Synthetic analogues of 2-oxo acids discriminate metabolic contribution of the 2-oxoglutarate and 2-oxoadipate dehydrogenases in mammalian cells and tissues

The biological significance of the DHTKD1-encoded 2-oxoadipate dehydrogenase (OADH) remains obscure due to its catalytic redundancy with the ubiquitous OGDH-encoded 2-oxoglutarate dehydrogenase (OGDH). In this work, metabolic contributions of OADH and OGDH are discriminated by exposure of cells/tissues with different DHTKD1 expression to the synthesized phosphonate analogues of homologous 2-oxodicarboxylates. The saccharopine pathway intermediates and phosphorylated sugars are abundant when cellular expressions of DHTKD1 and OGDH are comparable, while nicotinate and non-phosphorylated sugars are when DHTKD1 expression is order(s) of magnitude lower than that of OGDH. Using succinyl, glutaryl and adipoyl phosphonates on the enzyme preparations from tissues with varied DHTKD1 expression reveals the contributions of OADH and OGDH to oxidation of 2-oxoadipate and 2-oxoglutarate in vitro. In the phosphonates-treated cells with the high and low DHTKD1 expression, adipate or glutarate, correspondingly, are the most affected metabolites. The marker of fatty acid β-oxidation, adipate, is mostly decreased by the shorter, OGDH-preferring, phosphonate, in agreement with the known OGDH dependence of β-oxidation. The longest, OADH-preferring, phosphonate mostly affects the glutarate level. Coupled decreases in sugars and nicotinate upon the OADH inhibition link the perturbation in glucose homeostasis, known in OADH mutants, to the nicotinate-dependent NAD metabolism.

OGDH, despite a rather high level of the OGDH activity in reaction with 2-oxoadipate in vitro. Delineation of the physiological role(s) of the DHTKD1-encoded protein in in vivo studies is complicated by tissue specificity of the DHTKD1-dependent changes 24 , significant variation in the DHTKD1 expression in a population 27,35 and specific environmental conditions when the effects of the DHTKD1 expression acquire significance 24,27 . For instance, in liver, the DHTKD1-encoded protein is shown to be a primary regulator of 2-aminoadipate 27 , but this may be different in other tissues. In fact, the DHTKD1 knockout in mice elevated 2-oxoadipate and 2-aminoadipate in liver, but did not affect these metabolites in the brain 24 . As a result, understanding the organization and significance of the OADHC-involving pathway(s) obviously requires studies at a less complicated than organismal level. Therefore, in this work the problem is addressed using metabolomics of cultured cells with natural variations of the DHTKD1 expression, supported by in vitro studies of the enzymes from the rat tissues with similar variations in the DHTKD1 expression.
To further develop tools for discriminating biological functions of the OGDH(L) and DHTKD1 gene products, we have taken into account the successful application of the synthetic OGDH inhibitor succinyl phosphonate (SP) for the enzyme regulation in vivo 1,36,37 and the better accommodation of bulkier substrates by the active site of OADH, compared to OGDH 3,38 . For the present work, a series of SP homologues have been synthesized to compare their inhibitory potential towards OGDHC and OADHC in vitro and in cell cultures. Activities of the enzyme-enriched fractions along with metabolism and viability of the control and phosphonates-treated cultured cells have been assessed in the systems with varied OADH expression levels. Comparative analysis of the action of the homologous phosphonates indicates that the longest of the substrate analogues, adipoyl phosphonate (AP), is a specific in vivo inhibitor of the DHTKD1-encoded protein. Metabolomics experiments reveal that cells with varied DHTKD1 expression possess specific metabolic features and respond differently to the phosphonate analogues of 2-oxoglutarate and 2-oxoadipate. Based on the metabolomics data of the control and phosphonates-treated cells, specific cellular action of the OADH inhibitor AP is shown, in good accord with the inhibition studies in vitro. Unlike SP and GP, AP is able to decrease cellular glutarate and glucose. Based on the coupled decrease in cellular glutarate, glucose and nicotinic acid after inhibition of OADH by AP, the nicotinate-dependent NAD metabolism is suggested to mediate the perturbed levels of glucose. Thus, the synthesis and specific action of the OADH inhibitor AP, characterized in this work, provide a useful tool for deciphering cell-specific molecular mechanisms which underlie the known associations between the DHTKD1 expression with diabetes, obesity and cancer.

Results
Activity and stability of OADHC in tissue homogenates depends on relative expression of enzymatic components of the OGDH and OADH complexes. To choose the best source for assays of the DHTKD1-encoded protein and differentiation of its activity from that of OGDH(L), the relative abundance of the enzymes in different tissues has been analysed. The western blot detection of the DHTKD1 protein (OADH) agrees well with the transcriptomics and proteomics data ( Fig. 2A), and our own mass-spectrometry estimations (Fig. 2B). All the approaches demonstrate that the DHTKD1 protein is expressed in the liver much more than in the other tissues shown in Fig. 2. In particular, the OADH-specific peptides are much better detectable in homogenates of the rat liver than brain, whereas the OGDH-and DLST-specific peptides are equally detectable in both tissues (Fig. 2B).
