Whole organ, venation and epidermal cell morphological variations are correlated in the leaves of Arabidopsis mutants


J. L. Micol. E-mail: jlmicol@umh.es


Despite the large number of genes known to affect leaf shape or size, we still have a relatively poor understanding of how leaf morphology is established. For example, little is known about how cell division and cell expansion are controlled and coordinated within a growing leaf to eventually develop into a laminar organ of a definite size. To obtain a global perspective of the cellular basis of variations in leaf morphology at the organ, tissue and cell levels, we studied a collection of 111 non-allelic mutants with abnormally shaped and/or sized leaves, which broadly represent the mutational variations in Arabidopsis thaliana leaf morphology not associated with lethality. We used image-processing techniques on these mutants to quantify morphological parameters running the gamut from the palisade mesophyll and epidermal cells to the venation, whole leaf and rosette levels. We found positive correlations between epidermal cell size and leaf area, which is consistent with long-standing Avery's hypothesis that the epidermis drives leaf growth. In addition, venation parameters were positively correlated with leaf area, suggesting that leaf growth and vein patterning share some genetic controls. Positional cloning of the genes affected by the studied mutations will eventually establish functional links between genotypes, molecular functions, cellular parameters and leaf phenotypes.


A major challenge in the post-genomic era is the functional characterization of all of the genes in the genomes of model species. To this end, large-scale approaches toward the targeted inactivation of every gene have been undertaken in animals such as Caenorhabditis elegans (Kamath et al. 2003), Drosophila melanogaster (Winkler et al. 2005), mouse (Wu et al. 2007) and zebrafish (Wang et al. 2007), as well as in plants such as Arabidopsis thaliana (Alonso et al. 2003), maize (Fernandes et al. 2004) and rice (He et al. 2007). The emerging field of phenomics pursues the systematic study of phenotypes caused by individual mutations in every gene of a genome, an approach that is expected to provide a mechanistic link between genotypes and phenotypes (Lussier & Liu 2007). In these large-scale mutant screens, the phenotypes of mutants from gene-indexed collections must be recorded and quantified under standard conditions or in response to altered environments.

Forward genetic screens for mutations affecting the size and shape of Arabidopsis vegetative leaves have resulted in the isolation of hundreds of mutants (Berná, Robles & Micol 1999; Serrano-Cartagena et al. 1999; Horiguchi et al. 2006; Micol 2009). However, the causal genes for the phenotypes of most of these leaf mutants have yet to be identified. Thus far, we have positionally cloned 33 of the 76 genes initially identified in our laboratory based on the leaf phenotype of their mutant alleles (Robles & Micol 2001) and found that their wild-type products participate in processes as diverse as polar cell expansion, the transduction of hormonal signals, gene regulation, plastid biogenesis and chromatin remodelling (Pérez-Pérez et al., 2009b). Strikingly, in several cases, mutations that we initially classified together based on macroscopic phenotype actually affect genes involved in a single pathway or molecular mechanism, implying that gene–morphology relationships are highly reproducible, such as in the denticulata (den) mutants with pointed leaves, where at least four of which carry alleles of genes encoding ribosomal proteins (Horiguchi et al. 2011). In addition, the elongata (elo) mutants, with narrow and elongated leaves, led to the identification of genes encoding subunits of the transcription Elongator complex (Nelissen et al. 2005). In this way, it is becoming possible to begin mapping the organizational levels that connect genes with the whole leaf phenotype, starting with the cellular and tissue levels.

Several methods have been developed for the automated quantification of shape and size variations in biological structures such as leaves (Langlade et al. 2005; Weight, Parnham & Waites 2008). Using these methods, a principal component analysis (PCA) of quantitative data measured from a collection of leaf shape mutants of Arabidopsis and Anthirrinum majus uncovered key variations in leaf shape and size (Bensmihen et al. 2008). In addition, two groups have recently screened thousands of sequence-indexed Arabidopsis T-DNA insertion mutants for morphological phenotypes in leaves (Ajjawi et al. 2010; Myouga et al. 2010). However, the characterization of the cellular basis of leaf phenotypes in extensive mutant collections has not been reported so far.

We present here the results of a morphometric analysis of cellular parameters in the first- and the third-node vegetative leaves in 111 Arabidopsis mutants, which broadly represent the variation of mutational origin previously reported for leaf morphology in this species. Our detailed characterization of the cellular processes affected in this set of mutants will help us to identify some of the developmental pathways involved in leaf growth, and to determine the contribution to final leaf size of the different tissue layers within the leaf. Molecular cloning of the genes affected by the studied mutations will uncover novel functions required during leaf development. In addition, the quantification of size and shape parameters at the organ, tissue and cell levels in this set of mutants provides a quantitative baseline for leaf trait variation that will be useful for the study of the leaf phenotypes observed in other, gene-indexed, collections of loss-of-function mutants.


