Mitochondrial NAD⁺ gradient sustained by membrane potential and transport

Abstract

SLC25A51 is required for the replenishment of free nicotinamide adenine dinucleotide (oxidized form) (NAD⁺) into mammalian mitochondria. However, it is not known how SLC25A51 imports this anionic molecule to sustain elevated NAD⁺ concentrations in the matrix. Understanding this would reveal regulatory mechanisms used to maintain critical bioenergetic gradients for cellular respiration, oxidative mitochondrial reactions, and mitochondrial adenosine triphosphate (ATP) production. In this work, mutational analyses and localized NAD⁺ biosensors revealed that the mitochondrial membrane potential (ΔΨm) works in concert with charged residues in the carrier’s inner pore to enable sustained import of NAD⁺ against its electrochemical gradient into the matrix. Dissipation of the ΔΨm or mutation of select residues in SLC25A51 led to equilibration of NAD⁺ from the matrix. Corroborating data were obtained with the structurally distinct mitochondrial NAD⁺ carrier from Saccharomyces cerevisiae (ScNdt1p) and mitochondrial ATP transport suggesting a shared mechanism of charge compensation and electrogenic transport in these mitochondrial carrier family members.

Mitochondrial carriers can use voltage to control directional transport and create critical metabolic gradients.

INTRODUCTION

The mitochondrial matrix is a privileged intracellular compartment for numerous metabolic and signaling pathways. Hence, the functions and capacities of this organelle can be dynamically influenced by the local availability of specific fuels, cofactors, signaling intermediates, and products. Thus, controlling the import and export of these metabolites represents a type of regulation for mitochondria (1–4).

Much of this specific transport depends on dedicated transporters embedded in the inner membrane of mitochondria, including the family of mitochondrial carriers (MCF or SLC25 transporters) (5, 6). Many MCF transporters have been recently deorphanized and paired with cognate ligands (7, 8). However, with a few notable exceptions, the key parameters and cues that govern the activities of specific transporters in cells are still unknown.

In the case of mitochondrial nicotinamide adenine dinucleotide (oxidized form) (NAD⁺), corroborating data from multiple groups have shown that SLC25A51 is the primary and dedicated NAD⁺ transporter in human cells (9–11). SLC25A51 is required for oxidative tricarboxylic acid (TCA) cycle flux, mitochondrial adenosine triphosphate (ATP) production, and cellular respiration (9–11). At its core, SLC25A51 forms a characteristic MCF fold with six transmembrane helices organized in pseudotrifold symmetry (12–15). It is only ~300 amino acids and has no obvious regulatory domains (13, 14).

SLC25A51 is required to maintain the matrix NAD⁺/NADH (reduced form of NAD⁺) ratio through the selective import of oxidized NAD⁺ in cells (15). Unknown, nevertheless, is how SLC25A51 imports NAD⁺ into the matrix against its concentration gradient (16, 17) and amid opposition from approximately a −180-mV differential across the mitochondrial inner membrane (18, 19).

To study the activity of SLC25A51, we developed biosensor-based assays to compare the uptake activities of ectopically expressed wild-type or mutant transporters in intact cells. We use previously published fluorescent biosensors that are selective for free NAD⁺ and localized to the mitochondrial matrix (16, 20). By monitoring fluctuations of matrix NAD⁺ levels following ectopic expression of SLC25A51 and variants, we can obtain readouts of uptake activity in intact cells. Advantages of these assays include assessment of transporter activity in a native lipid environment and amid retention of metabolite gradients, endogenous protein density, cell respiration, and endogenous mitochondrial membrane potential (ΔΨm) (1, 21–25).

In this work, we address how cells import NAD⁺ into the mitochondrial matrix against the barriers of an opposing voltage and concentration gradient. Through mutational analyses, we found that an intact ΔΨm was required and functioned in concert with a specific ensemble of interior pore charges in both human SLC25A51 and ScNdt1p to sustain import of NAD⁺ required for mitochondrial function. We observed equilibrated NAD⁺ levels with electroneutral transporter variants and rescued uptake activity with complementary mutations that restored interior pore charges. We thus identified two distinct roles for charged residues in the pore: a subset required for transport activity itself and another that influences directional uptake activity. To our knowledge, this study provides the first evidence in cells that mitochondrial carriers use a charge compensation and electrogenic mechanism to leverage the import of ionic molecules. Overall, the data demonstrate that mitochondrial NAD⁺ import is governed by net charges within the binding site of the transporter, relative concentrations of the ligand, and strength of the membrane potential.

