Fruit weight and tissue morphology tomato (S. lycopersicum L.) mutants were isolated from a highly mutagenized EMS mutant collection produced in the miniature cv. Micro-Tom as previously described (Just et al., 2013; Petit et al., 2014). In a first step, 35 mutant lines previously identified as fruit size and/or morphology mutants were phenotyped to confirm the observed phenotypic alterations (six plants/mutant line); this was done on M2 or M3 plants. In a second step, each selected mutant line was self-pollinated and M4–M8 plants were used for detailed phenotypic analysis. Mutant phenotypes are therefore considered as fixed because genome homozygosity is in the ∼93 to 99% range. Phenotyping was carried out during 4 years (2012–2015) on tomato plants grown year-round (3–4 cycles/year) in greenhouse in controlled conditions as described in Rothan et al. (2016). In order to take into account the fruit phenotypic plasticity in changing environmental conditions (mainly due to seasonal variations), the non-mutagenized Micro-Tom parental line, thereafter called WT, was cultivated side-by-side with the mutant lines and used as a reference. In addition, for all the lines, the first flower from each inflorescence was removed to take into account the high incidence of abnormal fruits produced by this flower in cv. Micro-Tom. A total of 39 different parameters thoroughly describing the plant development (3), yield components (2), organ (6), tissue morphology (11), cell morphology (8) and nucleus ploidy (9) were used for phenotyping the mutants (Supplementary Table 1). Ovary analyses were performed at anthesis (i.e., fully opened flower) before fruit set. Fruit analyses were performed at breaker stage at the onset of ripening. Breaker stage was defined as the first appearance of yellowish traces on the fruit, which takes place at about 30–35 DPA when the fruit has reached its final size. To take into account the likely influence of photoassimilate availability on fruit growth and tissue morphology, most parameters (Supplementary Table 1) were measured in standardized conditions. To this end, fruit load on the plant was reduced to six fruits distributed on two fruit trusses (controlled load) by flower pruning. In addition, to allow comparison between controlled and unrestricted fruit loads, several parameters (Supplementary Table 1) were also measured from plants in which fruit load was left free and allowed to reach up to 20 fruits per plant (unrestricted load). In addition to ovary and fruit phenotypes, plant traits with possible effect on fruit growth (fruit yield, vegetative-to-reproductive phase transition) were also considered.

Fruit tissue morphology features were determined from fresh fruit equatorial sections by scoring the proportion of each fruit tissue. Fruit equatorial sections were analyzed using the Tomato Analyser 3.0^® software (Rodríguez et al., 2010). Whole fruit, pericarp, radial pericarp, columella and locular tissue area measurements were done according to Sun et al. (2015) (Supplementary Figure 1A). Values given for the proportion of pericarp (%P), radial pericarp (%RP), locular tissue (%LT) and columella (%C) were relative to the whole fruit and were thus independent from the variations in fruit weight observed amongst the various mutants.

Fresh equatorial sections of ovaries (30 μm thickness) and breaker fruits (150 μm) were obtained using a vibration microtome (Microm HM 650 V, Thermo Scientific) prior to staining with 0.01% calcofluor or 0.05% toluidine blue (TB). Three to eleven ovaries or fruits were analyzed for each mutant line (Supplementary Table 1). Sections were then observed under an epifluorescence microscope (Zeiss Axiophot, Carl Zeiss) for calcofluor staining or with a stereomicroscope (Olympus SZX16, Olympus) for TB staining. Pericarp measurements were performed on the three layers (exocarp, mesocarp, and endocarp), excluding the vascular bundles.

Average equatorial ovary area (O_A), ovary wall thickness (OW_thick), pericarp thickness (P_thick) and number of pericarp cell layers (Cell_layer) were determined using these sections (Supplementary Figure 1B). Measurements (5 per trait) were done and averaged using the Image-Pro PLUS software (Media Cybernetics, Silver Spring, MD, United States).

These sections were also used to measure the cell area within the ovary wall and the fruit pericarp. Cell segmentation was performed using the CellSeT software (Pound et al., 2012) to optimize automatic cell counting. Area quantification was done using the Image-Pro PLUS software (Supplementary Figure 1C). For area quantification, 100 to 300 cells were observed per section to evaluate the maximum cell area (OWCell_max, PCell_max), the mean cell area (OWCell_mean, PCell_mean) and the average cell area of the 25% larger cells (OWCell_25, PCell_25). The same segmentations were used to evaluate cell shape of the 10 and 25% largest cells by scoring the X/Y cell ratio where (X) corresponds to the adaxial–abaxial and (Y) to the medio-longitudinal axes.

