Mitochondrial NADP+ is essential for proline biosynthesis during cell growth

Nicotinamide adenine dinucleotide phosphate (NADP+) is vital to produce NADPH, a principal supplier of reducing power for biosynthesis of macromolecules and protection against oxidative stress. NADPH exists in separate pools, in both the cytosol and mitochondria; however, the cellular functions of mitochondrial NADPH are incompletely described. Here, we find that decreasing mitochondrial NADP(H) levels through depletion of NAD kinase 2 (NADK2), an enzyme responsible for production of mitochondrial NADP+, renders cells uniquely proline auxotrophic. Cells with NADK2 deletion fail to synthesize proline, due to mitochondrial NADPH deficiency. We uncover the requirement of mitochondrial NADPH and NADK2 activity for the generation of the pyrroline-5-carboxylate metabolite intermediate as the bottleneck step in the proline biosynthesis pathway. Notably, after NADK2 deletion, proline is required to support nucleotide and protein synthesis, making proline essential for the growth and proliferation of NADK2-deficient cells. Thus, we highlight proline auxotrophy in mammalian cells and discover that mitochondrial NADPH is essential to enable proline biosynthesis. NADPH exists in separate cellular pools within the cytosol and mitochondria. Tran et al. find that mitochondrial NADPH is essential to enable proline biosynthesis during cell growth.


NADK2 deficiency renders cells proline auxotrophic.
To understand the influence of mitochondrial NADP(H) on cellular metabolism, we used CRISPR-Cas9 to generate NADK2 deficiency (ΔNADK2) in six different cancer cell types: human embryonic kidney 293E (HEK293E) cells, cervical cancer cells (HeLa), myelogenous leukaemia cells (K562), non-small cell lung cancer cells (A549), breast cancer cells (T47D) and melanoma cells (A375). NADK2 deficiency reduced cell proliferation in all six cell lines (Fig. 1a) and decreased tumour growth in A549 and K562 xenograft mouse models (Fig. 1b-d and Extended Data Fig. 1a). This was in striking contrast to NADK deficiency, which had only marginal effects on proliferation (Extended Data Fig. 1b). Reconstitution of ΔNADK2 HEK293E cells with wild-type NADK2 cells completely restored proliferation (Extended Data Fig. 1c). NADK2 loss did not induce a compensatory increase in NADK expression, and overexpressing NADK did not compensate for loss of NADK2 (Extended Data Fig. 1c), indicating that NADK and NADK2 serve nonredundant functions in cells. To determine the cause of the decreased cell proliferation of ΔNADK2 cells, we attempted to rescue proliferation with metabolic intermediates thought to depend on NADPH availability. Treatment with the ROS scavenger N-acetylcysteine (NAC), nucleosides or fatty acids did not restore proliferation of ΔNADK2 cells ( Fig. 1e and Extended Data Fig. 1d,e), nor did supplementation with supra-physiological levels of pyruvate or aspartate, which support mitochondrial respiration and de novo nucleotide synthesis, respectively (Fig. 1e,f and Extended Data Fig. 1e,f). Surprisingly, supplementation with a mixture of nonessential amino acids (NEAAs) completely restored NADK2-deficient cell proliferation and even surpassed the growth of NADK2-expressing cells ( Fig. 1e and Extended Data Fig. 1e).
To identity the amino acids responsible for this rescue, we treated cells with seven individual NEAAs either at their physiological concentrations present in human plasma-like medium (HPLM 20 ; Fig. 1f and Extended Data Fig. 1f) or at ten times higher concentrations (Extended Data Fig. 1g). Strikingly, only proline rescued the proliferation of ΔNADK2 cells, and did so at both concentrations ( Fig. 1f and Extended Data Fig. 1f,g). Remarkably, proline rescued cell proliferation to the same extent as reconstitution with wild-type NADK2 in all the cell lines tested (Fig. 1f,g and Extended Data Fig. 1f-h). Similarly to monolayer growth conditions, we found that proline supplementation also supports the growth of NADK2deficient cells grown under anchorage-independent conditions (Fig. 2a-c). Of note, while proline supplementation fully rescued the growth defect of ΔNADK2 spheroids in HEK293E and T47D cells, we observed a significant, but incomplete rescue of that from K562 cells, indicating that NADK2 in some contexts is needed to support other processes besides proline synthesis. Together, these findings indicate a general requirement of proline for ΔNADK2 cells.
The ability of proline to restore cell proliferation is surprising because ΔNADK2 cells cultured in medium supplemented with proline still displayed a drastic decrease in mitochondrial NADP + and NADPH synthesis (Extended Data Fig. 1i), measured using rapid mitochondrial purification (MITO-Tag) 21 coupled with stable isotope tracing with [ 13 C 3 -15 N]nicotinamide 16 . These data suggest that proline can circumvent the requirement of mitochondrial NADP(H) for cell growth.
To gain insight into the metabolic connection between NADK2 and proline metabolism, we performed unbiased metabolomic profiling in ΔNADK2 HEK293E cells stably expressing either empty vector or NADK2. Interestingly, proline was the most significantly depleted metabolite after NADK2 deletion (Fig. 3a,b), with proline concentrations dropping below 90% in NADK2-deficient cells within 24 h in proline-free medium (Fig. 3c,d). Moreover, a reduction in proline levels, but not in many other amino acids, was also observed in NADK2-deficient A549 and K562 xenograft tumours, indicating the importance of NADK2 to maintain proline levels in tumours (Extended Data Fig. 2a,b).
