Branched-chain amino acid catabolism depends on GRXS15 through mitochondrial lipoyl cofactor homeostasis

Iron-sulfur (Fe-S) clusters are ubiquitous cofactors in all life and are used in a wide array of diverse biological processes, including electron transfer chains and several metabolic pathways. Biosynthesis machineries for Fe-S clusters exist in plastids, the cytosol and mitochondria. A single monothiol glutaredoxin (GRX) has been shown to be involved in Fe-S cluster assembly in mitochondria of yeast and mammals. In plants, the role of the mitochondrial homologue GRXS15 has only partially been characterized. Arabidopsis grxs15 null mutants are not viable, but mutants complemented with the variant GRXS15 K83A develop with a dwarf phenotype. In an in-depth metabolic analysis, we show that most Fe-S cluster-dependent processes are not affected, including biotin biosynthesis, molybdenum cofactor biosynthesis and the electron transport chain. Instead, we observed an increase in most TCA cycle intermediates and amino acids, especially pyruvate, 2-oxoglutarate, glycine and branched-chain amino acids (BCAAs). The most pronounced accumulation occurred in branched-chain α-keto acids (BCKAs), the first degradation products resulting from deamination of BCAAs. In wild-type plants, pyruvate, 2-oxoglutarate, glycine and BCKAs are all metabolized through decarboxylation by four mitochondrial lipoyl cofactor-dependent dehydrogenase complexes. Because these enzyme complexes are very abundant and the biosynthesis of the lipoyl cofactor depends on continuous Fe-S cluster supply to lipoyl synthase, this could explain why lipoyl cofactor-dependent processes are most sensitive to restricted Fe-S supply in GRXS15 K83A mutants. One-sentence summary Deficiency in GRXS15 restricts protein lipoylation and causes metabolic defects in lipoyl cofactor-dependent dehydrogenase complexes, with branched-chain amino acid catabolism as dominant bottleneck.


GRXS15 K83A causes retardation in growth 172
To complete embryogenesis, GRXS15 is essential in plants. To bypass embryo lethality,173 Arabidopsis grxs15 null mutants were complemented with the GRXS15 K83A variant which are 174 able to grow, but the plants have small rosette leaves (Moseler et al., 2015). Based on that 175 observation we aimed to further analyze the growth phenotype and compare with published 176 records of grxs15 knockdown mutants. A dwarf phenotype of the GRXS15 K83A 177 complementation lines #1 to #5 becomes apparent at the early seedling stage (Fig. 1A, B). 188 C: Germination rate of GRXS15 K83A lines #3 and #4 compared to WT. All seeds were initially stratified 189 at 4°C in the dark for 1 d (n = 6 with 20-25 seeds each; means ± SD). Germination was assessed with the 190 emergence of the radicle. No statistically significant differences were found using Student's t-Test 191 analysis.

193
Although only minor differences in seedling size could be observed, line #3 was the best 194 growing complementation line and line #4 the weakest ( Fig.1C; Moseler et al., 2015). This effect 195 was stable and consistent over several generations. The phenotype is similar to GRXS15amiR 196 knockdown lines reported by Ströher et al. (2016) (Supplemental Fig. S1). A T-DNA insertion 197 line grxs15-1 carrying a T-DNA in an intron within the 5'-UTR (Moseler et al., 2015), which had 198 been reported to display a short root phenotype (Ströher et al., 2016) cannot be clearly 199 distinguished from the WT in our hands, neither at seedling stage nor at rosette stage 200 (Supplemental Fig. S1). This allele was excluded from further analysis. To test whether the 201 reduced growth of GRXS15 K83A-complemented null mutants was true growth retardation or 202 caused by delayed germination, the two lines #3 and #4 were scored for the timing of radical 203 emergence. The absence of any difference between WT and the two mutants suggests that the 204 growth phenotype reflects a genuine growth retardation (Fig. 1C).