Cell ploidy quantification was determined by flow cytometry (CyFlow Space, Partec) using pericarp tissue from the fruit equatorial region according to Cheniclet et al. (2005). The relative proportion of each nucleus population (4C to 256C) was calculated together with the Ploidy Index (PI) and the Endoreduplication Factor (EF). Both are commonly used to estimate the mean ploidy level and the average number of endoreduplication cycles, respectively (Bertin et al., 2009). The 2C peaks were not quantified because of their low level in Micro-Tom at breaker stage (L. Fernandez, pers. obs.).

Multivariate and univariate analyses were performed using BioStatFlow application implemented using R packages (v2.7.7 INRA³) in order to identify differences and relationships between mutants or traits.

A Volcano plot was used to visualize significant phenotypic variations for fruit weight and pericarp thickness corresponding to the initial criteria used for mutant screening. Data from the different experiments were plotted in a single Volcano plot to insure consistency between the different environmental conditions. Mean comparisons between mutants and the WT were performed using Student’s t-test and choosing the false discovery rate (FDR) cut-off of 0.05 (Benjamini and Hochberg, 1995).

Principal component analysis (PCA) was used to have a global view of the data. As some parameters were analyzed over different experiments, for consistency, for each experiment the mutant values were expressed as relative to the WT (ratio between the mean value of the mutant and the mean value of the WT). The data used for PCA corresponded to the average mutant/WT ratios standardized to unit variance.

Clustering of the mutants was performed using correlation coefficients calculated on 24 fruit traits measured in controlled fruit load conditions. Spearman rank correlations were chosen in order to capture non-linear relationships. The correlations between traits were calculated separately for each mutant cluster based on the same fruit parameters together with ovary parameters. Correlations were calculated on the average mutant/WT ratios for each trait and considered significant when P-value < 0.01. Network reconstruction was performed using BioStatFlow application and visualization was done using Cytoscape software v3.1 using spring embedded layout (Shannon et al., 2003).

To investigate the induced variations in fruit weight and tissue morphology, we screened a tomato mutant collection previously generated in the miniature cultivar Micro-Tom by EMS mutagenesis. About 3500 M2 mutant lines had previously been phenotyped for over 150 phenotypic traits; within this collection the fruit shape and size diversity observed was found to be considerable (Just et al., 2013). For example, we observed in the collection 1253 multi-locular mutant families. Increased carpel number and fruit shape alterations are indeed of considerable interest to investigate variations in fruit weight and tissue morphology in cultivated tomato (van der Knaap et al., 2014) but may also hinder the existing relationships between different fruit traits (e.g., the link between pericarp cell ploidy and fruit size; Cheniclet et al., 2005). We therefore excluded fruit carpel number mutants and fruit shape mutants (e.g., round, flat, rectangular, ovoid or heart shaped fruits; Supplementary Figure 2) from the analysis. Likewise, as fruit growth patterns are likely modified in the parthenocarpic fruits which often lack locular tissue and in addition are very difficult to multiply (Serrani et al., 2007), strict parthenocarpic mutants were excluded from the analysis. However, mutants with a reduced number of seeds that could be easily maintained through sexual propagation were included in the study. At last, care was also taken to exclude mutants showing major variations in plant height and leaf attributes (number, size, shape, and color), since the carbon status of the plant can considerably impinge on fruit growth.

Phenotypic selection centered on fruit size and pericarp thickness produced, respectively, as many as ∼ 2300 and ∼1000 mutants in the collection. Taking into account the criteria indicated above, 35 mutants were selected and grown in a greenhouse. To investigate the developmental processes involved in tomato fruit growth, 12 mutants displaying robust and stable fruit phenotypes over the different environmental conditions experienced by the plants were then further characterized (Figure 1). All these mutants were comparable to the wild type in term of vegetative growth (data not shown).

To constitute the reference dataset to which all mutant phenotypic data could be subsequently compared, the tissue morphology of ovary and fruit (breaker stage) from cv. Micro-Tom was thoroughly investigated in WT plants. In Micro-Tom the final fruit size is reached after 30–35 days post anthesis (DPA), at breaker stage (Figure 2A). From ovary (1 mm diameter) to breaker stage (2.5 cm diameter), the equatorial fruit diameter increases by more than 2000-fold (Figures 2B,C) resulting in a mean fruit weight of 3 g (unrestricted load, 20 fruits per plant) to 5–6 g (controlled load, six fruits per plant). Fruit usually comprises three carpels separated by a radial pericarp (RP; 7% of the fruit area in equatorial section) fused to the central axis called columella (C; 11% of the fruit area) (Figure 2C and Supplementary Figure 1A). The pericarp (P) that develops from the ovary wall, and the locular tissue (LT) that differentiates from the placenta are the main fruit tissues, representing, respectively, 30 and 50% of the total fruit area in equatorial section (Figure 2C and Supplementary Figure 1A). The number of seeds is proportional to the final fruit weight, ca. 5 to 7 seeds per gram of fruit weight.