Proline is predominantly synthesized in the pancreas. Proline can be synthesized in both the mitochondria and the cytosol through a series of reactions that are dependent on NADP(H) or NAD(H) (Fig. 4a) 22 . Mitochondrial proline synthesis involves the NADPH-dependent formation of pyrroline-5-carboxylate (P5C) from glutamate. P5C can be further reduced to proline by NAD(P) H-dependent P5C reductase enzymes, through either PYCR1 or PYCR2 isoforms in mitochondria or the PYCRL isoform in the cytosol 22 . Intravenous infusions with 13 C 5 -labelled glutamine and liquid chromatography with tandem mass spectrometry (LC-MS/ MS) analysis in mice revealed highest enrichment of the newly synthesized [ 13 C 5 ]proline (M + 5) in the pancreas, with most of the other tissues displaying fractional enrichment similar to that in the blood (Fig. 4a,b), suggesting that most tissues have the ability to obtain proline from blood in addition to (or instead of) synthesizing it (Fig. 4a,b). Consistent with this, infusions with [ 13 C 5 ]proline (M + 5) in mice showed 43% enrichment of proline (M + 5) in blood, 10-20% enrichment in most tissues and the lowest enrichment in brain (3%; Fig. 4c). The [ 13 C 5 ]glutamate enrichment was very low in all tissues (0.5% or less), indicating little conversion of proline to glutamate under these conditions (Fig. 4c). Overall, our data suggest that many tissues obtain proline from the blood, while some, such as the pancreas, synthesize it from glutamine ( Fig. 4a-c).

NADK2 activity is required for de novo proline synthesis.
To determine whether loss of NADK2 affects proline biosynthesis, we used [ 13 C 5 ]glutamine tracing (M + 5) to measure the newly synthesized [ 13 C 5 ]proline (M + 5) generated from P5C synthase (P5CS) and P5C reductase (PYCR) in the context of NADK2 loss ( Fig. 5a and Extended Data Fig. 3a). Strikingly, knocking out NADK2 in five cancer cell lines nearly eliminated 13 C transfer from glutamine to proline, and this effect was completely reversed by expression of NADK2 (Fig. 5b,c and Extended Data Fig. 3b-e). Moreover, shRNA-mediated depletion of NADK2 or P5CS showed comparable levels of reduced proline synthesis (Extended Data Fig. 3f-h). These changes occurred without any visible alteration in the abundance of proteins in the proline biosynthesis pathway (Extended Data Fig. 4a,b). Importantly, fibroblasts derived from an individual with NADK2 deficiency due to a mutation in the start codon (p.Met1Val) 18 were also devoid of proline synthesis in this assay ( Fig. 5d and Extended Data Fig. 4c).
To determine whether NADK2 catalytic activity is required for the observed effect, we mutated a conserved catalytic residue on human NADK2 (D161A) that results in loss of its activity 23 and measured the synthesis of proline from glutamine ( Fig. 5e and Extended Data Fig. 4d,e). Consistent with previous results, expression of wild-type NADK2 restored proline (M + 5) levels to values similar to those in the wild-type cells, whereas cells expressing a catalytically dead NADK2 (D161A) mutant failed to synthesize proline (M + 5), indicating that the catalytic activity of NADK2 is essential for proline biosynthesis. To determine whether NADK2-dependent Fig. 1 | loss of NADK2 renders cells dependent on proline for cell proliferation. a, Cell proliferation and immunoblots of wild-type (WT) or single-cell-derived knockout cells of NADK2 (ΔNADK2) in HEK293E, HeLa, K562, A549, T47D and A375 grown in DMEM with 10% serum. Relative proliferation rate was assessed 72 h after plating and normalized to day 0. n = 3 biologically independent replicates for HEK293E, HeLa, K562 and A549 and n = 5 for T47D and A375. b, Experimental design of the tumour study. c, A549 ΔNADK2 cells stably reconstituted with either empty vector (Vec) or NADK2 were injected subcutaneously into athymic nude mice (n = 7 biologically independent animals). Tumour growth was monitored after tumour onset over time. Data are presented as the mean ± s.e.m. *P < 0.05 for comparisons were calculated using a one-sided Student's t-test. d, K562 ΔNADK2 cells stably reconstituted with either empty vector or NADK2 were injected subcutaneously into athymic nude mice (n = 3 or 4 biologically independent animals). Tumour growth was monitored as in c. Data are presented as the mean ± s.e.m. *P < 0.05 for comparisons were calculated using a one-sided Student's t-test. e, Relative proliferation rate as in a from WT or ΔNADK2 HEK293E cells. Cells were grown in 10% dialysed serum for 72 h or supplemented with NAC (5 mM), nucleosides (inosine, 0.1 mg ml −1 ; uridine, 0.1 mg ml −1 ), NEAA mixture (1×), pyruvate (10 mM) or aspartate (10 mM). n = 3-9 biologically independent samples. f, Relative proliferation rate as in e from isogenic ΔNADK2 HEK293E cells stably expressing either empty vector or NADK2 and supplemented with the indicated NEAAs at their concentrations in HPLM. n = 3 biologically independent samples. g, Relative proliferation rate in the presence or absence of proline (0.2 mM) was assessed in HEK293E, HeLa and K562 cells in three consecutive days. n = 3 biologically independent samples. Data are presented as the mean ± standard deviation (s.d.) and are representative of at least two independent experiments (a and e-g). ***P < 0.001 for comparisons were calculated using a two-sided Student's t-test (a) and one-way analysis of variance (ANOVA) test and Tukey's post hoc test (e-g).
supply of reducing power NADP(H) is the major contributor to proline synthesis, we tested the role of nicotinamide nucleotide transhydrogenase (NNT), a transmembrane enzyme that transfers electrons from NADH to NADP + and has been implicated in maintaining a reduced NADPH pool in mitochondria 24 . We generated double knockouts of NADK2 and NNT in HEK293E cells        and measured proline synthesis from [ 13 C 5 ]glutamine after expressing either NNT or NADK2 individually or together ( Fig. 5f and Extended Data Fig. 4f). NADK2 expression restored proline synthesis, while NNT failed to do so. These data indicate the NADPH production from NNT does not compensate for loss of NADK2, and imply that NADK2 is the primary source of mitochondrial NADP(H) for proline biosynthesis ( Fig. 5f and Extended Data Fig. 4f).

Mitochondrial NADP(H) fuels pyrroline-5-carboxylate production.
