Transcriptome analysis of gynoecium morphogenesis uncovers the chronology of gene regulatory network activity

Abstract

The gynoecium is the most complex organ formed by the flowering plants. It encloses the ovules, provides a surface for pollen contact and self-incompatibility reactions, allows pollen tube growth, and, post fertilization, develops into the fruit. Consequently, the regulation of gynoecium morphogenesis is complex and appropriate timing of this process in part determines reproductive success. However, little is known about the global control of gynoecium development, even though many regulatory genes have been characterized. Here, we characterized dynamic gene expression changes using laser-microdissected gynoecium tissue from four developmental stages in Arabidopsis. We provide a high-resolution map of global expression dynamics during gynoecium morphogenesis and link these to the gynoecium interactome. We reveal groups of genes acting together early and others acting late in morphogenesis. Clustering of co-expressed genes enables comparisons between the leaf, shoot apex, and gynoecium transcriptomes, allowing the dissection of common and distinct regulators. Furthermore, our results lead to the discovery of genes with putative transcription factor activity (B3LF1, -2, DOFLF1), which, when mutated, lead to impaired gynoecium expansion, illustrating that global transcriptome analyses reveal yet unknown developmental regulators. Our data show that genes encoding highly interacting proteins, such as SEPALLATA3, AGAMOUS, and TOPLESS, are expressed evenly during development but switch interactors over time, whereas stage-specific proteins tend to have fewer interactors. Our analysis connects specific transcriptional regulator activities, protein interactions, and underlying metabolic processes, contributing toward a dynamic network model for gynoecium development.

High-resolution temporal transcript profiling during Arabidopsis thaliana gynoecium morphogenesis reveals the chronology of gene regulatory network activity and identifies developmental regulators.

Introduction

Broadening our understanding of flower development is important as most of the terrestrial life is either directly or indirectly dependent on flowering plants (Sauquet et al., 2017) and agricultural advancements are required to feed a growing global population. Carpels, the female reproductive organs of the flowering plants, begin to develop after the plant has reached its generative maturity and flowering is initiated. Carpels are located in the innermost whorl of the flower and their sum is defined as gynoecium. The gynoecium bears the developing ovules, receives pollen grains, and allows their passage through specialized tissue to enable fertilization of the ovules. Subsequently, these develop into seeds while the gynoecium is converted into a fruit.

In Arabidopsis thaliana, flowers arise on the flanks of the inflorescence meristem. The flower consists of four concentric whorls of different organs: the outermost are sepals followed by petals, stamina, and the gynoecium formed in the center. Gynoecium development commences approximately 4 d after floral development initiation when the previously undifferentiated central dome in the middle of the flower starts to elongate and forms a hollow, oval shape. This tube-like gynoecium consists of two congenitally fused carpels (Smyth et al., 1990). Inside the gynoecium, the carpel margin meristem (CMM) initiates as the inner adaxial margins first bulge inward forming a boundary surface inside the hollow structure. The CMM then gives rise to the carpel marginal tissues from where placenta, ovules, false septum, and transmitting tract form (Bowman et al., 1999; Reyes-Olalde et al., 2013; Reyes-Olalde and de Folter, 2019). The septa primordia fuse and form the false septum through postgenital fusion. After approximately 11 d of flower development, stigmatic papillae start to appear at the tip of the developing organ. One day later the papillae fully cover the tip of the gynoecium and the open-ended structure closes by postgenital fusion, while style and transmitting tract differentiate, leading to the mature gynoecium (Smyth et al., 1990).

The initiation of the gynoecium requires activation of the class C and E homeotic genes AGAMOUS (AG) and SEPALLATA3 (SEP3; Bowman et al., 1989; Honma and Goto, 2001; Pelaz et al., 2000). These proteins form a tetramer protein complex with the active sites binding to a plethora of promoter regions in the Arabidopsis genome regulating the expression of the downstream genes to provide carpel organ identity and initiate carpel development (Theissen and Saedler, 2001; Sablowski, 2010; Smaczniak et al., 2017).

Post initiation, the dome-shaped floral meristem differentiates into several tissue types. These require specification and orientation toward the adaxial/abaxial and apical/basal axes, processes controlled by transcriptional regulators (TRs) such as PHABULOSA (PHB), REVOLUTA (REV), PHAVOLUTA (PHV), NUBBIN (NUB), and JAGGED (JAG; McConnell et al., 2001; Bowman et al., 2002; Dinneny et al., 2006). Induction and differentiation of the CMM tissues is regulated by TRs including SPATULA (SPT), CUP-SHAPED COTYLEDON1-2 (CUC1-2), HECATE1-3 (HEC1-3), and INDEHISCENT (IND; Heisler et al., 2001; Aida and Tasaka, 2006; Gremski et al., 2007; Kay et al., 2013) and differentiation of stigma and style by those such as NGATHA3 (NGA3), STY1, and STY2 (Sessions and Zambryski, 1995; Kuusk et al., 2002; Trigueros et al., 2009). A complex interplay of many additional genes, phytohormones, peptides, microRNAs, and epigenetic factors ultimately lead to the complete organogenesis of the gynoecium (reviewed in further detail in Alvarez-Buylla et al., 2010; Krishnamurthy and Bahadur, 2015; Wils and Kaufmann, 2017; Moubayidin and Østergaard, 2017).

Whereas genetic and protein interactions of many of the TRs coordinating carpel development are known, in addition to the genes directing the development of the other whorls of the flower (reviewed in Ferrándiz et al., 1999; Reyes-Olalde et al., 2013; Chávez Montes et al., 2015; Chen et al., 2018; Zúñiga-Mayo et al., 2019; Becker, 2020), we lack a comprehensive picture of expression dynamics of these TRs during carpel development. So far, the major transcriptomic studies of flower development in A. thaliana have focused on either the later stages of the developed flower organs (Klepikova et al., 2016) or complete buds at early to late stages (Mantegazza et al., 2014; Ryan et al., 2015). Here, we provide a high-resolution temporal transcription time-scale map of gynoecium development in A. thaliana. Our data are based on laser-microdissection (LMD) with subsequent RNAseq analysis of four different stages of carpel development starting from the initiation of carpel development to maturation, excluding the ovules. We show that specific genetic modules exist in a temporally precisely regulated manner and identify consecutively acting protein interaction networks key to gynoecium development. Further, we identify four putative transcription factors (B3 LESSER FERTILITY1-2, B3LF1, -2; and DOF LESSER FERTILITY, DOFLF1) based on their specific temporal expression during gynoecium development and show that they contribute to gynoecium longitudinal growth and seed formation.

