Lysine metabolism conveys kidney protection in hypertension


 Hypertension and kidney disease, two related, common, and severe disease entities, have been repeatedly associated with genomic variants and metabolic alterations of lysine metabolism. Here, we developed a stable isotope labeling strategy compatible with untargeted metabolomics acquisition to investigate the physiology and molecular spectrum of lysine’s metabolic fate in vivo. Mice received 13C6 labeled lysine through the diet over two months to track more than 100 lysine metabolites across various organs and body fluids. Lysine reacts rapidly with molecules of the central carbon metabolism, as opposed to slow or incorporation into proteins and metabolization into acylcarnitines. The kidney rapidly forms these lysine conjugates and lysine metabolism is decreased in early stages of hypertension. Lysine administration completely diminished the development of salt-sensitive hypertension and kidney injury in the Dahl salt-sensitive rat model. Administration of lysine leads to diuresis, acceleration of 13C6 lysine conjugate formation, and inhibition of albumin uptake, thereby protecting from nephron injury and metabolic stress. Lysine conjugates with malonyl-CoA to form a novel metabolite Ne-malonyl-lysine to inhibit fatty acid synthesis. Formation of Ne-malonyl-Lysine, and acetyl-Lysine during lysine treatment depletes malonyl-CoA and acetyl-CoA, respectively, a process that occurs at the expense of protein malonylation and acetylation. A significant fraction of lysine molecules was metabolized in the kidney and excreted as fructoselysine and saccharopine, via the urine, leading to an overall depletion of central carbon metabolites from the organism. A ketogenic diet also ameliorated hypertension, yet to a lower extent, and increased several lysine conjugates, including Ne-malonyl-lysine. In conclusion, isotope tracing of orally administered lysine illuminated lysine metabolism, and L-lysine protects the kidneys in metabolically defined subtypes of kidney disease.


Introduction
In the past decade, the role of amino acid metabolism has extended from a provider of protein building blocks to a decisive element in various physiological processes and diseases 1,2 . The amino acid lysine is an essential amino acid that is at the crossroads between proteins and metabolism and is of interest as both drug target and modi er of protein activity [3][4][5] . Lysine incorporated into proteins is heavily modi ed with an increasing number of molecules such as organic carboxylic acids [6][7][8][9][10] . Lysine from protein residues also gives rise to carnitines with a central role in beta oxidation. Multiple molecular lysine degradation and modi cation reactions are known. In this context, lysine's metabolism and its related bioactivity 11 has not been studied in a systematic fashion. Untargeted metabolomics offers the ability to identify altered and previously undescribed physiologically relevant molecules 11 . When applied together with isotope-labeling (using 13 C labeled molecules), kinetics of synthesis and degradation can be quanti ed, and novel metabolites can be discovered.
The kidney is a central hub of mammalian metabolism, but its metabolic importance is relatively understudied. With up to one out of ten persons suffering from kidney disease, kidney diseases pose a signi cant socioeconomic burden 12 . Hypertension, potentially caused by kidney disease, is also an epidemic, and both entities are strongly associated 13 . Multiple genome-wide association studies have highlighted the role of carnitine metabolism and lysine transport (via kidney-speci c transporter SLC7A9) [14][15][16] in kidney disease. Metabolomic biomarker studies in the Framingham cohort, one of the most important cohort for prediction of cardiovascular outcomes, showed that high abundance of lysine in the urine is protective 17 . Three randomized controlled trials (RCTs) in humans performed untargeted metabolomics of serum and urine under dietary conditions bene cial for hypertension (so-called DASH diet, or sodium restriction) [18][19][20] , and found association with high signals of lysine, or modi ed lysine, with protective interventions.
The aim of this paper was to establish an untargeted metabolomics work ow to interrogate ten body organs of a mouse that were fed with an isotope labeled diet 13 C 6 lysine chow for up to two months, and to utilize this knowledge to intervene with hypertensive kidney disease. The analysis revealed that the kidney is strongly involved in handling lysine metabolites, with a speci c bias towards conjugating free lysine to modi cations at a faster speed. Notably, because of the identity of lysine conjugated molecules formed, and lysine's metabolic activity, strong metabolic alterations driven by excess lysine orchestrate a kidney-protective effect in hypertension.

Results
Adult male mice were fed with normal diet and were switched to food containing exclusively 13 C 6 lysine for 0 (control), 1, 2, and 8 weeks (Fig. 1A). Unlabeled mice served as a control. A signi cant shift in global metabolomics features occurred depending on the duration of the diet (Fig. S1A). The analysis consisted of two approaches: 1) a mass-difference approach that looks for coeluting peaks (modi ed from previous protocols 21 ), and 2) a correlation-based approach that extracts the mass trace of provided compounds and uses correlations to call out isotopologues. The analysis revealed 66 metabolites (approach 1) as well as 133 metabolites (approach 2) that were validated using standards and fragmentation MS2 spectra (Fig. 1B). A signi cant overlap of validated metabolites was found by both methods. Overall correlation of extracted and veri ed metabolites was higher than other, non-veri ed metabolites ( Fig. S1 B, C). Analysis of isotopologue patterns revealed that virtually all these metabolites were labeled with 13 C 6 or 13 C 4 atoms, meaning their entire carbon backbone was incorporated (Fig. 1C).