To better understand which step in proline synthesis requires mitochondrial NADP(H), we also considered an alternative pathway that produces P5C from ornithine via ornithine aminotransferase (OAT; Fig. 5a). Supplementation of ΔNADK2 cells with OAT substrates (l-ornithine and a cell permeable α-ketoglutarate (α-KG)) restored cell proliferation to levels similar to those achieved by proline supplementation (Fig. 5g and Extended Data Fig. 4g). The ornithine-mediated rescue in cell proliferation of NADK2-deficient cells was not due polyamines, small polycations synthesized from ornithine that promote cell proliferation 25 , as supplementation of cells with a mixture of polyamines failed to restore cell proliferation in ΔNADK2 cells (Fig. 5g). Importantly, we found that l-ornithine was sufficient to restore proline levels ( Fig. 5h) and flux through proline biosynthesis via OAT ( Fig. 5i and Extended Data Fig. 4h), suggesting that the generation of P5C through P5CS is the bottleneck step that requires mitochondrial NADPH and NADK2 activity. Together, these data indicate that NADK2 depletion results in proline auxotrophy, making proline essential for the growth and proliferation of NADK2-deficient cells ( Proline does not alter the NADK2-dependent redox or bioenergetics. We next explored which proline-dependent pathways were most related to the loss of cell proliferation in NADK2-deficient cells. We considered whether PRODH-mediated proline oxidation, which has been implicated in ATP production, is involved in rescuing cell proliferation 26 . Inhibition of PRODH1/2 using a PRODH inhibitor did not affect the ability of proline to rescue cell proliferation in ΔNADK2 HeLa cells (Extended Data Fig. 5a), nor did it significantly affect the growth of NADK2-expressing cells (Extended Data Fig. 5b). These findings indicate that restoration of NADK2-deficient cell proliferation by proline is independent of proline oxidation by PRODH1/2.
In addition to its role in protein synthesis, proline has been implicated in the control of redox homeostasis and regulation of cellular bioenergetics (Extended Data Fig. 5c) 22 . To determine the direct effect of proline in redox and energy metabolism, we assessed several metabolic parameters in the context of NADK2 loss. Although depletion of mitochondrial NADPH after NADK2 loss increases intracellular ROS levels, exogenous proline supplementation did not influence ROS levels (Extended Data Fig. 5d,e). Notably, proline did not produce any discernible changes in reduced GSH, oxidized glutathione (GSSG) or in the GSH/GSSG ratio, which act in concert with NADPH to maintain the cellular redox state (Extended Data Fig. 5f,g) 27 . Although NADK2 loss reduced mitochondrial respiration as judged by decreased oxygen consumption rate (OCR) in HeLa, HEK293E, K562 and A549 cells (Fig. 6a-d), no substantial changes were observed in extracellular acidification rate (ECAR; Extended Data Fig. 5h,i) or 13 C 5 -labelling of tricarboxylic acid (TCA) cycle intermediates from [ 13 C 5 ]glutamine (Fig. 6e,f). Moreover, supplementing with proline at levels that restored proliferation did not alter these parameters, indicating a decoupling between proline-dependent cell proliferation and bioenergetics ( Fig. 6a-f). Together, these findings suggest that proline is not limiting for redox or bioenergetics under monolayer cell growth conditions. Although NADK2 appears to contribute to redox homeostasis, its requirement for cell proliferation is related to proline biosynthesis.

Proline supports nucleotide synthesis in NADK2-deficient cells.
Analysis of the cell cycle from G1-synchronized cells revealed a general delay in cell cycle progression of NADK2-deficient cells, which was restored by supplementation with proline or NADK2 expression (Extended Data Fig. 6a). Additionally, asynchronous NADK2-deficient cells showed an S-and G2/M-phase arrest that could also be rescued by supplementation with proline or NADK2 (Extended Data Fig. 6b). We hypothesized that the cell cycle regulation impairment was likely due to a decrease in cellular nucleotide or protein abundance. To test whether nucleotide synthesis is affected by loss of NADK2, we performed [ 13 C 6 ]glucose tracing analysis via LC-MS/MS to label nucleotides (Extended Data Fig. 7a). Remarkably, we found that the fractional abundance of newly synthesized purines (AMP (M + 5) and GMP (M + 5)) and pyrimidines (CMP (M + 5) and CDP (M + 5)) was decreased in ΔNADK2 cell lines and was restored by expressing NADK2 or simply by supplementing with proline ( Fig. 7a,b and Extended Data Fig. 7a). We did not see NADK2-dependent changes in intermediates of the one-carbon metabolism that support nucleotide synthesis following labelling with [3-13 C]serine 28 , which can be traced into formyl units, and then into nucleotides, such as deoxythymidine monophosphate (Extended Data Fig. 7b,c) 6,7 .
Interestingly, exogenous proline supplementation did not alter the synthesis of nucleotides in NADK2-expressing cells, indicating that proline is limiting for nucleotide synthesis in ΔNADK2 cells. To validate these results, we also used a surrogate assay to specifically measure the de novo purine and pyrimidine synthesis rates via [ 14 C]glycine or [ 14 C]aspartate incorporation into RNA, respectively 29,30 . Consistent with the [ 13 C 6 ]glucose tracing analysis, loss of NADK2 also displayed a decrease in 14 C incorporation into RNA, which was completely rescued by addition of proline (Fig. 7c,d  was observed, suggesting that the nucleotide salvage pathways and RNA synthesis were not affected (Extended Data Fig. 7e).
To gain insight into regulation of nucleotide synthesis by proline availability, we profiled the protein levels of genes that contribute to the nucleotide synthesis in HeLa, K562 and HEK293E cells (Fig. 7e and Extended Data Fig. 7f). Interestingly, we found that multiple enzymes in the de novo and salvage nucleotide synthesis pathway were under-expressed in NADK2-deficient, proline-depleted cells and their levels could be restored by proline supplementation or NADK2 expression (Fig. 7e). The most prominent proline-dependent changes across cell lines were observed in the levels of PRPS1 and PRPS2, the enzymes responsible for generating activated sugar     (PRPP) for nucleotide synthesis; APRT and HPRT, purine salvage enzymes; and de novo pyrimidine synthesis enzyme CAD (Fig. 7e). Akin to protein levels, we also observed proline-dependent changes at the transcript levels of these genes (Extended Data Fig. 8). Other enzymes that feed into nucleotide pathway, including the pentose phosphate pathway enzymes TKT, TALDO or PGD, or the purine synthesis enzyme GART, were not altered at the protein or transcript levels in a proline-or NADK2-dependent manner (Extended Data Figs. 7f and 8). These data indicate that proline levels regulate the expression of a subset of key nucleotide synthesis genes, and further work is required to fully elucidate the underlying mechanisms of the proline-mediated gene signature.
