Evolutionary trends in the floral transcriptome: insights from one of the basalmost angiosperms, the water lily Nuphar advena (Nymphaeaceae)


For correspondence (fax +1 515 294 1337; e-mail yoom@iastate.edu).


Current understanding of floral developmental genetics comes primarily from the core eudicot model Arabidopsis thaliana. Here, we explore the floral transcriptome of the basal angiosperm, Nuphar advena (water lily), for insights into the ancestral developmental program of flowers. We identify several thousand Nuphar genes with significantly upregulated floral expression, including homologs of the well-known ABCE floral regulators, deployed in broadly overlapping transcriptional programs across floral organ categories. Strong similarities in the expression profiles of different organ categories in Nuphar flowers are shared with the magnoliid Persea americana (avocado), in contrast to the largely organ-specific transcriptional cascades evident in Arabidopsis, supporting the inference that this is the ancestral condition in angiosperms. In contrast to most eudicots, floral organs are weakly differentiated in Nuphar and Persea, with staminodial intermediates between stamens and perianth in Nuphar, and between stamens and carpels in Persea. Consequently, the predominantly organ-specific transcriptional programs that characterize Arabidopsis flowers (and perhaps other eudicots) are derived, and correlate with a shift towards morphologically distinct floral organs, including differentiated sepals and petals, and a perianth distinct from stamens and carpels. Our findings suggest that the genetic regulation of more spatially discrete transcriptional programs underlies the evolution of floral morphology.


Gene expression profiling using microarray and other techniques is providing valuable insights into transcriptional programs, and has identified genes with potentially important roles in many developmental processes in plants as well as animals. Profiling provides a broader and deeper view of the genetic control of a given process by extending beyond those key transcription factors typically identified through mutant analyses to genes downstream of the transcription factors, and to genes involved in related pathways. The resulting large set of genes revealed via transcription profiling can be further analyzed to develop hypotheses of cascades, pathways or networks that may ultimately explain a given process. An excellent example of this approach in plants, and in Arabidopsis thaliana in particular, is the use of microarray techniques to examine the transcriptional programs regulated by the ABCE model of floral organ identity determination. According to this model, A, B, C and E functions interact in spatially overlapping domains to promote the development of floral organs: A- and E-function genes control sepal identity; A-, B- and E-function genes control petal identity; B-, C- and E-function genes control stamen identity; C- and E-function genes control carpel identity (Coen and Meyerowitz, 1991; Pelaz et al., 2000; Ditta et al., 2004).

Some functional aspects of the ABCE model have been shown to be conserved in other angiosperms [e.g. Petunia hybrida (Angenent et al., 1993); Oryza sativa (Kang et al., 1995, 1998; Fornara et al., 2004; Yamaguchi et al., 2006); Gerbera hybrida (Yu et al., 1999); Zea mays (Ambrose et al., 2000); and pea (Pisum sativum) (Taylor et al., 2002)], and gene expression assays across a much broader spectrum of angiosperm diversity suggest that it may represent the ancestral developmental program of flowers. For example, the expression domains of BCE homologs are essentially conserved in the eudicots [e.g. white campion (Silene latifolia) (Hardenack et al., 1994); three-leaf akebia (Akebia trifoliata) (Shan et al., 2006); Taihangia rupestris (et al., 2007); grape (Vitis vinifera) (Poupin et al., 2007)], monocots [e.g. oil palm (Elaeis guineensis) (Adam et al., 2007)], the magnoliids [e.g. pawpaw (Asimina longifolia), southern magnolia (Magnolia grandiflora) (Kim et al., 2005), avocado (Persea americana) (Chanderbali et al., 2006)], as well as the ANITA grade angiosperms, Florida anise (Illicium floridanum), water lilies (Cabomba caroliniana, Nuphar advena and Nymphaea odorata; Yoo et al., 2010), and Amborella trichopoda (Kim et al., 2005). (In this article, the term ‘basal angiosperms’ refers to angiosperms that are not members of either the eudicot or monocot clades, including Amborella, Nymphaeales, Austrobaileyales, Chloranthaceae and magnoliids. Basal angiosperms may have their own apomorphic traits, and are not necessarily less derived than eudicots and monocots.) Expression and heterologous complementation data from gymnosperms further suggest that the ABCE model may have evolved from an original BC program that operated in the common ancestor of the seed plants (Theissen and Melzer, 2007).