Results

Arabidopsis transcriptome data of four stages of carpel development

We sequenced laser-microdissected Arabidopsis carpel RNA samples at four different developmental stages: S1, initiation of carpel development after the differentiation of the central dome corresponding to stage 5 of A. thaliana flower development (Smyth et al., 1990); S2, elongation of carpel walls (stage 9); S3, during the female meiosis (stage 11, Smyth et al., 1990; Armstrong and Jones, 2001); and S4, between female meiosis and anthesis (stage 12). Sample preparation, RNA-seq, transcriptome assembly, and quality control are described elsewhere (Kivivirta et al., 2019). Four biological replicates were sequenced for all the four developmental stages and three were used for this analysis. 33 Mio paired-end reads were sequenced with read length of approximately 76 bp and annotated, resulting in expression information of all A. thaliana genes during gynoecium development. Supplemental Figure S1 shows examples of sample preparation with LMD and the FastQ sequence read counts of the sequenced libraries. A comparison of the four carpel transcriptomes to previously published data of complete flower buds (Klepikova et al., 2016; Chen et al., 2018) showed a highly similar but not set of genes (Supplemental Figure S2). A comparison of transcriptome data of carpel development with the whole flower buds 6–18 of Klepikova et al. (2016) showed over 13,000 in genes in the datasets. However, we found between 84 (S1) to 241 (S3) genes only in our datasets, indicating a higher sequencing depth or specificity. Among the carpel genes unique to our datasets were those such as IND, NGA2, NGA4, BEE3, and SHY. The genes not present in our carpel datasets were KNU, CKX3 and -5, and SUP (Supplemental Figure S2). A comparative analysis between the four carpel transcriptomes and complete bud transcriptomes (Chen et al., 2018) at stages 4 and 8 (Smyth et al., 1990) also revealed a high number (>13,000) of shared genes; however, we also found a high number of genes (1,931) absent in early stage carpel transcriptomes, likely regulating the development of the other whorls of Arabidopsis flowers. 604 genes unique to our early carpel transcriptomes were found, including genes like HEC1 and -2, SHP1 and -2, and NGA4. These comparisons suggest that organ-specific expression profiling indeed provides more specific datasets.

Expression dynamics of carpel developmental regulators

We were interested in the temporal expression profiles of known carpel developmental genes to learn if the timing of their expression matches with their known role in development. We analyzed carpel regulatory genes by generating an expression heatmap (Figure 1 and Supplemental Table S1). Expression strength of known carpel regulatory genes at each stage ranged from completely silent (e.g. CLV3 and CRC at S4) to approximately 300–500 transcripts per million (TPM; e.g. SEP3 and CRC at S1). Among the genes most important for floral organ identity, initiation, and maintenance were the MIKC MADS-box transcription factors SEP1–4 and AG (Figure 1 and Supplemental Figure S3). Whereas SEP1–4 showed strong differences in expression dynamics, AG was expressed evenly at a low level throughout the developmental stages. The APETALA3 (AP3) and PISTILLATA (PI) genes required for stamen and petal, but not gynoecium, organ identity exhibited expression in the first two stages of gynoecium development, confirming earlier observations (Goto and Meyerowitz, 1994). Interestingly, some late acting MADS-box genes required for fruit dehiscence, such FRUITFULL (FUL) and SHATTERPROOF2 (SHP2), were expressed throughout gynoecium development. The data obtained for MADS box genes are corroborated by earlier work, for example by Parenicová et al. (2003).

Figure 1.

The carpel regulome heatmap. Heatmap of carpel developmental genes illustrating the strength of gene expression during four developmental stages; strongly expressed genes are bright yellow and weakly expressed genes are in dark blue. Similarly expressed genes were clustered with Euclidean distance for the absolute values of gene expression. A, Heatmap of MIKC type MADS-box genes transcriptionally active in the carpel (the full set is available in Supplemental Figure S3). B, Genes involved in phytohormone signaling, homeostasis, perception, or biosynthesis, or related to phytohormone pathways, with known regulatory functions during carpel development. C, A collection of other regulatory genes required for carpel development. The table to the right of the heatmap indicates the gene’s contribution to carpel developmental regulation: organ identity, development of stigma/style, apical/basal and adaxial/abaxial patterning, and CMM development (references for gene functions in Supplemental Table S1) and, in B and C, their gene family membership. Genes with average expression of TPM <1 were omitted from the analysis.

Hormonal signaling is an integral part of carpel development, with crucial functions for signaling pathways such as auxin and cytokinin, but also others like brassinosteroids and gibberellins (Marsch-Martinez and de Folter, 2016; Zúñiga-Mayo et al., 2019). We observed expression of many of the genes and transcription factors related to these hormonal pathways (Figure 1). The genes that presented the highest expression levels were those involved in auxin and cytokinin regulation and response, auxin biosynthesis, and brassinosteroid regulation.

Genes related to different steps in the auxin pathway were identified, such as those coding for TAA and YUC (biosynthesis); PIN1, PIN3, and PIN7 (transport); PID (transporter regulation); and response factors including ARF5/MONOPTEROS, ARF3/ETTIN, ARF6, and ARF8. Also, transcription factors such as ANT and AIL6, among others, which are closely related to the auxin pathway, were found in the transcriptome data (Krizek, 2009). Moreover, we observed various transcription factors well known for their regulatory role in carpel development (Figure 1) that also affect auxin signaling, such as STY1, STY2, NGA, SPT, and CRABS CLAW (CRC), or that respond to auxin, such as the CUC1–3 genes.