Proline deficiency is sensed through the GCN2-ATF4 pathway. To determine whether the NADK2-dependent control of proline synthesis leads to altered protein synthesis rate, we measured the kinetics of expression of mScarlet red fluorescent protein (RFP) and observed that proline was able to increase protein synthesis in the NADK2-knockout cells to a level similar to that of NADK2-expressing cells (Fig. 8a). In addition to regulating nucleotide and protein synthesis, reduced proline levels also triggered a strong response in the GCN2-eIF2α-ATF4 pathway (Fig. 8b), which senses depletion of amino acids and the uncharged tRNAs 31 . Two hours of proline removal was sufficient to induce activation of this pathway in HeLa cells, as measured by autophosphorylation of GCN2 (Thr 899), phosphorylation of eIF2α (Ser 51) and increased ATF4 protein levels (Fig. 8b). Activation of the GCN2-eIF2α-ATF4 pathway was sustained for 12 h after proline removal, but the nutrient-sensing mTORC1 pathway did not respond during this time (Fig. 8b), in contrast to its rapid inactivation in amino acid-free medium (Fig. 8c) 32 . Nevertheless, prolonged proline removal (48 h) from NADK2-deficient HeLa and K562 cells induced a decrease in mTORC1 signalling, as well as an activation of the GCN2 pathway, indicating that proline levels are primarily sensed by the GCN2 pathway (Fig. 8d).

Discussion
Here we reveal that NADK2 and mitochondrial NADPH pools are essential to enable proline biosynthesis for cell growth and proliferation (Fig. 8e). While mitochondrial NADPH is a cofactor for many metabolic reactions 2,7,9,33,34 , including ROS mitigation, IDH2 function, lysine catabolism, fatty acid oxidation and proline metabolism, we discovered that only proline can circumvent the requirement of mitochondrial NADPH for cell growth and proliferation. This finding elucidates an unexpected hierarchy in the NADPH-dependent functions of mitochondria in proliferating cells.
Analogously to studies demonstrating that the main role of the electron transport chain in proliferating cells is to enable aspartate synthesis 35,36 , we find mitochondrial NADP(H) is vital to generate proline. We demonstrate that NADK2-deficient cells are proline auxotrophs and entirely unable to synthesize proline from glutamine in cancer cell lines. While proline auxotrophy was not previously reported in mammalian cells, Saccharomyces cerevisiae mutants defective in the proline synthesis genes, PRO1, PRO2 and PRO3, the mammalian homologues P5CS and PYCR, exhibit proline auxotrophy 37 .
We found that NADK2 activity is vital for P5CS function, the NADPH-dependent step of the proline synthesis pathway. Our data suggest that only the P5CS enzyme, but not PYCR in the proline synthesis pathway, depends exclusively on mitochondrial NADPH availability, as supplementing NADK2-deficient cells with ornithine, which bypasses the P5CS step but still requires PYCR to generate proline, restores proline production and cell proliferation. This finding is also in agreement with previous reports suggesting that mitochondrial PYCR preferentially uses NADH when assayed in vitro 38 .
Our infusion experiments with 13 C-labelled glutamine (M + 5) and proline (M + 5) indicated that many tissues can obtain proline from circulation, in addition to synthesizing it from glutamine. While most of the organs did not display high levels of glutamine-derived proline, interestingly, the pancreas showed the highest enrichment of the newly synthesized proline. Since pancreatic acinar cells synthesize and secrete more protein than any other cell type in the body [39][40][41] , the high rate of proline synthesis in the pancreas could be required to support the high demand for protein synthesis. Our findings should further stimulate the investigation of whole-body proline metabolism at a tissue-specific level, as individuals with NADK2 deficiency are reported to have normal or even excessive levels of proline in the blood, despite clear defects in proline biosynthesis in patient-derived fibroblasts.
Even though most tissues can acquire proline from the circulation, NADK2 deficiency decreased tumour growth and reduced proline levels by at least 50% in tumours. These data suggest that circulatory proline does not fully compensate for NADK2-dependent proline synthesis and that tumours employ cell-intrinsic biosynthesis of proline. Future strategies targeting NADK2 and simultaneously restricting proline levels in the diet could be used to further investigate the importance of the NADK2-proline axis in tumorigenesis of various cancer types.
While we find that NADK2 loss and proline deficiency limits nucleotide synthesis, at least in part, likely due to a decreased expression of several nucleotide synthesis genes, including PRPS1/2, HPRT, APRT and CAD, we cannot exclude the possibility that other mechanisms could be involved in controlling nucleotide synthesis in a proline-or NADK2-dependent manner. Additionally, it would be important to determine the relevance of proline sensing in NADK2-deficient tumours by the GCN2-eIF2α-ATF4 pathway, given the critical role of the GCN2 pathway in mediating metabolic adaptation to nutrient stress and aiding cancer cell survival 42,43 While we show that mitochondrial NADP(H) and NADK2 primarily control cell growth and proliferation via enabling proline biosynthesis and anabolic metabolism under cell culture conditions, this does not rule out other contexts, including complex physiological and pathological settings in vivo, in which mitochondrial NADP(H) is required for redox homeostasis and other functions beyond proline synthesis.
Taken together, our data indicate that NADK2 and mitochondrial NADPH pools are vital for generating proline levels required for both nucleotide and protein synthesis to restore cell proliferation. Thus, we have identified an example of mammalian proline auxotrophy that is directly dependent on the levels of mitochondrial NADP(H).

