Genotype-dependent response to carbon availability in growing tomato fruit.

Abstract

Tomato fruit growth and composition depend on both genotype and environment. This paper aims at studying how fruit phenotypic responses to changes in carbon availability can be influenced by genotype, and at identifying genotype-dependent and -independent changes in gene expression underlying variations in fruit growth and composition. We grew a parental line (Solanum lycopersicum) and an introgression line from Solanum chmielewskii harbouring quantitative trait loci for fresh weight and sugar content under two fruit loads (FL). Lowering FL increased fruit cell number and reduced fruit developmental period in both genotypes. In contrast, fruit cell size was increased only in the parental line. Modifications in gene expression were monitored using microarrays and RT-qPCR for a subset of genes. FL changes induced more deployments of regulation systems (transcriptional and post-transcriptional) than massive adjustments of whole primary metabolism. Interactions between genotype and FL occurred on 99 genes mainly linked to hormonal and stress responses, and on gene expression kinetics. Links between gene expression and fruit phenotype were found for aquaporin expression levels and fruit water content, and invertase expression levels and sugar content. In summary, the present data emphasized age- and genotype-dependent responses of tomato fruit to carbon availability, at phenotypic as well as gene expression level.

Full text

Page 1

1
Genotype-dependent response to carbon availability in growing tomato
1
fruit
2
Short running title: Responses to carbon availability in tomato fruit
3

4
Marion PRUDENT1,2,*, Nadia BERTIN1, Michel GENARD1, Stéphane MUÑOS2, Sophie
5
ROLLAND2, Virginie GARCIA3, Johann PETIT3, Pierre BALDET3, Christophe ROTHAN3,
6
Mathilde CAUSSE2
7
1 INRA, UR1115 Plantes et Systèmes de culture Horticoles, F-84000 Avignon, France
8
2 INRA, UR1052 Génétique et Amélioration des Fruits et Légumes, F-84000 Avignon, France
9
3 INRA, UMR619 Biologie du fruit, F-33883 Villenave d’Ornon, France
10
11
* Author to whom correspondence should be sent: 12
Marion PRUDENT 13
Address: INRA, UMR Génétique et Ecophysiologie des Légumineuses à graines, 17 rue de Sully, 14
21000 Dijon, France 15
Tel : 00 33 380 693 681 16
Fax : 00 33 380 693 263 17
E-mail : marion.prudent@dijon.inra.fr 18
19
20
hal-00600426, version 1 - 15 Jun 2011
Author manuscript, published in "Plant Cell & Environment (2010) 33, 1186-1204"

Page 2

2
Abstract
1
Tomato fruit growth and composition depend on both genotype and environment. This paper 2
aims (i) at studying how fruit phenotypic responses to changes in carbon availability can be 3
influenced by genotype and (ii) at identifying genotype-dependent and -independent changes in 4
gene expression underlying variations in fruit growth and composition. To this end, we grew a 5
parental line (S. lycopersicum) and an introgression line from S. chmielewskii harboring QTL for 6
fresh weight and sugar content under two fruit loads (FL). Lowering fruit load increased fruit cell 7
number and reduced fruit developmental period in both genotypes. In contrast, fruit cell size was 8
increased only in the parental line. Modifications in gene expression were monitored in expanding 9
fruits using microarrays and RT-qPCR for a subset of genes. FL changes induced more deployments 10
of regulation systems (transcriptional and post transcriptional) than massive adjustments of whole 11
primary metabolism. Interactions between genotype and FL were especially noticeable for 99 12
genes mainly linked to hormonal and stress responses, and on gene expression kinetics during 13
fruit development. Links between gene expression and fruit phenotype were found for aquaporin 14
expression levels and fruit water content, and invertase expression levels and sugar content during 15
fruit ripening phase. In summary, the present data emphasized age- and genotype-dependent 16
responses of tomato fruit to carbon availability, at phenotypic as well as at gene expression level. 17
Keyword index
18
Fruit growth, gene expression, genotype x environment interaction, hormone, metabolism, 19
regulations, Solanum lycopersicum, stress response, sugar, transcriptome 20
21
hal-00600426, version 1 - 15 Jun 2011