Catalysis of the overall OADH ( Fig. 2C) or OGDH (Fig. 2D) reactions in different tissues requires the assembly of the multienzyme complexes formed by non-covalent interactions of multiple copies of three catalytic subunits (Fig. 1). It is therefore not surprising that the assayed activities (Fig. 2C,D) do not correspond to the expression of only the OADH and OGDH components of the complexes (Fig. 2E,F). Obviously, the activities depend on the expression of all the components of the complexes (Fig. 2E,F). It is known that the expression of components of multienzyme structures in vivo occurs at the ratios enabling successful formation of the needed levels of the multienzyme structures 39 . Competition of the OADH and OGDH components for their common partner, the DLST-encoded E2o component, also implies that the components ratios should be in accord with relative affinities of OADH and OGDH to E2o, which are known to differ 38,40 . Transcriptomics (Fig. 2E) and proteomics (Fig. 2F) data indicate that in all the tissues the expression of E2o and E3 is comparable to the expression of the OGDH gene, and does not significantly increase when the DHTKD1 expression approaches that of OGDH. The ensuing competition between the OGDH and OADH components of the multienzyme complexes for the same DLST-encoded E2o is obvious from our experimental data. At very similar expressions of the OGDH, DLST, and DLD transcripts (Fig. 2E) and proteins (Fig. 2F), the fully solubilized OGDHC activity in liver is an order of magnitude lower than that in skeletal muscles (black bars in Fig. 2D). This is obviously due to the stronger competition between the OGDH and DHTKD1-encoded proteins for E2o in the liver, where the expression of the two proteins is comparable, than in the skeletal muscles, where OGDH is expressed much more than DHTKD1 (Fig. 2E,F).
The competition between the OADH and OGDH components for binding site on E2o may also lead to a lower stability of the overall OADHC vs OGDHC activity under conditions promoting dissociation of the complexes. This may occur upon solubilization of mitochondrial proteins by sonication and detergents. Comparison of the tissue expression of the components of the complexes (Fig. 2E,F) with their relative stabilities to the dissociating factors (Fig. 2C,D), reveals the components ratios under which the OADHC activity is better preserved (Figs. 2C ≥18 independent series of experiments for each tissue. (F) Relative levels of the DHTKD1, OGDH, OGDHL, DLST and DLD proteins in the mouse tissues. Except for spinal cord (1 experiment), the protein abundances were extracted from ≥2 independent experiments. D/O is the ratio between the expression of DHTKD1 and OGDH + OGDHL genes at the mRNA and protein levels. n.d. -Not determined. and 3A). This is observed when the DLST/DHTKD1 and DLD/DHTKD1 ratios (i) are higher than 10 and (ii) exceed at least by an order of magnitude the DLST/OGDH(L) and DLD/OGDH(L) ratios (Fig. 3B,C). In other words, at a comparable expression of the OGDH and E2o components of the multienzyme complexes, which is observed in all the tissues (Fig. 2E,F), at least a 10-fold excess of E2o over OADH is required to preserve the OADHC activity during the mitochondrial solubilization. In contrast, the activity of OGDHC is not lost during the solubilization at the comparable expression of OGDHC and E2o, given the expression of DHTKD1 is low (Fig. 2D). Pointing to a much weaker binding of E2o to OADH (E1a) than to OGDH (E1o), these data are in good accordance with the differences of the E2-binding regions between the DHTKD1-and OGDH-encoded protein sequences 3 and the relative binding affinities recently shown in studies of the recombinant proteins 38,40 .
According to the data on the recombinant OADHC and OGDHC 9,21 , the 2-oxo substrate concentrations (2 mM) in our assays of homogenates are saturating. Hence, the assayed activities of OADHC and OGDHC (Fig. 2C,D) correspond to the maximal rates of the overall OADH and OGDH reactions, catalysed by the complexes in the analysed samples. Our data thus reveal that OADHC and OGDHC in mammalian tissues possess a much stronger specificity to their cognate 2-oxo substrates, compared to the 20-30% cross-reactivity of the recombinant complexes, based on the k cat values with these substrates 9,21 . In fact, in the heart tissue before the solubilization by sonication and detergents, the OADH reaction is well-measurable, whereas the OGDH reaction rate is only about 3% of the OADH reaction rate (Fig. 2C,D). In contrast, when the proteins are fully solubilized by sonication and detergents, the preparation may possess a high OGDHC activity devoid of the OADHC activity, as in the liver and spinal cord (Fig. 2C,D). The different substrate specificity of the native and recombinant complexes may be due to the artificial structures of the recombinant complexes, which have been obtained at the ratios of the components, strongly deviating from those known for the native OGDHC 9,21 .

Analysis of the inhibition of OADH and OGDH reactions by the phosphonate analogues of homologous 2-oxodicarboxylic acids.
For the kinetic analysis of the inhibition of the OADH and OGDH reactions by the phosphonate analogues of their 2-oxo acid substrates (Fig. 4A), rat tissues with different OADH expression have been fractionated to enrich the studied activities. To preserve the OADHC activity, the solubilization of mitochondrial complexes was done by 3% Triton X-100 without sonication. The inhibitory potential of the phosphonates towards OADHC and OGDHC has been compared at the physiologically relevant levels of 2-oxo substrates (Fig. 4). In mammalian tissues, the concentrations of 2-oxoglutarate are about 0.1-0.2 mM 41 , while 2-oxoadipate normally does not exceed 0.01 mM 11,24 . Due to low reaction rates at this concentration of 2-oxoadipate, we determined the inhibition also at 0.02 mM substrate (Fig. 4B). We also compared the relative inhibitory power of the phosphonates at a fixed (0.2 mM) concentration of 2-oxoglutarate vs 2-oxoadipate (Fig. 4C).
The results in Fig. 4 demonstrate that the relative inhibitory power of the phosphonates in the OADH and OGDH reactions strongly depends on the relative expression of the DHTKD1 protein, which is up to three orders of magnitude higher in the rat liver, compared to the heart ( Fig. 2A). In fact, at 0.01-0.02 mM 2-oxoadipate, the longest of the phosphonates, AP, does not inhibit the OADH reaction catalysed by the cardiac preparation with a low DHTKD1 expression, but does affect this reaction in case of catalysis by the liver preparation with a high DHTKD1 expression (Fig. 4B). Under the same conditions, the shortest of the phosphonates, SP, exhibits the opposite behaviour, significantly inhibiting only the OADH reaction rate catalysed by the cardiac preparation (Fig. 4B). The intermediate length of GP causes its reactivity in the OADH reaction catalysed by both the cardiac and liver preparations (Fig. 4B). These data show that AP and SP discriminate the active sites of OGDH and OADH better, than GP. Besides, one may conclude that at the low concentrations of 2-oxoadipate the OADH reaction is predominantly catalysed by different enzyme complexes, i.e. OADHC and OGDHC, in the liver and heart preparations, correspondingly.