In Arabidopsis, the A-function is provided by APETALA1 (AP1) and APETALA2 (AP2), the B-function by APETALA3 (AP3) and PISTILLATA (PI), the C-function by AGAMOUS (AG) and the E-function by multiple SEPALLATA genes (SEP1–SEP4) (Coen and Meyerowitz, 1991; Honma and Goto, 2001; Theissen and Saedler, 2001). These are all transcription factors (only AP2 is not of the MADS-box family), but relatively few downstream targets have been identified via mutant and candidate-gene analyses, possibly in part because loss-of-function mutants have slightly altered phenotypes (Wellmer et al., 2006). However, microarray analyses have identified thousands of potential downstream targets in Arabidopsis (Schmid et al., 2003, 2005; Hennig et al., 2004; Wellmer et al., 2004, 2006; Gomez-Mena et al., 2005; Wellmer and Riechmann, 2005; Zhang et al., 2005; Alves-Ferreira et al., 2007).

Here, we investigate the transcriptional programs operating in flowers of the water lily Nuphar advena, a member of the Nymphaeales (comprising Nymphaeaceae, Cabombaceae and Hydatellaceae; Saarela et al., 2007), a phylogenetically pivotal lineage that lies sister to all extant angiosperms except Amborella (Figure 1) (Soltis et al., 1999, 2000; Leebens-Mack et al., 2005; Qiu et al., 2005; Jansen et al., 2007; Moore et al., 2007). We have used floral expressed sequence tags (ESTs) for Nuphar collected by the Floral Genome Project (Albert et al., 2005; Soltis et al., 2007) to conduct the investigation of the floral transcriptional program in one of the basalmost angiosperm lineages.

Figure 1.

 Summary tree for angiosperms. This tree is modified from the plastid genome trees of Jansen et al. (2007) and Moore et al. (2007), and only a few representatives of asterids and rosids are included. Nuphar advena belongs to Nymphaeales (red), which are the sister to all extant angiosperms except Amborella (Amborellales). Persea and Arabidopsis are members of Laurales (blue) and Brassicales (green), respectively.

Nuphar lies sister to all other genera of Nymphaeaceae and has trimerous flowers of moderate size, which is the inferred ancestral state of Nymphaeales (Les et al., 1999; Borsch et al., 2007, 2008; Löhne et al., 2007). Its flowers consist of two perianth whorls followed by numerous staminodes, stamens and carpels in a syncarpous gynoecium (Figure 2). The outer perianth whorl is green and sepaloid, the inner is yellow and petaloid, and staminodes are narrower than the perianth members, but broader than the stamens, with a nectary on the abaxial surface (Figure 2). We have assessed gene expression levels in floral buds of early developmental stages and mature floral organs and seeds, relative to leaves, to provide an assessment of the spatial and temporal expression patterns of transcripts encoding genes involved in Nuphar floral development. Through comparative analyses with similar data sets for Arabidopsis (Schmid et al., 2005) and Persea (Chanderbali et al., 2009), the latter a member of the magnoliid clade of basal angiosperms, we infer general evolutionary trends in floral transcriptomes.

Figure 2.

 The flower of Nuphar advena.
(a) Side view of whole flower.
(b) Three sepaloid tepals are arranged in the outer whorl, and three petaloid tepals are arranged in the inner whorl.
(c) Staminodes with a nectary (left, adaxial view; right, abaxial view).
(d) Stamens with anther (left two, adaxial view; right two, abaxial view).


Genes expressed differentially in reproductive organs

Custom microarrays targeting 6220 genes collected from Nuphar floral buds identified 3333 significantly differentially expressed genes (P < 0.05; false discovery rate = 0.38%) among eight sampled tissues: young (Ybd) and medium-aged (Mbd) floral buds from the pre-microsporangia initiation stage and the pre-meiotic stage, respectively, ‘sepaloid’ tepals (outer perianth whorl), ‘petaloid’ tepals (inner perianth whorl), stamens, carpels, seeds and leaves. Hierarchical clustering of log2 ratios relative to leaves assembled differentially expressed genes into two primary clusters, harboring those deployed primarily during early and late developmental stages, respectively (Figure 3). Within the latter cluster, three major subsets accommodate genes expressed primarily in: (i) seeds (module 1); (ii) carpels (the AGAMOUS-LIKE 6, AGL6, module); and (iii) broadly across the flower, but primarily in stamens and tepals (the AP3/PI/SEP1 module), respectively (Figures 3 and S1). Homologs of AG and AP1 were placed in the ‘early’ cluster (the AG module and the AP1 module, respectively), where they assembled with genes primarily expressed in carpels and stamens, and during the earliest developmental stages sampled, respectively (Figures 3 and S1).

Figure 3.

 Hierarchical clustering displays ‘early’ or ‘late’ gene clusters based on similarity of expression patterns. The color scale is presented in the right-hand bar: yellow indicates high expression; blue indicates low expression. Top, array tree; left, gene tree. Mbd, medium-aged floral buds at the pre-meiotic stage; Ybd, young floral buds at the pre-microsporangia initiation stage.