RESULTS

Identification of active SLC25A51 variants that result in loss of mitochondrial NAD⁺ when overexpressed

Our previous work identified a conserved and negatively charged glutamate-132 (E132) residue in the ligand-binding site of SLC25A51 important for its activity and selectivity (15). Localized in this pocket are three positively charged residues lysine-91, arginine-182, and arginine-278 that are also conserved (Fig. 1A) (15). To study the role of these basic residues, we systematically mutated each to either retain or lose its positive charge. Many mutations at these positions, including alanine mutation, destabilized SLC25A51, and so, among the tested variants, we identified K91Q, K91R, R182Q, R182K, R278L, and R278K as variants whose expressions were most similar to wild type (Fig. 1B). We used a single fluorescent protein biosensor localized to the mitochondrial matrix to assess the NAD⁺ uptake activity of each variant compared to wild-type SLC25A51 in HeLa cells (fig. S1A) (16). In each case when the positive charge was retained (K91R, R182K, and R278K), we observed that the variant imported NAD⁺ similarly to wild type (Fig. 1C, red). Individual loss of the positive charge at every position resulted in loss of robust NAD⁺ uptake (Fig. 1C). As previously reported (15), expression of the R182Q variant resulted in mitochondrial NAD⁺ levels comparable to expression of empty vector alone and had negligible transport activity (Fig. 1C, black) (15). However, the expression of K91Q or R278L variants consistently resulted in lowered mitochondrial NAD⁺ levels (Fig. 1C, blue). No discernible changes in ΔΨm or mitochondrial pH were detected under the experimental conditions, as estimated by MitoTracker CMXRos dye (Fig. 1D) and circularly permutated monomeric Venus (cpVenus) fluorescence (Fig. 1E) (16).

Fig. 1. Identification of at least two subclasses of uptake-deficient SLC25A51 mutants.

(A) Charged residues (cyan, positive; red, negative) in the ligand-binding pocket of human SLC25A51, AlphaFold2 model. (B) Western blot for ^Flag-HASLC25A51 variants transiently transfected in HeLa cells and detected with anti-Flag immunoblotting. Heat shock protein 60 (HSP60), loading control. (C) Mitochondrial NAD⁺ levels measured using a single–fluorescent protein (single-FP) NAD⁺ biosensor in HeLa cells 24 hours posttransfection of empty vector (n = 26), ^Flag-HASLC25A51 (n = 26), and indicated variants (n = 4 to 7). Data as box and whisker plot; analysis of variance (ANOVA), P < 0.001; post hoc Dunnett’s test compared to empty vector control (*) and wild-type ^Flag-HASLC25A51 (#), *P < 0.05, ***P < 0.001, and ###P < 0.001. (D) ΔΨm estimated by MitoTracker Red CMXRos in HEK293 cells (n = 6) with indicated ^Flag-HASLC25A51 variants (n = 4). ANOVA, P = 0.197. (E) Mitochondrial pH changes estimated by the pH sensitivity of circularly permutated mVenus in cells expressing vector (n = 25), ^Flag-HASLC25A51 (n = 25), and indicated variants (n = 4 to 7). Data as box and whisker plot; ANOVA, P = 0.196. (F and G) Mitochondrial NAD⁺ levels measured using a FRET ChemoG-NAD⁺_JF635 biosensor in HEK293 (F) and HEK293 SLC25A51 KO (G) cells with stable expression of ^Flag-HASLC25A51 variants (n = 6 to 7). Data as box and whisker plot; ANOVA, P < 0.001; post hoc Dunnett’s test compared to HEK293 or HEK293 A51 KO cells (*) and cells expressing wild-type ^Flag-HASLC25A51 (#), *P < 0.05, ***P < 0.001, and ###P < 0.001. (H) Mitochondrial NAD⁺ levels (single-FP NAD⁺ biosensor) in HeLa cells 24 hours posttransfection of vector (n = 19), ^Flag-HASLC25A51 (n = 18), and indicated variants (n = 4 to 9). Data as box and whisker plot; ANOVA, P < 0.001; post hoc Dunnett’s test compared to vector control (*) and wild-type ^Flag-HASLC25A51 (#), ***P < 0.001 and ###P < 0.001.

To substantiate these findings, we stably expressed the same variants in a different cell line and monitored NAD⁺ levels with a semisynthetic Förster resonance energy transfer (FRET)–based NAD⁺ sensor localized to the mitochondrial matrix (20) (fig. S1A). Using the ChemoG-NAD⁺_JF635 sensor, we confirmed that ectopic expression of K91Q and R278L resulted in diminished matrix NAD⁺ levels in human embryonic kidney (HEK) 293 cells and that their effects were distinct from wild-type SLC25A51 and R182Q variants (Fig. 1F, blue, and fig. S1B). Consistent with previously described residual activity in these variants (15), we observed increased NAD⁺ above basal levels when these same variants K91Q and R278L were expressed in cells genetically lacking SLC25A51 (A51 KO) (Fig. 1G, red, and fig. S1C).