Moco-dependent nitrogen metabolism is not limiting upon impaired GRXS15 function 265
The Moco precursor cyclic pyranopterin monophosphate (cPMP) is synthesized in the 266 mitochondrial matrix by CNX2 (At2g31955) and the cyclic pyranopterin monophosphate 267 synthase CNX3 (At1g01290) and is exported to the cytosol for subsequent biosynthesis steps 268 (Bittner, 2014;Kruse et al., 2018 (Sarasketa et al., 2014), but mutants deficient in Moco biosynthesis can be rescued by providing 274 ammonium as a nitrogen source to bypass nitrate reductase (Wang et al., 2004;Kruse et al., 275 2018), revealing NR as the main recipient of Moco. While the preference for nitrate (KNO 3 ) over 276 ammonium ((NH 4 ) 2 SO 4 ) could be confirmed in wild-type plants, we found that the growth 277 retardation of GRXS15 K83A roots is more pronounced on nitrate than on ammonium as sole 278 nitrogen source (Fig. 3A). Similar results were obtained when seedlings were grown on NH 4 Cl 279 instead of (NH 4 ) 2 SO 4 to control for possible impacts of the respective counter anions on the 280 growth behavior (Supplemental Fig. S2A).

285
A: Primary root length of GRXS15 K83A lines #3 and #4 as well as atm3-1 seedlings compared to WT 286 grown on vertical agar plates containing 5 mM KNO 3 or 2.5 mM (NH 4 ) 2 SO 4 as N-source for 8 d under 287 long-day conditions (n = 30; means ± SD). Student's t-Test analysis showed significant differences 288 between the growth on the different inorganic N-sources in all lines ***: P < 0.001. The pronounced growth retardation on nitrate could be indicative of severe NR 308 deficiency similar to nia1 nia2 mutants lacking functional NR (Wilkinson and Crawford, 1993). A 309 similar NR deficiency has been described for mutant alleles of the ABC transporter ATM3 that is 310 involved in Moco biosynthesis (Bernard et al., 2009;Teschner et al., 2010;Kruse et al., 2018). 311 atm3-1 mutants display a severe growth phenotype and are chlorotic (Fig. 3B). While GRXS15 312 K83A mutants are also smaller than WT, they are not chlorotic and thus do not phenocopy 313 atm3-1 (Fig. 3A, B). Despite NR activity being diminished to 50% of WT, root growth of atm3-1 314 was still better on nitrate than on ammonium (Fig. 3A, C). NR activity was not altered in the 315 GRXS15 K83A mutants #3 and #4 (Fig. 3C). Despite the unaffected NR activity, both grxs15 316 mutants contained significantly less nitrate than WT seedlings (Fig. 3F). Nitrite and other 317 inorganic anions like chloride, sulfate or phosphate were not altered between the mutant lines 318 and WT (Supplemental Fig. S2B). All other tested Moco-dependent enzymes such as AO or 319 XDH showed no decrease in activity in the grxs15 mutants compared to WT (Fig. 3E cluster (Couturier et al., 2015;Meyer et al., 2019). Thus, we measured the respiration of 341 detached roots and dissected the capacity of complex I and II-linked electron flow. Indeed, roots 342 of line #3 displayed a decreased respiration rate of 1.31 ± 0.35 nmol O 2 min -1 (mg DW) -1 343 compared with the wild-type rate of 2.92 ± 0.62 nmol O 2 min -1 (mg DW) -1 (Fig. 4A). This is 344 similar to root tips of GRXS15 amiR knockdown plants which were reported to consume less 345 oxygen than wild-type plants (Ströher et al., 2016). Addition of the cytochrome c oxidase 346 inhibitor KCN decreased the rate of both lines down to similar values. The remaining rates are 347 accounted for by the presence of alternative oxidases (AOXs), since they could be inhibited by 348 propylgallate (pGal). Interestingly, the AOX capacity appeared unchanged in line #3, even 349 though AOX is highly inducible by mitochondrial dysfunction. Next, we investigated if the 350 decreased root respiration is due to defects in the respiratory machinery or due to restricted 351 metabolite turnover, or both. First, we compared the abundance of respiratory complexes in 352 isolated mitochondria from GRXS15 K83A line #4, GRXS15 amiR by BN-PAGE. None of the 353 respiratory complexes including the Fe-S cluster containing complexes I, II and III was 354 decreased in abundance in either mutant (Fig. 4B). Additionally, we purified mitochondria from 355 whole seedlings of the GRXS15 K83A line #3 and supplemented them with succinate or 356 pyruvate/malate, respectively, as respiratory substrates. Succinate provides electrons to the 357 ubiquinone pool of the mETC via complex II, whereas pyruvate/malate predominantly provides 358 NAD(P)H mainly generated by malate dehydrogenase and the PDC. NADH is subsequently 359 oxidized mainly by complex I of the mETC and NAD(P)H by matrix-exposed alternative NADH-360 dehydrogenases. No differences in the respiration of isolated mitochondria were found with 361 supply of succinate or pyruvate/malate (Fig. 4C, D), suggesting that the differences in 362 respiration observed in whole roots cannot be accounted for by decreased capacities of the Fe-363 S cluster-containing complexes. In summary, similar total respiratory activities of WT and 364 mutants further indicate that the in vivo difference in respiration rate is not due to a defect at the 365 level of the mETC, but rather upstream or downstream.