Unlike the tissue-specific action of the phosphonates on the OADH reaction (Fig. 4B), the pattern of inhibition of the OGDH reaction does not strongly depend on the tissue, with SP and GP significantly inhibiting the OGDH reaction catalysed by both the cardiac and hepatic preparations (Fig. 4C). Remarkably, with both tissues, the inhibitory action of AP is absent in the OGDH reaction at 0.2 mM 2-oxoglutarate (Fig. 4C). However, AP still inhibits the OADH reaction at 0.2 mM 2-oxoadipate, if this is catalysed by the liver preparation (Fig. 4C). A higher contribution of OADHC in the liver, compared to heart, preparation to the catalysis of OADH reaction at 0.2 mM 2-oxoadipate is also obvious by the degree of the inhibition of this reaction by SP and GP. At 0.2 mM 2-oxoadipate, the OADH reaction catalysed by cardiac preparation is significantly inhibited by SP (60%) and GP (75%) much more than the same reaction catalysed by the liver preparation, where the inhibition by SP and GP is only 30% (Fig. 4C).
Remarkably, as a competitive inhibitor of OGDHC, SP should be more inhibitory at lower 2-oxoadipate concentrations. This is true for the cardiac preparation, but with the hepatic one the inhibitory power of SP increases along with the elevation of the 2-oxoadipate concentration from 0.02 mM (no observed inhibition by SP) to 0.2 mM (30% inhibition by SP) (Fig. 4B,C). The finding indicates a higher contribution of the SP-inhibited OGDHC to the OADH reaction at the non-physiological 2-oxoadipate concentration (0.2 mM), in good accord with a lower affinity of OGDHC vs OADHC to 2-oxoadipate 9,21 .
Thus, the inhibitory action of AP and SP demonstrates their preferential binding to OADH (DHTKD1-encoded protein) and OGDH (OGDH(L)-encoded protein), correspondingly, whereas efficiency of the OGDH and OADH inhibition by GP depends on the substrate, its concentration and relative expression of OGDH and OADH.
Study of the interactions of the homologous phosphonates with other enzymes transforming 2-oxo acids or their analogues. When SP was introduced as a highly specific OGDHC inhibitor in vivo, its reactivity to other enzymes transforming 2-oxoglutarate and its structural analogues was studied to ensure the specific action on OGDHC 42,43 . To confirm the specificity of the SP homologues introduced in this work, we have performed similar studies with GP and AP. Comparison of the influence of the three phosphonates on the activities of different enzymes of central metabolism transforming 2-oxo acids and/or their analogues, is presented in Table 1. The concentration of the phosphonates used in these studies (0.1 mM) causes strong inhibition of OGDHC by SP and OADHC by AP at the physiological levels of the substrates ( Table 1). The comparable concentration of the 2-oxo/2-hydroxy substrates (0.2 mM) was used to study potential off-target effects, unless the reliable activity assay required a higher concentration. For instance, 2 mM 2-oxoglutarate was required to assay the transaminase reactions, whose K m for 2-oxoglutarate is in the mM range [44][45][46] . The results of Table 1 confirm the previously known specificity of SP 37,42,43 and indicate that the activities of the enzymes from different classes are not significantly affected by the new phosphonates AP and GP either.
Different DHTKD1 expression in cell lines correlates with different abundance of the saccharopine pathway intermediates or nicotinate. In view of cellular heterogeneity of mammalian tissues, metabolic significance of the DHTKD1-encoded OADH has been studied in cell lines which differ in OADH abundance. The C6 rat glioma and MCF-7 human breast adenocarcinoma cell lines (Fig. 5A) are similar to tissues with relatively low and high DHTKD1 level, respectively (Fig. 2E). The critical ratios of DLST to DHTKD1 vs www.nature.com/scientificreports www.nature.com/scientificreports/ DLST to OGDH(L) also strongly differ in the C6 and MCF-7 lines (Fig. 5B), modelling the ratios existing in the tissues (Fig. 3B).
The strong difference in the DHTKD1 expression in the MCF-7 and C6 cells (Fig. 5A,B) is accompanied by the differences in cellular metabolomes (Fig. 5C,D). The 2-oxoadipate-linked metabolites ( Fig. 6) are of special interest in this regard. Levels of lysine and tryptophan, whose degradation occurs through the common intermediate 2-oxoadipate, are proportional to the DHTKD1 expression, which is higher in the MCF-7 cells, compared to the C6 cells. In contrast, the level of 2-aminoadipate is lower in MCF-7 vs C6 cells, in good accord with the known negative correlation between the level of 2-aminoadipate and the DHTKD1 expression 27 . This finding confirms a higher flux of the oxidative decarboxylation of 2-oxoadipate in the cells with a higher DHTKD1 expression (MCF-7). Association of the high DHTKD1 expression with the saccharopine pathway of lysine catabolism ( Fig. 6) is further supported by the high levels of 2-oxoglutarate and saccharopine in MCF-7 cells, whereas these metabolites are not detectable in C6 cells with a low expression of DHTKD1 (Fig. 5D). In contrast, a metabolite of the NAD salvage pathway, nicotinate, is abundant in C6 cells and not detectable in MCF-7 cells (Fig. 5D). As shown in Fig. 6, nicotinate interacts with the tryptophan pathway of de novo NAD biosynthesis from quinolinic acid. Because this pathway is alternative to the tryptophan degradation to 2-oxoadipate, the low DHTKD1 expression in C6 cells may increase the tryptophan flux to NAD synthesis (Fig. 6). As a result, the cell-specific upregulation of the saccharopine pathway is associated with upregulated expression of DHTKD1, whereas higher levels of nicotinate at a low expression of DHTKD1 link OADH reaction and metabolism of NAD, which is also known to depend on tryptophan.