Rates of 13 C 4 incorporations were signi cantly slower incorporated as compared to C6 units (Fig. S1 D).
We then separated the labeled peaks into different classes of lysine metabolites that included lysine, lysine degradation products such as saccaropine, pipecolic acid and aminoadipic acid, modi ed lysines such as N-and E-acetylation, methylation, hydroxylation and many others, carnitine synthesis products such as trimethyllysine, lysine-containing peptides and several acetylcarnitines. Detected lysine metabolites as well as chemical information are listed in TABLE S1 and S2. In general, we found using both methods that lysine had the highest incorporation slope among the metabolites, followed by lysine degradation and modi cation, and by peptides. In general, incorporation of lysine into acylcarnitines was slower than other metabolic processes (Fig. S1D). We calculated the slope of different classes of metabolites using a linear model during the rst two weeks. Notably, we did not observe any labeling in 13 C 2 units (such as glycine), and targeted analysis of the TCA cycle metabolites revealed no signi cant incorporation of TCA metabolites in the TCA cycle in any tissue (summary in Fig. S2). Analysis of the diet using untargeted metabolomics revealed that the food had no detectable amount of carnitines, and other lysine metabolites in the diet (Fig. S1). Comparison of different organs was performed next, and slopes of incorporation were summed and plotted. Organ-speci c incorporation rates (plotted as slope of difference of leading C 13 /C 12 isotope) of lysine metabolite classes were distributed to the different organs (Fig. 1D).
In the thymus, we observed several lysine incorporated peptides that were represented in this organ. The analysis revealed that the kidney cortex was one of the fastest incorporating organs, especially for modi ed lysines, peptides, and acylcarnitines. (Fig. 1D). Kidney and liver showed both the highest incorporation rates, but the kidney had a higher incorporation rate for Ne-Fructosyl-lysine (Fig. 1E).
We then sought to integrate metabolome and proteome in an independent experiment. Kidney cortex tissue from mice fed with 13 C 6 lysine for 1, 2, and 3 weeks were analyzed. Since lysine also gets incorporated into proteins, we determined both the metabolome and the proteome from these samples. We plotted the logarithmic ratios of the maximum isotopologue (peptides labeled with a mass shift of 6 Da) over control (Fig. 1F). The isotopologue ratio for lysine was higher than any other molecule, followed by metabolites from the lysine degradation pathway, and modi ed lysines. Proteins were incorporated signi cantly slower as compared to lysine metabolite, and histone and DNA-binding proteins had the lowest incorporation rate among proteins (Fig. 1F). Metabolites ratio of labeled over unlabeled molecules had a wider distribution compared to proteins (Fig. 1G). To further investigate if protein incorporation is different in the medulla, we micro-dissected the kidneys in four macroscopic areas: outer stripe cortex, outer stripe medulla, inner stripe outer medulla, and inner medulla (FIG. S3A). We found that the lysine protein incorporation was slower in the medullary areas, consistent with the known metabolic role of the proximal tubule (FIG S3B). Taken together, these data suggests that the cortex, and speci cally the proximal tubules that comprise more than 80% of the cortex, is likely to be the major hub within the kidney for lysine metabolism.
We had earlier reported a de cit of lysine metabolites in the kidney cortex in a hypertensive kidney disease model 22 . Based on these ndings, and on a potential bene cial role of lysine in the Framingham cohort 17 and 3 randomized controlled dietary intervention trials [18][19][20] , we hypothesized that administration of lysine can be helpful in a model of hypertensive kidney disease. We subjected hypertensive rats to high abundance of lysine in the drinking water ( Fig. 2A). Dahl Salt-Sensitive (D/SS) rats are a physiological relevant model of hypertension and hypertensive kidney disease that generates high blood pressure when fed a high salt diet. We monitored real-time blood pressure using telemetry over the entire time course of the disease (Fig. 2B). Both male and female rats on different salt diet regimens receiving lysine supplementation were largely protected with attenuated development of hypertension ( Fig. 2B, Fig. S4A). Albuminuria, a readout of hypertensive kidney damage, was decreased in various regiments (Fig. 2C, Fig. S4B). The excretion of potassium, as well as excretion of urinary metabolites/electrolytes did not change between lysine treated and non-treated rats, suggesting that hypertension-inducing salt diet was not taken up to a different extent (FIG. S4C). Automated histology analysis using convoluted neural nets 23 revealed a signi cant improvement in ultrastructure of glomeruli by lysine (based on quanti cation of more than 3500 glomeruli/group) (FIG. S4D).