Page 3

3

1
Introduction
2
Carbohydrate availability is a major factor limiting plant growth, in particular for sink organs such 3
as fruits. In tomato, carbohydrates needed for fruit growth come from photosynthetic sources 4
such as mature leaves, fruits having a low photosynthetic activity (Farrar & Williams 1991). 5
Increasing carbohydrate availability to the reproductive organs by reducing fruit load on the plant 6
enhances cell division in the ovary and thereby the final fruit size (Baldet et al. 2006). Conversely, a 7
low carbohydrate supply during the phase of rapid cell expansion leads to a reduction of fruit size 8
(Bertin et al. 2003; Heuvelink 1997) and even of dry matter (Gautier, Guichard & Tchamitchian 9
2001). In the case of severe carbon stress, sugar, protein and amino-acid contents can all be 10
reduced (Baldet et al. 2002; Gary et al. 2003). Fruit sugar content is the consequence of sucrose 11
import, carbohydrate metabolism, and dilution by water (Ho 1996). Sucrose enters the cells either 12
via the apoplasm after conversion to hexose by a parietal invertase, or via the symplasm. In the 13
cytoplasm, sucrose may be converted into fructose and glucose by invertase, or into fructose and 14
UDP-glucose by sucrose synthase (Frommer & Sonnewald 1995; Yelle et al. 1988). Hexoses are 15
then transformed into starch by the successive actions of fructokinase, hexokinase, 16
phosphoglucoisomerase, phosphoglucomutase, ADP-glucose-pyrophosphorylase and starch 17
synthase (Schaffer & Petreikov 1997a; Schaffer & Petreikov 1997b; Damari-Weissler et al. 2006). 18
Starch is transiently stored in the amyloplasts, and constitutes a carbon reservoir for hexose 19
synthesis (Dinar & Stevens 1981). The accumulation of carbohydrates in the fruit leads to a 20
gradient of osmotic pressure, causing first a massive entrance of water, notably via aquaporins 21
(see Kaldenhoff et al. 2008 for review), and subsequently cell expansion (Ho, Grange & Picken 22
1987). This expansion relies on cell wall plasticity, determined partly by the activity of enzymes 23
related to the synthesis or degradation of cell wall components in the epidermis as well as in the 24
hal-00600426, version 1 - 15 Jun 2011

Page 4

4
pericarp (Thompson, Davies & Ho 1998). Many of these processes are affected by hormones. 1
Cytokinins may affect fruit sugar content (Martineau et al. 1995), while auxins and gibberellins can 2
be involved in the regulation of cell enlargement, by controlling the expression of genes encoding 3
cell wall modifying proteins like expansins (Catala, Rose & Bennett 2000; Chen & Bradford 2000; 4
Chen, Nonogaki & Bradford 2002; Guillon et al. 2008). 5
Large-scale quantification of gene expression in tomato has been shown to be a powerful tool for 6
characterizing plant response to a variety of conditions, including salt stress (Ouyang et al. 2007), 7
light (Facella et al., 2008), developmental changes (Alba et al. 2005; Vriezen et al. 2008), 8
differentiation of specialized tissues (Lemaire-Chamley et al. 2005), mutants (Kolotilin et al., 2007) 9
or introgressions of genomic segments (Baxter et al. 2005). To our knowledge, the effect of 10
carbohydrate availability on tomato transcriptome has never been investigated, although many 11
genes are known to be sugar sensitive (Koch 1996). In fruit, the effect of carbohydrate availability 12
was only investigated for target enzymes and genes at few developmental stages. For example, it 13
was shown in peach fruit that high crop load leads to increased acid invertase activity at a final 14
stage of fruit growth (Morandi et al. 2008). In tomato, an obscurity-induced carbohydrate 15
limitation led to changes in the expression of some sugar transporters, and of enzymes involved in 16
sugar- or amino-acid metabolism at two stages of fruit development (cell division and cell 17
expansion) (Baldet et al. 2002). However, although recent studies emphasized that source-sink 18
relationships between vegetative and reproductive organs are genetically controlled and are a 19
central hub for controlling fruit metabolism in tomato (Schauer et al. 2006; Lippman, Semel & 20
Zamir 2007), comparative analysis of fruit response to carbohydrate availability in different 21
tomato genotypes has not been looked at in any depth. 22
This study aims at identifying genes and gene categories differentially regulated in growing fruit in 23
response to changes in carbon availability induced by alteration of fruit load on the plant. 24
Transcriptome analysis using tomato microarrays was carried out on fruit pericarps harvested 21 25
hal-00600426, version 1 - 15 Jun 2011