Remarkably, the differences in the 2-oxoadipate-linked metabolites are coupled to those in sugar metabolism: Compared to C6 cells, MCF-7 cells have more phosphorylated sugars and their derivatives (olive cluster at Fig. 5C,D), and less non-phosphorylated sugars (turquoise cluster at Fig. 5C,D). Besides, MCF-7 cells demonstrate higher levels of branched-chain amino acids, proline, serine, and ornithine, than C6 cells (olive cluster in Fig. 5C,D). Clusters of the metabolites, which are substantially lower in the MCF-7 relative to C6 cells (violet and turquoise clusters, Fig. 5C), include guanidine, polyamines, adipate and glutarate. Thus, the differences in the metabolomes of the cells with varied expression of DHTKD1 reveal that the cell-specific involvement of OADHC in the lysine and tryptophan catabolic pathways is associated with altered metabolism of (phospho)sugars and guanidine/polyamine compounds. www.nature.com/scientificreports www.nature.com/scientificreports/ cells by the phosphonate inhibitors of OADHC and OGDHC, the cells were incubated for 5 h with a moderate (0.5 mM) concentrations of SP, GP or AP in the minimal medium. These conditions minimize cellular ability to compensate for the enzyme impairments by rearranging the metabolic processes due to the inhibition. Metabolic www.nature.com/scientificreports www.nature.com/scientificreports/ changes in the phosphonates-exposed cells point to the phosphonate-specific action on the overall metabolic profiles. As seen from Fig. 7A,B, the changes induced by SP are clustered separately from those caused by GP or AP. Thus, metabolic action of SP is different from that of either GP or AP. This finding on cell cultures is in good agreement with the preferred binding of SP to OGDHС and AP to OADHC in vitro (Fig. 4).

Effects of the phosphonate analogues of 2-oxodicarboxylic acids on metabolism of the cells with different
The calculated distances of the phosphonate-induced changes in the metabolic profiles of the cells indicate that the difference in the actions of SP vs GP or AP is higher in the C6 cells ( Full comparison of the four investigated groups to each other was done for selected metabolites, using their values normalized to internal standard and protein. Choice of the indicators, which were robustly determined in both cell lines, enabled additional comparison of the specificity of the phosphonates action, dependent on the metabolic type of a cell. As shown earlier, the studied cells demonstrated specific differences in the pathways (Fig. 6) involving nicotinate (detected only in C6 cells) or saccharopine (detected only in MCF-7 cells). Statistical significance in this comparison was assessed by Friedman's test applicable for the data which do not have a normal distribution. The test relies not only on the average differences between the control and treated groups, but also on the consistency of changes in the repeated experiments. For instance, the mean values for adipate are reduced by both GP and AP in Fig. 8A, but AP reduces adipate only in the three out of five experiments, whereas GP does it in the four out of five experiments. Therefore, the GP action has a higher statistical significance according to Friedman's test, despite the lower, compared to the GP treatment, mean level of adipate after the AP treatment (Fig. 8A). Likewise, statistical significance of the phosphonate-induced decreases in glutarate in MCF-7 cells (Fig. 8D) is lower, compared to that estimated by t-test (Fig. 7B), because the changes in separate experiments are not well coinciding. However, the finding of more profound AP-induced changes in C6 vs MCF-7 cells is supported by both the t-test (Fig. 7) and Friedman's ( Fig. 8) analyses. As seen from the p-values characterizing significance of the changes in the cells treated with the phosphonates, C6 cells are affected by the treatment Figure 6. The OADHC-dependent pathways of the lysine and tryptophan catabolism, interacting with NAD metabolism. Lysine condenses with 2-oxoglutarate into saccharopine leading to 2-oxoadipate, whose oxidative decarboxylation by OADHC generates glutaryl-CoA. Glutarate is formed from glutaryl-CoA by hydrolysis or transacylation reactions 50 . Tryptophan is catabolised through alternative pathways leading to either 2-oxoadipate or quinolinic acid (QA). QA is an intermediate for the de novo biosynthesis of NAD from tryptophan 115,116 . In addition to its metabolic functions, NAD has signalling significance for glucose homeostasis, in particular through the Ca 2+ -mobilizing action of nicotinic acid adenine dinucleotide phosphate (NAADP), whose synthesis from NADP + requires nicotinate. Dependent on cell-specific expression of the OADH-involving pathways, the OADH inhibition may downregulate saccharopine pathway and/or increase QA for biosynthesis of NAD and its signalling derivatives from tryptophan. The quantified metabolites of the presented pathways are shown in grey. Solid and dashed arrows indicate the single and more than two enzymatic reactions within a metabolic pathway, correspondingly. NA -nicotinic acid (nicotinate); NAADPnicotinic acid adenine dinucleotide phosphate; NAM -nicotinamide; NAMN -nicotinic acid mononucleotide; QA -quinolinic acid. www.nature.com/scientificreports www.nature.com/scientificreports/ much more than MCF-7 cells. The group comparison in each cell line reveals specific action of the phosphonates on glutarate, adipate, glucose and 4-aminobutyrate (Fig. 8). The significance of the adipate reduction, which is pronounced in both C6 and MCF-7 cells, is higher after the treatments with the OGDHC-targeting inhibitors SP and GP (Fig. 8A,B). Hence, inhibition of OGDHC is primarily responsible for the decrease in adipate, in good agreement with earlier findings that OGDHC regulates the mitochondrial entry of fatty acids 47 , whose β-oxidation inversely correlates with the adipate levels 48,49 . In contrast to adipate, glutarate decreases significantly only in C6 cells (Fig. 8C,D), and the most pronounced change in glutarate is induced by the OADH-preferring AP (Fig. 8C). As shown in Fig. 6, the reduced generation of an