To understand lysine handling in high and low lysine conditions, we performed in vivo isotope labeling of 13 C 6 lysine in lysine treated and non-treated hypertensive rats. We injected lysine-treated and lysineuntreated rats with a de ned bolus of 13 C 6 lysine ( Fig. 2D) and analyzed urine (collected 24 hrs), serum and kidney tissue for incorporated lysine metabolites. More 13 C 6 labeled lysine was detected in urine of ad libitum lysine-treated rats (Fig. 2E). We also found an increased abundance of 13 C 6 labeled saccharopine, and fructoselysine in urine (normalized by creatinine), while 13 C 6 incorporation in the kidney was reduced (Fig. 2E). Consistent with an increased lysine clearance, physiological studies measuring 24 hrs urine revealed that lysine induced diuresis (Fig. 2F), a nding observed after only 1 day of lysine treatment (Fig. S4E).
Hypertension results in proteinuria, and proteinuria itself is further detrimental for kidney function when urinary proteins, especially albumin, is taken up by the proximal tubule, the predominant cell type within the kidney cortex 24 . Therefore, we hypothesized that albumin uptake in the proximal tubule was blocked or altered by lysine, supported by previous in vitro and in vivo ndings 25 in a physiological context. Consistent with this hypothesis, lysine administration itself in D/SS rats on a normal salt diet induced an increase in urinary albumin (Fig. 3A). Cell culture studies using Opossum Kidney (OK) proximal tubule cells revealed that even moderate addition of lysine decreased uptake of uorophore labeled albumin (Fig. 3B), a key function of proximal tubules. Basolaterally-applied lysine (2 hr, 50 mM) inhibited uptake of albumin and the uid-phase marker dextran (not shown), even when the cells were assayed 2 h after lysine washout (Fig. S5A). This is remarkable, because the receptor LRP2 responsible for albumin uptake is localized apically 26 . In addition, we observed that overnight incubation with lower doses of lysine resulted in a profound and dose-dependent decrease in albumin uptake (half max inhibition at ~ 3.5 mM lysine) (Fig. S5B). Next, we examined hypertensive kidneys after dietary high-salt exposure in vivo using dual photon microscopy and uorescently labeled albumin (Fig. 3C). Proximal tubules took up albumin and accumulate excessive albumin in luminal protein casts (arrows), suggesting that an overloading of the albumin uptake machinery occurs in hypertension. Consistently, lysine reduced albumin casts in kidneys as observed in Masson's trichrome staining (Fig. 3D). In immunohistochemistry staining, LRP2 (megalin) abundance, the chief lysine uptake mechanism protein inhibited by lysine, was decreased in dilated non-functional proximal tubules damaged by albumin 25 . This effect was completely diminished by lysine administration (Fig. 3D). Expression of kidney injury molecule-1 (KIM-1), a marker of proximal tubule injury, was strongly decreased with lysine as well, suggesting protection from tubule damage by lysine (Fig. 3D).
Combined, in vivo and in vitro observations argue for an effect of lysine beyond simple competitive inhibition of albumin binding to the apically-localized megalin and cubilin receptors 26 . To further investigate the mechanism, we examined the cortex metabolome (Fig. 3E). Kidneys from healthy rats (on a low-salt diet) did not change lysine or kidney metabolism at all when challenged with a high lysine intervention (q < 0.05). However, with increased disease severity, there were more alterations in lysine metabolites, and in general metabolism. Lysine was increased in hypertension, as also supported by HPLC analysis (Fig. S5C). We found that especially metabolites that had a faster incorporation in the initial experiment (such as Ne-Acetyllysine, saccharopine, and Fructose-lysine), were increased through the lysine challenge, suggesting these metabolites serve as an "over ow" (Fig. 3E). Further analysis of the dataset suggested that free fatty acids (FFA) were altered by lysine supplementation (Fig. 3F), with a strong decrease in the most severe damage condition. The entire class of FFA (n = 22 quanti ed molecules) showed a strong and distinct downregulation, with linoleic acid being decreased by lysine treatment in every single condition. Further examination of the dataset revealed a reduction of glucose and other sugar metabolites. NAD + , a kidney protective molecule associated with decreased oxidative stress, was increased with lysine treatment in the severe conditions (Fig. 3G).