Genes related to the cytokinin pathway include those encoding the response regulators ARR1, ARR10, and ARR12, and the cytokinin degradation enzymes CKX3 and CKX5. All these genes have been reported to be expressed during gynoecium development, particularly in meristematic tissues. Mutations in these genes cause reduced or increased meristematic activity, respectively (Bartrina et al., 2011; Reyes-Olalde et al., 2017). Also, gene encoding transcription factors such as the KNOX family members STM, BP, and KNAT2, and TCP14 and TCP15 were expressed at different stages, which play important roles in gynoecium development and have been associated with the cytokinin pathway (Lucero et al., 2015).

Brassinosteroids also play important roles in gynoecium development. In the transcriptome data, the brassinosteroid-related genes HALF FILLED (HAF/CES) and BEE1–3 were also expressed, especially at the intermediate and late stages of development. This is in line with their function in transmitting tract development later during gynoecium development and previously published expression data (Crawford and Yanofsky, 2011).

Gibberellins have recently been implicated in the negative modulation of ovule number (Gomez et al., 2018; Barro-Trastoy et al., 2020). DELLA proteins are negative regulators of gibberellin signaling, and their activity correlates positively with ovule number. Genes encoding DELLA proteins, such as GAI, RGA, and RGL2, were also found in the transcriptomes. Of these, only RGA was strongly expressed in the later stages, whereas the other showed weak expression, decreasing in time.

Some of the genes in the transcriptomes take part in networks that connect different pathways. For example, HEC1-3 induces auxin signaling and represses cytokinin signaling in the style (Schuster et al., 2015). Another example is SPT, which, besides inducing auxin biosynthesis and transport, activates the cytokinin response regulator ARR1 that in turn activates auxin biosynthesis and transport (Reyes-Olalde et al., 2017).

Chromatin remodeling is an essential component of plant development (Ojolo et al., 2018) but its involvement in gynoecium development has received little attention. We were thus interested in exploring whether known chromatin remodelers are differentially expressed during gynoecium development (Supplemental Figure S3). HISTONE DEACETYLASES1 and -2 (HDA1/2) were strongly expressed during gynoecium development whereas HDA3 showed only little expression. ACTIN-RELATED PROTEIN4 (ARP4), BRAHMA (BRM), SPLAYED (SYD), and CHC1/SWP73B showed expression largely restricted to the latter two stages. By contrast, GIF1/AN3 expression was mainly confined to the two early stages.

Several other TRs, which are not members of MADS-box genes, chromatin remodelers, or phytohormone-associated, contribute essential functions to carpel morphogenesis (Figure 1). Among these, CRC, FILAMENTOUS FLOWER (FIL), AS1, LUG, SEUSS (SEU), SEUSS-like2 (SLK2), LEUNIG-HOMOLOG (LUH), PHB, and ALC were most strongly expressed. By contrast, many other important regulators, such as CUC1, CUC2, or WUS were expressed at lower levels, suggesting that even genes expressed at low level may have a profound impact on gynoecium development.

In summary, our high-resolution data confirm previously reported expression data for individual genes and show differentiation of expression of regulatory genes, even between closely related homologs, such as SHP1 and SHP2 or the SEP1–4 genes. Moreover, we can now identify temporal changes in regulatory gene activation during gynoecium development.

Temporal dynamics of protein interactions

TRs often interact in dimers or higher order multimers. For A. thaliana gynoecium development, many protein interactions of TRs have been identified. However, we were interested in the temporal dynamics of these protein interactions. Thus, a comprehensive carpel protein interactome was generated based on protein interactions previously published as verified by Yeast Two-Hybrid (Y2H), Bimolecular Fluorescence Complementation (BiFC), and/or Co-Immunoprecipitation (Co-IP) analyses (Figure 2 and Supplemental Table S2 for references to the individual interactions). We overlaid these interactions with expression data to illustrate the transient nature of some gynoecium TR interactions.

Figure 2.

Temporal dynamics of the carpel regulatory interactome. A, Interaction map illustrating protein–protein interactions and homodimerization of important carpel developmental regulators based on experimentally verified interactions (Franz et al., 2018; Supplemental Table S2). An expression map (A) with the node color indicating the trend of expression (logarithmic change of expression values) through the four carpel developmental stages (blue–yellow–red): blue node color indicates decreasing expression during development, yellow indicates stable expression, and red indicates an increase in expression strength through the four stages. Circles indicate membership of interacting proteins to larger transcription factor families. B, Illustration of the most interactive hub proteins and their interaction partners. C and D, Expression data showing only highly expressed interactions (TPM >10) in developmental stages S1 (C) and S4 (D). The color coding indicates the strength of expression; darker red nodes indicate strong expression, light red nodes indicate weak expression. The nodes with a blue frame show protein interaction partners unique to the respective stage. Proteins without experimentally verified interactions with TRs related to gynoecium development were omitted from the interaction maps.

Figure 2 shows the contribution of single proteins and TR families to the carpel interactome. A group of several MADS-box proteins formed a highly interactive cluster, as did the bHLH, B3, and homeodomain transcription factor families. These families showed different levels of connectivity among each other and with regulators outside of their family: The MADS-box proteins were highly connected to each other but interacted with only five unrelated proteins. By contrast, the homeodomain proteins were less connected within their family, but interacted with nine proteins outside their family.

Several hub proteins with five or more interactions were identified from the network analysis (Figure 2): the bHLH protein SPT; the B3 AUXIN RESPONSE FACTOR6 (ARF6); the transcriptional repressor INDOLE-3-ACETIC ACID INDUCIBLE 27 (IAA27) of the AUXIN/INDOLE-3-ACETIC ACID protein family; the WD40 transcriptional corepressor TOPLESS (TPL); and the homeodomain proteins BELL1 (BEL1), KNAT1/BREVIPEDICELLUS, REPLUMLESS (RPL), and BEL1-LIKE HOMEODOMAIN9 (BLH9). Moreover, the MADS-box proteins AG, PI, AP3, SEP1–3, AGAMOUS-LIKE6 (AGL6), FUL, SHP1, and APETALA1 (AP1) acted as hubs. Interestingly, the majority of hub protein-encoding genes (BLH9, TPL, SPT, AGL6, AP1, SEP1, AG, and AP3) were expressed (TPM >10) throughout all stages (Supplemental Table S1). Only PI was expressed strongly exclusively in early developmental stages (Figure 2), and also SHP1 and SEP3 were expressed mainly in late developmental stages (Figure 2).