387
The capacity for electron flow in isolated mitochondria does not allow conclusions about 388 the actual mETC activity in planta. Hence, we tested whether the decreased respiration rate 389 may result in a change of the ATP status of the cells. For analyses of the MgATP 2level wild-390 type plants as well as the grxs15 mutants #3 and #4 were transformed with the MgATP 2-391 biosensor ATeam1.03-nD/nA (De Col et al., 2017) targeted to the cytosol. As cytosolic ATP is 392 predominantly provided by the mitochondria (Igamberdiev et al., 2001;Voon et al., 2018), any 393 disturbance in the mitochondrial ATP synthesis will also affect the ATP level in the cytosol. more efficient FRET between the sensor subunits and hence higher MgATP 2levels were found 396 in cotyledons compared to roots (Supplemental Fig. S3). However, no differences in the 397 Venus/CFP emission ratio could be observed between WT and GRXS15 K83A mutants 398 indicating similar cytosolic ATP levels (Supplemental Fig. S3). It should be noted though that the 399 energy charge of the adenylate pool cannot be deduced from these results as it would require 400 also analysis of AMP and ADP. 401

410
Previously we reported a 60% decrease in aconitase activity (Moseler et al., 2015), 411 which at last partially explain the decreased respiration rate, but a decrease in aconitase was 412 not seen in Ströher et al., 2016. To clarify the situation, we measured the activity of ACO, a 413