To determine whether the phenotypes seen with the K91Q or R278L variants depended on transporter activity, we used a previously identified null mutation N183V, which resided in the pocket and was not a charged residue (15). When we combined N183V with previously active mutations, such as K91R, we found that N183V neutralized transport activity (Fig. 1H and fig. S1D). We then evaluated the effect of N183V in combination with either K91Q or R278L variants and found that mitochondrial NAD⁺ levels were now indistinguishable from empty vector (Fig. 1H, black). That is, the N183V mutation neutralized the effects of K91Q and R278L variants, preventing diminishment of mitochondrial NAD⁺.

Together, the data suggest that K91Q and R278L variants, but not R182Q, retained transport activity, and the consequence of this activity was context dependent. Furthermore, it indicated at least two types of uptake-deficient SLC25A51 mutants: one class that is simply null for transport activity and another class that is deficient for the sustained import of NAD⁺ into the matrix, where NAD⁺ levels are already relatively high.

The net charge in the binding pocket is conserved

We analyzed the extent that inward facing central residues in binding pockets of SLC25A51 homologs are conserved among metazoans. We compiled ~330 protein sequences designated as putative human SLC25A51 homologs in the UniProt Database (release 2025 January) and used multiple sequence alignment analysis (data S1) to tally the residues at each of the 12 positions, two inward facing positions per transmembrane helix (Fig. 2A, left). We found strong conservation of the residues in the binding pocket, particularly when we classified the residues by their expected charge as neutral, acidic, or basic (Fig. 2A, right). For the vast majority of SLC25A51 homologs, the electrostatic characteristic of each residue was conserved. An unexpected exception was identified in the genus of Trichinella roundworms that instead have a negatively charged glutamate residue at position 44 (numbering based on human SLC25A51). When we superimposed the homologs harboring an E44 residue with a phylogenetic analysis of SLC25A51, we uncovered that the E44 residue evolved uniquely in the nematode clade (Fig. 2B). This gave us confidence that the identified E44 residue was not an error in annotation. No other identified deviations were consistently maintained throughout any clade.

Fig. 2. Net charge in the binding pocket is conserved and controls mitochondrial NAD⁺ uptake.

(A) Left: Residues in the NAD⁺ binding site plane are shown for human SLC25A51 (AlphaFold2): positive residues in cyan, negative residues in red, and neutral residues in dark gray. Top right: Conservation of binding site residues among 331 SLC25A51 homologs. Bottom right: Tally of occurrence for each residue type. (B) Phylogenetic tree with 21 representative SLC25A51 protein homolog sequences constructed using a maximum likelihood Jones-Taylor-Thornton substitution model, 500 bootstrap replicates, nodes with a bootstrap value (bold line) > 70, and relative evolutionary distance (scale bar). (C) Residues in the NAD⁺ binding site plane are shown for T. zimbabwensis SLC25A51 (AlphaFold2); numbering is based on relative amino acid positions in human SLC25A51. (D) Multiple sequence alignment with highlighted glutamate at the 44th position and arginine at the 95th position. (E) Residues in the NAD⁺ binding site plane for human SLC25A51 (AlphaFold2): positive residues in cyan, negative residues in red, and neutral residues in dark gray. (F) Mitochondrial NAD⁺ (single-FP NAD⁺ biosensor) in HeLa cells 24 hours posttransfection of empty vector (n = 22) or plasmid expressing ^Flag-HASLC25A51 (n = 22) and indicated variants (n = 4 to 8). Data as box and whisker plot; ANOVA, P < 0.001; post hoc Dunnett’s test relative to vector control (*) and wild-type ^Flag-HASLC25A51 (#), ***P < 0.001 and ###P < 0.001.

To study the E44 binding pocket, we generated an AlphaFold2 structural model using the sequence from Trichinella zimbabwensis (Fig. 2C) (26). We observed that the plane directly above E44 held a compensatory positive charge arising from residue arginine-95 (Fig. 2, C and D). Compellingly, R95 was conserved among the analyzed nematode homologs and only existed in homologs where E44 was present (Fig. 2D). The implication of R95 is that it may have coevolved to neutralize the charge of E44 so that the electrostatics of the pocket would be conserved. Given that SLC25A51 is an essential gene (27, 28), there would be strong evolutionary pressure to restore its import activity. This, in turn, suggested that a critical aspect of SLC25A51 activity and the import of mitochondrial NAD⁺ depended on the conservation of net charge in the pocket. Analysis of the binding pockets between human and Trichinella SLC25A51 revealed that each has a net charge of +2 (Fig. 2, C and E).