Page 5

5
days after anthesis, a stage of rapid fruit growth, cell expansion and storage of major 1
carbohydrates such as starch. To get an insight into interactions between genotype and 2
environment on gene expression, two closely-related tomato genotypes were grown at two 3
contrasted fruit loads, one (high load, HL) inducing competition for carbon among fruits (trusses 4
were not pruned), and the other (low load, LL) inducing low or no competition (trusses pruned to 5
one fruit). The two accessions differed by an introgressed chromosome fragment on chromosome 6
9, carrying several quantitative trait loci whose expression is either independent of the fruit load 7
(for fruit developmental duration, fresh weight, seed number, dry matter and sugar content) or 8
specific to one fruit load (cell number, cell size or fruit cracking) (Prudent et al. 2009). In a last 9
step, the expression of a subset of genes related to carbon metabolism, cell wall modification, and 10
water fluxes were analyzed using RT-quantitative PCR along fruit expansion and ripening. 11
Materials and methods
12
Plant growth, fruit thinning and sampling
13
Two tomato genotypes, Solanum lycopersicum cv. Moneyberg (hereafter called M) and an 14
introgression line (hereafter called C9d) carrying a fragment from the bottom of chromosome 9 of 15
Solanum chmielewskii in the M genetic background, were grown under controlled greenhouse 16
conditions at a day-night temperature set point of 25/15 °C during spring 2007 in Avignon, France. 17
The position of the introgression of genotype C9d as well as the quantitative trait loci (QTL) 18
previously identified, are described in Prudent et al. (2009). For both genotypes, all trusses were 19
pruned to one fruit when flower 2 was at anthesis (low fruit load condition or LL) on 40 randomly 20
selected plants, while trusses of 16 other plants were not pruned (high fruit load condition or HL). 21
Under HL condition, the average number of fruit sets per truss was similar in C9d and M (around 22
seven fruits). Anthesis was recorded three times a week, allowing the determination of fruit age 23
and fruit developmental duration. Fruits were harvested at five different developmental stages 24
hal-00600426, version 1 - 15 Jun 2011