We hypothesized that lysine's tendency to form conjugates was responsible for reduction in the FFAs. We examined the isotope labeling dataset for lysine conjugates that could be responsible for a decrease in fatty acid building blocks. Careful examination of the unknown peaks in the untargeted metabolomics isotope study revealed the presence of a mass consistent with Nε-malonyl-lysine in the kidney and liver (Nε-malonyl-lysine, m/z [H+] of 233.11 and m/z as a Na adduct of 255.10) (Fig. 4A). This molecule is a previously undescribed novel metabolite without a Pubchem entry. Synthesis of a standard, together with retention time comparison (Fig. S6A,B, Supplemental material and methods) con rmed the identity of the compound as Nε-malonyl-lysine (Fig. 4B), and targeted analysis using MRM showed labeling only the 13 C 6 lysine containing kidney tissue (Fig. S6F). In order to understand if malonyl-lysine would form enzymatically or chemically, we analyzed whether it would form in alkaline conditions from the educts malonyl-CoA and lysine (1:10 w/w) ratio. We observed that < 1% of lysine molecules were malonylated (Fig. 4B). We found that Nε-malonyl-lysine was signi cantly increased in the hypertensive kidneys with lysine ( Fig. 4C), suggesting that supraphysiological lysine supplementation could deplete the abundance of malonyl-CoA, a substrate of fatty acid synthesis and an inhibitor of fatty acids. Consistently, we observed signi cant decrease of malonyl-CoA (Fig. 4D). In addition, Nε-acetyl-Lysine was increased by lysine treatment (Fig. 4E). Acetyl-CoA was decreased in at least one lysine treatment condition and trended to be reduced in the other two conditions (Fig. 4F). While the free acetyl-lysine and Nε-malonyllysine were increased, protein modi cation through these metabolites decreased as evidenced by immunoblotting (Fig. 4G). These ndings suggest that the chemical modi cation of lysine occurs in the kidney to deplete important anabolic molecules Acetyl-CoA and malonyl-CoA.
We then analyzed the metabolic changes in a more integrative fashion, evaluating both matched tissue, urine, and serum from the same animals, and calculating correlation coe cients between individual metabolites in different compartments. Integrative analysis of correlations between tissue and urine metabolites revealed that urinary lysine was strongly associated with tissue lysine and tissue lysine degradation products (Fig. 5A). High lysine in the urine was also negatively correlated with proteinuria ( Fig. 5B). Interestingly, urinary lysine was also negatively correlated with the abundance of sugar molecules, such as glucose, glucose-6-phosphate, and sucrose. In fact, targeted analysis revealed a decrease in glucose abundance in the cortex. Together with an increase of Nε-fructoselysine in the urine this suggests a net glucosuric effect. Interestingly, this effect was stronger in severely damaged cortices. Lysine treatment also induced loss of other amino acids under these conditions, consistent with previous reports (Fig. 5C). The strongest excretion of lysine was as a sugar conjugate (fructoselysine), acetyl-CoA (acetyl-lysine), and α-ketoglutarate (saccharopine) in the damaged conditions (Fig. 5C). Analysis of lysine stoichiometry (mol/per ug creatinine) revealed that most of the excreted lysine was in the unmetabolized form. However, the excretion of lysine in the form of saccharopine, fructoselysine was more signi cant when kidneys were protected by lysine (Fig. 5D). Using targeted metabolomics, we con rmed the decrease of glucose and glucose-6-P in the tissue, but there was no altered abundance of these metabolites in the urine. The tissue abundance of the nal metabolites of the TCA cycle, fumarate, and malate, were signi cantly depleted by lysine (Fig. 5E), although their urinary excretion was not altered (not shown). Metabolomic analysis of serum revealed that in hypertension, there were no major changes in fructoselysine, saccharopine, acetyl-lysine, and others (Fig. S7), as opposed to control rats. Taken together, these results suggest that excessive lysine reacts with active metabolites of the central carbon mechanism to excrete them, suggesting metabolic alterations may play a role in the bene cial effects of lysine in this model.
Lysine administration caused a net loss of glucose by forming Nε-fructoselysine, and a net loss of metabolites of anabolic processes, such as acetyl-CoA, malonyl-CoA and TCA cycle intermediates. This situation could be consistent with a local catabolic state, like ketosis. Therefore, we checked whether a ketogenic diet was able to yield a protective effect in the same model, and alter lysine metabolism (Fig. 6A). In fact, the rats were protected from hypertension (Fig. 6B, Fig. S8), but still developed proteinuria. Metabolic analysis of the tubules showed that lysine, acetyl-lysine, saccharopine, adipic acid and galactosyl-hydroxylysine were signi cantly increased along with precursors of ketone bodies, while central sugar metabolites were expectedly decreased (Fig. 6C). Consistent with the behavior of these lysine conjugates, Nε−malonyl-lysine metabolite was also increased, suggesting that reduced availability of hexose to bind lysine was causing shifts to other lysine conjugates (Fig. 6D). The data support the role of metabolic interventions with lysine in kidney disease (Fig. 6E), and demonstrate lysine-dependent processes controlling physiology and metabolism of the kidney.