The net charge in the pocket controls mitochondrial NAD⁺ uptake

The identification of roundworm E44;R95 SLC25A51 variants revealed an opportunity to directly test how the net charge in human SLC25A51 contributes to its uptake function. Human SLC25A51 encodes two neutral residues at the positions asparagine-44 and leucine-95. Previous computational analyses of the binding pocket indicated that N44 was not critical for NAD⁺ uptake by SLC25A51 (15), and this was supported experimentally here by the ability for an N44A variant to import NAD⁺ similarly to wild type (Fig. 2F and fig. S2A). Previously, we reduced the net charge in the pocket by mutating either a positive lysine or arginine residue. Now, we kept all original charged residues in place and instead introduced an additional negative charge at position 44 (N44E or N44D), mimicking the roundworm variant. Introduction of either mutation resulted in reduced rather than increased mitochondrial NAD⁺ levels upon ectopic expression of the variant (Fig. 2F, blue, and fig. S2B). When we combined N44E with L95R, mimicking the evolved compensatory mutation in the roundworm, we found that this restored robust uptake activity and mitochondrial NAD⁺ levels were increased (Fig. 2F and fig. S2B). The combination of neutralizing mutation N183V with N44D eliminated the N44D-dependent reduction of mitochondrial NAD⁺ levels. With N183V + N44D SLC25A51, mitochondrial NAD⁺ levels were similar to empty vector, suggesting loss of previous activity (Fig. 2F and fig. S2C).

We observed a pattern that when a variant had a net charge in its pocket of +1—reduced from the original +2—its ectopic expression would result in lowered mitochondrial NAD⁺ levels as long as the variant retained activity; otherwise, it would have negligible effect similar to ectopic expression of empty vector (Figs. 1, C and F, and 2F). Expression of L95R alone, which changed the net charge in the pocket to approximately +3, did not interfere with the variant’s uptake activity (Fig. 2F and fig. S2B). Thus, we conclude that SLC25A51 requires an interior net charge of approximately +2 for the robust import of mitochondrial NAD⁺ in cells.

Contribution of charges from the plane of the matrix gate

To determine whether the charges that contribute to mitochondrial NAD⁺ import extended beyond the centrally located ligand binding pocket, we considered the exposure of different features of SLC25A51 during its transport cycle based on the experimental adenosine diphosphate (ADP)/ATP carrier (AAC) structures (Fig. 3A) (12, 29–31). The cytoplasmic and matrix gates are two defined regions with charged residues. For NAD⁺ import, we realized that the matrix gate residues are positioned, such that they could continuously influence the NAD⁺ ligand from its position in the binding pocket in the cytoplasmic open state (c-state) to its exit into the matrix. In contrast, cytoplasmic gate residues are exposed only to the intermembrane space at all times. While cytoplasmic gate residues would be positioned to influence ligand binding, they are blocked to the matrix side by a hydrophobic barrier as the transporter transitions to a matrix open state (m-state) and thus may not have as strong an influence on NAD⁺ import specifically (Fig. 3A).

Fig. 3. Contribution of matrix gate charges in NAD⁺ uptake.

(A) Schematic showing relative positions of transporter features for NAD⁺ import. (B) Matrix gate plane and inward facing residues in human SLC25A51 (AlphaFold2): positive charge in cyan, negative charge in red, hydrophobic residues in yellow, and hydrophilic residues in white. (C) Mitochondrial NAD⁺ levels (single-FP NAD⁺ biosensor) in HeLa cells 24 hours posttransfection of empty vector (n = 21), ^Flag-HASLC25A51 (n = 21), and indicated variants (n = 5 to 13). Data as box and whiskers plot; ANOVA, P < 0.001; post hoc Dunnett’s test compared to vector control (*) and wild-type ^Flag-HASLC25A51 (#), **P < 0.01, ***P < 0.001, and ###P < 0.001. (D) Cytoplasmic gate plane and inward facing residues in human SLC25A51 (AlphaFold2): positive charges in cyan and negative charges in red. (E) Mitochondrial NAD⁺ (single-FP NAD⁺ biosensor) in HeLa cells after expression of empty vector (n = 20), ^Flag-HASLC25A51 (n = 20), and indicated variants (n = 4 to 10). Data as box and whiskers plot; ANOVA, P < 0.001; post hoc Dunnett’s test compared to vector control (*) and wild-type ^Flag-HASLC25A51 (#), **P < 0.01, ***P < 0.001, ##P < 0.01, and ###P < 0.001.

The matrix gate in human SLC25A51 is composed of a single ionic interaction between K236-E139 (Fig. 3B), and this salt bridge is surrounded by inward facing neutral residues. We introduced additional negative charges at residues Q52 and N233. We observed lowered mitochondrial NAD⁺ levels that were rescued by introduction of a compensatory positive charge on a nearby residue (Fig. 3C, Q52E + Q142K and N233D + L55K, and fig. S3, A and B).