430
Diminished GRXS15 activity does not lead to any major signs of oxidative stress 431 Yeast ∆grx5 mutant as well as a Arabidopsis grxs14 null mutant are sensitive to 432 oxidative stress and at least for the ∆grx5 it was shown that specific proteins are oxidized in this 433 mutant (Rodríguez-Manzaneque et al., 1999;Cheng et al., 2006). Aconitase is highly sensitive 434 to oxidative stress and redox metabolism in the matrix (Verniquet et al., 1991;Navarre et al.,  To investigate any other metabolic defects in the GRXS15 K83A mutant, we measured 476 the concentrations of several organic acids in the GRXS15 K83A mutants. We found each of the 477 analyzed organic acids in the complemented grxs15 mutants #3 and #4 to be increased. 478 Pyruvate showed the most pronounced change, increasing by more than four-fold from 31.5 ± 479 2.4 pmol (mg FW) -1 in the WT to 131.76 ± 3.8 and 153.97 ± 16.5 pmol (mg FW) -1 in line #3 and 480 #4 (Fig. 6). The accumulation of citrate and isocitrate was significant in line #4, but not in line 481 The pronounced pyruvate accumulation may be caused by a backlog of metabolites due 500 to a lower TCA flux or by a diminished activity of PDC, which catalyzes the decarboxylation of 501 pyruvate to acetyl-CoA (Yu et al., 2012). The E2 subunit of this multi-enzyme complex uses a 502 lipoyl cofactor, the synthesis of which was shown to be compromised in GRXS15 amiR mutants 503 (Ströher et al., 2016). In plant mitochondria, the lipoyl moiety is an essential cofactor of four 504 protein complexes: PDC, OGDC, BCKDC, and GDC (Taylor et al., 2004). Ströher et al. (2016) 505 showed decreased lipoylation of PDC E2-2 and E2-3 but no effects on E2-1. On the other hand, 506 a pronounced decrease was observed in all GDC H protein isoforms and differences in the 507 degree of lipoylation were explained by different modes of lipoylation. To get insight into the 508 metabolic effects of diminished GRXS15 activity, we tested for protein lipoylation in the weakest 509 complementation line #4 and directly compared the results to lipoylation in GRXS15 amiR and WT. 510 Furthermore, the complementation lines #3 and #4 were characterized for metabolites 511 dependent on lipoyl cofactor-dependent enzymes. Immunodetection of the lipoyl group with 512 specific antibodies to the cofactor indicated that the amount of lipoate bound to the H subunit 513 isoforms of GDC was decreased in the GRXS15 K83A mutant to a similar extent as in 514 GRXS15 amiR (Fig. 7A). In contrast, the H protein levels were largely unchanged in all tested 515 lines. GRXS15 was barely detectable in GRXS15 amiR while in line #4 the mutated GRXS15 516 K83A was present at even higher amounts than the endogenous protein in wild-type plants. 517 To further test whether the accumulation of pyruvate was due to a less active PDC, we 518 measured the activity of the PDC in isolated mitochondria. Interestingly, there was a 22% 519 reduction in activity. While the WT had a PDC activity of 92.7 ± 6.5 nmol NADH mg -1 min -1 the 520 GRXS15 K83A line #3 had a significantly lower activity of only 72.40 ± 6.2 nmol NADH mg -1 min -521 1 (Fig. 7B). 522 The pronounced increase of pyruvate and several TCA intermediates (Fig. 6) may have 523 further effects on downstream metabolites. Given that intermediates of glycolysis and the TCA 524 cycle are hubs for synthesis of amino acids and because mutants defective in PDC subunit E2 525 show an increase in the pools of nearly all amino acids (Yu et al., 2012), we profiled the 526 abundance of amino acids. Most amino acids were increased in the mutants compared to WT 527 seedlings, with more pronounced increases in line #4 compared to line #3 (Fig. 7C, 528 Supplemental Table S1). Particularly high increases in amino acid abundance of more than 529 200% were observed for glycine and serine derived from 3-phosphoglycerate, for alanine, 530 leucine and valine all derived from pyruvate, and for isoleucine (  BCAAs may not exclusively result from increased availability of their parent compounds, but 576 also from restricted BCAA degradation capacity. To test this hypothesis, we measured the 577 content of the respective keto acids resulting from deamination of the BCAAs by branched-chain 578 amino acid transaminase (BCAT; Supplemental Fig. 5A). The keto acids α-ketoisocaproic acid 579 (KIC), α-keto-β-methylvaleric acid (KMV) and α-ketoisovaleric acid (KIV) derived from the 580 BCAAs accumulated massively in both GRXS15 K83A mutants (Fig. 7D). Here, KIC 581 accumulated in the GRXS15 K83A mutants up to 15-fold, resulting in values of 3.5 ± 0.11 pmol 582 (mg FW) -1 in line #3 and 3.8 ± 0.6 pmol (mg FW) -1 in line #4 compared to 0.25 ± 0.032 pmol (mg 583 pronounced changes support the hypothesis of decreased BCKDC activity creating a bottleneck 585 in keto acid catabolism (Supplemental Fig. S5A). The higher accumulation of KIC can be 586 accounted for by the preference of BCKDC for the Val derivative (Taylor et al., 2004) resulting in 587 KIV to be metabolized faster and to accumulate less strongly. Despite the presumed bottleneck 588 in catabolism of BCAAs, the grxs15 mutants did not show enhanced Leu sensitivity 589 (Supplemental Fig. S5B). Similarly, ivdh mutants deficient in isovaleryl-CoA dehydrogenase did 590 not display an increased sensitivity to external supply of Leu compared to WT. 591 592 Supplementary