Page 6

6
from the cell expansion period until maturity (21, 28, 35, 42 days after anthesis (daa) and red ripe). 1
Fruits were weighed, and locular tissue and seeds were removed. The pericarp tissue was then 2
weighed, frozen in liquid nitrogen and stored at -80°C until nucleic acid manipulations or stored at 3
-20°C until phenotypic measurements.
4
Phenotypic measurements
5
At each developmental stage, six fruits per genotype and per fruit load were randomly harvested 6
between the fourth and the ninth truss of the plants, at proximal positions: flowers 2, 3 or 4 under 7
HL and only flower 2 under LL. Pericarp powders were lyophilized and weighed and pericarp water 8
content was then deduced. Soluble sugars (glucose, fructose, sucrose) and starch were extracted, 9
quantified by an enzymatic method (Gomez et al. 2007; Gomez, Rubio & Auge 2002) and 10
expressed in g per 100 g of pericarp dry matter. Additional phenotypic measurements were 11
carried out on fruits harvested at the red ripe stage: seeds were counted, and the number and the 12
mean size of pericarp cells were assessed according to the method described in Bertin, Gautier & 13
Roche (2002). An analysis of variance was performed on phenotypic measurements in order to 14
study the effects of fruit age, fruit load, genotype and their two- or three-way interactions with R 15
Software (http://www.r-project.org). 16
RNA extractions
17
All fruits used for RNA extraction were randomly harvested between the fourth and the ninth truss 18
of the plants, at proximal positions, similarly to phenotypic measurements. Fruit samples used for 19
microarray and quantitative real time PCR were different. Total RNA used for microarray 20
experiment was isolated from two biological pools of 25 fruits at 21 daa, following the procedure 21
described by Chang, Puryear & Cairney (1993). Total RNA used for quantitative real time PCR was 22
isolated from three biological pools of ten fruits (at 21, 28, 35, 42 daa and red ripe stage) with TRI 23
hal-00600426, version 1 - 15 Jun 2011

Page 7

7
Reagent® Solution (Ambion) following the procedure described by the manufacturer, with minor 1
modifications.
2
Microarray experiments
3
TOM2 cDNA glass slides were fabricated by arraying Array-Ready Oligo Set™ for the Tomato 4
Genome (Operon) onto Corning® UltraGAPS™ slides using a BioRobotics MicroGridII arrayer 5
(Genomic Solutions) in Instituto de Biología Molecular y Celular de Plantas (Valencia, Spain). Slides 6
contained 12,160 oligos corresponding to 11,862 unique and randomly selected transcripts of the 7
tomato genome. Information about the oligo set was described on the Operon website 8
(http://omad.operon.com/download/storage/lycopersicon_V1.0.2_datasheet. pdf). Fluorescent 9
probes were prepared from 50 µg of total RNA, using the Amino Allyl MessageAmpTM II aRNA 10
Amplification Kit (Ambion), and following manufacturer’s specifications. Purified Cy3- or Cy5-11
labelled AA-aRNAs were dried in a speed-vac, resuspended in 9 µL nuclease free water and 12
fragmented using RNA Fragmentation Reagents (Ambion). Probes were mixed with 88 µL of the 13
hybridization solution consisting of 0.5 µg.mL-1 denatured salmon sperm DNA (Stratagene), 5X SSC, 14
0.25% SDS and 5X Denhardt’s solution and 50% formamide. The solution was denaturated for 1 15
min at 100°C and cooled down to 37°C. Labelled AA-aRNAs were hybridized with slides for 16h at 16
37°C under agitation. After hybridization, slides were washed at 30°C with 2X SSC, 0.2% SDS for 5 17
min, with 0.2X SSC, 0.1% SDS for 5 min, with 0.2X SSC for 3 min, and with 0.02X SSC for 30 s. 18
Finally, slides were dried with nitrogen gas for 3 min at 30°C before scanning.
19
Data analysis of microarray experiments
20
The microarray experimental design consisted of two biological replicates, and for each biological 21
replicate the dyes were reversed (dye swap) for a total of four slides per comparison (Fig. 3a). 22
The raw data corresponding to the median spot intensities, with no background subtraction were 23
analyzed using the Bioconductor (http://www.bioconductor.org) package R/Maanova v1.4.1 (Cui, 24
hal-00600426, version 1 - 15 Jun 2011