Plant material and growth conditions

Seeds of the Arabidopsis thaliana (L.) Heynh. accessions Landsberg erecta (Ler), Columbia-0 (Col-0) and Enkheim-2 (En-2) were obtained from the Nottingham Arabidopsis Stock Centre (NASC). Ninety-six mutants were isolated in the laboratory of J.L. Micol in the Ler genetic background by either fast neutron bombardment (five mutants; P. Robles and J.L. Micol, unpublished results) or ethyl methanesulfonate (EMS; 91 mutants; Bernáet al. 1999) mutagenesis. Eight mutants from the Arabidopsis Information Service Form Mutants collection were generated in the En-2 background (Serrano-Cartagena et al. 1999). The aba2-1, aba3-3, axr2-1, axr4-1, cue1-5 and ga5-1 mutants were generated by EMS mutagenesis in the Col-0 background and were provided by NASC. The hve-1 mutation is a spontaneous mutation carried by the Ei-5 accession (Candela, Martínez-Laborda & Micol, 1999). Unless otherwise stated, we used homozygous lines in all of our analyses (Supporting Information Table S1).

Seeds were surface sterilized, stratified for 2 d at 4 °C to synchronize their germination, and plated under sterile conditions in 150 mm diameter Petri dishes containing 100 mL of half-strength Murashige and Skoog agar medium supplemented with 1% sucrose. Plants were grown at a density of 20 per plate, representing two mutant lines and a wild-type line (Fig. 1a). Sterile cultures were incubated in Conviron TC16 growth chambers at 20 ± 1 °C, 60–70% relative humidity and with continuous illumination of 78 µmol m−2 s−1. All genotypes were sown in triplicate plates.

Figure 1.

Phenotyping of Arabidopsis leaf mutants. (a) An example of the rosette morphologies of plants grown in vitro. The plate shown contained eight anu9-1 (top semi-circle) and eight anu10 (bottom semi-circle) homozygous seedlings. Four wild-type Ler plants were grown as growth references close to the edges of the plate. (b) A first-node vegetative leaf from Ler after clearing with chloral hydrate; the region of the lamina that was analysed is coloured red. (c) DIC micrograph of the adaxial epidermis of a third-node vegetative leaf from Ler, with the cells that were analysed highlighted in red. (d) Subepidermal palisade mesophyll underlying the pavement and guard cells shown in (c), with the cells that were analysed coloured red. Scale bars: (a) 2 cm, (b) 2 mm and (c,d) 50 µm. All samples were collected at 21 days after stratification.

Microscopy and morphometric analysis

Twenty-one days after stratification (das), Petri dish pictures were taken with a Panasonic DMC-FX9 digital camera (2816 × 2112 pixels; Osaka, Japan). Individual rosette information was extracted from these pictures using the NIS Elements AR 2.30 image analysis package (Nikon, Tokyo, Japan). Measurements were taken for rosette area (RA), perimeter (RP), and maximum (RXF) and minimum (RNF) Feret's diameters (i.e. the longest and the shortest distances between any two points in the measured object, respectively), as well as for the area (EA), perimeter (EP), and major (EXC) and minor (ENC) chord lengths of the best-fitting ellipse containing the rosette. Rosette compactness (RC) was calculated as RA/EA, and rosette evenness (RE) was calculated as RP/EP.

A minimum of five leaves from the first and third nodes were manually excised, immediately submerged in 70% ethanol and stored at 4 °C for 2 d in 24-well tissue culture polystyrene plates. Samples were then submerged in a clearing solution (80 g chloral hydrate in 30 mL water) until photosynthetic tissues became transparent and veins were visible, as previously described (Candela et al. 1999). Whole leaves were mounted on glass slides in solutions of 80 g chloral hydrate, 20 mL glycerol and 10 mL water. Single-leaf pictures to visualize leaf margins and veins were taken with a Nikon SMZ1500 stereomicroscope equipped with a Nikon DXM1200F digital camera (3840 × 3072 pixels). Microscopic visualization of leaf tissues was performed using differential interference contrast (DIC) optics on a Leica DMRB microscope (Leica Microsystems GmbH, Wetzlar, Germany) equipped with a Nikon DXM1200 digital camera. Pictures from the adaxial epidermis and the underlying palisade mesophyll were taken halfway along the primary vein and the leaf margin as previously described (Tsuge, Tsukaya & Uchimiya 1996; Serrano-Cartagena et al. 2000) for a minimum of four leaves per genotype.

Venation pattern, adaxial epidermis and palisade mesophyll diagrams were obtained by drawing leaf veins and cell borders, respectively, on the screen of a Wacom Cintiq 18SX Interactive Pen Display screen (Wacom Company Ltd., Tokyo, Japan) and using the Adobe Photoshop CS3 program, as previously described (González-Bayón et al. 2006).