OADHC product glutaryl-CoA may decrease glutarate, as the latter is generated from the former through non-enzymatic hydrolysis or the transacylation reaction with other acyl-CoA derivatives 50 . The most significant decreases in glutarate by AP, compared to those by SP and GP, confirm that the observed decreases in glutarate mainly originate from the inhibition of OADHC. Because the degree of the glutarate decrease should be inversely related to the spare OADHC capacity of the cells, lower expression of DHTKD1 in C6 vs MCF-7 cells leads to significant perturbation in the glutarate levels only in C6 cells. Remarkably, in C6 cells, not only the level of glutarate, but also that of glucose shows the strongest decrease after the treatment with the OADH-preferring AP, compared to the treatments with SP or GP (Fig. 8E). Moreover, the treatment of C6 cells with AP also significantly decreases 4-aminobutyrate, known to be strongly associated with perturbed glucose metabolism through the neurotransmitter function of this metabolite [51][52][53][54][55] (Fig. 8G). In www.nature.com/scientificreports www.nature.com/scientificreports/ contrast, neither glutarate, nor glucose, nor 4-aminobutyrate are significantly affected in MCF-7 cells by any of the phosphonates (Fig. 8D,F,H). The coupling of the AP-induced perturbations in the OADHC function and cellular glucose level is in good accord with the known association between the glucose homeostasis and DHTKD1 expression and/or function 23,27,56 . www.nature.com/scientificreports www.nature.com/scientificreports/ The results of statistical analysis of the phosphonates-induced changes in the selected metabolic indicators according to Friedman's test (Fig. 8) are supported by the results of clustering procedure presented in Fig. 7. The clustering ascribes glucose and 4-aminobutyrate to the same cluster not only in C6 cells, where the changes in these metabolites are statistically significant (Fig. 7A), but also in MCF-7 cells, where the metabolites are not significantly changed (Fig. 7B). However, only in C6 cells the cluster of glucose with 4-aminobutyrate belongs to the larger one including glutarate (Fig. 7A), whereas in MCF-7 cells the clustering does not reveal a relationship between these metabolites (Fig. 7B). In this regard, clustering of the changes in the C6 cell-specific metabolite nicotinate deserves attention. As shown in Fig. 7A, the phosphonates-induced changes in nicotinate are associated with those in glucose and 4-aminobutyrate, forming a larger cluster including also other sugars, i.e. maltose and fructose (green cluster in Fig. 7A). All metabolites of this cluster experience the most pronounced changes under the action of the bulkier phosphonates, GP or AP, and are affected much less significantly by SP. In contrast, specific action of SP increases the TCA-cycle-associated metabolites alanine and aspartate 36,57 significantly more than GP or AP do. Close positioning of the nicotinate-including cluster of sugars (green cluster in Fig. 7A) and the cluster including adipate and glutarate (blue cluster in Fig. 7A) is worth noting in view of the known interactions of nicotinate and DHTKD1 with regulation of pathways involving not only lipids, but also glucose 23,27,56,[58][59][60][61][62][63][64][65][66][67] .
Analysis of the correlated changes of all the quantified metabolites across the four experimental groups within specific metabolic networks of C6 or MCF-7 cells (Fig. S3) also reveals that in C6 cells the levels of nicotinate are coupled to those of glutarate and adipate. Moreover, this type of analysis indicates that all these metabolites form the cluster with 2-aminoadipate in C6 cells (Fig. S3A). In MCF-7 cells, the 2-aminoadipate cluster involves sugars and glutamate, but is distant from the cluster including lysine and saccharopine. Nevertheless, the cluster of lysine and saccharopine in MCF-7 cells is close to the one including the indicators of functions of the dehydrogenases of 2-oxoadipate and 2-oxoglutarate, i.e. glutarate and adipate (Fig. S3B), correspondingly.
As a result, the overall (Figs. 7 and S3) and specific (Fig. 8) changes in cellular metabolomes, which are observed after the phosphonates treatment of the cells with different DHTKD1 expression, reveal specific actions of the shortest (SP) and longest (AP) phosphonates targeting OGDHC and OADHC, respectively (Fig. 4). In the cells with a low DHTKD1 expression, significant impairment in the OADH function, manifested in reduced glutarate level, is associated with perturbed glucose homeostasis (Fig. 8). Detection of strong changes in metabolic profiles of cells in minimal medium, induced by the homologous phosphonates in accordance with their in vitro specificity to OADH and OGDH and affecting different groups of structurally unrelated metabolites, indicate that the phosphonates penetrate into the cells and interact with their expected enzyme targets.

Effects of the phosphonate analogues of 2-oxoglutarate and 2-oxoadipate on cellular NAD(P)H:MTT reductase activity. Cellular reducing power, which could be assayed using the NAD(P)
H-dependent reduction of artificial dyes, such as 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium (MTT), represents an integral indicator of cellular metabolism and proliferating ability. The changes in the reducing power of C6 and MCF-7 cells upon exposure to SP, GP or AP have been followed under the two sets of conditions: after a 5 h incubation with the analogues in a minimal medium (HBSS with glucose) and after a long (96 h) incubation in a complete growth medium (DMEM). The former conditions are the same as those where the cellular metabolic profiles presented in Fig. 7 have been assayed, whereas the latter conditions are used to assay the proliferating capacity of the cells. In addition to the action of the phosphonates on cellular metabolomes at a fixed concentration (Figs. 7 and 8), the concentration dependencies of cellular reducing power on the homologous phosphonates, which vary with the phosphonate structures ( Fig. 9), point to intracellular action of the phosphonates on cellular metabolism.