Page 8

8
Kerr & Churchill 2003) in R-2.6 (http://www.r-project.org). Data visualization, normalization, and 1
statistical analysis, including multiple test adjustments (FDR) were performed as described by 2
Mounet et al. (2009). F statistics computed on the James-Stein shrinkage estimates of the error 3
variance (Wu et al. 2003) were calculated and genes with a P value < 5.10-2, FDR < 0.05, fold 4
change ≥ 1.6 and average intensity > background mean + 2 background SD were selected. 5
In order to analyze specifically the interaction between the genotype and the fruit load, another 6
statistical study was performed on normalized data using Bioconductor LIMMA package v2.13.8 7
(Smyth 2005a). The data were normalized using the printtiploess (within-array normalisation) and 8
scale (between-arrays normalisation) functions (default parameters). Flagged spots were given a 9
weight of 0.1 using the weight function. A factorial design analysis (Smyth 2005b) was performed 10
and a linear model with a coefficient for each of the four factor combinations (C9dLL, C9dHL, MLL 11
and MHL) was fit. The interaction term (C9dLL-C9dHL) - (MLL-MHL) was extracted. The P values 12
resulting from moderated t test were corrected for multiple testing using the Benjamini-Hochberg 13
FDR adjustment. As above, genes were considered to be significantly differentially expressed if 14
adjusted P values were < 5.10-2, fold change was ≥ 1.6. For each of the four conditions, spots with 15
an average intensity higher than background mean + 2 background SD were considered as 16
detected. 17
Quantitative real-time PCR
18
The quantitative real-time PCR experiment followed the eleven golden rules proposed by Udvardi, 19
Czechowski & Scheible (2008). Reverse transcription was performed with 2 µg of total RNA from 20
each sample treated with DNAse in 50µL with oligo-dT (10 µM) and AMV Reverse Transcriptase 21
(10 U/µL) (Promega), according to the manufacturer’s instructions. The RT mix was diluted 5-fold 22
in water and 2 µL aliquots were stored before use. Quantitative real-time PCR was carried out 23
using a Stratagene Mx3005P® thermocycler (Stratagene, Cedar Creek, TX) in a reaction volume of 24
hal-00600426, version 1 - 15 Jun 2011

Page 9

9
20 µL, in 96-well plates. PCR were conducted using 7.5 µL SYBR Green mix (133 mM KCl, 27 mM 1
Tris HCl pH 9, 0.2 % Triton® X-100, 8 mM MgCl2, 2 µL of 1500-fold dilution SYBR®Green I Nucleic 2
Acid Gel Strain 10000X (Lonza), 0.5 mM each dNTP, 2 µL of the 5-fold dilution of RT mix, 1 U Taq 3
DNA polymerase, 9.8 µL H2O, and 0.12 µM of each primer. Primer sequences are detailed in 4
Supporting Information Table S1. PCR conditions were: 2 minutes at 95°C, followed by 40 cycles at 5
95°C for 20 sec, 20 sec at the primer specific temperature (~55°C) and at 72°C for 35 sec. A 6
thermal denaturation curve of the amplified DNA was carried out, in order to measure the melting 7
temperature of the PCR product. For each reaction, three technical replicates were run. Relative 8
gene expression was calculated by the
T
C

2
method (Livak & Schmittgen 2001), with the 9
eukaryotic translation initiation factor 4A-2 (eIF-4A-2) (U213502) as an internal control. The use of 10
this unigene as a reference was validated under our experimental conditions, as advised by 11
Gutierrez et al. (2008). For each genotype, and at each developmental stage, fruit load effect on 12
gene expressions was evaluated using a Student’s t test with R software, and the significance of 13
the interaction between genotype and fruit load was tested via a two-way analysis of variance in 14
R. Normalized gene expression data were analyzed by principal component analysis (PCA) using 15
the “princomp” function in R. The principal component scores were then plotted for individual 16
observations.
17
Functional categorization
18
Classifications of unigenes into functional groups were obtained from MapMan ontology 19
classifications (Thimm et al. 2004), gene family assignments from TAIR 20
(http://www.arabidopsis.org), and the literature when unigenes exhibited no homology with 21
Arabidopsis thaliana proteins. Functional categories were restricted to eleven: cell wall 22
modification, electron transport, hormonal responses, photosynthesis, primary metabolism, 23
hal-00600426, version 1 - 15 Jun 2011