Methods
Cell culture conditions. All cell lines were maintained in DMEM (Corning/ Cellgro, 10-017-CV) containing 10% FBS at 37 °C and 5% CO 2 . Cancer cell lines, including HeLa, K562, T47D, A549 and A375 were obtained from the American Type Culture Collection. HEK293E cells were a gift from C. Mackintosh. Primary fibroblasts from NADK2-deficient individuals have been described previously 18 and were provided by the laboratory of J.A.P. III. NADK2-knockout cells were maintained in 10% FBS in the presence of 0.2 mM proline. For cell proliferation experiments involving supplementation of medium with various metabolic intermediates, cells were plated in 10% dialysed serum.
Xenograft experiments. All animal procedures were performed in accordance with ethical regulations and guidelines approved by the Institutional Animal Care and Use Committee at University of Texas Southwestern Medical Center (protocol no. 2020-102880). A549 and K562 cells were deleted for NADK2 by CRISPR-Cas9 and stably reconstituted with constructs expressing either empty vector or NADK2 WT. Male athymic nude mice aged 6 weeks old were subcutaneously injected with NADK2-deficient or NADK2-expressing cells (5 × 10 6 for A549 cells and 4 × 10 6 for K562 cells). Tumour growth rates were monitored weekly. Mice were euthanized once the tumour size reached around 2 cm 3 . 13 C nutrient infusions in mice. All procedures were approved by the Institutional Animal Care and Use Committee at University of Texas Southwestern Medical Center (protocol no. 2017-101840). All mice were housed in a pathogen-free environment with a 12:12 light/dark cycle and fed a chow diet (Envigo, Teklad Global 16% Protein Rodent Diet) ad libitum. All infusions took place between 09:00 and 11:00 with no prior fasting of the mice. Healthy, 6-to 10-week-old mice (C57BL/Ka) were initially anaesthetized using ketamine and xylazine (75 mg kg −1 and 10 mg kg −1 , intraperitoneally (i.p.)) and maintained under anaesthesia using subsequent doses of ketamine (20 mg kg −1 , i.p.) alone as needed. A 25-gauge catheter was inserted into the tail vein, and isotope infusions began immediately after a retro-orbital blood draw to mark time zero.
For [ 13 C 5 ]glutamine infusions, a total dose of 2.5 g kg −1 was dissolved in 1,500 μl normal saline, supplemented with 50 U ml −1 heparin, and administered with a bolus of 150 µl min −1 for 1 min followed by an infusion rate of 2.5 μl min −1 for 5 h, as previously reported 44,45 . For proline infusions, mice were infused with a stock of 20 mM [ 13 C 5 ]proline first with a bolus of 125 µl min −1 for 1 min followed by an infusion rate of 0.1 µl min −1 g −1 for 3 h. Retro-orbital blood draws were taken throughout the infusion to monitor tracer enrichment in blood. After cessation of the infusion, tissue samples were taken and frozen in liquid nitrogen, and later extracted in 80% methanol for extraction. Blood samples obtained during the infusion were chilled on ice for 5-10 min and then flash frozen in liquid nitrogen. Aliquots of 10-20 μl were added to 80% methanol for extraction. To generate NADK2-knockout cells in different cells lines, NADK2 CRISPR-Cas9 knockout plasmids containing a GFP marker and derived from the GeCKO (v2) library were obtained from Santa Cruz (sc-414406). NNT sgRNA sequence was cloned into the GFP-expressing PX458 CRISPR vector (Addgene, 48138) using the following guides: sense: GTGTCCTCTACTTAGCAATT; antisense: AATTGCTAAGTAGAGGACAC. Two days after transfection of the CRISPR-Cas9 knockout plasmids, GFP-positive cells were subjected to single-cell sorting into 96-well plates via flow cytometry with BD FACS Aria Fusion. Clones were grown in DMEM containing 30% FBS and supplemented with 1× NEAA mixture (Gibco, 11140050), 1 mM pyruvate, 10 mM HEPES and 55 µM β-mercaptoethanol (Thermo Fisher Scientific, 21985023). Clonal cells were screened by immunoblotting with NADK2 or NNT antibody. NADK knockouts in HEK293E cells were described previously 16 .
Transient expressions were performed with polyethylenimine or Lipofectamine 3000 Transfection Reagent transfection methods using 2 µg of plasmid DNA per well of a six-well plate. Experiments were initiated 48 h after transfection.
Immunoblotting. Protein extracts were prepared according to previously described methods 16 . Protein concentrations were determined using the Bradford assay. Protein lysates (20 µg) were loaded on 4-15% Criterion TGX Precast gradient gels and subjected to immunoblotting with the indicated antibodies. Antibody dilutions were performed as follows: PYCR1, PYCR2 and PYCRL antibodies were used at 1:2,000; PRODH1 antibody was diluted at 1:200; HPRT antibody was diluted at 1:200; β-actin antibody was diluted at 1:10,000; and all remaining antibodies were diluted at 1:1,000 in 5% BSA in TBST buffer. Secondary antibodies were used at a 1:5,000 dilution in 5% milk in TBST. Proteins were detected with enhanced chemiluminescence. For gel source data, see Supplementary Fig. 1. Immunoblotting images were assembled in Adobe Fireworks 8. mRNA expression analysis. Total RNA was extracted from HeLa and K562 cells cultured in DMEM supplemented in 10% FBS in the presence or absence of proline (0.2 mM) for 48 h by RNeasy Plus Mini Kit (QIAGEN, 74136). RNA (1 µg) was subjected to reverse transcription with EcoDry Premix (Takara, 639545). The resulting cDNA was diluted in nuclease-free water (1:4) and amplified using Bio-Rad SsoAdvanced Universal SYBR (Bio-Rad, 1725274) and CFX384 Touch Real-Time PCR Detection System (Bio-Rad). Data are shown as the ratio between the gene of interest and the control gene (RPLP0) as previously described 29 . The quantitative PCR data were analysed with a Bio-Rad CFX Manager version 3.1.1517.0823. Primer sequences are shown in Supplementary Table 3.