Venation pattern and epidermal drawings were individually processed by C++ image analysis algorithms that make use of the SDC Morphology Toolbox (http://www.mmorph.com/cppmorph/). A detailed description of the image analysis methods will be published elsewhere. Measurements of the area (LA), perimeter (LP), length (LL) and width (LW) of the first and third leaf lamina, as well as the circularity (LC), were performed as previously described (Pérez-Pérez, Serrano-Cartagena & Micol, 2002). From the venation pattern diagrams of the first and third node leaves, the venation length (TVL), number (FVN) and mean length (FVL) of free-ending veins, number of vein branching points (BPN), and number (ALN) and mean area (ALA) of areoles were measured (Candela et al. 1999; Alonso-Peral et al. 2006). Vascular density (VAD) and complexity (VAC) were calculated as TVL/LA and BPN/LA, respectively.

The area (MCA), perimeter (MCP), and major (MCX) and minor (MCN) chord lengths of the best-fitting ellipse were measured for each palisade mesophyll cell using ImageJ 1.40 g (http://rsb.info.nih.gov/ij/). The total area and number of drawn cells were determined for each diagram from adaxial epidermis micrographs. Cells smaller than 50 µm2 were excluded from the morphometric analysis to minimize drawing errors. Stomatal pores were manually coloured, allowing the identification and measurement of individual guard cell areas. New diagrams including only pavement cells were automatically generated, and the area (PMA), perimeter (PMP), major (PMX) and minor (PMN) chord lengths of the best-fitting ellipse and circularity (PMC) were measured for each individual pavement cell using ImageJ. All macros used are available upon request.

Measurement of quantitative variation and statistical analysis

Descriptive statistics [mean, standard deviation (SD), maximum and minimum] and linear correlations were calculated using the StatGraphics Centurion XV (StatPoint Technologies Inc., Warrenton, VA, USA) and SPSS 16.0.2 (SPSS Inc., Chicago, IL, USA) programs. PCA (Pearson 1901) was used to reduce the dimensions of our sets of variables. PCA involves the calculation of the eigenvalue decomposition of a data covariance matrix, usually after mean centring the data for each attribute (Jolliffe 2002). Hierarchical agglomerative clustering was performed with Ward's method (Ward 1963), where the cutoff distances for the optimal number of clusters were determined by Mojena's criterion (Mojena 1977). To compare the data for a given variable among the various genotypes studied, we performed multiple testing analyses with the analysis of variance F-test or the Fisher's least significant differences (LSD) methods (Fisher 1918). Multiple comparisons between various sets of variables were performed by canonical correlation (Hotelling 1936) and multiple regression (Pearson 1908) analyses. One-sample Kolmogorov–Smirnov tests (Massey 1951) were performed to analyse the goodness-of-fit between the distribution of the data and a theoretical distribution (normal, log-normal, normal-inverse logistic, log-logistic, gamma or Rosin–Rammler).


Growth conditions for phenotyping

We first analysed the effects of seedling density on the overall growth of the Landsberg erecta (Ler), Columbia-0 (Col-0) and Enkheim-2 (En-2) accessions, which represent the wild-type genetic backgrounds for the mutants studied in this work. Densities above 20 plants per plate inhibited leaf growth (Supporting Information Fig. S1a). Therefore, we ultimately chose a grid containing 20 squared areas for the sowings, such that each plate included two mutant genotypes (eight plants each) and their wild-type accession (Fig. 1a; see Materials and methods). To account for the morphological differences between different leaves, we determined the areas of first- and third-node vegetative leaves in Ler and Col-0 (Supporting Information Fig. S1b), as previously described for the En-2 accession (Pérez-Pérez et al. 2002). As bolting time did not substantially differ between all the studied lines (data not shown), we collected leaf samples for morphometry at 21 d after stratification (das). In the wild-type genetic backgrounds at 21 das, the first- and the third-node leaves had ended their rapid growth phase and reached about 75 and 70%, respectively, of their final sizes at bolting (Supporting Information Fig. 1b). Leaf venation and margins were drawn by hand from the micrographs of cleared leaves (Fig. 1b), and their morphological parameters were automatically extracted from the drawings and measured as described in Materials and methods. Despite the spatial variation of cell size and cell shape within the leaf lamina (Donnelly et al. 1999), a representative area was selected halfway between the primary vein and the leaf margin as previously described (Tsuge et al. 1996; Serrano-Cartagena et al. 2000). Cell borders were drawn by hand from DIC micrographs of the adaxial epidermis and the palisade mesophyll (Fig. 1c,d). Cell sizes and shape parameters were subsequently quantified as described in Materials and methods.

We characterized 111 Arabidopsis mutants, 103 of which carried recessive mutations. Five mutants were originally generated by fast-neutron bombardment, 97 by EMS treatment, one was spontaneous and eight were of unknown origin (Supporting Information Table S1).

Rosette morphologies

To reduce the number of parameters describing rosette morphology, we performed a PCA (see Materials and methods). Three principal components (PCs) accounted for 97% of the variation among the studied samples (Supporting Information Table S3A). PC1 explained 78.1% of the total variance and largely corresponded to the rosette size. PC2 and PC3 accounted for 11.1 and 7.7% of the variance, respectively, and were mostly related to RC and RE (this is a measure of rosette serration), respectively (Supporting Information Table S3B; Fig. 2a). Consistently with our PCA, we found highly significant and positive correlations for most rosette parameters related to size, while the correlations found involving RC and RE were not significant (Fig. 2b and Supporting Information Table S3C). To visualize the effects of PC1, PC2 and PC3 on rosette morphology, representative rosette diagrams are depicted in Fig. 2c, where the PC values vary plus or minus two SDs from the mean. Hereafter, we use RA, RC and RE as the best descriptors for PC1, PC2 and PC3, respectively.