Many of the interacting hub proteins were generally rather strongly expressed (e.g. SEP3 with TPM peak 297, SEP1 with 190, and ARF6 with 246), but their dynamics and interactivity changed during development (Figure 2). The MADS box protein-encoding genes SEP4 and PI especially were expressed in S1 but upon gynoecium maturation, their expression was reduced and other genes encoding highly interacting proteins like SEP3, SHP1, SHP2, STK, and FUL showed increased expression.

Interestingly, not only hub genes, but also some proteins with few interactors, showed stable expression throughout carpel development (Figure 2). The expression of each member in the cluster of the interacting proteins NGA2-TPL-AP2-AS1-IAA27-ARF5-ARF6-ARF8 remained remarkably stable (Figure 2). However, this cluster was complemented by the interaction of TPL with a complex of KAN1, KAN2, and ULT1 at S1, which is not found in S4. Conversely, in S4, interactions of HAT1/JAIBA, NGA1, and ARF31 with TPL and BEE2 with ARF6 were established.

In addition, the networks of TCP15-TCP14-SMU1, SLK1-LUH-SLK2, and LUG-SEU-SEP3-PKL-FUL-SHP2-SEP1-AG were stable throughout carpel development. Interestingly, other proteins also supplemented these stable networks in different stages: The SLK1-LUH-SLK2 network is connected to CUC1/2/3 in S1 and exchanges this connection with PIF1 in S4. The LUG-SEU-SEP3-PKL-FUL-SHP2-SEP1-SEP2-AG network is modified by the addition of SEP4, AP3, and PI in S1 and by STK, and BEL1 in S4.

The interactome of S1 of carpel development includes 12 stage-specific proteins, whereas the S4 interactome includes 9 stage-specific proteins, and 37 proteins participate in the interactomes of both stages. This suggests that initiation and early morphogenesis of the carpel require more TR interactions than the later stages, when tissue differentiation is completed.

Another interesting group of proteins included hub proteins (Figure 2) that have interaction partners with a generally or temporally very low level of expression. For example, BEL1 interacts with KNAT2, but KNAT2 expression was at a low level and at different stages, such that chances are high that the proteins never meet in the gynoecium. The same may apply to interactions with AGL6 and AP1, as the former has 6 and the latter 13 protein interactors, but they were hardly expressed in the gynoecium (Figures 1, 2). Similar scenarios apply to KNAT1, SPT, and IAA27, which were all expressed at a low level. An extreme example is RPL, which has six interaction partners but was expressed at a very low level throughout gynoecium development and may be active mainly after fertilization during fruit development. However, an analysis of longevity of the proteins is beyond the scope of this manuscript, but would be required to clarify if these protein interactions are of biological relevance.

In summary, protein interactions directing carpel morphogenesis are temporally very dynamic. Only a few hub proteins maintain a high number of interactions throughout carpel development, such as SEP3, AG, and SEP1. Some components of the network, such as the one centered on TPL, were shown to be active throughout carpel development but change few interacting partners during morphogenesis. Further, differences in connectivity between transcription factor families were observed: whereas MADS-box proteins mainly interacted among themselves, the bHLH family proteins were highly connected with members of other TF families.

Co-expression analyses provide comprehensive information on expression patterns and resulting shifts in biological processes

We were then interested to identify genes that were co-regulated with the previously described carpel regulators to identify clusters of co-expressed and possibly co-regulated genes. Further, we aimed to learn if the carpel transcriptomes share more similarity with those of the leaf or shoot apical meristem (SAM). Automatically partitioned clusters were generated to visualize co-expressed genes (Figure 3, for the full list of clusters and genes see Supplemental Table S3) within the four carpel development stages in comparison to leaf and SAM (RNAseq data of leaf and SAM tissue were obtained from Klepikova et al., 2016).

Figure 3.

Clusters of co-expressed genes during gynoecium development. The strength of expression (Y-axis) is illustrated for the different tissues and developmental stages of gynoecium development (X-axis). The transcriptome files include the leaf blade (L), SAM (S) (from Klepikova et al., 2016), and four stages of gynoecium development (S1, S2, S3, S4, this study). Examples of known regulators are shown below each cluster. For the complete list of genes and GO analyses, see Supplemental Table S3.

The largest cluster consisted of genes exclusively upregulated throughout gynoecium development (C1: 2,847 genes) and included several well-known gynoecium developmental regulators such as AG, SEP1–4, SHP1 and -2, SEU, SLK1-3, CES, LUG, LUH, HAF, and FUL. The second largest cluster (C2: 2,570 genes) included genes that are down regulated during gynoecium development when compared with SAM or leaf tissue. Cluster C3 (1060 genes) included genes with putative roles in both SAM and gynoecium development, such as ALCATRAZ (ALC), ARF5, DORNROSCHEN-LIKE (DRNL), HEC3, NTT, SQN, ULTRAPETALA (ULT), BEL1, and JAG. Cluster C8 included 628 genes highly expressed in the leaf and gynoecium, but downregulated in the SAM, with CNA being the only known carpel regulator member. The cluster containing genes with SAM-only expression (C7) was surprisingly small with only 758 genes, as was the cluster C10 combining all 393 genes strongly expressed only in the last two stages of gynoecium development. Genes with high expression in the first two gynoecium development stages and lower expression in leaf and SAM tissues were contained in cluster C9 (550 genes) and included PIN1, PID, FIL, ANT, ETT, and ARF4.

Next, we were interested in the biological processes reflected by the clusters and identified overrepresented GO terms (Figure 3 and Supplemental Table S3). Cluster C1 including genes upregulated throughout gynoecium development showed enriched terms related to metabolic and transcriptional regulation, whereas the contrasting cluster C2 showed enriched terms related to general biosynthesis and metabolism. Metabolic processes were depleted in C3, a cluster similar to C1, but with the weakest expression in leaves. GO terms related to metabolic processes were also depleted in C6 that contained genes with the highest expression in S3 and S4, suggesting weaker metabolic activity during gynoecium development compared with the SAM and leaf tissues. In C6, terms related to fertilization and zygotic development were enriched. Cluster C9 including genes highly expressed during S1 and S2 showed enriched terms related to the cell cycle and nucleic acid metabolism. Cluster C10 including genes that are nearly exclusively expressed during S3 and S4 showed enriched terms related to the import of nutrients, mainly sugars.