Rapid isolation of mitochondria for measurement of NADP + and NADPH by LC-MS/MS. ΔNADK2-HA-Mito HEK293E cells were generated by stably expressing 3XHA-EGFP-OMP25 (HA-Mito) in the ΔNADK2 HEK293E cells as previously described 46 . ΔNADK2-HA-Mito cells were then reconstituted by lentivirus delivery of empty vector or NADK2, both in the Lenti-III-PGK backbone plasmid. As the negative control, HEK293E cells expressing 3XMyc-EGFP-OMP25 (Control-Mito) were used.
Cells from a confluent 15-cm plate were labelled for 3 h with [ 13 C 3 -15 N] nicotinamide and HA-tagged mitochondria were isolated as previously described 46 from an equal amount of protein for each condition and metabolites. Metabolites were extracted in 50 µl of 80% methanol and analysed by targeted LC-MS/MS without drying with a QExactive HF-X hybrid quadrupole orbitrap high-resolution mass spectrometer (HRMS; Thermo Fisher Scientific) coupled to a Vanquish ultra-high-pressure liquid chromatograph (UHPLC) system. Cell proliferation assays. Crystal violet assay was used to determine cell proliferation. Briefly, cells were rinsed with PBS and stained with 0.1% crystal violet solution in 10% ethanol (Sigma, C6158) for 20 min at room temperature. After staining, cells were rinsed three times with PBS and air dried for 30 min, before extraction with 10% acetic acid. Absorbance was measured at 590 nm by SpectraMax iD3 Multimode Microplate Detection Platform (Molecular Devices). Cells were plated in 24-well plates in DMEM supplemented with 10% dialysed FBS or as indicated in the figure legends as follows: 4 × 10 4 HeLa, 5 × 10 4 HEK293E, 1 × 10 5 K562, 4 × 10 4 A375, 2.5 × 10 4 A549 or 3 × 10 5 T47D cells. For HEK239E and K562 cells, 24-well plates were first coated with 0.01% poly-l-lysine (Sigma, P4707). Seeding density of untreated adhered cells was also measured 12 h after plating and used to calculate the relative proliferation rate. For treatment with amino acids, the indicated concentrations shown in Supplementary Table 1 were used.
Tumour spheroid growth assay. Spheroid growth assay was used to determine the anchorage-independent cell growth and adapted from a previous report 47 . Briefly, white, clear-bottom 96-well plates (Corning, 3903) were coated with 50 µl of 1.5% low-melting agarose (Lonza, 50101) and allowed to dry. Isogenic NADK2-deficient or NADK2-expressing HEK293E, T47D and K562 cells (1,000 cells per well) were seeded in 150 µl of DMEM supplemented with 10% dialysed FBS, with or without proline (0.2 mM). The plates were centrifuged at 2,000 r.p.m. for 10 min. Spheroid images were taken in a Zeiss Primovert microscope with an AxioCam 208 colour camera at ×4 magnification, and image size was analysed with ImageJ software version Java 1.8.0_172 (National Institutes of Health). Spheroid cell growth was monitored and recorded within 7 d after plating and as indicated in the figure legends.
Cell cycle analysis. Cells were synchronized in G1 via a double thymidine block method, as previously described 48 . Briefly, isogenic NADK2-deficient or NADK2-expressing HeLa cells were plated in DMEM supplemented with 10% serum in the presence or absence of proline as indicated in the figure legends. First, cells were treated with 3 mM thymidine (Sigma, T9520) for 24 h, followed by release from thymidine for 12 h in fresh 10% serum containing DMEM and another 24-h treatment with 3 mM thymidine. Cells were released from the G1 thymidine block and collected every 2 h for a 10-h period. Cells were collected and immediately fixed with ice-cold ethanol (70%). Cells were subjected to staining solution (0.5 ml) containing 0.1% Triton X-100, 0.2 mg ml −1 RNase A (Sigma, R4875) and 20 µg ml −1 propidium iodide (BioLegend, 421301) in PBS. Stained cells were analysed using a BD LSRFortessa flow cytometer (BD Biosciences), and the collected data were analysed using FlowJo (BD Biosciences, version 10.7.1 for Windows).
Protein synthesis rate measurements. Isogenic NADK2-deficient or NADK2expressing HeLa cells were plated in six-well plates in the presence and absence of 0.2 mM proline. Next, 2 µg of mScarlet RFP was transfected with Lipofectamine 3000 reagent. Cells were collected at 0, 2, 4, 6, 8, 10 and 12 h after transfection, fixed with 4% paraformaldehyde and analysed by flow cytometer (BD Biosciences LSRFortessa). Unstained cells were used as the negative control, and the percentage of RFP-positive cells was calculated from biological duplicates and representative of two independent experiments. Intracellular reactive oxygen species measurement by CellROX green. HeLa and HEK293E cells were cultured in six-well plates in DMEM containing 10% dialysed FBS and supplemented with or without l-proline as indicated in the figures. After 48 h, cells were treated for 30 min with 5 µM CellROX green reagent (Invitrogen, C10444) in FluoroBrite DMEM containing 10% dialysed FBS (Thermo Fisher Scientific, A1896701). Cells were washed with PBS twice, collected, and live cells were analysed immediately by flow cytometry (BD Biosciences LSRFortessa). Results were analysed using FlowJo (BD Biosciences, version 10.6.1 for Windows). Unstained cells were used as the negative control. Data are presented from biological triplicates and are representative of at least two independent experiments.

Quantification of intracellular concentrations of proline and glutamate.