Figure 2.

Principal component analysis of rosette morphometry. (a) Representation of PC1, PC2 and PC3 in a tridimensional space. (b) Scatter plots of rosette parameters. (c) Rosette size and shape variation among mutant samples. For each PC, a representative rosette diagram corresponding to plus or minus two times the standard deviation (+2SD and −2SD) is shown. The rosette diagrams of the ven6-1, exi9-1, api2, icu3-1, ven4, exi5 and sca3-1 mutants are shown from left to right.

To estimate the phenotypic variability within a given genotype, we performed hierarchical cluster analysis of RA, RC and RE data (Supporting Information Table S3D) and we calculated the distribution of each individual measurement in a tridimensional space (Supporting Information Table S3E; see Materials and methods). We named this factor as morphometric heterogeneity (Supporting Information Table S3F), which could have values from 0 (phenotypically homogeneous line) to 1 (phenotypically heterogeneous line). Only in a few cases all of the rosette data for a given genotype belonged to the same tridimensional cluster, indicating low morphological heterogeneity. For instance, most data for the as1-12, elo2 and sea2-1 genotypes fit into clusters 15, 20 and 10, respectively (Supporting Information Table S3F). Our results indicate that numerical quantification of rosette morphologies is not sufficient to discriminate between any two unrelated genotypes due to the morphological heterogeneity in their rosette phenotypes.

RA, RC and RE were normalized to the wild-type values to allow direct comparisons between all the mutants studied, regardless of their genetic backgrounds. On average, rosette mutant sizes were approximately 60% smaller (128.9 ± 76.5.0 mm2, n = 947) than those of their wild-type accessions (340.4 ± 95.7 mm2, n = 107), and only the axr4-1 rosettes were significantly larger (417.1 ± 45.0 mm2) than those of Col-0 (302.0 ± 41.5 mm2). The extreme values for RC corresponded to icu5, as1-12 and axr2-1 with increases of 1.90-, 1.53- and 1.48-fold compared with the RC values of their wild-type counterparts, respectively. The icu3-1, elo1-1, elo2-1, elo3-1 and elo4-1 mutants displayed the lowest compactness values (0.65- to 0.50-fold) compared with those of Ler. RE ranged from 0.58- (icu5) to 1.24-fold (den1). We found that average values for RC and RE are good descriptors for the morphology of individual leaves; hence we propose using RA, RC and RE parameters in case of a high-throughput phenotypic characterization of mutant phenotypes is foreseen.

Leaf morphometry and venation patterns

Regarding the leaf lamina and vein parameters, we found highly significant and positive correlations between size-related lamina parameters (LA, LP, LL and LW) and venation length (TVL) (Supporting Information Fig. S3 and Supporting Information Tables S4C and S5C), which to us suggest that leaf growth and vein patterning share some genetic controls. In addition, parameters related to the topology of the venation [number of vein branching points (BPN), free-ending veins (FVN) and areoles (ALN)] were also highly significant and positively correlated with venation length, which is consistent with the branching net-like pattern of Arabidopsis leaf venation (Candela et al. 1999). The highest correlation between pairs of parameters measuring venation was found between ALN and BPN (89.9 and 94.3% for the first- and third-node leaves, respectively).

A PCA of leaf lamina and vein parameters allowed us to identify four main PCs (PC1 to PC4) that accounted for 88.2 and 90.0% of the variation in first- and third-node leaves, respectively (Supporting Information Tables S4A and S5A). Interestingly, we found highly significant and positive correlations for PC1, PC2 and PC3 between first- and third-node leaves (Table 1) which reflect a similar behaviour in trait variation regardless of the leaf node (first or third) studied. PC1 positively correlated with variations in overall lamina size parameters, as well as with TVL, BPN, ALN and FVN in both first- (PC1.1; Supporting Information Table S4B) and third-node leaves (PC1.3; Supporting Information Table S5B). PC2 positively correlated with ALA and FVL, but negatively correlated with VAD and VAC. PC3 mostly accounted for LC, while FVL and FVN negatively and positively influenced PC4, respectively.

Table 1.  Linear correlations between principal components of first- (PCX.1) and third-node (PCX.3) leaves
  1. Data analysis was performed as indicated in Materials and methods. Highly significant correlations (P values < 0.005) are shown in bold.