We were subsequently interested to see which processes change during gynoecium development and how the gynoecium differs from leaf and SAM tissues. We compared co-expression clusters with contrasting patterns (Figure 4) to elucidate the differences in enriched GO-terms between the set of genes expressed in the carpel when compared with other tissues, as well as between early (S1, S2) and late (S3, S4) carpel development (for the complete analysis, see Supplemental Table S3). GO terms related to hormone response were overrepresented only in late carpel development stages (Figure 4). Cell cycle-related genes were underrepresented in cluster C2 but highly overrepresented in those genes upregulated in early stages of carpel development. Photosynthesis-related genes were overrepresented in C2 and in late stages of carpel development, whereas they were under-represented in cluster C1 and early carpel development. RNA-splicing-related genes were overrepresented in cluster C1 and early carpel development suggesting that differential splicing may play a role in carpel morphogenesis. Genes involved in the regulation of gene expression were overrepresented throughout carpel development as were floral organ development genes.

Figure 4.

Over-representation of GO terms in clustered co-expression data. Co-expression clusters of the four carpel developmental stages with (blue clusters) and without (yellow and green clusters) expression data from SAM and leaf tissue of contrasting patterns shown to the left. Log2-fold enrichments of relevant GO terms, representing larger sets of semantically similar terms, are shown on the right (*P-value < 0.005 − 10⁻¹⁰, **P-value < 10⁻¹⁰).

In summary, our data show a succession of events, starting from upregulation of photosynthesis and downregulation of cell cycle activity in leaf and SAM. In early stages of carpel development, photosynthesis seems to play no major role but genes involved in cell cycle, regulation of gene expression, and floral development are upregulated. In late stages, phytohormone response and photosynthesis-related genes are upregulated.

Digital gene expression approaches can identify developmental regulators

Transcriptome analysis is a useful tool to clarify co-expression of gene clusters and networks, but we were interested to know if it could also identify genes of hitherto unknown function that can be assigned as gynoecium developmental regulators. As proof of concept, seven genes with specific expression patterns were selected for reverse genetic analysis. Whereas three SALK insertion lines showed no obvious fertility defects, four had significantly decreased fertility (Figure 5). For three of these lines, genotyping by sequencing showed single T-DNA insertions in the target genes. The associated loci were named B3LF1, -2, for B3 LESSER FERTILITY 1 and 2, and DOFLF1 DOF LESSER FERTILITY 1 (Figure 5, A). doflf1 has a single T-DNA insertion in the only exon of a DOF binding transcription factor-encoding gene (AT5G66940, also named ATDOF5.8) and, as B3LF1, was restricted in its expression to the first two developmental stages. B3LF1 encodes an AP2/B3 transcription factor (AT3G17010, also named REM22) with high S1 and moderate S2 expression and the single T-DNA insertion in b3lf1 is located in the second exon. B3LF2 also codes for an AP2/B3 transcription factor (AT3G46770) and was strongly expressed in S2 and S3 and the insertion is in the first exon. The siliques of b3lf1, -2, and doflf1 were 9.4%–16.6% shorter and with 5.7%–20.3% fewer seeds than the wild type (Figure 5), all significantly different from Col-0. We were then interested if the B3LF1, -2, and DOFLF1 genes are integrated in the regulatory network shown in Figure 2 and searched the upstream regions for transcription factor binding sites identified by a scan of binding site motifs in the promoter regions (Figure 5). Each gene is most likely regulated by one MADS-box protein complex including AG and at least two MADS-box protein complexes bind to each promoter, suggesting that the B3LF1, -2, and DOFLF1 genes are under direct control of floral homeotic protein complexes. Genotyping by sequencing (Supplemental Table S4) revealed homozygous insertions in target genes of B3LF1, -2, and DOFLF1. In addition, we measured significantly lower silique size and seed number in SALK_063793C in a B3 domain transcription factor (At5g46915). This mutant showed several T-DNA insertions making it difficult to pinpoint the cause for the phenotype.

Figure 5.

Candidate mutant analysis. A, An overview of the characterized SALK insertion lines, gene families, and expression. B, Intron–exon structure of B3LF1, B3LF2, and DOFLF1 genes and locations of T-DNA insertion in each SALK-line as red arrow. C, Silique length and seed counts for all seven SALK lines (n = 30 siliques). Given are the means and the error bars indicated standard deviation. A star denotes significant difference of α > 0.05 to Col-0. D, Phenotypes of Col-0 wild type and the b3lf1, b3lf2, and doflf1 mutants showing significant defects in silique length and seed number. Siliques were digitally extracted for comparison. E, B3LF1, B3LF2, and DOFLF1 locus promoter analysis of TF-binding motifs (http://www.chip-hub.org/). Y-axis labeling: find individual motif occurrences (FIMOs) are calculated by a log-likelihood ratio score for each position of the binding motif in the sequence database and includes Q-values for each position of putative binding sites by false discovery rate analysis. Shown are binding sites for TRs 1.5-kb upstream and highlighted genes with known regulatory functions in flower development. Areas of high binding activity in the promoter region are indicated as dark, high bars.

Discussion

Here, we use LMD RNAseq to generate expression data with high temporal resolution to resolve global transcriptional dynamics specific to gynoecium development. However, high specificity in transcriptome analysis is often achieved at the expense of sensitivity (probability of representing a particular transcript in the library), accuracy (how well the read quantification corresponds to actual mRNA concentration), and precision (technical variation of the quantification; Ziegenhain et al., 2017). Here, we can show that transcription factor genes with known low levels of expression, such as NGA2, NGA3, NGA4, and HEC1 (with RPKM of 14, 8, 7, and 1, respectively; Klepikova et al., 2016) are detected by our approach and show TPM values of 29, 9, 2, and 6, respectively (Supplemental Table S1). These numbers compare well in magnitude with the RPKM values taken from Klepikova et al. (2016), demonstrating a high level of accuracy. A comparison was made between of the four carpel transcriptomes and previously published data of complete buds unique genes for all transcriptomes (Supplemental Figure S2). However, the unique genes from Klepikova et al. (2016) or Chen et al. (2018) did not include any known gynoecium developmental regulators. Conversely, among the unique genes from the microdissected carpel transcriptomes were several known carpel regulators, showing that our approach provides a better sequencing depth suited to find even those transcription factors expressed at a very low level. Interestingly, genes expressed at a low level, such as DOFLF1 with a maximum TPM value of 19, may also be associated with severe mutant phenotypes (Figure 5).