Page 10

10
protein metabolism, secondary metabolism, signalling, stress responses, transcription and 1
transport. 2
Results
3
Fruit load effect on fruit phenotype
4
Tomato fruits of the two lines, Moneyberg (M) and the introgression line C9d, were harvested at 5
five developmental stages from 21 daa to the red ripe stage, and their phenotype was described 6
either throughout fruit development (Fig. 1), or only at maturity (Fig. 2). The analysis of variance 7
(not shown) underlined significant interactions between fruit age and fruit load for all traits 8
measured kinetically. At maturity, seed number was the only trait which was not affected by fruit 9
load. For the other traits, the significant fruit load effect was either genotype-independent or 10
associated to significant genotype x fruit load interactions. When fruit load was reduced to one 11
fruit (LL), fruit development duration decreased and cell number increased, without any 12
interaction with genotype (Fig. 2b). On the other hand, larger fruits (Fig. 1a), associated with 13
higher cuticular macrocracks (Fig 2a), lower pericarp water and sugar contents (Fig. 1b and 1d) 14
and higher starch content (Fig. 1c) were consistently observed during fruit development in fruits 15
grown under LL, when compared to HL fruits in both genotypes. However sugar contents were 16
similar at maturity. For these traits, the effect of fruit load interacted significantly with genotype, 17
for at least one fruit developmental stage. It was also the case of cell size, which was higher under 18
LL conditions for M, but not for C9d (Fig. 2b). 19
At 21 daa, which corresponds to the period of rapid fruit growth and to the developmental stage 20
chosen for microarray transcriptome analysis, modification of fruit composition in response to 21
changes in fruit load was significant in both genotypes for sugar content (Fig. 1d) and in M for 22
water and starch content, indicating a strong significant interaction between genotype and fruit 23
load (G x FL) for these two traits. 24
hal-00600426, version 1 - 15 Jun 2011

Page 11

11
1
Fruit load effect on the transcriptome from 21 daa fruits and genotype-fruit load
2
interactions
3
Microarray experimental design and data analysis
4
In the tomato genotypes studied, the 21 daa stage of fruit development corresponds to the cell 5
expansion phase during which rapid fruit growth occurs and starch accumulation peaks (Dinar & 6
Stevens 1981). At 21 daa, fruit load effect was obvious for starch and soluble sugar contents; no 7
significant effect on fruit weight and water contents could be observed (Fig. 1a and 1b). This stage 8
of fruit development appears therefore critical for the control of fruit growth and of the 9
concomitant changes in fruit composition. 10
In order to reduce the biological variability between samples, we used 40 randomly 11
selected plants for the low load (LL) condition and 16 plants for the high load (HL) condition. For a 12
given genotype and developmental stage, up to 50 fruits were randomly picked from the plants. 13
From these, large pools of 25 fruits each were constituted at random, as previously described 14
(Buret, Duby & Flanzy 1980). This sampling and pooling strategy effectively reduces the variability 15
between biological replicates e.g. by excluding the environmental effects linked to plant location 16
in the field, and allows thus to focus only on the effects of genotype and fruit load. 17
In addition, the microarray experiment was designed as a loop where C9dLL was compared 18
to C9dHL itself compared to MHL itself compared to MLL itself compared to C9dLL (Fig.3a). 19
Compared to single comparisons, this design decreases considerably the variance of estimated 20
effects (Churchill 2002; Yang & Speed 2002), allowing much more confidence in the estimation of 21
the differentially expressed genes. Four technical replicates (including 2 dye swaps) were done per 22
comparison i.e. 8 hybridizations per comparison (2 biological repeats X 4 technical repeats). To 23
further increase the significance of the results from statistical analyses, the threshold for log2 fold-24
hal-00600426, version 1 - 15 Jun 2011