HEK293E and K562 cells were cultured in six-well plates in DMEM containing 10% FBS and 0.2 mM proline. After 48 h, cells were incubated in proline-free DMEM containing 10% dialysed serum for 24 h or collected at time 0 h. Metabolites were extracted on ice with 500 µl of extraction buffer containing 80% methanol (−80 °C) supplemented with isotope-labelled internal standards (125 ng [2,3,3,4,4,5,5-D7] Oxygen consumption rate and extracellular acidification rate. OCR and ECAR were measured by an XFe96 Analyzer (Seahorse BioScience). Isogenic NADK2-deficient and NADK2-expressing HeLa, HEK293E and A549 cells were cultured overnight in Seahorse XF96 cell culture microplates (Agilent, V3-PS). K562 cells were seeded on Cell-Tak-coated plates and analysed on the same day. Before measurement, cells were rinsed three times with DMEM (Sigma, D5030) supplemented with 2 mM l-glutamine, 1 mM pyruvate and 10 mM glucose (pH 7.4), and incubated for 30 min at 37 °C in an CO 2 -free incubator. Oligomycin, CCCP and antimycin A were injected for a final concentration of 2 µM, 1 µM and 1 µM, respectively. The assay programme was set up in Wave software (Agilent Technologies, version 2.6.1 for Windows), programmed for four cycles of 2-min mixing followed by 3 min of measurements. For normalization, cells were fixed with 10% neutral buffered formalin (Thermo Fisher Scientific, 5701) and stained with crystal violet solution (Sigma, C6158). OCR values were normalized to optical density from the crystal violet assays.

[U-14 C]glycine, [U-14 C]aspartate and [ 3 H]uridine incorporation into RNA.
Incorporation of 14 C and 3 H radiolabel into RNA was performed as previously described 29 . Cells were seeded in six-well plates in DMEM containing 10% serum and supplemented with or without l-proline (0.2 mM). Two days after plating, cells were labelled for 6 h with 1 μCi of uniformly labelled 14 C (U-14 C) glycine, [U-14 C] aspartate or [U-3 H]uridine in 10% dialysed serum in the presence or absence of proline. RNA was isolated using RNeasy Plus Mini Kit (QIAGEN, 74136) according to the manufacturer's protocol and eluted with 50 μl of water. Radioactivity was measured by liquid scintillation counting from 35 μl of eluted RNA using a BD, Beckman Coulter LS 6500 Liquid Scintillation Counter. Data are representative of biological triplicates from at least two independent experiments. To achieve comparable seeding density between NADK2-deficient and NADK2-expressing cells, the following cell numbers were seeded: HeLa and HEK293E cells: 6 × 10 5 for ΔNADK2 without proline, 4 × 10 5 for ΔNADK2 with proline, and 4 × 10 5 cells for isogenic ΔNADK2 expressing NADK2 in the presence or absence of proline; K562 cells: 1 × 10 6 for ΔNADK2 without proline, 7 × 10 5 for ΔNADK2 with proline and 7 × 10 5 cells for isogenic ΔNADK2 expressing NADK2 in the presence or absence of proline.
Rapid isolation of mitochondria from measurement of NADP + and NADPH by LC-MS/MS. ΔNADK2 HEK293E cells stably expressing 3XTag-EGFP-OMP25 (HA-Mito) were stably reconstituted with either empty vector or NADK2. HEK293E cells expressing 3XMyc-EGFP-OMP25 (Myc-Mito) were used as the negative control. Cells were labelled for 3 h with [ 13 C 3 -15 N]nicotinamide, and HA-tagged mitochondria were isolated as previously described 46,49 from an equal amount of protein for each condition and metabolites. Metabolites were extracted in 50 µl of 80% methanol and analysed by targeted LC-MS/MS without drying with a QExactive HF-X hybrid quadrupole orbitrap HRMS (Thermo Fisher Scientific) coupled to a Vanquish UHPLC system.
Targeted metabolic flux analysis. For metabolite analysis by LC-MS/MS, cells were seeded in six-well plates and experiments were performed at ~90% confluency, with biological triplicates or quadruplicates for each condition. For glutamine tracing experiments, cells were incubated in glutamine-free DMEM containing 10% dialysed FBS and 2 mM [ 13 C 5 ]glutamine; for glucose tracing experiments, cells were incubated with glucose-free DMEM containing 10% dialysed FBS and 10 mM [ 13 C 6 ]glucose; and for ornithine tracing experiments, cells were incubated with DMEM containing 10% dialysed FBS and 0.5 mM [ 13 C 5 -15 N 2 ] ornithine (Cambridge Isotope Laboratories, CNLM-7578-H-PK). All labelling experiments were performed for 3 h. Metabolites were extracted on ice with 500 µl of 80% methanol (−80 °C) as previously described 16 . Metabolite extracts from the pooled supernatants were dried down in a SpeedVac concentrator. Dried samples were resuspended either in water, and run in a 6550 iFunnel Q-TOF (Agilent Technology), or in 80% acetonitrile aqueous solution, and run on an AB QTRAP 5500 (Applied Biosystems SCIEX).
Isotopologue analysis on a 6550 iFunnel Q-TOF. Approximately 5 μl of analyte was run with a flow rate of 250 μl min −1 by reverse-phase chromatography, using a 1290 UHPLC coupled to a HRMS 6550 iFunnel Q-TOF mass spectrometer (MS; Agilent Technology). Raw data files were analysed using Profinder software (Agilent Technologies, version B.08.00 SP3). Peak integration results were manually curated in Profinder for improved consistency and exported as a spreadsheet.
Isotopologue analysis on an AB QTRAP 5500. Mass spectrometry was performed on an AB QTRAP 5500 (Applied Biosystems SCIEX) with electrospray ionization source in the multiple reaction monitoring mode. Data acquisition was achieved by injecting 10 µl of analytes with a flow rate of 0.2 ml min −1 on a SeQuant zip_HILIC column (150 × 2 mm) while the column and samples were kept at 4 °C using a Nexera UHPLC system (Shimadzu). The solvents for the mobile phase were: 10 mM ammonium acetate aqueous solution at pH 9.  Table 2 and were obtained using Analyst software (Applied Biosystems SCIEX, version 1.6.1). Data and peak area integration were analysed using MultiQuant software (Applied Biosystems SCIEX, version 2.1.1).

Isotopologue analysis of the TCA cycle intermediates by gas chromatography.
