mQTL profiling of the tomato RIL population
The calculation of the coefficient of variation (CV) showed that most of the metabolites possess a CV value higher than 40%, which indicates that there is considerable variation within the RIL population for the metabolite levels in the dry seeds (Figure 8, Table S9 ). In order to investigate if such a high level of variability within metabolites could be explained by differences in alleles and genetic factors, a metabolic quantitative trait locus (mQTL) analysis was performed with the obtained metabolite data. Each metabolite is in general controlled by several pathways and regulators. Thus, as expected, we hardly identified metabolites for which a single genetic locus significantly explained the metabolite levels.
In our study we performed mQTL analysis for each maternal environment to evaluate the genetic variation within each sub-population. Furthermore we used the whole set of RILs to detect mQTLs explained by a genetic component (G) and the genotype by environment interaction (G×E).
We identified mQTLs across all conditions (Table 3 ). Regarding the sub-populations 66 and 129 mQTLs were detected for seeds from LN and HP environments, respectively. The heatmap of the LOD profiles and characteristics of the mQTLs in each environment are presented inFigure S4 and Table S10 , respectively. In both maternal environments several mQTLs were detected which were hardly detected for other metabolites. For example, in the seeds developed under HP conditions a single strong QTL on chromosome 9 was detected, regulating asparagine. Another independent significant QTL was identified on the top of chromosome 11 for phenylalanine under the same environmental conditions (Figure S4A, Table S10 ). Detection of such specific mQTLs in our data reveals the tight and independent genetic regulation of metabolite biosynthesis in seeds (Keurentjeset al. 2008). Under the same maternal condition some organic acids such as benzoate, gluconate, glycerate and glycolate mapped to a similar position on chromosome 5 (Figure S4A, Table S10 ). On chromosome 9, we detected mQTLs for TCA cycle intermediates including citrate and malate which were co-locating with the one regulating F6P as one of the precursors of the TCA cycle. There is also a QTL on the top of chromosome 1 affecting amino acids in seeds from the HP environment. Despite the strong correlation that has been found between amino acids and TCA cycle intermediates in seeds from HP conditions, no co-located QTLs were identified for them. This might be due to several smaller QTLs regulating variation of the metabolites, each of them explaining a small part of the variation and therefore not reaching the threshold LOD score. Regarding the seeds grown in the LN maternal environment we found more than one QTL for some of the metabolites such as GABA, citrate and malate. The vital role of these metabolites has been reported in relation with the alleviation of environmental stress effects (Kaplanet al. 2004; Kinnersley & Turano 2000; Krasensky & Jonak 2012; Obata & Fernie 2012). For the LN environment many of the amino acids have co-locating QTLs at the bottom of chromosome 4 and in the middle of chromosome 5 (Figure S4B ). Such strong co-locating QTLs for amino acids was expected since they showed a high connection in the correlation network of the LN environment (Figure 5A ). In general, such co-localizing QTLs for metabolites suggest that, in addition to the single independent QTLs regulating metabolite contents, some general regulatory loci and genes are involved in the regulation of metabolite synthesis (Keurentjes et al. 2008).
Combining the sub-populations and using the whole set of RILs leads to an increase in the number of detected QTLs with 382 and 146 QTLs for G and G×E effects, respectively. An overview of the detected QTLs is provided by the heatmap of the LOD profiles (Figure 9 ). On the top and bottom of chromosome 4 there are two QTLs that explain the variation for many amino acids such as aspartate, GABA, glutamine, methionine, serine and threonine. Similarly, a co-located QTL was detected for galactarate and malate on chromosome 10 (Figure 9 ). Co-localization of these mQTLs is not surprising since galactarate is the precursor of 2-oxoglutarate and 2-oxoglutarate is one of the intermediates of the TCA cycle and is generally converted to malate in a couple of subsequent reactions. Our results show that myo-inositol and galactinol are highly associated with each other and closely grouped together. Therefore, it is not surprising that they both have a co-locating QTL on chromosome two (Figure 6, Figure 9 ). The robust correlation between raffinose pathway metabolites including galactinol and myo-inositol has also been reported for seeds of other species that developed under environmental stress (Cook et al.2004; He et al. 2016). These metabolites are known for their protective role for cellular structures of embryos during seed development and desiccation (Taji et al. 2002). Furthermore, they are able to play a key role in protecting plants from the effects of stress resulting from reactive oxygen species (ElSayed, Rafudeen & Golldack 2014). Some of the organic acids including gluconate, glycerate and glycolate, together with two of the TCA cycle intermediates (malate and succinate), had a co-locating QTL on chromosome 9 (Figure 9 ). Glutamate and GABA showed a shared QTL on chromosome 4 which has previously been detected in the same population developed under standard conditions (Kazmi et al. 2017). Metabolites belonging to the same functional class are often highly correlated and can have co-locating mQTLs (Kazmi et al. 2017). Although several mQTLs were detected at similar positions, in general more co-located mQTLs would be expected due to the strong correlation that has been observed between the metabolites. This could be related to the fact that several small QTLs are involved in regulation of the metabolites and each of them is explaining only a small part of their variation. Such small QTLs are likely to escape the QTL significant threshold in the QTL analysis (Keurentjes et al. 2008).
A few mQTLs co-located with the phenotypic QTLs that have been detected in a previous study (Geshnizjani et al. 2020). For instance, the QTLs on the middle of chromosome 10 affecting galactarate and malate co-located with ones influencing uniformity of germination (U8416) at different germination conditions, such as high temperature, mannitol, water and NaCl. In addition, the QTL on chromosome 9, which is specifically regulating methionine, is located at the same position as QTLs affecting seed size, seed weight and fresh and dry weight of the seedlings. Despite the many strong correlations between metabolites and phenotypic traits (Figure 9 ), we could hardly detect co-locating QTLs for them. This might be due to the fact that each of the phenotypes may not be correlated with a specific metabolite but with a group of metabolites and thus the final metabolic balance between the groups of metabolites could affect phenotypic traits such as Gmax.