We also searched among the differentially expressed genes for organ-specific and organ-combination-specific expression, arbitrarily defined as upregulation by at least two-fold in target organ(s) alone (Table 1). These moderately stringent criteria identified 831 genes, most of which were expressed primarily in both perianth organs and stamens (124), followed by petaloid tepals (113), both perianth organs (101) and stamens alone (73). These top four expression patterns account for 49.2% of Nuphar genes, with at least twofold upregulation in floral tissues relative to leaves. Seventy-six genes were primarily expressed in pre-microsporogenic floral buds, 63 in seeds, 23 in carpels alone and 16 in all four floral organs.

Table 1.   The number of genes upregulated (i.e. a twofold higher expression level) in Nuphar floral tissue relative to leaves
OrganNo. of genes (%)OrganNo. of genes (%)
  1. Mbd, medium-aged floral buds at the pre-meiotic stage; Ybd, young floral buds at the pre-microsporangia initiation stage.

Sepaloid + petaloid tepals + stamens124 (14.9)Sepaloid + petaloid tepals + seeds13 (1.6)
Petaloid tepals113 (13.6)Stamens + seeds13 (1.6)
Sepaloid + petaloid tepals101 (12.2)Sepaloid tepals + stamens + seeds10 (1.2)
Stamens73 (8.8)Mbd9 (1.1)
Seeds65 (7.8)Carpels + seeds7 (0.8)
Ybd53 (6.4)Sepaloid + petaloid tepals + carpels5 (0.6)
Sepaloid + petaloid tepals + stamens + seeds48 (5.8)Sepaloid tepals + seeds4 (0.5)
Sepaloid tepals43 (5.2)Petaloid tepals + stamens + seeds3 (0.4)
Petaloid tepals + stamens32 (3.9)Sepaloid tepals + carpels2 (0.2)
Sepaloid tepals + stamens30 (3.6)Stamens + carpels2 (0.2)
Sepaloid + petaloid tepals + stamens + carpels + seeds25 (3.0)Petaloid tepals + seeds1 (0.1)
Carpels23 (2.8)Petaloid tepals + stamens + carpels + seeds1 (0.1)
Sepaloid + petaloid tepals + stamens + carpels16 (1.9)Stamens + carpels + seeds1 (0.1)
Ybd + Mbd14 (1.7)  

Many of the genes upregulated in Nuphar flowers are genes that encode transcription factors, such as the bZIP, G2-like, NAC, MADS and MYB-related gene families (Table S1). All of the Nuphar MADS-box genes examined, except the AP1 homolog, are upregulated in reproductive organs, and their expression patterns generally agree with previously reported relative quantitative RT-PCR (RQ-RT-PCR) data for Nuphar homologs of AG, PI, AP3, AP1, AGAMOUS-LIKE 6 (AGL6) and SEP1 (Kim et al., 2005; Zahn et al., 2005; Yoo et al., 2010) (Figure S2). bZIP transcription factors are diverse in their functions, including light and stress signaling, floral transition and seed development (Jakoby et al., 2002; Nijhawan et al., 2008). Eight members of the bZIP family are differentially expressed in Nuphar, and are all mainly expressed in young buds. NAC-LIKE (NAP, ACTIVATED BY AP3/PI) plays a role in cell morphogenesis, leaf senescence (Li et al., 2004; Guo and Gan, 2006) and floral development in Arabidopsis, particularly in petals and stamens (Sablowski and Meyerowitz, 1998), but the Nuphar homolog is expressed in the early floral stages (young and medium bud), and not in mature stamens and petaloid tepals (Table S1). KANADI1 (KAN1), a member of the G2-like family, is known to regulate leaf and carpel polarity in Arabidopsis, and is expressed temporarily on the abaxial side of initiating floral-organ primordia (Kerstetter et al., 2001). The Nuphar KAN1 homolog is expressed in petaloid tepals, stamens and carpels (Table S1).

Among the genes downregulated in Nuphar flowers are several related to leaf morphogenesis in Arabidopsis: for example, homologs of ANGUSTIFOLIA (Kim et al., 2002), CURLY LEAF (Kim et al., 1998) and UV-B-INSENSITIVE 4 (Hase et al., 2006) (Table S2). However, several genes related to flower development in Arabidopsis are also downregulated in some reproductive organs (Table S2). For example, a homolog of LEUNIG, required for flower organ formation and development in Arabidopsis (Chen et al., 2000; Conner and Liu, 2000), is downregulated in all floral tissues. Others include transcription factors of uncertain function, such as members of the bHLH, homeobox and WRKY families (Table S2).