GRXS15 function limits growth 602
Null mutants of mitochondrial GRXS15 are embryo-defective but can be partially 603 complemented with a mutated GRXS15 protein compromised in its ability to coordinate a [2Fe- and Moco, the mETC, and the TCA cycle. While biotin feeding experiments clearly excluded 618 biotin biosynthesis as the limiting factor, the picture was less clear for Moco, which is an 619 essential cofactor for several cytosolic enzymes (Schwarz and Mendel, 2006). Nitrate 620 assimilation, which is dependent on Moco-containing nitrate reductase, initially showed the 621 expected nitrate sensitivity. Measurements of extractable nitrate reductase activity, however, 622 showed no defects. Because, similarly xanthine dehydrogenase and aldehyde oxidases did not 623 show changes in their activities between mutants and the WT, deficiencies in Moco supply can 624 be excluded as a putative metabolic bottleneck in GRXS15 K83A mutants. Nitrate sensitivity in 625 grxs15 mutants leaves us with the conundrum of a different link between mitochondrial functions 626 of GRXS15 and nutrient assimilation, which deserves further investigation in the future. 627

GRXS15 does not affect energy balance and ROS levels 629
Diminished growth correlates with decreased root respiration rates in both, severe 630 GRXS15 amiR knockdown mutants (Ströher et al., 2016)  would be expected to affect electron flow along the mETC. In humans, it was observed that a and hence activity of complex I (Ye et al., 2010). In yeast, Δgrx5 mutants displayed a decreased 636 complex II activity, albeit an unaffected protein abundance in this complex (Rodríguez-637 Manzaneque et al., 2002). In contrast, we found no changes in abundance of any mETC 638 complexes in severe grxs15 mutants of Arabidopsis (Fig. 4B). Consistently, feeding of 639 mitochondria isolated from GRXS15 K83A mutants with succinate revealed that SDH, which 640 contains three different Fe-S clusters (Figueroa et al., 2001), does not constitute a bottleneck in 641 mitochondrial metabolism of grxs15 mutants. Generally, the respiratory capacity is not affected 642 in the mutants compared to WT, which indicates that supply of Fe-S clusters to components of 643 the mETC is not compromised in grxs15 mutants. The lower respiratory rate in GRXS15 K83A 644 mutants also does not lead to changes in ATP levels. This, however, may also partially be due 645 to decreased consumption of ATP with restricted growth and also the activity of adenylate 646 kinase that contributes to formation of ATP (and AMP) from ADP to buffer the ATP level (De Col 647 et al., 2017). Our overall conclusion to this point is that reduced respiration is likely due to 648 restricted substrate supply rather than assembly of complexes in the mETC and their supply 649 GRXS15 was detected as part of higher order protein assemblies in a mitochondrial 661 complexome analysis (Senkler et al., 2017). A particularly strong interaction between GRXS15 662 and mitochondrial isocitrate dehydrogenase 1 (IDH1) was observed in yeast two-hybrid screens 663 with IDH1 as bait and this interaction was subsequently confirmed by bimolecular fluorescence 664 (BiFC) assays (Zhang et al., 2018). Consistent with a suspected role of GRXS15 in IDH1 665 function, the isocitrate content was decreased significantly in a grxs15 knockdown mutant, while 666 the relative flux through the TCA cycle increased (Zhang et al., 2018). IDH1 has recently been 667 reported to contain several redox-active thiols that can change their redox state depending on 668 substrate availability for the TCA (Nietzel et al., 2020). The IDH1-GRXS15 interaction thus could 669 point at a possible function of GRXS15 as a thiol-switch operator for regulatory thiols. This is activity (Moseler et al., 2015;Begas et al., 2017). The increase in all analyzed metabolites of the 672 TCA cycle is rather consistent with metabolite patterns found in knockdown mutants of 673 mitochondrial MnSOD, in which increased levels of organic acids correlated with a decrease in 674 ACO activity (Morgan et al., 2008). Aconitase contains a [4Fe-4S] cluster and has frequently 675 been used as an enzymatic marker for defects in Fe-S cluster assembly and transfer in yeast 676 and human cells (Rodríguez-Manzaneque