Discussion
Alterations in lysine metabolism are commonly observed in patients obtaining diets protective from hypertension [18][19][20] , and lysine transport also associates with kidney function across multiple GWAS and mGWAS studies. High abundance of lysine in the urine (associated with variants in the tubular lysine uptake transporter SLC7A9) associated with a good prognosis 17 , and these ndings were replicated in another CKD cohort 14 . The important role of lysine for kidney disease is also supported by animal studies suggesting that lysine supplementation does not alter lysine in the serum, but has effects on secondary disease entities in the bone 27 . Together, these data demonstrate that differential amino acid metabolism can be a potential target for diseases at the interface of hypertension and kidney disease.
Here, we develop a technology to utilize untargeted metabolomics work ows in order to quantify lysine fate across tissues. Isotope tracing methods usually use targeted metabolomics in order to develop hypothesis-driven views or models of metabolism. To catalogue as many molecules as possible, we performed global untargeted metabolomics analyses to identify labeled metabolites 28,29 . This work ow compares lysine metabolites in extension of previous studies that focused on protein incorporation [30][31][32] . The approach used here comes with some limitations of untargeted metabolomics, but also has several key advantages. First, one can catalogue the entity of metabolites of a distinct molecule at a broader depth as targeted approaches. Second, it allows to discover new and biologically important molecules from untargeted isotope labeled metabolomics data. The use of labeling at high percentages (over 95% of labeled lysine molecules were labeled) enables identi cation of signi cantly incorporated metabolites in vivo. The feeding through the diet (as opposed to parenteral administration) enables analysis of microbiome-derived metabolites. The combination of untargeted isotope-based metabolomics and proteomics suggested that lysine is replaced faster with 13 C 6 units before it is incorporated into the proteins and carnitines. This suggests a major network of lysine modi cation ("epimetabolome") independent of the incorporation of metabolites into proteins, and that the main pool of free modi ed lysines derive from direct chemical and enzymatic reactions 5 , and not from protein modi cations such as histones.
The observation of molecular Nε−malonyl-lysine, a previously unidenti ed metabolite, was only possible through the in vivo isotope labeling approach. We suggest that the malonyl-Lysine metabolite is formedenzymatically or non-enzymatically -via the reaction with malonyl-CoA, the elongator of fatty acid synthesis, and an important anabolic metabolite. In fact, we observed lysine-dependent decrease in free fatty acids in the kidney cortex, but not in serum and urine, suggesting a renal speci c effect. This also suggests that concentrations of lysine in the kidney are predisposing for the formation of the metabolite.
The alteration of FFA synthesis by NE-malonyl-lysine might also be of importance to general mechanisms of brosis development 33 .
We used lysine administration to investigate kidney-speci c lysine handling in hypertension. Lysine strongly decreased blood pressure, proteinuria, tissue damage and increased NAD+, a kidney-protective metabolite [34][35][36] , while at the same time increasing diuresis. The proteinuria is particular relevant because lysine itself blocks albumin uptake, and induces a mild proteinuria in normal conditions (Fig. 3). Several previous studies suggested the presence of detrimental lysine modi cations in hypertension 37,38 . Our study suggests that administration of lysine leads to conjugation of lysine to central carbon molecules, while it prevents these molecules to modify proteins. Mechanistically, these processes lead to diuresis that ushes out the protein plaques formed in the tubule system. Metabolically, this is bene cial because the lysine is excreted in modi ed forms: one prominent and fast lysine over ow system is the lysineconjugation to central carbon metabolites. Acetyl-CoA, and glucose, form acetyl-lysine and fructoselysine (Amadori reaction), respectively. Formation of saccharopine consumes NADH and alpha-ketoglutarate and generates NAD and Glutamate. Together, this depletes metabolites of the central carbon metabolism from the kidney. The depletion of central carbon metabolites from the kidney is reminiscent of the effect of SGLT2 inhibitors, a recent breakthrough intervention to delay the outcome of cardiovascular kidney disease 39,40 . In fact, the quantitatively most important excretion mode of lysine was Fructosyl-lysine. Similar to the lysine challenge, ketosis of the entire body led to a depletion of sugars in the kidney cortex and increased Nε−malonyl-lysine, as well as saccharopine, presumably because more lysine is freely available. Notably, this metabolic intervention could also ameliorate hypertension, but kidney damage was altered to a lesser degree. Ketogenic diet has been a successful intervention to kidney diseases such as polycystic kidney disease 41 commonly associated with high blood pressure.
In conclusion, the analysis of labeled metabolites demonstrated the kidney-speci c role of lysine metabolites, and its physiologic activity. Kidney protective effects in hypertension occur via lysine's unique physiology in the tubules and via its chemically modi ed entities that provide a sink for important sugar metabolites of the central carbon metabolism and fatty acid metabolism. Thus, lysine's metabolic activity alters the epimetabolome for kidney protection. Methods 13 C 6 Lysine isotope labeling study.
The protocol for labeling was described previously 30 . After 1, 2, and 8 weeks (protocol 1) or 1, 2, and 3 weeks (protocol 2), mice were sacri ced and perfused with ice-cold PBS. All denominated organs were snap-frozen from the same mouse and stored at -80 °C degrees. Blood was separated in erythrocyte and plasma by centrifugation.