Comparative floral transcriptomics among Nuphar, Persea and Arabidopsis

Among the differentially expressed Nuphar genes, 2449 have homologs in the AtGenExpress data set for Arabidopsis, of which 957 are also differentially expressed in the Persea data set. Hierarchical clustering of Nuphar-Arabidopsis (NA) and Nuphar-Persea-Arabidopsis (NPA) data sets sorted floral organs into groups in accordance with species, with Nuphar and Persea flowers transcriptionally closer to each other than either is to Arabidopsis. These relationships are supported by gene clusters characterized by upregulation in Nuphar, but by downregulation in Arabidopsis, and vice versa (Figure 4a), and by upregulation in Nuphar, Persea or both, but by downregulation in Arabidopsis, and vice versa (Figure 4b). We observed that only a very small proportion of the genes examined show similar expression patterns between Nuphar and Arabidopsis; these genes are involved in metabolism, energy production, transcription activity and biogenesis of cellular components, and this same subset of genes is also found in the NPA data set (Table S3). In particular, MADS-box gene homologs from three species showed similar expression patterns (Table S3; Chanderbali et al., 2009).

Figure 4.

 Hierarchical clustering results of combined data sets of Nuphar, Persea and Arabidopsis.
(a) Clustering result from the NA data set: I, upregulation of Arabidopsis genes; II, downregulation of Arabidopsis genes.
(b) Clustering result from the NPA data set: I, upregulation of Arabidopsis genes; II, downregulation of Arabidopsis genes. The color scale is presented in the right-hand bar: yellow indicates high expression; blue indicates low expression. Top, array tree; left, gene tree.

Within flowers, the inferred hierarchies of organ similarities are identical in Nuphar and Persea. In both, the perianth organs cluster together, followed by stamens, with carpels more distant, as in the individual analyses of the Nuphar (Figure 3) and Persea (Chanderbali et al., 2009) data sets (Figure 4b). By contrast, Arabidopsis petals always cluster with stamens, and either sepals are placed distant from the other floral organs (Figure 4a; NA data set) or carpels occupy this position (Figure 4b; NPA data set). Moreover, the branch lengths separating individual floral organs are appreciably shorter in the two basal angiosperms, suggesting greater similarities in the transcriptional profiles of their floral organs. Scatter plots comparing the transcriptional profiles of adjacent floral organ categories (Figure 5) also indicate that Arabidopsis floral organs are more divergent from each other than are those of Nuphar and Persea. Petals and stamens share the highest similarity among Arabidopsis organs, but with a rather low correlation (r = 0.66), compared with the almost linear correspondence between inner tepals and outer tepals for Nuphar (= 0.96) and Persea (r = 0.98). Perianth organs and stamens are also highly correlated in the two basal angiosperms, with r = 0.85 and 0.89 for stamens versus inner and outer perianth whorls, respectively, in Nuphar, and r = 0.82 and 0.79 for stamens versus inner and outer tepals, respectively, in Persea.

Figure 5.

 Scatter plots of spatial gene expression patterns of Nuphar, Persea and Arabidopsis. For these graphs we used NPA data sets, and the log2 transformed expression data were used. Pearson correlations (r) of gene expression level between floral organs are presented.

When we numerically compared the genes that are upregulated by at least two-fold in floral organ(s) in all three species, similar results were obtained. In Nuphar, the highest proportion of these genes is found in both perianth organs and stamens (24.9%), whereas in Arabidopsis only a small number of genes are upregulated in the same organs (4.9%). By contrast, in Arabidopsis most of the upregulated genes are found in stamens (23.3%) and carpels (23.9%; Table 2). Analysis of the NA data set also shows similar results to that of the whole data set, although the number of genes upregulated in floral organ(s) differs. In Persea, like Nuphar, most genes are expressed in perianth organs alone, or in both perianth organs and stamens (Table 2), although many more genes are expressed in stamens and carpels of Persea (22.9 and 13.8%, respectively) compared with Nuphar (Table 2). However, considering differentially expressed genes regardless of folder changes, there is no difference in gene proportion in any floral tissue of Nuphar and Persea (Table S4).

Table 2.   Number of genes upregulated in any floral tissue of Nuphar, Persea and Arabidopsis
 NupharPerseaArabidopsisNuphar from NAArabidopsis from NA
Number (%)Number (%)Number (%)Number (%)Number (%)
  1. The proportions of genes are calculated from twofold upregulated genes in any floral tissue relative to leaves; sepals, sepaloid outer tepals, outer tepals and sepals; petals, petaloid inner tepals, inner tepals and petals. Major differences among three species are shown in bold.

  2. aThis percentage is calculated as the number of twofold upregulated genes over the whole number of probes in Nuphar (690 out of 6220; 491 out of 2449 for the NA), Persea (955 out of 6086) and Arabidopsis (7794 out of 22 592; 1081 out of 2449 for the NA).