et al., 2002;Bandyopadhyay et al., 2008;Liu et al., 677 2016). It came as a surprise that ACO was reported to be unaffected in mitochondria of 678 Arabidopsis grxs15 mutants, both in abundance and activity (Ströher et al., 2016). Consistent 679 with this report, we also found no change in abundance of mitochondrial ACOs, but did find 680 reduced activity (Fig. 5). This decrease in activity may well reflect decreased supply of   lipoylation may thus render lipoyl cofactor-dependent-enzymes indirectly sensitive to defects in consistent with previous observations on GRXS15 amiR mutants by Ströher et al. (2016). Similar 708 to the Arabidopsis mutants also humans carrying mutations in mitochondrial GLRX5 are 709 deficient in lipoylation of mitochondrial proteins (Baker et al., 2014). A critical restriction through 710 lipoylation deficiency is further supported by increased amounts of pyruvate and 2-OG as well 711 as several other organic acids and amino acids derived from these precursors (Figs. 6 and 7C). 712 Similar increases in pyruvate as well as the accumulation of most amino acids were also shown 713 for Arabidopsis plants with a mutated PDC-E2 subunit resulting in only 30% PDC activity (Yu et 714 al., 2012). A much more pronounced increase of alanine in PDC-E2 mutants than in GRXS15 715 K83A mutants may be attributed to a higher severity of the metabolic bottleneck if PDC activity 716 is down to 30%. Of all metabolites analyzed in this study, the 4 to 15-fold increases of BCKAs in 717 GRXS15 K83A mutants were the most pronounced relative changes compared to the WT. The 718 findings that these increases were stronger in more severe mutants, point at BCKDC as a 719 critical bottleneck. The keto acids KIC, KIV and KMV are products of transamination of the 720 BCAAs leucine, isoleucine and valine (Hildebrandt et al., 2015). Further degradation of the keto 721 acids in GRXS15 K83A mutants is limited because BCKDC relies on efficient lipoylation of the 722 E2 subunit. Like GDC, PDC and OGDC, BCKDC consists of different subunits, which may not 723 be present in stoichiometric amounts. Recently, Fuchs et al. (2020) reported quantitative data 724 for the abundance of proteins in single mitochondria (Fig. 8). These data indicate low 725 abundance of BCKDC-E2 compared to GDC-H and particularly PDC-E2. Given that all four 726 dehydrogenase complexes rely on the same dihydrolipoyl dehydrogenase subunits, i.e. E3 or L 727 subunits, it is obvious that the relative abundance of subunit proteins will have some impact on 728 assembly of functional complexes. With the assumption that all different E2 and H subunits 729 compete with each other and with the same chance of getting lipoylated, the absolute number of 730 lipoylated PDC-E2 proteins is expected to be higher than that of BCKDC-E2 proteins. If even 731 non-lipoylated E2 or H formed complexes with E3 or L, very little functional BCKDC could be 732 formed. Furthermore, the very low copy number of BCKDC-E1 subunits compared to BCKDC-733 E2 implies that under lipoyl cofactor-limiting conditions, E1 subunits are more likely to form 734 complexes with non-lipoylated and hence non-catalytic E2. Vice versa, the few E2 copies that 735 do get lipoylated may not be those that assemble with E1 subunits to form active complexes. 736 Selective transcriptional upregulation of several nominally lipoylated subunits in GRXS15 amiR as 737 reported by Ströher et al. (2016) would additionally increase the imbalance and tighten the 738 metabolic bottleneck even further. 739 LIP1 was estimated to be present with 85 copies in a single mitochondrion compared to 740 4200 copies of ACO2 and 9900 copies of ACO3 (Fig. 8)   by Fuchs et al. (2020). E3 and L subunits are formed by the closely related and highly similar proteins 782 mtLPD1 (4876 copies) and mtLPD2 (10114 copies), The total of both isoforms is given but it should be 783 noted that a preference of GDC for mtLPD1 and of the other three complexes for mtLPD2 has been 784 proposed (Lutziger and Oliver, 2001). Deficiencies of these enzymes generates metabolic bottlenecks 785 and causes an increase of their respective substrates and particularly for PDC and OGDC also a severe 786 limit in carbon supply to the TCA cycle.