A strong concentration-dependent increase in cellular NAD(P)H:MTT reductase activity after a 5 h treatment by all the phosphonates is higher in MCF-7 cells (3-fold) than in C6 cells (2-fold) (Fig. 9A,B). In addition to the higher amplitude, the increase in MCF-7 cells is observed at lower concentrations of the phosphonates (from 0.2 mM, dashed line in Fig. 9B), compared to C6 cells (at 10 mM, Fig. 9A). As a result, at 5 mM concentration of the analogues, C6 cells do not change their reducing power, whereas MCF-7 cells exhibit almost maximal increase. A higher increase in MCF-7 cells is also obvious after a 96 h incubation of the cells with the phosphonates (Fig. 9C,D). Under these conditions, it also becomes obvious that the increase in the cellular NAD(P) H:MTT reductase activity is much more pronounced after the treatment by SP, than GP or AP. In MCF-7 cells SP also causes the most pronounced decrease in the levels of adipate (Fig. 8B) manifesting increased β-oxidation of fatty acids. It is therefore probable that cellular NAD(P)H:MTT reductase activity in the SP-treated cells increases due to contribution of β-oxidation of fatty acids. The contribution may be increased upon impaired production of NADH in the TCA cycle whose flux is limited by the SP-inhibited OGDH 36 .
After a prolonged (96 h) incubation, cellular viability is strongly reduced in both cell lines by all the three phosphonates (Fig. 9C,D). Remarkably, the strongest antiproliferative action in both C6 and MCF-7 cells is exhibited by AP, whereas SP still induces compensatory changes manifested in temporary growth of cellular reducing power at 0.02-0.2 mM SP in C6 cells and 0.2-1 mM in MCF-7 cells (Fig. 9C,D). The data testify to more limitations for compensating perturbations in the OADH-than OGDH-dependent metabolic pathways in both C6 and MCF-7 cells.

Discussion
Our study shows that OGDHC and OADHC extracted from mammalian tissues, are able to discriminate 2-oxoglutarate and 2-oxoadipate better than the complexes reconstructed from recombinant subunits 9,21 do. While the recombinant complexes exhibit significant level of cross-reactivity to the 2-oxoglutarate and 2-oxoadipate 38 , the high selectivity to the cognate 2-oxo substrates is demonstrated for OADHC in the non-solubilized liver and spinal cord homogenates or for OGDHC in the solubilized heart homogenate (Fig. 2C,D) www.nature.com/scientificreports www.nature.com/scientificreports/ results on the complexes in the brain 68 . In vivo, the multienzyme organization of OADHC and OGDHC should be adjusted to the pathways where the complexes perform their specific biological roles. In this regard, the native organization of OADHC and OGDHC as separate molecular machines has more advantages than a hybrid complex incorporating different E1 components. Existence of the 2-oxo substrate-specific multienzyme structures is supported by different stability, localization and/or regulation of the OADHC and OGDHC-catalysed reactions, observed in this work and earlier upon animal supplementation with thiamine 68 or mitochondrial production of reactive oxygen species 16 . Besides, the native structures may undergo different post-translational modifications and form heterologous complexes with other proteins of the specific metabolic pathways where OADHC and www.nature.com/scientificreports www.nature.com/scientificreports/ OGDHC participate. All these factors are known to affect the multienzyme structure of the 2-oxo acid dehydrogenase complexes 2 . The observed difference between the cross-reactivity to the 2-oxo substrates of the complexes present in animal tissues (this work) and those formed from recombinant components 9,21 , underlines significance of the study of native structures in addition to those assembled in vitro from the recombinant components.
New regulators of 2-oxo acid dehydrogenases have been developed in this work to specifically target the 2-oxo acid dehydrogenases isoenzymes transforming homologous substrates 2-oxoglutarate and 2-oxoadipate. Although polar molecules, such as 2-oxo phosphonates, do not diffuse across biological membranes, they may be transported by cellular carriers for similar compounds, such as 2-oxodicarboxylate carriers. In particular, slow transport of a phosphonate analogue of 2-oxoglutarate was suggested from the experiments with purified transporter of 2-oxoglutarate 36 . Intracellular action of the phosphonates is supported by the metabolic and viability effects of the phosphonates in cell lines (Figs. 7-9), correlating with in vitro specificity of their inhibitory action on OADHC and OGDHC (Fig. 4). Preferential inhibition of OGDHC by SP and OADHC by AP is revealed (Fig. 4) along with no significant effects of GP or AP on the enzymes transforming 2-oxo acids or their analogues are found (Table 1). Thus, confirming the previously established specificity of SP action in vivo 1,37,42,43 , the data of Table 1 show such a specificity for GP and AP as well. The specific binding of the 2-oxo substrates and phosphonates to the complexes in vitro agrees well with the results of cellular studies (Figs. 7-9), demonstrating that the metabolic and viability effects of the 2-oxoadipate mimics GP and AP strongly differ from those of the 2-oxoglutarate mimic SP (Figs. 7-9). Our recent structural data on the OGDH interaction with SP 69 point to the role of specific catalysis-related conformational transitions in the strong inhibition exerted by the phosphonates on their cognate 2-oxo acid dehydrogenases complexes. In view of the competitive inhibition by 2-oxo phosphonates vs 2-oxo acids 2,37,69 , a much lower concentration of 2-oxoadipate vs 2-oxoglutarate in cells and tissues 11,70 may increase the selectivity of the bulkier phosphonates GP and AP in targeting OADHC vs OGDHC. Thus, although one can never exclude an off-target effect of a drug, several lines of evidence provided in this work, strongly support the specific in vivo action of the phosphonate analogues of 2-oxoacids on the cognate dehydrogenases of 2-oxo acids. The supporting data include (i) screening of the enzymes from different classes and metabolic pathways regarding their reactivity to the phosphonates (Table 1), (ii) the metabolic actions of the homologous phosphonates (Figs. [7][8][9] in accordance with the preferential binding of SP to OGDHC and AP to OADHC (Fig. 4) and (iii) modelling of independent data on perturbed glucose metabolism in the DHTKD1 mutants 23 by AP better than by SP or GP (Fig. 8E).