All floral parts41 (5.9)127 (13.3)451 (5.8)34 (6.9)59 (5.5)
Sepals47 (6.8)39 (4.1)516 (6.6)15 (3.1)38 (3.5)
Sepals + petals114 (16.5)121 (12.7)128 (1.6)81 (16.5)10 (0.9)
Sepals + petals + stamens172 (24.9121 (12.7)380 (4.9)123 (25.1)27 (2.5)
Sepals + petals + carpels5 (0.7)27 (2.8)131 (1.7)4 (0.8)21 (1.9)
Sepals + stamens40 (5.8)9 (0.9)287 (3.7)14 (2.9)32 (3.0)
Sepals + stamens + carpels1 (0.1)5 (0.5)42 (0.5)0 (0.0)7 (0.6)
Sepals + carpels2 (0.3)2 (0.2)83 (1.1)2 (0.4)8 (0.7)
Petals114 (16.5)33 (3.5)566 (7.3)96 (19.6)103 (9.5)
Petals + stamens34 (4.9)24 (2.5)397 (5.1)32 (6.5)50 (4.6)
Petals + stamens + carpels1 (0.1)16 (1.7)358 (4.6)1 (0.2)64 (5.9)
Petals + carpels0 (0.0)9 (0.9)545 (7.0)0 (0.0)96 (8.9)
Stamens86 (12.5)219 (22.9)1819 (23.3)64 (13.1)172 (15.9)
Stamens + Carpels3 (0.4)71 (7.4)228 (2.9)2 (0.4)36 (3.3)
Carpels30 (4.3)132 (13.8)1863 (23.9)23 (4.7)358 (33.1)
Total690 (11.1)a955 (15.7)a7794 (34.5)a491 (20.0)a1081 (44.1)a


Global gene expression patterns in Nuphar

We have analyzed spatial and temporal gene expression patterns using custom microarrays targeting transcripts encoding genes involved in Nuphar floral development. Hierarchical clustering of differentially expressed genes revealed two distinct transcriptional programs in the data. One program involves genes that are expressed almost exclusively during the early stages of floral development (pre-microsporangial initiation and/or pre-meiotic), whereas the other harbors a transcriptional program with a later temporal pattern. Within this ‘later’ program, one subset (module 1) appears to be primarily deployed at the latest developmental stage sampled here, seed development, whereas the other is associated with floral organ development. Most of the genes in this subset (the AP3/PI/SEP1 module) are primarily expressed in the perianth organs and stamens, with weaker expression in carpels, and a smaller number (the AGL6 module) is expressed primarily in carpels. Homologs of AG and AP1 are both expressed during the early floral bud stages, and are placed in the ‘early’ transcriptional program, but although AP1 is not readily detectable in mature floral organs by our measurements, the transcript abundance of the AG homolog increases in mature stamens and carpels.

The spatiotemporal expression patterns of the genes noted in the AP3/PI/SEP1 module and the AGL6 module are therefore consistent with the ABCE model. For example, although the AP3/PI/SEP1 module shows the broadest expression areas across all floral organs (expression areas of E-function genes; Figure S1), genes in this module are mainly expressed in the perianth whorls and/or stamens, which agrees with the main expression domain of B-class gene homologs. In the AP1 module, more than half of the genes are strongly expressed in early developmental stages, which is consistent with the expression pattern of A-class gene homologs, but not in tepals (whorls 1 and 2), which represent the expected expression area of A-function genes. Considering that the presence of a ‘true’ A function is currently questionable (Kramer and Hall, 2005; Davies et al., 2006), this result cannot be considered as a deviation from the ABCE model. Genes in the AGL6 module are mainly expressed in carpels and seeds, but Nuphar AGL6 is expressed in all floral organs (see below), and AGL6 homologs are expressed in inflorescence buds, perianth and female reproductive tissues in other angiosperms (Ma et al., 1991; Rounsley et al., 1995; Hsu et al., 2003; Fan et al., 2007; Rijpkema et al., 2009; Schauer et al., 2009). In gymnosperms, however, AGL6 homologs are mainly expressed in the megasporangium and integument (Mouradov et al., 1998; Shindo et al., 1999; Winter et al., 1999; Reinheimer and Kellogg, 2009). It is therefore tempting to speculate that the expression of AGL6 homologs in carpels stems from an ancient role in the female reproductive unit of seed plants, and that they have acquired a different role in the perianth of angiosperms (see also Reinheimer and Kellogg, 2009).