Distance correlation analysis between biological replicates analyzed in this study shows that the transcriptomes of the two early and the two late stages are clearly distinct (Kivivirta et al., 2019). However, of the four LMD RNAseq replicates that were analyzed for each stage, three clustered closely together and those were used for the analysis. Most likely, we have reached the morphological and genetic limit of differentiation between stages, and more fine-grained analysis by LMD would be sub-optimal in terms of accuracy. Single-cell transcriptome analyses would be more suitable to, e.g. identify transcripts of a specific, small-scaled tissue type, such as the HEC1–3 genes that are expressed mainly in the few cells of which one will later form the transmitting tract (Gremski et al., 2007), but their overall contribution to the transcriptome is very low (Figure 1). Single cell RNA-sequencing (scRNA-seq) of developing gynoecia can improve sensitivity but relies in many cases on fluorescence-assisted cell sorting (FACS). However, where fluorescent marker lines labeling specific tissues or cell types are limited, detection of cell types is difficult, and even more so for those cell types that form only a very small proportion in a tissue (Rich-Griffin et al., 2020). Here, this would apply to, e.g. the dwindling stem cell population at early stages of gynoecium development or the placenta formation.

Also, the role of protein turnover on morphogenesis requires attention when assessing the dynamics of transcriptional activity in developmentally active tissues. Stability varies among plant proteins, ranging from several hours to months with an average total protein half-life of 4–6 d (Li et al., 2017; Scheurwater et al., 2000); however, more specific data for TR turnover during developmental processes are not available. Thus, some TRs may be active for prolonged periods even though their transcripts can no longer be detected. However, the effect of these stable proteins may be limited as dilution occurs following increases in tissue cell number. For example, ANT, KNAT2, CRC, CUC3, NUB, and ETT are required for CMM tissue differentiation (Sessions and Zambryski, 1995; Dinneny et al., 2006; Alonso-Cantabrana et al., 2007; Krizek, 2009; Gross et al., 2018), but are mainly expressed at early stages and the proteins they encode possibly persist for long periods. More generally, a study by Mergner et al. (2020) concludes that protein synthesis constitutes a time delay for developmental stage expression markers. For example, gene families associated with specific flower stages showed proteins one stage later than their transcripts. Moreover, whereas HEC1–3 are expressed at very low levels throughout gynoecium development, their proteins may be particularly stable as their phenotypes are striking (Gremski et al., 2007; Schuster et al., 2015).

High-resolution transcriptome analyses reveal subfunctionalization between closely related homologs

The SEP genes are known for their importance in flower development and organ and meristem identity (Pelaz et al., 2000; Ditta et al., 2004), but so far, little research has been published regarding each gene’s specific role in gynoecium development. SEP genes act partially redundantly in flower development, such that only the quadruple sep1 sep2 sep3 sep4 mutant fails to form floral organs (Ditta et al., 2004). SEP3 is thought to be most important for floral organ identity as it forms the most protein interactions with other MADS-box proteins (Immink et al., 2009; Smaczniak et al., 2012). Moreover, it mediates ternary complex formation between AG and STK, AG and SHP1, AG and SHP2, SHP1 and SHP2, STK and SHP1, and STK and SHP2, all involved in carpel and ovule development (Favaro et al., 2003). However, our transcriptome analysis shows substantial differential dynamics between the SEP genes (Figure 1), suggesting subfunctionalization of this gene family in gynoecium development. SEP4 is generally expressed at a low level, but SEP1 and SEP2 are expressed strongly in the two early stages whereas SEP3 is most strongly expressed in the two later stages. Whereas the ternary complex formation of SEP3 is well researched, the role of SEP1 and SEP2 has not been elucidated in much detail and they have fewer interactors among MADS proteins. Moreover, their ability for cooperative DNA binding differs between individual SEP proteins (Jetha et al., 2014). Whereas the sep1 sep2 sep3 mutant fails to form carpels (Pelaz et al., 2000), adding a single functional SEP1 allele to the triple mutant restores carpel formation (Favaro et al., 2003). However, based on their strong expression during early carpel development, we suggest important, but hitherto unknown, roles for SEP1 and SEP2 in gynoecium development and a high degree of redundancy based on their sequence similarity and expression pattern, and possibly dimerization of SEP1 and SEP2 with non-MADS proteins. Severe subfunctionalization and extreme reduction in expression of SEP genes was also observed in several plant species, for example in Gerbera hybrida, whose genome includes seven SEP genes. Whereas one of them, GRCD6, is hardly expressed, the other six genes diverge strongly in their expression pattern and function, and several distinct phenotypes were observed in the gynoecium when individual SEP genes were downregulated (Zhang et al., 2017). Gene duplication followed by subfunctionalization thus seems to be common to SEP homologs. The MADS-box genes SHP1 and SHP2 serve as second example for expression divergence of highly redundant genes. Neither of the single mutants displays a phenotype, but the double mutants are defective in dehiscence zone formation (Liljegren et al., 2000). Possibly, SHP2 has an earlier function in gynoecium development as it expressed also in the early stages of gynoecium development and even stronger in late stages. By contrast, SHP1 is hardly expressed in early stages and only moderately in the late stages (Figure 1). In addition, their interaction partners for dimerization differ; whereas SHP1 interacts with SEP3, SEP1, STK, AGL13, and AG, SHP1 only interacts with SEP3, SEP1, and AGL6 (Figure 2 and Supplemental Table S3). However, SEP3 mediates interaction of SHP2 and STK as well as SHP2 and AG (Favaro et al., 2003), suggesting that subfunctionalization based on different dimerization partners is overridden by ternary complex formation.