The data obtained in this work pointed out the cell-specific involvement of OADHC not only into the saccharopine pathway, where the enzyme function has been studied earlier 24,27 , but also into metabolism of NAD. The two patterns of the DHTKD1 expression in animal tissues and cells are revealed by our transcriptomics analysis (Figs. 2E and 5A). Cellular expression of DHTKD1 is elevated along with the upregulated saccharopine pathway of lysine catabolism. Independent data confirm the prevalence of the saccharopine pathway in liver and kidney [71][72][73][74][75][76] , where the high level of DHTKD1 expression is observed (Fig. 2). By contrast, the saccharopine pathway is downregulated during the brain development 71,74,77 , and the DHTKD1 expression is not abundant in the adult rat brain (Fig. 2). However, a low OADH abundance, observed in most tissues ( Fig. 2A), may be relevant for regulation of NAD metabolism (Fig. 6). Our data show that in the cells with a low expression of DHTKD1, the OADHC inhibitor AP changes the levels of both glucose and nicotinate, a NAD precursor (Fig. 7A). It is also known that NAD biosynthesis de novo occurs through the tryptophan transformation into quinolinic acid 78 (QA, Fig. 6), which is alternative to the tryptophan catabolism generating 2-oxoadipate (Fig. 6) [78][79][80][81] . It may thus be suggested that the low abundance of OADH is involved in the regulation of the tryptophan fluxes through these alternative catabolic pathways. The OADH inhibition may increase the tryptophan flux to QA, thus elevating nicotinamide mononucleotide for biosynthesis of NAD (Fig. 6). This may alter the usage of cellular nicotinate in the pathway, increasing its involvement in biosynthesis of the signalling derivatives of NAD, such as nicotinic acid adenine dinucleotide phosphate and cyclic ADP-ribose phosphate 79,80 . Regulation of glucose homeostasis via the synthesis of these NAD derivatives 82-84 from tryptophan is supported by the diagnostic value of plasma tryptophan levels in diabetes and obesity 85 , and altered DHTKD1 expression under these pathologies 27,56,66,67 .
Thus, our study shows that incubation of C6 cells with a specific OADH inhibitor AP causes coupled changes in the levels of glucose, glutarate (the marker of OADH inhibition) and nicotinate (the marker of altered NAD metabolism). In addition to glucose, the levels of eight other sugars and their derivatives are significantly affected by AP in C6 cells (Fig. 7A). The findings support the AP-induced perturbations in sugar metabolism, coupled to alterations of NAD pathways.
As a result, the cell-specific features of the saccharopine pathway of lysine degradation or tryptophan-supported NAD biosynthesis may be addressed by different levels of DHTKD1 expression. The present work demonstrates that the adipoyl and succinyl phosphonates have their preferred target enzymes, OADH and OGDH, respectively, and are able to discriminate metabolic contributions of their complexes in mammalian cells and tissues. Ability of adipoyl phosphonate to specifically regulate the DHTKD1-encoded protein in vivo may be useful to develop combinatorial therapies correcting pathological states associated with changed DHTKD1 expression, among which are insulin resistance, obesity, and cancer.

Materials and Methods
Synthesis of the phosphonate analogues. All 1 H, 13 C and 31 P NMR spectra were recorded at 400, 100.6 and 161.9 MHz, respectively with a Bruker Avance 400 spectrometer.
After standing overnight, the pellets were briefly sonicated and clarified by centrifugation at room temperature. The resulting samples were used for kinetic assays and stored on ice at 4 °C with app. 20-30% OGDHC and OADHC inactivation within two weeks. The heart preparations also contained the activity of pyruvate dehydrogenase complex and liver preparation -the activity of branched-chain 2-oxo acid dehydrogenase complex, which were utilized in the studying of the specificity of the inhibitory effects of 2-oxo phosphonates.
Protein concentration in fractions was measured using Bradford assay 87 with modification of Zor & Selinger 88 .
Tissue homogenization for enzyme activity measurement and Western blotting. Tissue samples were homogenized in 50 mM MOPS, pH = 7.0, 20% glycerol and 2.7 mM EDTA in IKA T10 basic homogenizer. Sonication and detergent extraction of the homogenates was performed as previously described for the rat brain homogenates 68 . For Western blotting, the non-sonicated homogenates were diluted to app. 8.5 mg tissue per mL, and proteins were separated on hand-cast 10% acrylamide gels containing 0.5% 2,2,2-trichloroethanol. For detection of the DHTKD1-encoded protein, gels were blotted onto PVDF membrane using PowerBlotter (Thermo Scientific) and probed with anti-DHTKD1 antibodies (PA5-24208, Invitrogen, 1:1000) and HRP-conjugated secondary antibodies (P-GAM Iss, Imtek, Russia, 1:1000) in iBind Western Device (Thermo Scientific). Proteins were visualized using Clarity Western ECL substrate (Bio-Rad) and ChemiDoc MP Imaging System (Bio-Rad). Band intensities were calculated using ImageLab software (Bio-Rad) and normalized by relative whole protein level, measured as combined signal in the 20-250 kDa region of the corresponding gel lane using stain-free gel mode in ChemiDoc MP and ImageLab. Raw images of anti-DHTKD1-stained membrane and of total protein visualized in gel are given in Fig. S1A,B, respectively.
Protein detection by mass-spectrometry. Crude mitochondria-enriched fractions, prepared from fresh rat livers and cerebral cortices were obtained according to published procedures 4,89 . The proteins were separated in polyacrylamide gels and subjected to in-gel digestion by trypsin and analysed by LC-MS/MS as described earlier 90 . Protein identification was performed by PEAKS Studio 8.0 (Bioinformatics Solutions) with false discovery rate set to <1%. The number of identified peptides for DHTKD1, OGDH or DLST proteins was used as a semi-quantitative measure of their expression, as previously described 90 , since the two variables roughly correlate with each other 4 .