PC2.1 −0.003−0.0290.1110.987−0.1070.001
PC3.1  −0.0130.0330.0360.6590.765
PC4.1   −0.2860.0430.727−0.628
PC1.3    −0.019−0.2240.060
PC2.3     −0.0180.005
PC3.3      0.056

We normalized the leaf lamina and venation data to allow direct comparisons between all mutants, irrespectively of their leaf size and genetic background. Although most of the mutants have been isolated based on their abnormal leaf morphology (Bernáet al. 1999; Serrano-Cartagena et al. 1999), we identified new venation phenotypes within the studied mutants. Plotting the average values for VAD uncovered the already known extreme vascular phenotype of hve-1 (Fig. 3g), with a 35% decrease in VAD compared with the Ler background (Fig. 3c,k) both in the first- (VAD1) and third-node (VAD3) leaves. The anu6 (Fig. 3f) and ond4-1 (Fig. 3h) mutants displayed only low VAD1 values, while ang4 (Fig. 3n) displayed significant decreases only in VAD3 values. Mutants with higher vein densities were also identified as having high VAD values: orb1-1 (Fig. 3d) displayed a 1.48-fold increase in VAD1, while sea1 (Fig. 3l) and ond3 (Fig. 3m) displayed 1.62- and 1.55-fold increases, respectively, in VAD3. In addition, ang4 (Fig. 3n), den14 (Fig. 3i), ron1 (Fig. 3j) and api7 exhibited increased numbers of disconnected veins, as shown by the FVN/TVL ratio (Supporting Information Fig. S4). With our detailed analysis of venation characteristics in the collection of leaf mutants studied we aimed to identify novel phenotypes affecting vein patterning.

Figure 3.

Vein phenotypes found among the leaf mutants. (a–j) First-node leaf margin (orange) and venation (black) diagrams. (a) En-2 and (c) Ler wild-type venation patterns. (b,d,e) Increased vein densities found in (b) icu2-1 (compared with En-2), (d) orb1-1 and (e) as3-1 (compared with Ler). (f–h) Reduced vein densities exhibited by (f) anu6, (g) hve-1 and (h) ond4-1. (i,j) Poor peripheral vein connectivity, which increases the number of free-ending veins, shown by (i) den14 and (j) ron1. (k–o) Third-node leaf margin and venation diagrams. (k–m) Higher vein densities in (l) sea1 and (m) ond3 than in (k) Ler. (n) Reduced vein density and increased number of free-ending veins in ang4. (o) Prominent hydathodes shown by anu8-2. Scale bar: 2 mm.

Palisade mesophyll analysis

Mesophyll cell area (MCA) was highly significantly and positively correlated with all other mesophyll parameters except circularity (MCC; Supporting Information Table S6A). Wild-type MCA ranged from 500 to 3500 µm2 (Fig. 4a). Six mutants (sea2-1, ond3, exi9-1, icu9-1, orb1-1 and exi3) displayed average MCA values smaller than 325 µm2. Only four mutants (anu9-1, den16, den15 and anu2) exhibited average MCA values higher than 2300 µm2 and included very large mesophyll cells of up to 9000 µm2 (Supporting Information Table S6B), which is a characteristic trait of mutants exhibiting the ‘compensated cell-enlargement’ mechanism that is thought to operate during leaf growth (Tsukaya 2006). Considering the average MCA values, a 24-fold range in palisade mesophyll cell size was found between the extremes, represented by exi1-1 (Fig. 4e) and anu9-1 (Fig. 4f). We iteratively tested several statistical functions (see Materials and methods) to determine the best-fitting distribution for each genotype data (Supporting Information Table S6B). The MCA values for the three wild-type genotypes fit the gamma distribution (Supporting Information Table S6B), and representative histograms are shown in Supporting Information Fig. S5a–d. From these results, we concluded that non-parametric methods should be tested when we want to significantly compare the differences between MCA data from different genotypes and their genetic backgrounds.

Figure 4.

Palisade mesophyll morphometry. Box-plots of (a) mesophyll cell area (MCA) and (b) mesophyll cell circularity (MCC) data as estimates for cell size and shape, respectively. Mutants exhibiting extreme values are outlined. (c) Scatter plot of the average values for LA3 and MCA of the studied lines. Mutants with small leaves that also contain small mesophyll cells are depicted in blue. Mutants with large mesophyll cells and small leaves are represented in red. The wild types are shown in green. (d–f) Mesophyll micrographs from (d) Ler and (e) exi1-1 with small mesophyll cells, and from (f) anu9-1 with very large mesophyll cells. Scale bars (d–f): 50 µm.

We found no significant correlation between leaf area (LA3) and MCA (r2 = 0.023, P value > 0.01) (Fig. 4c), which supports our hypothesis that mesophyll growth poorly contributes to leaf expansion (González-Bayón et al. 2006). Indeed, mutants with small leaves (<30 mm2) displayed a wide range of average MCA values: some of these mutants (sea2-1, exi9-1 and icu1-9) had small mesophyll cells, while others (anu9-1, den16, den15 and anu2) contained very large mesophyll cells compared with their Ler background (Fig. 4d). Average values for MCC, which estimates the cell shape, exhibited low variation (Fig. 4b) and ranged from 0.64 in exi5 to 0.83 in den8. This result was rather predictable because the shape of mesophyll cells in most of the studied lines is regularly cylindrical. Four mutants (api5, exi5, icu3-1 and sea2-1) displayed the highest range of values for MCC, which is due to the heterogeneities in the morphology of their mesophyll cells.

