]. Counts per million (CPM) were plotted in dotplots using ggplot2 (v3.5.1) [89]. Overlaps of DEGs were visualized using the eulerr package (v7.0.2) [90].

RNA-seq analysis sep1234q, fa, an, and sep4t inflorescence tissue

Meristem tissue from sep1234q and fa mutants for mRNA extraction was collected at 13:00 hrs from mature plants grown in a greenhouse. Per genotype, three individual plants served as biological replicates, three inflorescences (~5 mm) were pooled per replicate, and flash-frozen in liquid nitrogen. Total RNA was extracted using the RNeasy Plant Mini Kit (QIAGEN) following the manufacturer’s protocol and assessed for quality using a Bioanalyzer. Sequencing libraries were prepared from 100 ng of total RNA at the Genomic Technologies Facility (GTF), University of Lausanne. In brief, indexed libraries were generated using the WATCHMAKER mRNA Library Prep Kit (Watchmaker Genomics) according to the manufacturer’s instructions, and library fragment size distribution and concentration were evaluated using a Bioanalyzer. Libraries were sequenced on two Element AVITI lanes, yielding between 17,145,097 and 20,335,750 paired-end 150 bp reads per sample. Meristem tissue from sep4-triple (sep4t; j2 ej2 lin) and an mutants for mRNA extraction was collected at 12:00 hrs from mature plants grown in a greenhouse. Per genotype, two individual plants served as biological replicates, three inflorescences (~5 mm) were pooled per replicate, and flash-frozen in liquid nitrogen. Total RNA was extracted using the RNeasy Plant Mini Kit (QIAGEN) following the manufacturer’s protocol and assessed for quality using a Bioanalyzer 2100 (Agilent). Libraries were prepared from 100 ng of total RNA at the Genome Center of Cold Spring Harbor Laboratories, Cold Spring Harbor, using the Kapa mRNA HyperPrep Kit (Kapa Biosystems) according to the manufacturer’s instructions. Libraries were sequenced on one NextSeq500 lane, yielding between 22′916′312 and 25′812′432 paired-end 75 bp reads per sample.

Read quality controls, alignments and counts were produced as described in the previous section. Lowly expressed genes were discarded by retaining only genes with ≥10 counts in at least 3 samples. Differential expression analysis and variance-stabilizing transformation (VST) was conducted in DESeq2 (v1.46.0) [91]. The transformed expression matrix was corrected for batch effects (WT samples [7] versus mutant (sep1234q and fa) samples) using the removeBatchEffect function from the limma (v3.62.2) [86]. Estimation of developmental stage was conducted using real-age prediction from transcriptome staging on reference (RAPToR, v1.2.0) [52]. A reference model was built on the staged WT meristem data using the functions ge_im() (formula = X ~ s(time, bs = “ts”, k = 5), nc = 3) and make_ref() (n.inter = 100). The WT reference model was used to stage mutant samples using the ae() function. Correlation profiles and stage estimates were visualized with plot() and plot_cor(), respectively, and mean stage estimates in ggplot2 (v3.5.2).

Gene ontology enrichment analysis

Gene ontology enrichment analyses for biological processes were performed among differentially expressed genes using a functional annotation database for the ITAG4.0 tomato genome from PLAZA [92] and the ClusterProfiler package [93] with a pvalueCutoff = 0.05 and pAdjustMethod = BH. All annotated genes in the tomato genome (ITAG4.0) were used as background.

Phylogenetic analysis

To reconstruct the phylogenetic tree of the SEP family across angiosperm species, a total of 22 proteomes were collected from the Orthologous MAtrix (OMA; release 12/2024) [48] database and the Conservatory CNS project [78] (S10 Table). Splice variants were removed to only retain primary protein isoforms and reduce redundancy. Hierarchical Orthologous Groups (HOGs, gene families) were obtained using FastOMA (v1.2.0) [94] and a species tree from the NCBI Taxonomy Database [95] as the guide phylogeny (S2 Data). The SEP and AGL6 gene families were identified using the proteins from S. lycopersicum and A. thaliana (S9 Table). The 135 protein sequences of this gene family were aligned with MAFFT (v7.505) [73], MUSCLE (v3.8.1551) [96] and Kalign (v3.3.2) [97] using default parameters. Resulting alignments were combined to a consistency-based alignment using T-Coffee (v13.46.0.919e8c6b) [98,99]. The maximum likelihood (ML) tree was inferred using IQ-TREE (v1.6.12) with 1,000 ultrafast bootstrap replicates [100]. This tree was pruned using the ETE4 (Environment for Tree Exploration) [101] toolkit to include only representatives of the Solanaceae family and Arabidopsis. To identify gene duplications, the gene trees were reconciled with the species tree via the species-overlap method [102] using the ETE4 toolkit.