Correlation between metabolites and seed phenotypic traits
In order to assess the relationship between metabolites and seed phenotypic traits, seed performance phenotypes which were previously assessed for the same seeds (Geshnizjani et al. 2020), were integrated into the metabolic correlation matrix (Figure 7 ). We found many positive and negative correlations between metabolites and phenotypic traits. The results revealed that seed size and weight are positively correlated with most of the amino acids and TCA cycle intermediates such as succinate, citrate and malate (Figure 7, Table S8 ). The strongest positive correlation was found between seed size and amino acids including pyroglutamate, leucine and isoleucine (R≥0.4, p -value<0.0001). Among all the seed germination traits maximum germination percentage (Gmax) showed the highest number of significant correlations with metabolites of which most are negative. Gmax under osmotic stress (mannitol and NaCl) has a significant positive correlation with 2-ketoglutarate which is one of the TCA cycle intermediates, involved in supplying the required energy for seed germination (Table S8 ). Gmax under optimal and sub-optimal germination environments showed strong negative correlation with many of the amino acids (such as pyroglutamate, GABA, methionine and leucine), organic acids (glycerate and malonate) and TCA cycle intermediates (malate and succinate).
Amino acids, are the precursors of protein synthesis and also precursors of some TCA cycle intermediates (e.g. citrate and succinate), serve as energy generation units for embryo growth as well as radicle protrusion (Lehmann & Ratajczak 2008; Ratajczak et al. 1996; Rosental, Nonogaki & Fait 2014). Since energy and proteins are two elements supporting germination, such a negative correlation between them and germination of tomato seeds is not expected. However, our results are in accordance with several foregoing studies which reported that accumulation of amino acids, such as methionine, lysine and GABA, may cause inhibition of seed germination (Amir 2010; Angelovici et al. 2011). In some other reports, amino acids were considered as one of the biological methods to control weeds since the external application of many amino acids decreased the seed germination percentage for some species such as broomrape (Vurro et al. 2006; Wilson & Bell 1978). Such a negative effect of amino acids on seed germination could be related to accumulation of certain amino acids in the seeds and subsequent reduction of some other metabolites such as TCA cycle intermediates which may play vital roles in seed germination (Angeloviciet al. 2011; Rosental et al. 2016). For example, the biosynthetic pathway of lysine uses pyruvate which is the central component of the TCA cycle. Depletion of pyruvate from the TCA cycle will ultimately result in a decrease in the production of TCA cycle intermediates. Hence, such a decrease in TCA cycle input results in declined levels of available energy, which in turn negatively affects seed germination (Angelovici et al. 2011; Day et al. 1994; Shedlarski & Gilvarg 1970). A strong negative correlation was found between Gmax in water and methionine content of the seeds (R=0.42, p -value<0.001). Similar results have been found in different species such as lettuce (Wilson & Bell 1978) and tomato (Rosental et al. 2016). Feedback inhibition of increased methionine on the upstream enzymes activity such as cystathionine γ-synthase (CGS) has been reported before (Chiba et al. 2003; Rosental et al. 2016). Hence high methionine content of seeds may limit the synthesis of sulfur-rich proteins which subsequently results in the reduction of seed germination (Amir 2010). However, our findings seem in contrast with a few other studies in which a high level of methionine did not lead to a decrease of germination which indicated that methionine was not negatively correlated with germination (Amir, Han & Ma 2012; Gallardo et al. 2002).
We also performed a correlation analysis between metabolites and seed phenotypic traits within each tomato RIL sub-population and two correlation heatmaps were generated (Figure S3 ). In general, substantial differences were not observed between the two maternal environments (HP and LN); however, correlations appeared stronger within HP as compared to LN conditions and some correlations were specifically observed in one of the environments. For example, the positive correlation observed between many amino acids and seed size and weight were either lost at LN or were not as strong as what was observed at HP (Figure S3 ). In addition, a limited number of metabolites (e.g. galactarate) showed a significant positive correlation with most of the phenotypic traits in LN; however, the same metabolite showed a weak negative correlation with the same seed phenotypic traits in HP (Figure S3 ). An association of germination percentage and metabolic content of the dry seeds may raise the possibility to predict germination behaviour using the metabolic signature of the dry seeds (Rosental, Nonogaki & Fait 2014).