Metabolomics sample preparation.
Frozen organs were kept on dry ice and 10 mg tissue was weighed. Then, 800 ul ice-cold extraction solution containing Acetonitril : Methanol : Water 2:2:1 was added. Samples were subjected to tissue homogenization using a multiplex-bead-beater (Storm) for 30 sec (liver), 1 min (lung), or 45 sec (all other tissues). The supernatant was transferred to a prechilled Eppendorf tube. Then, beads were washed with 200 ul of the extraction solution, and the wash solution was added to the remaining homogenate. The homogenate was incubated at -20 °C degrees for 2 hrs. Then, tissue was spun down at 4 °C degrees for 20 min, and the supernatant was transferred to another vial. The supernatant containing the extract was transferred into a speed vac and dried down. The next morning, the dried-down extract was once resuspended in 100 ul (per 10 mg tissue) of acetonitrile-water 1:1, and the solution was centrifuged at 4 °C degrees. Then, the solution was transferred to autosampler vials and stored at -80 °C until further use.
Untargeted metabolomics analysis and mass spectrometry.
LC-MS/MS analysis for metabolomics was performed as previously described 22 . Data was annotated with in-source fragments and adducts. Quality controls were run every 5 samples. The mass spectrometer was initially calibrated using NaFormate peaks and in addition post-run. For untargeted metabolomic analysis, we used a UHPLC-MS approach. For fractionation, we used hydrophilic interaction liquid chromatography (HILIC) fractionation and reversed-phase (RP) chromatography as previously described 22 . We used a quadrupole time-of-ight instrument (Impact II, Bruker, Bremen, Germany) coupled to a ultrahigh-performance liquid chromatography (UHPLC) device (Bruker Elute, Bruker, Billerica, MA), or to an Agilent In nity 1290 UHPLC device (Agilent, USA). The MS was calibrated using sodium formate (post-run mass calibration). Data were acquired over an m/z range of 50 to 1000 Da in positive ion mode and negative ion mode (HILIC only). Electrospray source conditions were set as follows: end plate offset, 500 V; dry gas temperature, 200 °C; drying gas, 6 liters/min; nebulizer, 1.6 bar; and capillary voltage, 3500 V.
To increase metabolome coverage and minimize ion suppression, we used a dual fractionation strategy. For RP separation, an ACQUITY BEH C18 column (1.0 × 100 mm, 1.7-µm particle size; Waters Corporation, Milford, MA) was used, and for HILIC fractionation, a ACQUITY BEH amide (1.0 × 100 mm, 1.7-µm particle size; Waters Corporation, Milford, MA) column was used. Flow was 150 µl/min, and a binary buffer system consisting of buffer A (0.1% FA) and buffer B (0.1% FA in acetonitrile) was used. The gradient for RP was: 99% A for 1 min, 1% A over 9 min, 35% A over 13 min, 60% A over 3 min, and held at 60% A for an additional 1 min. The gradient for HILIC consisted of 1% A for 1 min, 35% A over 13 min, 60% A over 3 min, and held at 60% A for an additional 1 min. The injection volume was always 2 µl. For molecule identi cation purposes, putative molecules of interest were fragmented using three different collision energies (10, 20 eV) or ramp collision energies (20 to 50 eV).
Bruker Raw les (*.d) were transformed into mzml les using the compassxport_converter.py script (Bruker). Then, data les were uploaded to XCMS online, and differential peaks were extracted in positive and negative ion mode 42 . Feature Detection was performed with the centwave method with the following options: ppm = 10, minimum peak width = 5, maximum peak width = 20, mzdiff = 0.01, Signal/Noise = 6, Integration method = 1, pre lter peaks = 3, pre lter intensity = 100, noise lter = 100. For Retention time correction, we used obiwarp method with profStep = 1. Alignment was perfomed with bw 05, minfrac = 0.5, mzwid = 0.015. mminsamp = 1, and max = 100. Statistical test was performed using an unpaired parametric t-test (Welsh), with post-hoc analysis = TRUE,. Statistical ltering was performed for priorization of features of interest, including a fold change of at least 1.5, and a p-value lower than 0.05, and a corrected p value (q-value) lower than 0.05. The analysis included PCA and multivariate analyses.
Metabolites were identi ed based on 1) unique mass, 2) MS 2 spectra comparison to authentic standard 3) coelution with authentic standard, and 4) isotopic pattern. The following adducts were routinely considered : Na, H, for positive mode, and Cl, formate for negative mode. Intensity data were plotted using instantclue 32 , ggplot2, or circosplot.R package.

Kidney dissection.
Kidney cortices were isolated under manual control (protocol 1, Fig. 1). Kidneys were manually dissected using a stereomicroscope. Based on anatomical criterions, the kidneys were dissected into cortex, inner stripe outer medulla, outer stripe outer medulla and inner medulla (protocol 2, Suppl. Figure 3). The samples were snap-frozen and stored at -80 °C or dry ice.