We tested the contributions of charged residues at the cytoplasmic gate position by individual removal of both positive and negative charges and comparing similarly expressed variants to wild type (Fig. 3D and fig. S3B). The data were variable and not as robust as wild type but indicated retention of some NAD⁺ uptake activity (Fig. 3E). Together, the data showed that net charge of the interior pore contributes to NAD⁺ import and that the strongest contributors are from the ligand-binding pocket and the matrix gate.

Ndt1p requires a positive net charge of approximately +2 to import NAD⁺ in cells

Ndt1p is a mitochondrial NAD⁺ transporter and a functional homolog of human SLC25A51 in Saccharomyces cerevisiae (9–11, 32). Ndt1p is predicted to have evolved NAD⁺ transport independently of the SLC25A51 family (8, 13, 15). Consequently, the binding site in Ndt1p is different from SLC25A51 and contains charged residues R346 and K289 one helical turn away (Fig. 4A). To determine how these charges contribute to Ndt1p import activity, we tested Ndt1p variants K289R, K289Q, R346K, R346Q, T347A, and T347D, which had similar expression to wild-type Ndt1p (fig. S4A). Retention of the charge in the pocket (K289R, R346K, and T347A) resulted in a functional Ndt1p capable of continuous mitochondrial NAD⁺ uptake (Fig. 4B). Loss of one positively charged residue either resulted in loss of activity (K289Q) or lowered mitochondrial NAD⁺ levels (R346Q); introduction of an additional negative charge (T347D) resulted in lowered mitochondrial NAD⁺ levels (Fig. 4B).

Fig. 4. Ndt1p requires a net charge approximating +2 to import NAD⁺ in cells.

(A) Binding site residues in yeast Ndt1p (AlphaFold2): positive charges in cyan and hydrophilic residues in white. (B) Mitochondrial NAD⁺ (single-FP NAD⁺ biosensor) in HeLa cells 24 hours posttransfection expression of vector (n = 19), ^Flag-HANdt1p (n = 15), and indicated variants (n = 4 to 12). Data as box and whisker plot; ANOVA, P < 0.001; post hoc Dunnett’s test compared to vector control (*) and wild-type ^Flag-HASLC25A51 (#), *P < 0.05, ***P < 0.001, and ###P < 0.001. (C) Residues in the matrix gate are shown for yeast Ndt1p (AlphaFold2): positive charges in cyan, negative charges in red, and hydrophobic residues in yellow. (D) Mitochondrial NAD⁺ (single-FP NAD⁺ biosensor) in HeLa cells with empty vector (n = 18), ^Flag-HANdt1p, and indicated variants (n = 4 to 8). Data as box and whisker plot; ANOVA, P < 0.001; post hoc Dunnett’s test compared to vector control (*) and wild-type ^Flag-HASLC25A51 (#), **P < 0.01, ***P < 0.001, and ###P < 0.001. (E) A model for facilitated import of NAD⁺ into the matrix.

We next tested the charged residues in the matrix gate of Ndt1p. Distinct from SLC25A51, the matrix gate of Ndt1p has three positive (K102, K201, and R303) and two negative (D99 and E300) residues (Fig. 4C). R303 participates in a cation-π interaction with W198 but retains its positive charge (33). A reported set of experiments in proteoliposomes showed that Ndt1p variants with K102C and R303C mutations were nonfunctional (33). We generated analogous K102Q and R303Q mutations, and these variants also showed negligible activity in cells in agreement with the published proteoliposome data (Fig. 4D and fig. S4B) (33). We next tested variant K201C, which was previously published to have ~40% activity in vitro (33)_. Variant K201C, as well as related variant K201Q, resulted in lowered mitochondrial NAD⁺ levels (Fig. 4D and fig. S4B). In contrast, variant K201R that retained the positive charge resulted in robust mitochondrial NAD⁺ uptake when tested in cells (Fig. 4D and fig. S4B). Together, the data indicate that Ndt1p functions similarly to SLC25A51 and control of NAD⁺ import also depends on its pore charges. Furthermore, Ndt1p similarly has two subclasses of charged residues in its pore, one subset of residues that is required for activity and another subset that contributes to the net charge for sustained import of NAD⁺.

Shielding ligand charges and facilitating directional movement through a transport cycle

To develop a testable model for these results, we considered a charge compensation mechanism previously proposed for the ATP/ADP transporter (34). In this model, pore charges in the binding site of the transporter would shield the charges originating from the ligand, and upon transport of the ligand, the charges also would be cotransported across the membrane. The evidence for this model includes measured current coinciding with directional transport of ADP⁻³ and ATP⁻⁴ by AAC homologs in vitro, with current amplitudes that would favor the export of ATP⁻⁴ and import of ADP⁻³ in cells (34–37). Site-directed mutagenesis of Thermothelomyces thermophila AAC (TtAAC) directly indicated roles for pore charges in this transport (37). Thus far, evidence for this mechanism has been limited to in vitro data. To determine whether our data supported the existence of charge compensation and electrogenic transport by SLC25A51 and Ndt1p in cells, we considered the requirements for import of anionic (NAD⁺)⁻¹ into the mitochondrial matrix.