Similarly, HEC genes, known for their function in phytohormone control during gynoecium development (Schuster et al., 2015), show a peculiar pattern of expression. HEC1 is expressed in the later stages especially at S3 where it interacts with SPT to control carpel fusion. The lesser known HEC2 starts with strong early expression but completely ceases after S2 and HEC3 is most expressed at the S2. The specific function of each HEC gene is still mostly unclear, but the transcriptomic data suggest a specific role for each of the three genes in carpel morphogenesis. Our data show a replacement of early interaction of SPT-HEC2 with SPT and HEC1, HEC3, IND, and ALC (Figure 2). Similar replacements can be observed in other hub proteins such as TPL and AG, which exchange interactions over time (Figures 1, 2).

Prediction of genetic interactions in gynoecium development

Negative or positive correlation of gene expression during gynoecium development can support predicted genetic interactions. For example, SEU and LUG together repress AG in the outer whorls of the flower (Franks et al., 2002), but we found strong expression of both of these genes also in the gynoecium with the highest expression during S1–3. This is in line with the seu lug phenotype in the gynoecium, characterized by lack of organ growth and carpel fusion (Franks et al., 2002), suggesting a continuous repression of hitherto unknown target genes during gynoecium development. Interestingly, both genes are also expressed significantly higher than AG in the gynoecium, suggesting that additional regulatory factors may be needed for the floral identity network regulation, protection of AG expression, and proper gynoecium formation.

The protein interaction map (Figure 2) provides a simplified overview of the regulatory dynamics during carpel development. Interactivity of the regulatory proteins is highly complex; however, relevant interactions are determined by presence and strength of expression at given time. Proteins like AG, PI, AP1, SEP1, SEP3, SHP1, SHP2, and STK interact with five or more regulatory partners each, and expression of most of these proteins is established at carpel initiation or even before that. Our data suggest that interactions are at their highest complexity when tissue determinacy is established at the initiation of organogenesis, as has been described previously (Ó’Maoiléidigh et al., 2014). This is suggested by more interactions and more stage-specific genes involved during the initiation of carpel development (Figure 2).

Co-expression clusters reveal temporal emphasis on gene expression regulation during gynoecium development

Comparing transcriptomes of four gynoecium stages, leaf, and SAM tissues by co-expression clustering (Figure 3) showed that the majority of genes highly expressed throughout carpel development is at most weakly expressed in leaf and SAM tissue, suggesting a high level of difference between these tissues. Further, more co-upregulated genes were shared between the SAM and gynoecium than in leaf and gynoecium development suggesting closer similarity of gynoecium and SAM tissue. However, this may be due to rapid expansion of the organ combined with later arising meristematic activity from the carpel margins. The evolutionary ancestor of the carpel is thought to be leaf-like (Moubayidin and Østergaard, 2017; Becker, 2020) and consequently, the transcriptional program of gynoecia should be more similar to that of leaves than the SAM. However, this might be an oversimplified view as gene regulation related to photosynthesis is a major contribution to the leaf transcriptome but plays only a minor role in gynoecium development, as photosynthesis-related genes are enriched only in a single cluster comprising leaf and late gynoecium stages. With regard to developmental regulation, the leaf primordium is meristematic at its inception and, during growth, meristematic potential is restricted to the margins, reminiscent of different SAM zones (Alvarez et al., 2016). However, the leaf transcriptome does not reflect these spatially differentiated leaf tissue types or developmental stages. Interestingly, the observation that at/after stage 9 of A. thaliana flower development, gynoecium development shifts from bilateral to radial growth (Moubayidin and Østergaard, 2017) are consistent with our data on TR expression (Figure 1). Several TRs changed their expression between S2 and S3. For example, the adaxial/abaxial regulators FIL, CRC, and KAN2 showed a decline in expression after S2 suggesting that abaxial/adaxial polarity required for bilateral growth is established in S1 and S2 and subsequent radial growth requires different regulators. However, many regulatory processes seem to require maintenance throughout gynoecium development. For example, the C1 cluster included genes upregulated throughout gynoecium development but not in leaf and SAM tissues and was most strongly enriched in regulators of gene expression, splicing, and floral development.

Moreover, the shift of activity between S2 and S3 becomes also obvious when overrepresented GO terms were compared between genes upregulated in early (Figure 4, C11) and late clusters (Figure 4, C12). At the early stages, genes related to cell division, RNA splicing, and regulation of gene expression were enriched, emphasizing the importance of early regulation of morphogenesis. By contrast, cluster C12 is enriched in photosynthesis-related genes. This includes upregulation of genes required for photosynthesis and carbohydrate transport, suggesting that the gynoecium may be a net sink organ but that also contributes energy to the reproductive effort. Also, previous work has shown that flowers and fruits are not merely a cost to the carbon budget of the rest of the plant, but also contribute to this (Bazzaz et al., 1979; Aschan and Pfanz, 2003; Gnan et al., 2017; Brazel and Ó’Maoiléidigh, 2019). Interestingly, in the case of gynoecium photosynthesis, developmental clues trigger upregulation of photosynthesis-related genes and not light availability, because S4 is around 36 h before anthesis (Smyth et al., 1990). Mizzotti et al. (2018) have shown that expression of photosynthesis, tetrapyrrole-biosynthesis, and plastid–ribosomal proteins is strongest between 3 and 6 d after pollination, and our data show that expression of many of these genes is already activated while the flower is still closed (Supplemental Table S3). Photosynthetic activity of reproductive organs has been more thoroughly reviewed in Brazel and Ó’Maoiléidigh (2019).

In summary, we describe a fine-scale map of transcriptional changes during gynoecium development as a resource to plant scientists. This provides a unique temporal perspective on global gene expression and protein complex formation potential, suggesting a large number of candidate developmental regulators orchestrating gynoecium development, three of which (B3LF1, -2, DOFLF1) we have confirmed play a functional role.