The expression patterns of homologs of MADS-box genes from Nuphar generally agree with those previously reported based on RQ-RT-PCR data (Kim et al., 2005; Zahn et al., 2005; Yoo et al., 2010). This is especially true for homologs of AP3, PI and SEP1. However, there may be differences in the levels of gene expression among floral organs (Figures 3 and S2). For example, the Nuphar AG homolog was detected in stamens and carpels by RQ-RT-PCR techniques (Yoo et al., 2010), and is strongly upregulated in stamens and carpels relative to leaves according to the microarray data, but is also upregulated, albeit at lower expression levels, in outer and inner perianth whorls (Figure 3). In all instances (AP1, AG and AGL6), microarray data suggest broader expression patterns than observed with RQ-RT-PCR. Whether these differences reflect greater sensitivity of microarrays, non-specific binding or experimental variation in cDNA populations remains unclear. RQ-RT-PCR uses sequence-specific primer sets to amplify target cDNA, whereas microarray experiments are based on hybridization between probes and target RNA. Therefore, if duplicate gene copies with sufficient sequence similarity exist in the cDNA population in microarray studies, the probe may bind to both gene copies, and wider gene expression patterns could be detected if duplicates collectively exhibit broader gene expression domains. Similar results would be obtained if microarray probes target conserved motifs present in multiple genes. Also, expression levels measured by RQ-RT-PCR are relative to that of an internal control; therefore, direct comparison between these two sets of experiments may only be partially reliable. When examined without an internal control, RQ-RT-PCR expression of Nuphar AP1 and AGL6 homologs was detected in all floral tissues, even though there is variation in signal intensities among organs, based on the microarray data. However, AG transcripts were not detected in outer and inner perianth whorls by RQ-RT-PCR. Expression levels of AG are relatively high in stamens and carpels (Figures 3 and S2), suggesting that expression levels in outer and inner perianth whorls may lie below the threshold sensitivity of RQ-RT-PCR. The broader expression pattern observed in microarrays might result from the non-specificity of the Nuphar AG probe, which was designed from sequences lying between the MADS-box and I regions. Also, BLAST analyses of the probes against the National Center for Biotechnology Information (NCBI) nr database only found other members of the AG gene lineage, supporting the interpretation that they only bind other AG homologs rather than more distantly related MADS-box genes.

The differentially expressed genes identified in the Nuphar floral transcriptome are similar to those of Arabidopsis, i.e. MADS-box genes and other transcription factors. Previous studies of the Arabidopsis transcriptome showed that many potential signaling components, such as protein kinases, transcription factors and other putative signaling proteins, were upregulated during reproductive development (Hennig et al., 2004; Schmid et al., 2005). Wellmer et al. (2006) also showed that in Arabidopsis closely related genes are highly correlated in their temporal gene expression patterns, and the majority of those genes are upregulated during certain developmental stages. In this study, we demonstrated that many transcription factors were differentially expressed in floral tissues, and as in Arabidopsis, their expression patterns are correlated. For example, genes encoding bHLH, homeobox and WRKY transcription factors are expressed in leaves, whereas genes encoding bZIP, MADS and MYB-related transcription factors are mainly expressed in floral tissues (Tables S1 and S2). In particular, all the Nuphar WRKY transcription factor homologs examined are downregulated in floral tissues, whereas MADS-box genes, except the AP1 homolog, are upregulated in reproductive organs. These expression patterns are also observed in Arabidopsis, suggesting functional conservation among major regulatory elements between these two species.

Comparative floral transcriptomics among Nuphar, Persea and Arabidopsis

Our analyses suggest distinct transcriptional programs are deployed in each of the three species examined here. Most of the homologs with similar expression patterns are involved in the basic, and perhaps understandably conserved, processes of cellular metabolism, energy production and biogenesis of cellular components. Also, homologous MADS-box genes from Nuphar, Persea and Arabidopsis exhibit similar expression patterns (Figure 4b; Table S3), suggesting that major components of their function may be conserved. However, homologs of other genes regulating flower development, such as NAP and AFO (ABNORMAL FLORAL ORGANS), are expressed in different floral organs in the three species, suggesting less conservation in broader genetic networks and, perhaps, illustrating one principal mechanism behind floral diversification across the angiosperms. The transcriptional patterns in the three species examined here have also revealed a fundamental difference separating flowers of basal angiosperms from those of eudicots. Both basal angiosperms, Nuphar and Persea, have transcriptional programs that are only weakly partitioned among floral whorls compared with Arabidopsis, and this regulatory shift appears to correlate with a shift in floral morphology. Arabidopsis has well-differentiated floral organs, and the number of genes expressed in more than one floral organ is relatively small, indicating that each floral organ has a nearly unique floral transcriptional program. The transcriptional programs in Nuphar and Persea correspond well with the ‘fading borders’ model of floral organ identity, which explains the morphological intergradation of floral organs in basal angiosperms through fading activities of floral organ identity genes towards the borders of their broad activity zones (Buzgo et al., 2004, 2005). Minor differences in pigmentation can distinguish the outer ‘sepaloid’ tepals from the inner ‘petaloid’ tepals of Nuphar (Warner et al., 2008, 2009), whereas the inner perianth organs of Persea bear glandular patches that are absent from the outer tepals (Chanderbali et al., 2006). The strong linearity between expression profiles of inner and outer tepals in both Nuphar (r = 0.96) and Persea (r = 0.98) evidently reflects their largely undifferentiated morphology. This linearity is also observed between perianth members and stamens of Nuphar (r = 0.85) and Persea (r = 0.82), supporting their grouping in hierarchical clustering analyses. In contrast, a sharp distinction in expression profiles (0.42 ≤ r ≤ 0.66) suggests that the floral organs of Arabidopsis develop under more divergent transcriptional programs, relative to those of the two basal angiosperms.