SLC25A51 has a single ligand-binding site and is thought to function via an alternating-access mechanism as a member of the SLC25 family (15). Hence, both cytoplasmic and matrix open states are expected to be thermodynamically equivalent, and, thus, the directionality of the transport cycle would be imposed by external cues (38). As an anionic molecule, the ligand NAD⁺ has at least two major challenges for its import into the matrix. It must be brought into a matrix pool that has a relatively higher concentration of NAD⁺ (16, 17), and NAD⁺ would also need to overcome a large voltage to enter the matrix. The latter presents the greater energetic challenge, as the mitochondrial matrix is typically held at −180 mV (18, 19).

We propose that the positive pore charges in SLC25A51 and Ndt1p may function to shield the −1 charge of the NAD⁺ ligand and enable its transfer into the matrix. Our data indicate that a +2 or +3 net charge, but not a +1 net charge, in the binding site of the transporter is sufficient for sustained import; thus, we conclude that the net charge must be approximately +2 (Fig. 4E). We do not yet know whether there is a molecule that is antiported in SLC25A51, but for Ndt1p, it can be either ADP⁻³ or adenosine monophosphate (AMP)⁻² (32), which should readily be exported out of the matrix due to charge. The net result from a transfer cycle would be a loss of a −3/−2 ligand for the gain of a −1 ligand in the matrix (Fig. 4E). Thus, there would be a resulting net movement of negative charge out of the matrix and toward the inner membrane space that would be aided by the ΔΨm (Fig. 4E). Notably a variation of this model could apply if SLC25A51 operates through a uniport mechanism, wherein the pore would simply shield the −1 ligand to facilitate its transfer into the matrix. This would entail that SLC25A51 switches conformation spontaneously to a cytoplasmic open state, a previously proposed model (39). While there may be some loss of membrane potential, the cost would be minimal, as there is less NAD⁺ transport relative to proton pumping mechanisms. Briefly, the data together suggest that a charge-based transporter mechanism facilitates the sustained import of NAD⁺ to enable generation of a mitochondrial NAD⁺ gradient.

ATP gradients are regulated by ΔΨm

The AAC has been found in proteoliposomes to leverage a charge compensation mechanism to control its transport. A net charge of +3.3 in its pore enables the directional import of ADP and export of ATP. This leads to a −1 charge being transported out of mitochondria in presence of ΔΨm (Fig. 5A) (34–37). The model predicts that a loss of ΔΨm would increase relative mitochondrial ATP levels.

Fig. 5. ATP gradients are established by membrane potential.

(A) A model for facilitated import of ADP and export of ATP across inner mitochondrial membrane. (B) Schematic for ratiometric mCardinal-iATPSnFR2 construct. (C) ATP levels measured using mCardinal-iATPSnFR2 biosensor (n = 6) normalized to mCardinal-cpSFGFP. Data as box and whisker plot; unpaired two-sided t test, ***P < 0.001. (D) ATP levels (mCardinal-iATPSnFR2) in HEK293T cells after indicated treatments (n = 3 to 4). Data are means ± SD; ordinary one-way ANOVA, P < 0.001; post hoc Tukey’s test, *P < 0.05 and ***P < 0.001. ns, not significant.

To test the consequence of this in cells, we used an established ATP biosensor, iATPSnFR2 (40) that we fused with mCardinal for ratiometric measurements (Fig. 5B and fig. S5A). Sensors localized to the cytosol and mitochondrial matrix retained similar responsivity to ATP as previously published (fig. S5B) (40). As expected, we found that both HEK293T and U2OS cells had a higher concentration of free ATP in the cytosol relative to the mitochondrial matrix (Fig. 5C).

When acutely treated with carbonyl cyanide p-trifluoromethoxyphenylhydrazone (FCCP), expected to abolish both ΔΨm and complex V–ATP synthesis, we found that mitochondrial matrix ATP levels increased within 4 hours (Fig. 5D). With no active ATP synthesis, the observation of increased mitochondrial matrix ATP levels suggests increased import, rather than export, when membrane potential was lost. This time-dependent increase in matrix ATP was not recapitulated under dimethyl sulfoxide (DMSO) vehicle control conditions or with oligomycin A, which also ablates complex V–ATP synthesis but retains ΔΨm (Fig. 5D). We did not observe significant changes in cytosolic ATP levels, but the larger pool likely masked smaller changes arising from altered transport.