In our study of a large set of leaf mutants, we found that mesophyll cell size is highly variable between genotypes and it was not significantly correlated with leaf size. With our mathematical tabulation of mesophyll cell morphologies, we spotted some leaf mutants that are candidate to be altered in the mechanism for cell-size compensation during leaf growth, whose study ultimately will lead us to a better understanding of mesophyll growth coordination.

Cellular parameters at the adaxial epidermis

We found a highly significant and positive correlation between pavement cell area (PMA) and all of the size parameters measured in these cells (PMP, PMX and PMN) except circularity (PMC; Supporting Information Table S7A). The sizes of adaxial pavement cells, measured as PMA, were highly heterogeneous in the wild-type lines (Fig. 5a and Supporting Information Table S7B). The PMA values did not fit normality for nearly all of the genotypes, and the most common best-fitting distributions were the log-normal (50 mutants; Supporting Information Fig. S5e) and the gamma (34 mutants; Supporting Information Fig. S5f). Indeed, when we analysed together all of the PMA values (36 946 cells), the data fit a log-normal distribution (Fig. 5b). Most mutants displayed PMA values lower than those of their wild types, and five (exi9-1, icu1-9, sea2-1, orb1-1 and icu9-1) displayed average PMA values lower than 550 µm2 (Supporting Information Table S7B). Only the icu13, hve-1, icu14, icu5, as2-11, icu12 and aba1-1 mutants displayed substantially larger pavement cells (measured as average PMA; Supporting Information Table S7B) than their wild types. Pavement cell densities (number of pavement cells per mm2) and lamina areas were used to estimate the total number of pavement cells on the third-node leaves for each genotype. The number of estimated pavement cells ranged from 2500 in icu1-9 to 70 000 in as3-1 (28-fold range). We plotted the average values for LA3 against log10PMA (Fig. 5f) and found a low but significant correlation between both variables (r2 = 0.085, P value = 0.002), suggesting that the size of adaxial pavement cells may contribute to leaf size. Interestingly, some mutants with small leaves belonging to the Exigua phenotypic class had small pavement cells that were of normal lobe shape (Fig. 5c). Only a few mutants with small leaves, such as exi9-1 (Fig. 5d), icu1-9 and sea2-1, displayed small pavement cells with fewer lobes. In contrast, mutants with very large pavement cells, such as icu4-1 (Fig. 5e) and as2-11, did not exhibit large leaves, likely due to a concomitant reduction in their pavement cell numbers. We found a highly significant and negative correlation between pavement cell size (PMA) and stomatal density (r2 = 0.370, P value = 0.000), confirming that the pavement cell/stomata ratio was not altered in the majority of the leaf mutants studied.

Figure 5.

Adaxial epidermal cell morphometry. (a) Box-plots of pavement cell area (PMA) data. (b) Histogram for the entire PMA dataset with an overlay of the theoretical normal distribution. (c–e) Representative micrographs of the adaxial epidermis from leaf mutants displaying small (c, exi5) and large (e, icu4-1) pavement cells with normal morphologies, or from mutants displaying small cells with altered morphologies (d, exi9-1). (f) Scatter plot of the average values for LA3 and PMA. Mutants with small leaves and small pavement cells are represented in blue. Mutants with large pavement cells are represented in red. Wild-type genotypes are shown in green. (g) Box-plots of guard cell area (GCA) data for the studied lines. Mutants exhibiting extreme values are outlined. (h) Scatter plot of the average values for stomatal densities and GCAs. Mutants with small or large guard cells are depicted in blue or red, respectively. Wild-type lines are shown in green. Scale bars (c–e): 50 µm.

From our analysis of the adaxial epidermis, we found a fourfold range in the average values for guard cell area (GCA) between the mutants with extreme values, represented by exi9-1 (47.32 ± 17.87 µm2) and aba3-3 (199.04 ± 42.71 µm2) (Fig. 5g). In contrast to that found for palisade mesophyll cells, most GCA data fit the normal distribution (Supporting Information Fig. S5g–h), with only a few exceptions (Supporting Information Table S8). Seventy-three leaf mutants displayed average GCA values ranging from 120 to 165 µm2, and only four (aba3-3, ang3, anu2 and den2; red dots in Fig. 5h) showed larger guard cells (>170 µm2) than their wild types. In addition to exi9-1, the icu6, exi2-1, ond3 and icu9-1 mutants characteristically included guard cells sized <90 µm2 (blue dots in Fig. 5h). We then estimated adaxial stomatal densities (i.e. the number of stomata per mm2) for each genotype. Three mutants (icu9-1, icu1-9 and orb1-1) displayed more than 520 stomata per mm2, while the stomatal densities for icu5, icu3-1 and exi8-1 were fewer than 115 stomata per mm2. In addition, the numbers of stomata per leaf were estimated from the average LA3 values. Third-node leaves from as3-1, ven4 and api2 were estimated to contain more than 20 000 stomata, while those of their wild-type Ler had only 12 000. In some cases, mild defects in stomatal patterning were found, such as in aba2-1 and ven1, which displayed increased frequencies of adjacent stomata. The variation found in guard cell size and patterning in the studied mutants suggested that their altered leaf size and shape is not primarily caused by defects in the development of stomata.