Page 12

12
change was set-up to 0.68 (1.6 fold change) i.e. a value higher than that commonly used in similar 1
studies (e.g. Wang et al. 2009). As a result of this combination of pooling strategy, experimental 2
design and statistical analyses, a smaller number of differentially expressed genes identified by 3
transcriptome analysis can be anticipated but more robust results are expected. Indeed, in a 4
previous study done in tomato using a similar design with fewer technical replicates, most of the 5
results from microarray analysis were further validated by qRT-PCR (Mounet et al., 2009). 6

Impact of fruit load on fruit transcriptome
7
8
Because fruit load effect on fruit development and composition is the main focus of our study, we 9
excluded from the statistical analyses the comparison of genotypes for the same load. The effect 10
of fruit load on tomato fruit transcriptome in each genotype was analyzed by comparing gene 11
expressions under HL and LL in M and in C9d (Fig. 3a). In total, 103 genes out of 11,862 (0.9 %) 12
were differentially expressed in M, versus 147 in C9d (1.2 %) (Fig. 3b). Among them, 56 13
differentially expressed genes were common to C9d and M with similar responses whatever the 14
genotype, except for one unigene showing an opposite response depending on the genotype. 15
To identify which biological processes are modified by a fruit load change, unigenes were classified 16
into functional categories. Figure 4a displays the distribution of the differentially expressed 17
unigenes into functional categories (Supporting Information Table S2 and Table S3). According to 18
the Mapman classification, all biological processes were affected by fruit load but to a variable 19
extent. As expected from the modification of the source/sink relationships under LL, the 20
expression of genes implicated in primary and secondary metabolism was indeed affected. 21
Accordingly, LL fruits also displayed changes in cell wall related genes (mostly up-regulated) and in 22
electron transport (mostly down-regulated). A large proportion of the differentially expressed 23
genes, mostly down-regulated, have no function attributed. However, the most striking 24
differences between HL and LL –in terms of expression level-, were observed for genes involved in 25
hal-00600426, version 1 - 15 Jun 2011

Page 13

13
signalling, in regulation of gene expression (hormonal responses, transcription) and in protein 1
metabolism (protein synthesis, post-translational modifications and degradation). Details of 2
several categories showing remarkable changes are displayed in Table 1. In primary metabolism, 3
the enzyme exhibiting the highest changes under LL conditions in both M and C9d genotypes (log2 4
fold-change = ~-1.6 in M, ~-1.5 in C9d) was the fructose-2,6-bisphosphatase (F26BPase). Several 5
differentially expressed genes with putative function in signalling exhibited high fold-changes 6
under low load conditions in both M and C9d (1.4 < log2 fold-changes < 4.8). While functions of 7
genes encoding proteins with calmodulin-binding motif and leucine rich repeats may remain 8
elusive, the COP9 signalosome is a key player of the machinery controlling protein degradation 9
that regulates a variety of processes in plants including light-regulated development and hormone 10
signalling (Chamovitz 2009). In addition, some ethylene, auxin and cytokinin-related genes with 11
roles in hormone biosynthesis, degradation or responses were also up- or down-regulated under 12
LL (Supporting Information Table S2 and Table S3). Among the 29 differentially expressed genes 13
encoding transcription factors and other proteins involved in gene regulation, 9 displayed very 14
consistent up-regulation (2 genes) or down-regulation (7 genes) in both M and C9d under LL. 15
Several transcription factor families were represented (HB, bHLH, AP2/EREBP, C2H2 zinc finger, 16
bZIP …). Of these, the gene encoding a TAZ zinc finger protein exhibited very high and strikingly 17
similar changes in M and C9d under low load. The fold change log2 value was ~-5.3 for both 18
genotypes, i.e. a ~40-fold reduction in transcript abundance in the 21 daa fruits from M and C9d 19
cultivated under LL conditions. With the notable exception of the elongation factor 1-alpha 20
implicated in the protein synthesis machinery and of protein phosphatase 2C, most of the genes 21
classified into the protein metabolism category and differentially expressed under LL are involved 22
in protein degradation. 23
24
Interactions between genotype and fruit load
25
hal-00600426, version 1 - 15 Jun 2011