Isolation of mitochondria 812
Arabidopsis mitochondria were purified from 2-or 4-week-old seedlings as described 813 before (Sweetlove et al., 2007) with slight modifications. All steps were performed on ice or at 814 4°C. Seedlings were homogenized using mortar and pestle and the homogenate was filtered 815 (Miracloth; Merck Millipore) before cellular debris was pelleted by centrifugation for 5 min at 816 1,200 g. The supernatant was centrifuged for 20 min at 18,000 g, and the pellet of crude 817 mitochondria was gently resuspended in wash buffer (0.3 M sucrose, 0.1% (w/v) BSA and 818 10 mM TES, pH 7.5) and centrifuged for 5 min at 1,200 g. The supernatant was transferred into 819 a new tube and centrifuged for 20 min at 18,000 g. The pellet was gently resuspended in final and centrifuged for 40 min at 40,000 g. Mitochondria were transferred into a new tube and 822 washed three times with final wash buffer (0.3 M sucrose, 10 mM TES pH 7.5). 823

Respiration Assays 824
Oxygen consumption of intact Arabidopsis roots and isolated mitochondria was 825 measured in Oxytherm Clark-type electrodes (Hansatech; www.hansatech-instruments.com) as 826 described before (Wagner et al., 2015). Whole roots from seedlings vertically grown on agar 827 plates were cut below the hypocotyl-root junction and assayed in a volume of 1.