Materials and methods

Transcriptome assembly, heatmaps, and interactome analysis

Raw sequencing reads of the four stages of A. thaliana gynoecium development (Kivivirta et al., 2019) were used to generate the transcriptomes (GenBank: Bioproject accession PRJNA549137). Trimming, quality testing, assembly, and annotation were carried out in CLC workbench version 11.0.1. (QIAGEN, Hilden, Germany) as previously described in Kivivirta et al. (2019) with the A. thaliana genome (Swarbreck et al., 2008). Gene expression heatmaps were constructed with Heatmapper: expression (Babicki et al., 2016). Euclidean distance of absolute values of gene expression was used with the expression values of a list of genes derived from Reyes-Olalde et al. (2013), Pfannebecker et al. (2017a, 2017b), Parenicová et al. (2003), and Ojolo et al. (2018). Gene functions and families were based on earlier publications for each gene. The genes with expression of TPM <1 were left out of the analysis. For the complete list of genes, their functions, and expression, see Supplemental Table S1. Protein–protein interaction maps were constructed with the GeneMANIA (Franz et al., 2018) app in cytoscape 3.8.0 (Shannon et al., 2003). Protein–protein interactions were searched for a set of carpel regulatory genes (Supplemental Table S1). Genes with no known interactions with other carpel regulatory genes were discarded. Information on gene families, change of gene expression, and intensity of expression was added to the interaction map after the analysis. Change of binary logarithmic of expression was applied for Figure 3, A. Expression strengths for Figure 3, B and C, are based on the absolute TPM values of gene expression. For the Venn diagrams (Supplemental Figure S2), TPM values were calculated as described in Kivivirta et al. (2019) from the whole-bud transcriptome data of Klepikova et al. (2016; Flower 6–8, Flower 9–11, and Flower 15–18) and Chen et al. (2018, stages 4 and 8). Venn diagrams were generated using InteractiVenn (www.interactivenn.net, accessed 23 September 2020; Heberle et al., 2015). Functional classifications were carried out using the functional classification tool of panther (www.pantherdb.org, accessed 16 October 2020; Mi et al., 2019) with sets of identifiers unique to each data set (TPM > 5).

Co-expressed clusters

Automatically partitioned co-expression clusters were generated with Clust front-end version 1.0.0 (Abu-Jamous and Kelly, 2018). The datasets were automatically normalized. Cluster tightness was set to 2 and minimum cluster size to 40 genes. Genes with flat expression were filtered out of the analysis. The SRA files additionally included in the analysis were SRR3581346 (SAM) and SRR3581838 (leaf blade; Klepikova et al., 2016). GO enrichment for gene sets was analyzed with PANTHER 15.0 gene ontology enrichment (Mi et al., 2019, GO version March 2020). The results of a Fisher’s Exact test using the A. thaliana reference list for overrepresentation analyses were corrected by calculating false discovery rates. The generated lists were then visualized using REViGO (Supek et al., 2011) in the version available in June 2020 (GO version January 2017). For the visualization of redundant GO terms with REViGO, uncorrected P-values of overrepresentation tests were grouped based on their semantic similarity utilizing the inbuilt SimRel function with the A. thaliana reference and an allowed similarity setting of 0.7. The generated lists and treemaps were further processed with the DrasticData Treemapping tool (drasticdata.nl, Delft, The Netherlands) to achieve a better visualization.

SALK mutant analysis

Arabidopsis thaliana cv Col-0 and the SALK mutant line plants were grown in peat/perlite mixture (3:1) in long-day conditions. The mutants were self-pollinated to achieve homozygous insertion lines (Supplemental Table S4). SALK mutant lines were verified to be homozygous by genotyping the lines with locus- and insert-specific primers. Genotyping was done in two separate reactions with locus-specific LbB + RP and LP + RP primers to verify presence of the specific insert and absence of the wild-type locus (Supplemental Table S4). Siliques from 30 developing flowers were analyzed for morphological abnormalities, gynoecium length, and seed number at the stage of silique ripening (Smyth et al., 1990: stage 17) for each mutant line. Statistical significance was evaluated using the single factor ANOVA and t test. The siliques were recorded with Leica DM550 (Leica Application Suite 4.3.0, Wetzlar, Germany) and the lengths of the siliques were measured with ImageJ (https://imagej.nih.gov/ij/). The intron–exon structures and protein-binding site motifs for TRs upstream B3LF1, -2, and DOFLF1 genes were based on ChIP-Hub (http://www.chip-hub.org). FIMO-binding maps use plant transcription factor database (http://planttfdb.cbi.pku.edu.cn/) as a source for protein-binding sites. SALK-line insertion locations were verified with genotyping-by-sequencing. 350-bp insert DNA libraries were created for all mutant lines with the lesser fertility phenotype; DNA was extracted with NucleoSpin Plant II (Macherey-Nagel, Düren, Germany). Plant whole-genome resequencing was carried out by Novogene (Cambridge, United Kingdom). T-DNA insertions were located using TDNAscan (Sun et al. 2019).

Accession numbers

Sequence data from this article can be found in the GenBank/EMBL data libraries under accession numbers XXX listed in Supplemental Table S1.

Supplemental data

Supplemental Figure S1. Sample preparation with LMD and FastQ analysis for raw sequence read files.

Supplemental Figure S2. A comparative analysis of genes (TPM > 5) between the four stages of gynoecium development and previously published data of complete buds at different growth stages.

Supplemental Figure S3. Heatmap of the MIKC type MADS-box genes and genes involved in chromatin remodeling.

Supplemental Table S1. Detailed information on expression and function of genes related to gynoecium development (Figure 1)

Supplemental Table S2. Protein interactions of known developmental regulators

Supplemental Table S3. Automatically partitioned co-expressed clusters

Supplemental Table S4. SALK-mutant analysis and genotyping

 

K.I.K. and A.B. wrote the manuscript. K.I.K. performed the digital gene expression analysis, mapping expression data to protein interaction data, and loss-of-function mutant analysis. C.R. conducted the co-expression clusters and GO enrichment analyses. D.H. assembled the transcriptomes and calculated the TPM values, and S.d.F. and N.M.-M. analyzed the phytohormone-related genes. K.I.K. and C.R. were involved in figure preparation. A.B. designed the study. All authors contributed, read and approved the final manuscript.

The author responsible for distribution of materials integral to the findings presented in this article in accordance with the policy described in the Instructions for Authors (https://academic.oup.com/plphys) is: Annette Becker (annette.becker@bot1.bio.uni-giessen.de).