Correlation between co-expression modules and physiological changes
The expression data and traits data of all samples were used in Weighted Gene Coexpression Network Analysis (WGCNA) and the relationships between them were visualized (Supplementary Figure S7 ). By using hierarchical clustering and dynamic tree cut method, co-expression modules were obtained (Supplementary Figure S8 ); after merging and filtering, a total of 28 modules remained; among them, the largest module (MEpaleturquoise) contains 2,691 genes while the smallest module (MEdarkorange2) only includes 181 genes (Supplementary Table S6 ).
The relationships between modules and physiological traits were examined (Figure 9 ). Firmness showed high correlation with the expression levels of genes from MEdarkorange2 and MEdarkred modules (R > 0.7) and anti-correlation with those from MElightyellow, MEdarkgrey and MEbrown4 modules (R < -0.7). Ethylene production was significantly correlated with six modules (P < 0.01), including MEsaddlebrown, MElightgreen, MEbisque4, MElightsteelblue1, MEskyblue3 and MEpaleturquoise. MEskyblue3 module was positively correlated with LOX2 (R = 0.86) but negatively correlated with ASA (R = -0.7). MEgrey module had no significant (P< 0.05) correlation with all physiological changes. Notably, the levels of AAO, AMS, LOX2, O2- and β-Gal displayed similar correlations with ethylene in many modules.
To further explore the biological function of each module, GO annotation and enrichment analysis were performed. As shown (Supplementary Figure S9 ), MEdarkgrey module was enriched in the GO term of endopeptidase Clp complex in cellular component category; MEskyblue3 module was enriched in the GO term of amino acid kinase activity; MElightyellow module was enriched in catalytic activity in molecular function category. Moreover, the sub-network involving all the above genes was extracted from the whole co-expression network (Supplementary Figure S10 ). The sub-network is roughly parted into two clusters: the cluster in the upper side contains genes mostly from MEskybule module and the cluster in the lower side contains genes from other modules. The genes AUX/IAA, SAUR and SAMS3 from the bisque4 module had high degrees in the sub-network, indicating that they played key roles in the correlations with other modules.