Starting from the abaxial (external) to the adaxial (internal) side of the fruit, the pericarp is classically subdivided into: (i) the exocarp, which is constituted by the epidermal cell layer and by several layers of small collenchyma cells; (ii) the mesocarp displaying the smaller cells close to the exocarp and to the endocarp while largest cells are located in the inner mesocarp; and (iii) the endocarp constituted of one epidermal cell layer (Figures 2A,C and Supplementary Figure 1B). Vascular bundles, which are constituted by very small phloem cells and by xylem cells, are regularly distributed within the mesocarp (Figure 2C and Supplementary Figure 1B). The ovary wall (100–120 μm in thickness) is composed of less than 10 cell layers of uniform small cells (Figure 2B). The pericarp (2 mm in final thickness) consists of 14 to 17 layers of cells with a mean area of 0.01 mm², which show considerable size heterogeneity (Figure 2C).

Final fruit weight was significantly different from that of WT in three large fruit mutants (P4E2, P1B3, and P40D11) and in six small fruit mutants (P12H11, P24C3, P6C7, P3C6, P24C17, and P8F7) (Figures 1, 3). Among these, the P3C6, P24C17, and P8F7 mutants displayed a >2-fold reduction in fruit weight (Figure 3). In addition, the P8F7, P24C3, P12G7, and P30A9 mutants produced fruits with a thicker pericarp while the only thin pericarp mutant identified was P3D3 (Figures 1, 3). Noteworthy, although P8F7 and P24C4 were small fruit mutants, their pericarp thickness (P_thick) and percentage of pericarp tissue per fruit (%P) were significantly higher than those of WT (Figure 3).

To further investigate the contribution of the various plant, fruit, tissue and cell characteristics to the variations in fruit weight and tissue morphology, we performed a principal component analysis (PCA) based on 37 phenotypic traits scored in the 12 mutants and in WT as a reference (Supplementary Table 1). PC1 clearly separated mutants according to pericarp-related traits, which are opposite to locular tissue-related traits in the PCA (Figures 4A,B). In contrast, PC2 separated mutants according to final fruit weight (Figures 4A,B). Ovary-related traits such as ovary wall thickness and ovary cell area had limited impact on total variation of PC1 and PC2 (Figure 4A). However, ovary wall traits together with nuclear ploidy (4C, 8C, and 64C values) of pericarp cells accounted for 14.6% of total variance on PC3 (Supplementary Figure 3). Because the plant-related traits scored (Node, FGD, VGD, Supplementary Table 1) poorly contributed to the total variance, they were excluded from further analyses focused on ovary and fruit traits.

To further investigate the contribution of each trait to the variations in fruit weight and tissue morphology, we built correlation networks using Spearman correlations between the 24 fruit traits scored in controlled conditions of fruit production (traits underlined in the PCA shown in Figure 4A). The correlation network identified positive and negative relationships between the mutants and, rather surprisingly given the large phenotypic diversity observed in the mutants, clearly separated the mutants in only two distinct major clusters (Figure 5A). Cluster 1 included the P4E2, P1B3, and P40D11 large fruit mutants negatively correlated to P6C7 and P12H11 small fruit mutants (Figures 4B, 5A). Cluster 2 included the P3C6, P8F7, P24C17, P30A9, P12G7, P24C3, and P3D3 mutants. Cluster 2 mutants are mainly thick pericarp mutants, except for the P3D3 thin pericarp mutant which is negatively correlated to the P24C3 mutant (Figures 4B, 5A). Cluster 1 encompasses fruit weight variations centered on PC2 while cluster 2 is related to pericarp thickness variations centered on PC1 (Figure 4A).

We next built, for each mutant cluster, a correlation network between the parameters describing ovary and fruit characteristics. This allowed us to explore their interrelations at organ, tissue and cellular levels and to analyze their contribution to fruit weight and tissue morphology (Figures 5B,C). For both clusters, ovary network was independent from fruit network except that ovary wall thickness and cell size were correlated to columella in cluster 1 and to fruit weight in cluster 2 (Figures 5B,C). Thus, fruit weight appears as independent from other fruit characteristics and as related to pre-anthesis ovary development in cluster 2 (Figure 5C). For fruit traits, in cluster 1, positive correlations were observed between fruit weight, % of locular tissue and columella, seed number, pericarp cell size and high ploidy (64C and 128C); negative correlations were observed between these traits and lower ploidy values (16C and 32C) and % of pericarp tissues (Figure 5B). The positive correlations between fruit pericarp characteristics (proportion in the fruit, thickness and cell size) and high (128C) to extreme (256C) ploidy levels were the main features of cluster 2; negative correlations were observed between these traits and % of locular tissue and columella (Figure 5C).