Samples were extracted in 80% methanol. The resulting supernatant was evaporated and resuspended in 30 µl of a 10 mg ml −1 solution of myristic acid in anhydrous pyridine. Samples were heated to 70 °C for 15 min, and then added to 70 μl of N-(tert-butyldimethylsilyl)-N-methyltrifluoroacetamide derivatization reagent for 1 h at 70 °C. Samples were analysed on a 7890 gas chromotagraph coupled to an Agilent 5975C mass selective detector. The observed distributions of mass isotopologues were corrected for natural abundance (Fig. 6e,f).
Sequence alignments. Amino acid sequences for NADK2 were obtained from Uniprot (http://www.uniprot.org/). Clustal Omega was used for multiple sequence alignments and Jalview was used for visualization for the alignments.

Data availability
Supplementary Information including Supplementary Tables 1-3 and a figure exemplifying the gating strategy for Extended Data Fig. 6 are provided with this paper. All other data are available from the corresponding author upon request. Source data are provided with this paper. Fig. 1 | NADK2-deficient cells require proline for cell proliferation. a, A549 and K562 ∆NADK2 cells stably reconstituted with either empty vector (Vec) or NADK2 were injected subcutaneously into athymic nude mice. Tumour growth curves are shown as average of all tumours from the data shown in Fig. 1c,d. Data are presented as mean ± SEM from 7 tumors (A549) or 3 or 4 tumours (K562). *P < 0.05 for pairwise comparisons calculated using a one-sided Student's t-test. b, Immunoblots and cell proliferation of wild-type or single cell-derived knockout cells of NADK in HEK-293E cells grown in DMEM with 10% serum. Proliferation rate was assessed 72 h post-plating and normalized to Day 0. c, Relative proliferation rate was assessed in three consecutive days from wild-type or isogenic ∆NADK2 cells stably expressing either empty vector, NADK2 or NADK. Immunoblots for NADK, NADK2 and β-actin are shown. d, Relative proliferation rate as in (b) from isogenic ∆NADK2 HeLa cells stably expressing either empty vector or NADK2. Cells were grown in 10% dialyzed serum or supplemented with oleate (0.1 mM). e, Relative proliferation rate as in (b) from wild-type or ∆NADK2 HeLa cells. Cells were grown in 10% dialyzed serum for 72 hours or supplemented with NAC (5 mM), nucleosides (inosine 0.1 mg/ml uridine, 0.1 mg/ml), non-essential amino acid (NEAA) mixture (1X), pyruvate (10 mM) or aspartate (10 mM). f, Relative proliferation rate as in (b) from isogenic ∆NADK2 K562 cells stably expressing either empty vector or NADK2 and supplemented with the indicated individual non-essential amino acids at concentration present in human plasma-like medium (HPLM). g, Relative proliferation rate as in (b) from isogenic ∆NADK2 HEK-293E cells stably expressing either empty vector or NADK2, and supplemented with the indicated individual non-essential amino acids at 10X concentration present in human plasma-like medium (HPLM). Related to Fig. 1f. h, Relative proliferation rate as in (b) from HCT116 cells with stable shRNA-mediated knockdown of NADK2 grown in the presence or absence of proline (2 mM). i, Normalized peak areas of mitochondrial NAD + (M + 4), NADP + (M + 4) and NADPH (M + 4) from labeling with 13 C 3 -15 N-nicotinamide are shown. ∆NADK2 HEK293E cells stably expressing HA-Mito were stably reconstituted with either empty vector or NADK2. HEK-293E cells expressing (Myc-Mito) were used as a negative control (1-Ctrl). HA-tagged mitochondria were isolated from equal amount of protein for each condition and metabolites were analyzed by targeted LC-MS/MS. Data are presented as the mean ± SD (b-i) from n = 4 of biologically independent samples for (b), n = 3 for (c,g,h,i), n = 12 for (d), n = 3-9 for (e), n = 3-5 for (f) and are representative of at least two independent experiments. *P < 0.05, **P < 0.01, ***P < 0.001 for comparisons calculated using a two-sided Student's t-test for (b) and with one-way ANOVA test and Tukey's post-hoc test for (c-i). Fig. 3 | Glutamine-dependent proline synthesis is dependent on NADK2 (Data supporting Fig. 5). a, Schematic of 13 C 5 -glutamine labelling. b, Fractional abundance (%) of glutamine (M + 5) from experiment presented in Fig. 5b. c, Fractional abundance (%) of glutamine (M + 5) from experiment presented in Fig. 5c. d, Fractional abundance (%) of glutamine (M + 5), glutamate (M + 5) and proline (M + 5) from isogenic ∆NADK2 A549 cells stably expressing either empty vector (-) or NADK2 cDNA and labeled for 3 hours with 13 C 5 -glutamine. Wildtype counterparts of each cell line are shown as controls. e, Fractional abundance (%) of glutamine (M + 5), glutamate (M + 5) and proline (M + 5) from isogenic ∆NADK2 K562 cells stably expressing either empty vector (-) or NADK2 cDNA and labeled for 3 hours with 13 C 5 -glutamine. Wildtype counterparts of each cell line are shown as controls. f, Immunoblots from HCT116 cells stably expressing a control shRNA (Ctrl), NADK2 shRNA, or two shRNAs against P5CS. g, Fractional abundance (%) of glutamine (M + 5), glutamate (M + 5) and proline (M + 5) from HCT116 cells stably expressing a control shRNA (Ctrl) or two shRNAs against P5CS. Immunoblots are shown in (f). h, Fractional abundance (%) of glutamine (M + 5), glutamate (M + 5) and proline (M + 5) from HCT116 cells stably expressing a control shRNA (Ctrl) or an shRNA against NADK2. Immunoblots are shown in (f). Data are presented as the mean ± SD from n = 4 of biologically independent samples for (b-e) and n = 3 or 4 for (g,h) and are representative of at least two independent experiments. *P < 0.05, **P < 0.01, ***P < 0.001 for comparisons was calculated using a two-sided Student's t-test for (h) and with one-way ANOVA test and Tukey's post-hoc test for (b-e,g).