Sustained NAD⁺ import depends on ΔΨm

An implication of the model for SLC25A51 and yeast Ndt1 is that the ΔΨm would be required to sustain import of NAD⁺ into the matrix. To determine whether potential is required for robust NAD⁺ uptake, we quantified the amount of trace-labeled ³²P-NAD⁺ imported in an hour using recombinant Escherichia coli cells that ectopically expressed Ndt1p or SLC25A51 (amino acids 29 to 297); E. coli lacks endogenous NAD⁺ import. We dissipated potential with ionophore FCCP, and this resulted in reduced uptake of ³²P-NAD⁺ for both transporters (Fig. 6, A and B).

Fig. 6. Direction of NAD⁺ transport depends on ΔΨm.

(A and B) ³²P-NAD⁺ uptake in E. coli–expressing ^Flag-HANdt1 (A) and SLC25A51 (amino acids 29 to 297) (B) with DMSO or 10 μM FCCP treatment (n = 3 to 4). Data as bar graphs, means ± SD; unpaired two-sided t test, **P < 0.01. (C) Mutated residues (red) in the NAD⁺-bound ChemoG-NAD⁺ sensor, AlphaFold3. (D) Emission spectra following excitation at 488 nm of ChemoG-NAD⁺_JF635 and Dead-ChemoG-NAD⁺_JF635 with NAD⁺. a.u., arbitrary units. Inset: Colocalization between Dead-Mito-ChemoG-NAD⁺ (green) and MitoTracker CMXRos (red) in live HEK293 cells. Scale bar, 10 μm. (E) Mitochondrial NAD⁺ levels (normalized FRET Mito-ChemoG-NAD⁺_JF635) in HEK293 cells (n = 5). Data as box and whisker; unpaired two-sided t test, ***P < 0.001. (F) Mitochondrial NAD⁺ levels (normalized FRET Mito-ChemoG-NAD⁺_JF635) in HEK293 SLC25A51 KO cells stably expressing yeast ^Flag-HANdt1p (n = 6). Data as box and whisker; unpaired two-sided t test, **P < 0.01. (G) Mitochondrial NAD⁺ levels (FRET Mito-ChemoG-NAD⁺_JF635 normalized to Dead-Mito-ChemoG-NAD⁺_JF635) relative to t = 0 min in HEK293T cells (n = 3). Data are means ± SD; two-way ANOVA, P < 0.01; post hoc Šídák’s test, **P < 0.01 and ***P < 0.001. (H) Mitochondrial NAD⁺ levels (single-FP NAD⁺ biosensor) in HeLa cells 24 hours posttransfection with ^Flag-HASLC25A51 variants and after DMSO (n = 8) or 10 μM FCCP (n = 6) treatment for 4 hours. Data as box and whisker; ANOVA DMSO, P < 0.001; ANOVA FCCP, P < 0.01; post hoc Dunnett’s test relative to vector control (*) and ^Flag-HASLC25A51 (#), *P < 0.05, **P < 0.01, ***P < 0.001, #P < 0.05, ##P < 0.01, and ###P < 0.001. (I) Mitochondrial NAD⁺ levels (single-FP NAD⁺ biosensor) in HeLa cells 24 hours posttransfection with ^Flag-HASLC25A51 and after 2 hours of treatments (n = 6). Data as box and whisker; unpaired two-sided t test, ***P < 0.001.

To determine the effects of FCCP on mitochondrial NAD⁺ in live cells, we needed an approach to monitoring mitochondrial NAD⁺ amid anticipated pH changes. We started with the less pH-sensitive ChemoG-NAD⁺_JF635 FRET sensor (20) and abolished its NAD⁺ binding based on an AlphaFold3-predicted NAD⁺-bound structure (Fig. 6C) (41). We confirmed that this sensor did not respond up to 1 mM NAD⁺ in vitro and that it could be localized to mitochondria (Fig. 6D). Using NAD⁺-dead ChemoG-NAD⁺_JF635, we found that our experimental treatment (10 μM FCCP in HEK293 cells for 4 hours) increased the appearance of the FRET/enhanced green fluorescent protein (EGFP) ratio independently from any NAD⁺ changes (fig. S6A). We quantified that untreated measurements were ~0.6-fold relative to FCCP-treated measurements and used this as a normalization factor to estimate ChemoG-NAD⁺_JF635 FRET sensor measurements with FCCP treatment compared to untreated conditions. We found that FCCP treatment resulted in a time-dependent diminishment of mitochondrial NAD⁺ levels as measured by Mito-ChemoG-NAD⁺_JF635 (Fig. 6, E and G). Analogously, restored mitochondrial NAD⁺ levels by ectopic expression of Ndt1p in HEK293 SLC25A51 KO cells were also sensitive to FCCP (Fig. 6F). We confirmed dissipation of ΔΨm under these experimental conditions (fig. S6B). We observed a trending increase (P = 0.1) in cytosolic NAD⁺ levels after 3 hours of FCCP treatment, relative to DMSO control conditions (fig. S6C).