We finally questioned whether a linear combination of some of the cellular descriptors measured in third-node leaves could explain the size differences observed for rosette areas. Thus, multiple regression analysis was performed between the average values for rosette area and these cellular parameters (see Materials and methods). The most plausible model indicates a low but highly significant correlation (r2 = 0.206, P value = 0.000) between RA and MCC, GCA, PMA and PMP. Taken together (and with only a few exceptions), these data are consistent with the hypothesis that cells at the epidermis contribute to the size of the leaf (Satina, Blakeslee & Avery 1940) and that any other alteration in epidermal cell size and/or number could modify the leaf size accordingly.


We developed a standardized method for phenotyping Arabidopsis plants grown in vitro and tested it by analysing rosettes, leaves and leaf cells in 111 mutants and three wild-type accessions.

Leaf area was negatively correlated with the number of seedlings grown in Petri dishes, and a significant reduction was observed in cell expansion in the leaves of plants grown at high densities. This observation likely reflects competition among neighbouring plants for mineral resources within the Petri dish, as previously reported for other plant species (Sadras & Denison 2009). Nearly all of the mutants analysed in this work displayed smaller rosettes with altered leaf shapes, and only the auxin resistant4-1 (axr4-1) mutant, whose wild-type allele encodes a positive regulator of the auxin influx (Dharmasiri et al. 2006), displayed an increased size of the rosette. The bias over small-sized mutants in our study reflects the screening approach by which most of these mutants were initially isolated (Bernáet al. 1999). Consistent with the role for auxin in regulating organ size, loss-of-function mutations in the AUXIN RESPONSE FACTOR2 (ARF2) gene increase leaf size by enhancing both cell division and cell expansion (Okushima et al. 2005; Schruff et al. 2006).

We defined several parameters to quantify the rosette morphologies of mutant lines (Supporting Information Table S2) and we experimentally validated some of them. Thirteen of the denticulata (den) mutants studied here were initially isolated because of their pointed leaf laminae (Bernáet al. 1999) and displayed higher rosette evenness than Ler (this work). Further, all four elongata (elo) mutants display lower rosette compactness values than Ler due to their long petioles and narrow leaves, while mutants with short petioles and broad leaves [e.g. incurvata5 (icu5), asymmetric leaves1-12 (as1-12) and axr2-1] had the highest compactness values. The rosette parameters that we have chosen (area, evenness and compactness) are useful for a high-throughput numerical characterization of phenotypes among distinct genotypes or in response to specific environmental conditions.

We found that the length of leaf venation was highly significant and was positively correlated with the number of pattern elements (free-ending veins, branching points or areoles), as well as with the lamina size. This finding is consistent with leaf venation (which is a net-like structure in Arabidopsis) being produced by an iterative process of increasing complexity (Candela et al. 1999). Our results demonstrate that screening for venation mutants on first-node leaves is informative and much less laborious than the analysis of more complex (adult) leaves. Additionally, gathering information on the vascular densities, number of areoles per area unit and number of free-ending veins per mm of venation allowed the direct comparison of the vascular patterns between the mutants, irrespective of the sizes of the leaves and their genetic backgrounds. We identified altered venation patterns within our leaf mutant collection, such as those in hemivenata-1 (hve-1) and ondulata4-1 (ond4-1) with lower vein densities, in den14 and angusta4 (ang4) with disconnected peripheral veins, and in orbiculata1-1 (orb1-1), serrata1 (sea1) and ondulata3 (ond3) with increased vein densities. Our results indicate that some venation traits, such as venation length and the number of patterning elements, are positively correlated with lamina area, suggesting that leaf growth and vein patterning share at least some genetic controls. A detailed cellular study of mutants with lower vein densities and small leaf size will help to unravel the genetic pathways involved. HVE encodes a CAND1 protein that negatively regulates ubiquitin-mediated auxin signalling (Alonso-Peral et al. 2006), and its loss-of-function causes simple venation patterns (Candela et al. 1999) and small leaves with lower cell numbers (this work). Further experiments are thus required to address the direct role of CAND1 in promoting cell division in non-vein leaf tissues. In agreement with our hypothesis that leaf growth and vein patterning share some genetic controls, we have identified in this work ANG4 as a novel regulator of venation pattern in Arabidopsis. ANG4 encodes the functional homolog of the yeast and human histone H2B-monoubiquitinating BRE1 RING E3 ligase, whose mutations reduce the number of cell divisions (Fleury et al. 2007). Further characterization of these and other vein mutants using in vivo timing of venation pattern formation (Rolland-Lagan, Amin & Pakulska 2009) and identification of the genes altered in the corresponding mutants is currently ongoing.