Page 14

14
The previous study showed independently the effect of fruit load on each genotype. Several genes 1
displayed very similar fold changes in the two genotypes (e.g. the F26BPase and the TAZ 2
transcription factor) while others were mostly affected in one or the other genotype (e.g. the AS2 3
or the CCAAT box transcriptional regulators preferentially up-regulated in C9d) (Table 1). In order 4
to specifically identify the LL affected genes displaying a genotype-dependent change in gene 5
expression, we analyzed the genotype x fruit load (G x FL) interaction using LIMMA. Ninety-nine 6
genes showing significant G x FL interactions (Supporting Information Table S4) were identified 7
and further classified into functional categories (Fig. 4b). Among them, a high proportion belonged 8
to hormonal and stress response categories albeit all biological processes were targets of G x FL 9
interactions. For all functional categories, most of the genes up-or down-regulated under low load 10
conditions showed a greater variation of gene expression in one or the other genotype 11
(Supporting Information Table S4). In this table, positive values indicates that the corresponding 12
gene was more up-regulated or down-regulated under LL relative to HL conditions in C9d than in 13
M while negative values indicate the opposite. Their absolute value gives an indication of the 14
extent of the differences between the two genotypes. The largest category comprised genes 15
related to various hormone biosynthetic pathways including ethylene (ACC synthase and ACC 16
oxidase), auxin (IAA hydrolase), jasmonic acid (lipoxygenase, ent-kaurenoic acid hydroxylase) and 17
phytosulfokines and salicylate (Table 2). All the hormone-related genes displaying G x FL 18
interactions were more expressed in C9d while cell wall related genes were more or less expressed 19
in one or the other genotype (Table 2).
20
Validation of microarray data
21
The abundance of transcripts from selected genes was monitored by qRT-PCR in order to validate 22
the microarray data, on independent fruit samples. Quantitative PCR was carried out on eleven 23
unigenes, and confirmed the changes in transcript abundance (Supporting Information Fig. S1). 24
hal-00600426, version 1 - 15 Jun 2011

Page 15

15
Because of the difference in sensitivity of the two techniques (already observed in Mohammadi, 1
Kav & Deyholos 2007 and Fernandez et al. 2008), in a lot of cases the ratio values based on 2
quantitative PCR were higher than those based on microarray but the Pearson’s correlation 3
coefficient between the two methods was high (r = 0.92, p = 7.10-16). 4
Fruit load effect on expression of selected genes along fruit development
5
To gain further insight into how genes reacted to the fruit load change in the two genotypes, we 6
performed qRT-PCR on five stages from the cell expansion phase to fruit maturity, in the two 7
genotypes and under the two fruit loads. The functions of the 15 selected genes were related to 8
processes potentially involved in fruit growth and carbohydrate accumulation. Some genes were 9
chosen based on microarray results while others were selected because of their importance in 10
metabolism (when absent from the DNA chip). The correspondence between gene codes, their 11
annotations and their sequence references are detailed in Table 3. As shown on Fig. 5, for each of 12
the 15 observed genes, we identified (i) a significant fruit load effect whatever the fruit 13
developmental stage, for M, for C9d or for both genotypes and (ii) a significant G x FL interaction 14
at least at two developmental stages. 15
Water flux-related genes
16
Two aquaporins: a delta-tonoplast integral protein (delta-TIP), and a plasma membrane intrinsic 17
protein (PIP1), involved in water flux across biological membranes, were particularly affected by a 18
change in fruit load. The down-regulation of delta-TIP under LL condition was confirmed for both 19
genotypes (except at 42 daa) and PIP1 was similarly affected by fruit load. Moreover, a shift in 20
expression with fruit load was observed for PIP1: its expression was down-regulated earlier under 21
LL than under HL. 22
hal-00600426, version 1 - 15 Jun 2011