The higher proportion of genes commonly expressed in perianth members and stamens compared with the number of genes expressed in stamens alone suggests that the staminal developmental program is not as unique in Nuphar and Persea as it is in Arabidopsis. Moreover, genes deployed primarily in Nuphar and Persea stamens tend to have broader expression domains than those expressed primarily in carpels (Table 2). These features of the floral transcriptomes of Nuphar and Persea may support the ‘out-of-male’ or ‘mostly-male’ hypotheses for the origin of the flower. According to these hypotheses, female structures and perianth organs evolved from male structures through the contraction of the male developmental program in the upper region of male cones, and sterilization of the outer stamens, respectively (Frohlich and Parker, 2000; Baum and Hileman, 2006; Theissen and Melzer, 2007). Therefore, perhaps these expression features are a vestige of an ancient event: the evolution of female organs and perianth organs from male organs.

In organ relationships based on expression data, Nuphar and Persea perianth organs clustered together with stamens, as do Arabidopsis petals and stamens (Figure 4). According to the traditional view, petals (or in some cases inner tepals) in basal angiosperms are thought to have derived from bracts (bracteopetals), whereas petals of eudicots are considered to be derived from stamens (andro-petals) (Hiepko, 1965; Takhtajan, 1991; Endress, 2001). However, the hierarchy of organ relationship indicates that Nuphar and Persea perianth organs share a genetic program with their respective stamens, and, therefore, may also have a staminal origin (i.e. their tepals would therefore be andropetals). Other data also support our hypothesis that the tepals in Nuphar are derived from stamens, and are hence andropetals (Yoo et al., 2010), as has also been proposed for Persea (Chanderbali et al., 2006).

Although spatial expression patterns and differentially expressed gene proportions are similar in Nuphar and Persea (Figure 4; Table S4), the carpels and stamens of Persea appear to have a slightly more distinct transcriptome than those of Nuphar, as many more genes are twofold upregulated in these organs in Persea than in Nuphar (Table 2). Perhaps, therefore, Nuphar has a less-defined transcriptional program than does Persea, and this broader expression profile might represent the more ancient (or ancestral) transcriptional pattern, considering the basal phylogenetic position of Nuphar relative to Persea (Figure 1), and the floral morphologies of these two genera, especially their different stamen features. The stamens of Persea possess well-differentiated anthers and filaments. Hence, more genes might be involved in the morphogenesis of stamens of Persea compared with stamens of Nuphar, where stamens are petal-like without well-differentiated anthers and filaments.

The stronger similarities in the transcriptional profiles of the floral organs of Nuphar and Persea compared with those of Arabidopsis indicate an evolutionary transformation across angiosperm history towards the organ-specific refinement of ancestrally overlapping transcriptional programs. However, Nuphar and Persea also show differences in some spatial gene expression patterns, although they are similar in some aspects of floral morphology (e.g. both have largely undifferentiated perianths). In Nuphar, most of the floral genes are involved in the development of perianth members and stamens, and the floral transcriptional programs of these organs overlap substantially. In contrast, in Persea, inner and outer tepals share similar gene expression patterns, with proportionally more genes expressed in either stamens or carpels. Thus, each floral organ category (tepal, stamen and carpel) of Persea is controlled by a slightly more specific transcriptional program compared with that of Nuphar. Phylogenetic data, floral morphology and expression profiling, taken together, indicate that the broad patterns of expression observed across floral organs of the basal angiosperms Nuphar and Persea are likely to represent the ancestral condition in angiosperms, whereas the organ-specific transcriptional profiles observed in Arabidopsis are derived. Homologs of the B- and C-function genes, namely PI, AP3 and AG, have similar expression patterns in Nuphar and Persea, but are more broadly expressed than in Arabidopsis. As a consequence, the transcriptional cascades they regulate in Nuphar and Persea are also more broadly deployed. Perhaps the regulatory changes that maintain the strict spatial domains of ABCE functions have led to more spatially discrete downstream transcriptional programs during the course of floral evolution and diversification.

Experimental Procedures

Sample preparation, probe labeling and microarray hybridization

Young leaf tissue, ‘young’ floral buds at the pre-microsporangial initiation stage, ‘medium-aged’ floral buds at the pre-meiotic stage, and outer ‘sepaloid’ tepals, inner ‘petaloid’ tepals, stamens and carpels, dissected from flowers at anthesis, were collected from four individuals (biological replicates) from Pennsylvania, USA (Landherr, PAC 95537). Total RNA was extracted from all tissue samples using the RNeasy Plant Mini kit (Qiagen, http://www.qiagen.com). Both the quantity and quality of RNA were assessed using a 2100 Agilent Bioanalyzer. RNA transcripts were labeled using the Low-input RNA Labeling kit (Agilent Technologies, http://www.agilent.com), and were then hybridized to the arrays according to the manufacturer’s protocol.