Proteomics sample preparation.
Kidney samples of 13 C 6 labeled kidneys were subjected to proteomics analysis using a tryptic in solution digestion protocol followed by nLC-MS/MS analysis. In brief, kidney pieces were minced and homogenized using a glass homogenizer in 8M urea containing 10 mM Ammoniumbicarbonate as well as protease and phosphatase inhibitor cocktail (Thermo). The homogenate was spun down at 6 °C for 20 min, and the supernatant was kept for further analysis. A small aliquot was subjected to protein measurement using BCA assay (Thermo). The proteins were reduced and alkylated using 5 mM DTT (30 min) and 10 mM IAA (1 hr) in the dark. Then, urea concentration was diluted to < 2M and LysC (1:100 w/w) ratio was added, and the mixture was digested for 16 hrs at 37 °C. The next day, the reaction was terminated by addition of 2% formic acid. Peptides were desalted using in-house made stage-tips and analyzed by nLC-MS/MS.

Proteomics analysis.
The peptides were separated by reverse-phase nano ow-LC-MS/MS analysis and sprayed into a quadrupole-orbitrap tandem mass spectrometer (qExactive plus, thermo scienti c) as previously described 43 . Raw Proteomics data were parsed with MaxQuant v 1.5.3.3. 44 , using a Uniprot RefSeq reference proteome database from Jan 2017, with using LysC (cuts after each Lysine) as a protease.
Multiplexicity of the analysis was 2, with 13C6 labels in proteins as a modi cation in the second channel.
The analysis has been previously described 45 . The non-normalized ratios (Heavy/13C6 over Light12C6) were used for further analysis using Perseus 46  Lysine treatment study.
Experimental animals received either vehicle (water) or L-Lysine (17 mg/ml) via drinking water (n = 6 per group). Blood pressure was measured by telemetry. D/SS rats were anesthetized with 2-3% (vol/vol) iso urane and a blood pressure transmitter (PA-C40; DSI) was surgically implanted subcutaneously, with the catheter tip secured in the abdominal aorta via the femoral artery. After a 3-day recovery period, blood pressure was measured with a DSI system ("telemetry") in conscious, freely moving SS rats under HS diet protocol, similar to those described previously [48][49][50] .
For urine collection, rats were placed in metabolic cages (no. 40615, Laboratory Products) for a 24 hrs urine collection. These urine samples were used to determine electrolytes, microalbumin, and creatinine. Whole blood and urine electrolytes and creatinine were measured with a blood gas and electrolyte analyzer (ABL system 800 Flex, Radiometer, Copenhagen, Denmark) 48 . Kidney function was determined by measuring albuminuria using a uorescent assay (Albumin Blue 580 dye, Molecular Probes, Eugene, OR) read by a uorescent plate reader (FL600, Bio-Tek, Winooski, VT).

Ketogenic diet study.
Male D/SS rats (n = 7 per group) were fed either control high salt diet (4% NaCl, # 113756, Dyets Inc.) or high salt ketogenic diet (Keto Diet 4% NaCl, TD.190564, Teklad Custom Diet) for 28 days. Food composition is depicted in supplementary Fig. 8A. Water and food were provided ad libitum. Blood pressure acquisition, urine and whole blood analyzes were performed as descried above in lysine treatment study. Pulsed 13 C 6 Lysine labeling in the SS rat.
Albumin uptake cell culture studies.
Confocal microscopy was used to detect uptake of uorescent albumin (AlexaFluor-647 albumin, 40 µg/ml) in con uent monolayers of OK proximal tubule epithelial cells 51 (visualized by uorescent Factin, 488 nm), grown on transwell lters under orbital shear stress as previously described and coincubated with lysine or glycine 52 . Cells were incubated for 1 hr in serum-free culture media containing uorescent albumin ), after pretreatment of 5-50 mM L-Lysine (or other amino acids) for 2 h, or after overnight treatment with 1-10 mM L-Lysine. Treatment was performed from the apical side unless otherwise indicated. Cell-associated albumin was quanti ed by spectro uorimetry.
All surgical, imaging, rat uorescent albumin, and image analysis procedures were performed as described previously 53,54 . Imaging was conducted using an Olympus FV1000 microscope adapted for an intravital two-photon microscopy with high-sensitivity gallium arsenide nondescanned 12-bit detectors.
Animals were anesthetized with pentobarbital sodium (50 mg/ml). A jugular venous line was used to introduce uorescent rat albumin (Texas Red labeled) and high-molecular weight dextran (150 kDa FITClabeled, TdB Consultancy, Uppsala, Sweden).

Immunohistochemistry.
Rat kidneys were xed in 10% formalin and processed for para n embedding as previously described 55 .
Bioinformatic image analysis via convolutional neural net analysis.