Enzyme activity assays. Enzyme activities of OADHC and OGDHC in tissue homogenates and enriched fractions were assayed in the overall reactions (Fig. 1B, Reaction 4) by spectrophotometric quantification of NADH production at 340 nm using Sunrise microplate reader (Tecan) and molar extinction coefficient of NADH 6220 M −1 *cm −1 , as previously described 68 . The assay medium (pH 7.0) contained 50 mM MOPS, 1 mM DTT, 1 mM MgCl 2 , 1 mM CaCl 2 , 1 mM ThDP, 2.5 mM NAD + , 0.05 mM CoASH, and 2 mM 2-oxoglutarate or 2-oxoadipate, if not specified otherwise. Initial rates of the OGDH and OADH reactions, catalysed by the tissue homogenates, were determined during 5 min. For the enzyme-enriched preparations, the time interval between 3 and 10 min from the reaction start, was found optimal, as the burst kinetics of the blank reaction lacking the 2-oxo substrate, was completed within 3 min. All the reaction rates were corrected for the rates of the blank reactions in the medium omitting 2-oxo acid substrate. Activities of other enzymes related to 2-oxo acid metabolism, were assayed in the following media: pyruvate dehydrogenase complex -in 50 mM MOPS (pH 7. Activities of the pyruvate and branched-chain 2-oxo acid dehydrogenase complexes were measured in enzyme-enriched preparations, obtained from the rat heart and liver as described above. Other enzymes were assayed in sonicated and detergent-treated homogenates of the same tissues and/or commercially available cellular studies. Dulbecco's modified Eagle's medium (DMEM; Sigma D6046, supplemented with 10% FBS, 100 IU/mL penicillin and 100 µg/mL streptomycin) and DMEM/F12 1:1 (Sigma D8437, supplemented with the same reagents) media were used for culturing C6 rat glioma and MCF-7 human breast adenocarcinoma cell lines, respectively. Both lines were obtained from American Type Culture Collection (LGC Standards, Poland). To grow cells for the viability tests, the cell-optimized densities were used: C6 cells were seeded in 96-well plates in 100 µl DMEM at a density of 5*10 3 cells/well for 5 h incubation or 1*10 3 cells/well for 96 h incubation; MCF-7 cells were seeded in 100 µl DMEM/F12 at a density of 1.5*10 4 cells/well for 5 h incubation or 5*10 3 cells/well for 96 h incubation. After 24 h the media were exchanged for HBSS, supplemented with 1 g/L glucose, 1.3 mM CaCl 2 , 1.0 mM MgSO 4 , and phosphonate analogues (0-10 mM) for 5 h incubation, as previously described 91,92 . For 96 h incubation, the media were exchanged for the same ones including the phosphonates. After 5 or 96 h incubation with the analogues, changes in the NAD(P)H levels were measured by using the redox indicator, 3-(4,5-dim ethylthiazol-2-yl)-2,5-diphenyltetrazolium, as previously described 92 , with sodium dodecyl sulphate used as a solvent for formazan crystals.
For measuring the levels of different metabolites, cells were seeded on 5 or 10 mL Petri dishes (5*10 5 or 1.5*10 6 cells/dish for C6 cells; 1.5*10 6 or 4*10 6 cells/dish for MCF-7 cells) in the same growth medium as described above for viability tests. 24 h later, the medium was exchanged for HBSS, supplemented with 1 g/L glucose, 1.3 mM CaCl 2 , and 1.0 mM MgSO 4 . The control cells and cells with 0.5 mM of SP, GP or AP added to the medium were further incubated for 5 h. After that, the cells were washed twice with PBS, and metabolites were extracted for quantification of metabolite abundance. 2 mL of ice-cold methanol containing 50 μM ribitol as internal standard 93 were used for extraction of cells grown on 5 mL Petri dishes; for cells grown on 10 mL dishes, 4 mL of the extracting reagent were used instead. The extracts were centrifuged for 20 min at 65536 g, 4 °C, with the supernatants and pellets used for the quantification of metabolites and total protein, correspondingly, after storage at −70 °C. www.nature.com/scientificreports www.nature.com/scientificreports/ metabolites during derivatization. After incubation for 2 h at 37 °C, 68.6 μL of N-methyl-N-(trimethylsilyl)trifluoroacetamide (for derivatization of polar groups) and 1.4 μL of the mixture of the methyl esters of various fatty acids (for calibration of the chromatograph) were added to the extracts, and the samples were derivatized for 30 min at 37 °C.
After the incubation, the samples were analysed using a gas chromatography-coupled mass spectrometer Pegasus BT (LECO). Chromatograms and mass spectra were analysed using the ChromaTOF (LECO) and OpenChrom (open-source) software. TagFinder software 111 together with the metabolite library of Max Plank Institute of Molecular Plant Physiology 112,113 were used for the peak annotation and quantification. Only the metabolites which were reliably detected in all experiments, were used for characterization of the phosphonate effects. The metabolite peak heights were normalized for that of ribitol and for total protein content, measured by the Bradford assay 87 in the cell pellet, obtained during methanol extraction procedure.
Influence of a phosphonate on metabolic profile in each experiment was estimated by normalization of the data obtained from the phosphonate-treated cells to the control cells in the same experiment. The ratios obtained from independent experiments were log2-transformed and used to find mean ± SEM of the changes in metabolic profiles. For comparison between non-treated C6 and MCF-7 cell lines, the mean values were determined from log2-transformed absolute levels. Heatmaps were built in RStudio 1.2 (RStudio Inc.).

Statistical analysis.
Fold changes in metabolite abundances were calculated within one cell batch, log2-transformed and averaged using the data obtained with different cell batches. One-sided unpaired t-test was applied to the log2-transformed ratios to reveal statistical significance of differences between the phosphonate-treated and control cells. To compare the untreated C6 and MCF-7 cell lines, the unpaired t-test was applied to ribitol-and protein-normalized and log2-transformed peak heights.
Multiple groups were compared either by one-way ANOVA or Friedman's tests, followed by Tukey's or Dunn's post-hoc tests, respectively, depending on the results of Shapiro-Wilk test for normality. The specific details of the analysis are indicated in the figure captions. For visualization of correlation matrices, Spearman's correlation coefficients and their p-values were calculated. Differences and correlations with p ≤ 0.05 were considered significant.