We next focused on main fruit parameters highlighted in the correlation network analysis (Figures 5B,C) to investigate more deeply the fruit developmental patterns that characterize the mutants in each cluster. Common features were identified amongst cluster 1 mutants regarding final fruit weight, pericarp cell size and pericarp nuclear ploidy (Figure 6A). In this cluster, the relative proportion of the pericarp and of other tissues (locular tissue, radial pericarp, and columella) remained similar to that of the WT, regardless of the final fruit weight (Figures 6A,B). Likewise, the seed number remained proportional to fruit weight. Surprisingly, in view of the commonly admitted hypothesis that cell number is a major determinant of fruit weight (Frary et al., 2000), the number of cell layers in the pericarp was barely affected despite the large variation in fruit weight (Figures 6A,C). In contrast, the main changes for large fruits were the strong increase in 128C nuclei content and to a lesser extent in 64C nuclei, to which was associated a slight increase in pericarp cell size. The opposite was true for small fruit mutants.

In fruits from cluster 2 mutants, a very different growth pattern was prevailing. In this cluster, fruit weight was either similar to that of the WT or was strongly reduced. However, in contrast to cluster 1, fruit weight reductions were not accompanied by the proportional reduction in all fruit tissues. On the contrary, in all cluster 2 mutants but one (P3D3), the proportion of the pericarp and radial pericarp was substantially increased to reach 40–49% of the fruit tissues (∼30% in WT). Conversely, the proportion of locular tissue was strongly reduced to 27–37% (∼50% in WT), as was the number of seeds (Figure 6A). The resulting thick-walled fruits had a bulky and fleshy appearance, reminiscent of processing tomatoes (Figure 6B). As already observed for cluster 1 mutants, the number of pericarp cell layers remained largely unaffected. Increased proportion of pericarp tissue in the fruit was associated with increased pericarp thickness and/or pericarp outgrowth in the inner part of the fruit (Figure 6B) and with high ploidy values (64C to 256C). Noteworthy, even for the most contrasted P3C6, P8F7, and P24C17 small fruit phenotypes, the pericarp cell area was at least equivalent or higher than that of the WT, while in cluster 1 small fruit phenotypes were associated with smaller pericarp cell area. The thin-pericarp P3D3 mutant, which has the same size as several thick-pericarp mutants displaying high ploidy values, exhibited the opposite trend. Compared to thick-pericarp mutants, the proportions of fruit tissues were inversed (20% of pericarp and 59% of locular tissue) and the fruit had the lowest nuclear ploidy and pericarp cell area values of all the mutants.

Because endopolyploidy is a major determinant of cell expansion and fruit growth in tomato (Cheniclet et al., 2005; Chevalier et al., 2014), we further examined the pericarp nuclear ploidy levels of the mutants. Mutants were classified according to the proportion of 64C to 256C nuclei into “low” (P3D3, P12H11, P6C7 mutants), “high” (P4E2, P24C17, P40D11, P3C6, P1B3, and P24C4) and “extreme” (P8F7, P12G7, and P30A9) ploidy categories (Figure 6A and Supplementary Figure 4). The most striking feature is the large increase in 64C to 256C nuclei in the “extreme” ploidy mutants, and in particular of 128C and 256C nuclei. It is noteworthy to mention that 256C nuclei are not detected in WT and other mutant pericarp cells and that there is a 10-fold increase in 128C nuclei in these mutants compared to the WT (Supplementary Figure 4).

Even though endopolyploidy is a major factor contributing to fruit tissue growth, and hence to fruit size, additional mechanisms are likely involved in the coordination of fruit growth at organ or tissue levels. As illustrated in fruit pericarp sections (Figure 6C), the largest cells appeared more elongated in the thick-pericarp mutants from cluster 2 than in WT or cluster 1 mutants. We therefore evaluated for these cells the X/Y ratio, where X and Y describe the cell dimensions according to the two main axes of fruit growth, along the medio-lateral axis (X, periclinal growth) and along the abaxial–adaxial axis (Y, anticlinal growth) (Figures 2A, 7A). In cluster 1 mutants, the largest cells within the pericarp harbored a round shape except for one mutant (P6C7) where cells were slightly flattened (Figure 7B). Strikingly, in cluster 2, several thick-pericarp mutants (P24C17, P8F7, P30A9, and P12G7) showed a significant increase in the cell elongation along the Y-axis, which may explain the formation of lobes at the inner face of the pericarp. The thin-pericarp mutant P3D3 displayed the opposite phenotype, with highly flattened cells.