Rat kidneys were cleared of blood, formalin xed, para n embedded, sectioned, and mounted on slides as previously described 56 . Slides were stained with Masson's trichrome stain. The localization and scoring of glomeruli was performed by a novel and robust application of convolutional neural nets 23,57 . A cumulative distribution plot was generated (OriginPro 9.0) using glomerular injury scores based on a scale of 0-4 as previously described 58 , and the probability for a corresponding score interval was calculated (more than 3,500 glomeruli per group). Cortex protein cast analysis was performed using a color deconvolution lter and Analyze Particles mode in the Fiji image application (ImageJ 1.51u, NIH).
Bioinformatic algorithms for mass-difference based isotope selection.
Labeled isotopologues were detected using an in-house isotracker script (see Data Availability and github submission) that operated as follows: rst, we search for groups of features that, according to their m/z difference (< 10 ppm), could correspond to an isotopic envelope composed of a light isotope and heavy isotopes. Only features within 2 seconds of retention time difference were allowed to be grouped into a single isotopic envelope. Next, we compared the isotopic envelope between labeled and unlabeled samples. The envelopes presenting at least one heavy isotope with statistically signi cant higher abundance in labeled samples compared to unlabeled samples were retained for further analysis.
Correlation-based isotope selection approach.
Labeled isotopologues were detected using an in-house script (see Data Availability and github submission) that operated as follows: The custom R script is designed to identify isotopes from stable labeled isotope LC-MS data based on an input list of compounds and sum formulas and is available online including documentation via https://github.com/hpbenton/targeted_isotopes. It uses mzR and MSnBase from the bioconductor repository to open and manipulate the data. Once the raw data is opened each le is independently searched for a list of possible compound hits. These compounds are searched by creating a small bin of a user chosen ppm range around the mass. The vector of data undergoes a smoothing using a Savitzky Golay lter. Any compound that is above a given threshold (default 1000 counts) and is also above the chosen signal to noise is selected. Then, since the formula is known, any and all isotopes are searched within the same ppm range and at that retention time range. If the isotope peak also satis es the above criteria the two vectors are correlated to help con rm a true positive isotope. Most isotopes are correlated above 0.9 59 , Supplemental Fig. 1. The script used a list of 400 lysine metabolites derived from KEGG and METLIN as input.
For a detailed description of Nε-malonyl-Lysine and Nα-malonyl-lysine isomers, please see the supplementary material and methods. Molecules were synthesized as described in the supplementary materal and methods and characterized by NMR and mass spectrometry. Volatiles were removed under reduced pressure and the crude product was puri ed by mass directed preparative reversed phase HPLC to give the formic acid salt which was directly used as an analytical standard and fragmented at 10, 20 and 20-50 eV in ESI in positive ion mode.
Malonyl-CoA and lysine in vitro reaction.
1 uM lysine and a 10 uM of malonyl-CoA was incubated together in 10 ul of PBS, pH = 8 at 37 degrees for 1 h. Both the isotope-labeled ( 13 C 6 ) and non-isotope-labeled ( 12 C 6 ) form of lysine were used, in order to exclude unspeci c molecule products. Mixtures were analyzed on a QQQ as well as a QTOF machine.
Speci c transitions were used in order to detect both heavy and light forms of N-e-malonyl-lysine as well as lysine, and malonyl-CoA.
Malonyl-CoA levels from tissue lysates were determined using a commercial rat ELISA assay (MyBiosource.com) according to the manufacturer's manual. injected. For analysis of lysine metabolites, identical column and chromatography conditions as in the "untargeted metabolomics" section were used. In all cases, a standard curve was recorded and integrated using the mass hunter platform (Agilent). The method for the TCA cycle including transitions was previously published 22 . The transitions used for malonyl-Lysine were as follows: 233 -> 84; 233 -> 129.09, 233 -> 147.10 (for non-labeled malonyl-Lysine) and 239 -> 89, 239 -> 134, 239 -> 153 for 13C6 labeled malonyl-Lysine).

Study approval.
All studies using D/SS rats were conducted at Medical College of Wisconsin and protocols were approved by the MCW Animal Care and Use Committees and were performed in accordance with the standards set forth by the NIH Guide for the Care and Use of Laboratory Animals (National Academies Press, 2011). All mice studies were conducted at the MPI for Heart Lung and Blood Research Bad Nauheim as previously described in accordance to the local authorities 31,60,61 .

Statistical analysis.
Two-tailed t-tests were used for comparison unless otherwise indicated. Analysis of metabolomics and proteomics data is described in the respective sections.

Declarations
Data availability.
In house Scripts are available through Github (see isotope selection approaches, for correlation based approach https://github.com/hpbenton/targeted_isotopes), and for mass-difference approach https://github.com/xdomingoal/isoTracker. Metabolomics data will be made available through XCMS online. Proteomics data is available through the PRIDE/proteomExchange repository 62