Microarray design

We used custom microarrays produced by Agilent Technologies, containing 10 187 in-situ synthesized and randomly arranged 60-mer oligonucleotide probes, targeting 6220 unique Nuphar floral transcripts collected from a pre-meiotic floral cDNA library by the Floral Genome Project (Albert et al., 2005; Soltis et al., 2007; http://fgp.huck.psu.edu/). For quality control checks of hybridization, 544 Agilent controls were included in the arrays. Expression data were collected from an interwoven double-loop design (Altman and Hua, 2006) for eight tissues with 16 arrays (Figure S3). The double-loop design minimizes the variance of pair-wise comparisons between any two tissues, and efficiently detects differentially expressed genes (Altman et al., 2006).

Data acquisition and statistical analysis

Microarrays were scanned with an Agilent DNA microarray scanner using Feature Extraction v9.1.3 (Agilent Technologies). Raw data were imported into the Bioconductor package Limma, and were then processed as previously described (Chanderbali et al., 2009). After quality control checks for hybridization, arrays were background-corrected and loess normalized within arrays, and A-quantile normalized between arrays (Smyth, 2004, 2005). To identify significantly differentially expressed genes among the eight tissue samples, we employed a one-way empirical Bayes anova using single-channel analysis while considering the correlation between channels at each spot (Smyth, 2004; Smyth et al., 2005). Genes showing differential expression were assembled by hierarchical clustering (Eisen et al., 1998), as implemented by de Hoon et al. (2002), into groups with similar gene expression patterns. Specifically, log2 ratios relative to leaf (P value < 0.05; false discovery rate = 0.38%) were read into cluster v3.0 (de Hoon et al., 2002), and hierarchical clustering was performed via the centroid linkage of Pearson correlations. The cluster results were visualized using Java TreeView (Saldanha, 2004). We also searched the expression data of differentially expressed genes for transcripts with at least twofold upregulation in floral organs relative to leaves, so as to identify genes significantly upregulated in particular floral organs and/or floral stages.

The family-wise error rate and other multiple comparison methods are meant to avoid false detections when we expect very minimal differential expression. Because we are looking at different organs, we expect a very high level of true detections and more false non-detections than false detections. Hence, the estimated false detection rate is smaller than the nominal P value. For example, in the overall comparison of tissue means, an adjusted P value of 0.05 corresponds to a P value of 0.068. We therefore use the unadjusted P value to determine statistical significance, and include the corresponding estimated false discovery rate.

Comparison of microarray data with RQ-RT-PCR data

We compared our microarray data with RQ-RT-PCR data for Nuphar from Kim et al. (2005), Zahn et al. (2005) and Yoo et al. (2010) for Nuphar homologs of the floral organ identity genes AP1, AP3, PI, AG, SEP1 and AGL6. We compared the raw level of gene expression from the arrays in each tissue with that in leaves to assess whether gene expression patterns from microarrays and RQ-RT-PCR are consistent with each other.

Comparative floral transcriptomics

To investigate evolutionary patterns in floral transcriptomes, we compared the expression patterns of differentially expressed Nuphar genes with those of their putative homologs in Arabidopsis, representing the derived eudicot lineage Brassicaceae, and Persea, of the magnoliid clade of basal angiosperms (Figure 1). In the absence of reliable gene family phylogenies, except for MADS-box genes (i.e. Kim et al., 2005), we based gene homology assignments on amino acid sequence similarity (best reciprocal BLAST E score < 10−5). We combined the AtGenExpress (Development) expression data for Arabidopsis (Schmid et al., 2005) and the Persea data set (Chanderbali et al., 2009) with our Nuphar data set based on putative homology to construct two separate data sets: (i) Nuphar and Arabidopsis homologs (NA data set); and (ii) Nuphar, Persea and Arabidopsis homologs (NPA data set). Each data set was subjected to numerical search (see above for details), hierarchical clustering analyses and pairwise plots of gene expression levels in floral organs within species to compare the extent of correlation in their transcriptional profiles.


We thank C. dePamphilis, J. Carlson, K. Wall, L. Mueller, D. Ilut and W. Farmerie for EST sequencing and analysis; and the staff of the Interdisciplinary Center for Biotechnology Research at the University of Florida for technical assistance with microarray processing. We also thank Jin Koh for help with the experimental procedures. This study was supported by National Science Foundation Grants PRG-0115684 (Floral Genome Project) and PGR-0638595 (Ancestral Angiosperm Genome Project).