To dissect the relationship between variations at the cellular and whole organ levels, we quantified cellular parameters in the adaxial epidermis and in the palisade mesophyll, as well as leaf size and venation characteristics in the third-node leaves of a collection of 111 leaf mutants. The leaves of angulata8 (anu8) and anu9 were similar in size in both mutants and significantly smaller than those in the wild type (Ler). Mesophyll cells in anu8 were similar in size to those in Ler. However, anu9 displayed very large mesophyll cells. Thus, the small leaf sizes observed in the anu8 and anu9 mutants are likely caused by reduced cell proliferation. Understanding why mesophyll cells in anu9, but not in anu8, display compensatory cell enlargement (Tsukaya 2008) requires the molecular identification of these two ANU genes. Probing the genetic interactions of anu with other mutants specifically affected in cell-size compensation, such as angustifolia3 (Horiguchi et al. 2006), would help to define the role of the ANU genes in organ growth control. Furthermore, six of the den mutants studied here display small leaves but very large mesophyll cells, suggesting that these mutations also trigger cell-size compensation.

We found no significant correlation between mesophyll cell size and leaf area, which is in agreement with a nutritional (photosynthetic) function for the mesophyll. Interestingly, the size of epidermal cells was positively and significantly correlated with leaf area, suggesting that it is the epidermis and not the mesophyll that contributes the most to the final size of the leaf (Satina et al. 1940). Consistent with this scenario, we previously found that the strong decrease in mesophyll cell numbers in reticulata (re) mutants does not significantly affect leaf size (González-Bayón et al. 2006). Here, we present additional evidence that other leaf mutants, such as those in the Venosa (Ven), Anu and Den phenotypic classes, with a large decrease in mesophyll cell numbers, display only slight modifications in leaf size and shape, confirming a secondary role of the inner mesophyll layer for leaf growth and development.

Another interesting finding of our study is that the size of adaxial pavement cells for a given genotype failed to fit normality, but instead followed a log-normal distribution. Because the leaves that we sampled were not fully expanded, the variation in pavement cell areas within a given genotype can be attributed to different timings for exiting mitosis or endoreduplication. Many of the leaf mutants studied are characterized by small pavement cells in their adaxial epidermis. The small leaves of most mutants in the Exigua (Exi) class contained small cells in both the epidermis and the mesophyll and displayed no other defects in their shapes. One plausible explanation for this result is that the EXI genes are positive regulators of cell expansion, a possibility that is currently being addressed in our laboratory.

As regards the average size of guard cells among the studied mutants, those of the aba3-3 mutant were the largest found. ABA3 encodes a putative molybdenum cofactor sulfurase required for the last step of the biosynthesis of ABA (Xiong et al. 2001), a hormone that regulates stomatal closure (Desikan et al. 2004). Thus, the effects of ABA loss-of-function alleles in guard cell size are likely indirect because ABA directly regulates guard cell turgor. Interestingly, the sizes of pavement and guard cells were uncoupled in only a few cases. The epidermis of orb1 contained small and unlobed pavement cells, while those in the as2-11 mutant were instead very large. Both mutants displayed normal-sized stomata, suggesting that the expansions of pavement and neighbouring guard cells are independently regulated in these mutants. AS2 is required for leaf primordium specification (Semiarti et al. 2001) and for the establishment of the adaxial-abaxial polarity within the growing leaf (Wu et al. 2008). The molecular identification of ORB1 would shed light on the specific mechanism controlling the growth of pavement cells.

The quantitative data obtained from this large collection of leaf mutants is a powerful means to dissect functional relationships between whole rosette phenotypes and their cellular parameters. The power of our leaf mutant dataset is underlined by the observation that entire classes of mutants selected at the rosette scale displayed very similar changes at the cellular level. In the few cases that were analysed in detail (Nelissen et al. 2005), this effect even extends to the molecular processes that are involved. We hypothesize that the same applies to other groups of mutants in our and other mutant collections. This hypothesis will be tested by the molecular cloning of all of the genes whose mutations lead to visible leaf phenotypes (as identified in our laboratory) and by analysing the cellular phenotypes of known mutations from T-DNA insertion collections (Alonso et al. 2003). Our phenomics approach will help us to functionally dissect the synergistic phenotypes obtained when combining mutants affected in different cellular processes (Pérez-Pérez, Candela & Micol, 2009a). Once the molecular nature of the genes and their interactions are known, one could begin elucidating the entire molecular networks required for leaf growth.


The authors wish to thank Héctor Candela for comments on the manuscript. This work was supported by the Ministerio de Ciencia e Innovación of Spain [BIO2007-30797-E, BIO2008-04075, and CSD2007-00057 (TRANSPLANTA) grants] and the Generalitat Valenciana (PROMETEO/2009/112), by a pre-doctoral fellowship from the Agency for Innovation through Science and Technology to S.D. and by the European Commission [LSHG-CT-2006–037704 (AGRON-OMICS)].