Determination of metabolite levels via HPLC 850
Aliquots (45-55 mg) of freshly ground plant tissue were used for absolute quantification 851 of amino acid, α-keto acid and organic acid content each. 852 Free amino acids and α-keto acids were extracted with 0.5 mL ice-cold 0.1 M HCl in an 853 ultrasonic ice-bath for 10 min. Cell debris and insoluble material were removed by centrifugation supernatant were mixed with an equal volume of 25 mM OPD (o-phenylendiamine) solution and 856 derivatised by incubation at 50°C for 30 min. After centrifugation for 10 min, the derivatised keto 857 acids were separated by reversed phase chromatography on an Acquity HSS T3 column 858 (100 mm x 2.1 mm, 1.7 µm, Waters) connected to an Acquity H-class UPLC system. Prior 859 separation, the column was heated to 40°C and equilibrated with 5 column volumes of solvent A 860 (0.1% (v/v) formic acid in 10% (v/v) acetonitrile) at a flow rate of 0.55 mL min -1 . Separation of 861 keto acid derivatives was achieved by increasing the concentration of solvent B (acetonitrile) in 862 solvent A (2 min 2% B, 5 min 18% B, 5.2 min 22% B, 9 min 40% B, 9.1 min 80% B and hold for 863 2 min, and return to 2% B in 2 min). The separated derivatives were detected by fluorescence 864 (Acquity FLR detector, Waters, excitation: 350 nm, emission: 410 nm) and quantified using 865 ultrapure standards (Sigma). Data acquisition and processing were performed with the 866 Empower3 software suite (Waters). Derivatisation and separation of amino acids was performed 867 as described by (Yang et al., 2015). 868 Total organic acids were extracted with 0.5 mL ultra-pure water for 20 min at 95°C. 869 Organic acids were separated using an IonPac AS11-HC (2 mm, Thermo Scientific) column 870 connected to an ICS-5000 system (Thermo Scientific) and quantified by conductivity detection 871 after cation suppression (ASRS-300 2 mm, suppressor current 95-120 mA). Prior separation, 872 the column was heated to 30°C and equilibrated with 5 column volumes of solvent A (ultra-pure 873 water) at a flow rate of 0.38 mL min -1 . Separation of anions and organic acids was achieved by 874 increasing the concentration of solvent B (100 mM NaOH) in buffer A (8 min 4% B, 18 min 18% 875 B, 25 min 19% B, 43 min 30% B, 53 min 62% B, 53.1 min 80% B for 6 min, and return to 4% B 876 in 11 min). Soluble sugars were separated on a CarboPac PA1 column (Thermo Scientific) 877 connected to the ICS-5000 system and quantified by pulsed amperometric detection (HPAEC-878 PAD). Column temperature was kept constant at 25°C and equilibrated with five column 879 volumes of solvent A (ultra-pure water) at a flow rate of 1 mL min -1 . Baseline separation of 880 carbohydrates was achieved by increasing the concentration of solvent B (300 mM NaOH) in 881 solvent A (from 0 to 25 min 7.4% B, followed by a gradient to 100% B within 12 min, hold for 882 8 min at 100% B, return to 7.4% B and equilibration of the column for 12 min). Data acquisition 883 and quantification was performed with Chromeleon 7 (Thermo Scientific). 884

Nitrate Reductase assay 897
Nitrate reductase (NR) assay was performed as described previously (Scheible et al., 898 1997)  added. Samples were allowed to stand for 15 min at RT in the dark and the absorbance of the 906 produced azo-dye was measured at 540 nm. 907

Aconitase assay 908
Aconitase activity was analyzed in a coupled assay measuring NADPH formation by 909 monitoring the increase in absorbance at 340 nm using a plate reader (CLARIOstar ® ; BMG). 910 The reaction mixture contained 50 mM HEPES pH 7.8, 2.5 mM NADP + , 5 mM MnCl 2 , 0.1% (v/v) 911 Triton X-100 and 0.05 U isocitrate dehydrogenase. The mixture was allowed to come to 912 equilibrium after addition of protein extract. The reaction was started by adding 8 mM cis-913 aconitic acid. 914

Pyruvate dehydrogenase complex assay 915
To estimate the activity of pyruvate dehydrogenase complex, mitochondria were isolated 916 as described previously and reduction of NAD + was measured at 340 nm in a reaction mixture 917 containing ~10 µg mitochondria in 100 mM MOPS pH 7.4, 1 mM CaCl 2 , 1 mM MgCl 2 , 4 mM 918 cysteine, 0.45 mM thiamine pyrophosphate, 0.18 mM Coenzyme A, 3 mM NAD + and 0.1% (v/v) 919 Triton X-100. The reaction was started with 7.5 mM pyruvate. 920

Fatty Acid Methyl Ester (FAME) Measurement 921
The analysis of fatty acids was performed by quantification of their respective fatty acid 922 methyl esters (FAMEs) via gas chromatography coupled with a flame ionization detector as 923 described before (Browse et al., 1986). 1 mL 1 N HCl in MeOH was added to 5 seeds or 0.9% (w/v) NaCl and 1 mL hexane were added. Samples were mixed vigorously and centrifuged 927 with 1,000 g for 3 min. The hexane phase was transferred to a GC vial. FAMEs were quantified 928 using pentadecanoic acid as internal standard. 929

Statistical analysis 956
Statistics and error bars were applied for independent experiments with at least three 957 biological replicates using the program GraphPad Prism 6. 958