We found that the FCCP-driven loss of matrix NAD⁺ resulted in equilibration of mitochondrial NAD⁺ levels despite ectopic expression of SLC25A51 variants, suggesting that observed effects from SLC25A51 depended on ΔΨm (Fig. 6, H and I). Differences in mitochondrial NAD⁺ levels across variants (increases from overexpression of wild-type SLC25A51 and decreases from expression of K91Q or R278L variants) were eliminated upon dissipation of ΔΨm, and with membrane potential, all variants equilibrated to similar mitochondrial NAD⁺ levels (Fig. 6H). No variant was able to sustain import of NAD⁺ into the mitochondrial matrix without an intact ΔΨm. We observed a similar equilibration of mitochondrial NAD⁺ with valinomycin treatment, which dissipated ΔΨm as a potassium-selective ionophore but not with complex V blocker oligomycin A (Fig. 6I). Acute treatment with oligomycin A slightly increased relative mitochondrial NAD⁺ levels (Fig. 6I), likely due to the observed increase in ΔΨm (fig. S6D).

Sustained charge-dependent import of NAD⁺ is required for mitochondrial function

We identified two SLC25A51 variants T94V and R278L that displayed similar activity in vitro for ³²P-NAD⁺ uptake in the absence of intact membrane potential (Fig. 7A, left). What distinguished the variants was that T94V retained a +2 net charge in its pore and R278L had a +1 charge due to loss of the arginine. In the presence of membrane potential, we found that the +2 T94V variant was able to increase its rate of ³²P-NAD⁺ import. In contrast, the +1 charged R278L variant did not show any increased uptake, and on the contrary, it slightly decreased uptake (Fig. 7A, right). Expression of these variants in HEK293KO cells confirmed that only +2 T94V was able to sustain higher matrix NAD⁺ levels similar to wild-type SLC25A51 (Fig. 7B, right). The variants had comparable effects on matrix NAD⁺ levels when ΔΨm was dissipated (Fig. 7B, left). Together, the data indicate that the elevated and sustained mitochondrial NAD⁺ levels arose from ΔΨm-driven import.

Fig. 7. Sustained electrogenic import of NAD⁺ is required for mitochondrial function.

(A) ³²P-NAD⁺ uptake after 1 hour in E. coli cells expressing SLC25A51 (amino acids 29 to 297) with either R278L (n = 4) or T94V (n = 3) mutation. Data as bar graph, means ± SD; two-way ANOVA, P < 0.001; post hoc Tukey’s test, *P < 0.05 and ***P < 0.001. (B) Mitochondrial NAD⁺ levels (FRET ChemoG-NAD⁺_JF635) after 4 hours of 10 μM FCCP treatment in HEK293 SLC25A51 KO with wild-type or indicated variants. Data as box and whisker; n = 5 to 7 biological replicates; ANOVA, P < 0.001; post hoc Dunnett’s test relative to basal (*) and ^Flag-HASLC25A51 (#), *P < 0.05, ***P < 0.001, ##P < 0.01, and ###P < 0.001. (C) Left: Cell count on the indicated days. Right: Calculated proliferation rate. Data as box and whisker; n = 9 biological replicates; ANOVA, P < 0.001; post hoc Tukey’s test, **P < 0.01 and ***P < 0.001. (D) Pan–expansion microscopy expanded HEK293 A51 KO cells expressing HA-SLC25A51 and pan-stained with NHS ester dye and anti-HA immunostaining.

NAD⁺ import by the +2 T94V SLC25A51 variant ought to enable sustainment of the mitochondrial NAD⁺ gradient needed for proliferation of cells in galactose medium. Among the tested variants, we found that only the +2 SLC25A51 variant (T94V), and not the +1 SLC25A51 variant (R278L), was able to proliferate in galactose medium (Fig. 7C). Both variants proliferated similarly in glucose medium (fig. S7A).

SLC25A51 is exposed to physiological fluctuations in ΔΨm

For ΔΨm to regulate SLC25A51 activity, the expectation is that the transporter would need to reside in locales such as mitochondrial cristae that would be exposed to physiological fluctuations in ΔΨm. We expressed Flag-hemagglutinin (Flag-HA) epitope–tagged SLC25A51 in HEK293 A51 KO cells and confirmed that this rescued mitochondrial NAD⁺ levels (Figs. 1G and 7B). We performed expansion microscopy and stained with an N-hydroxysuccinimide (NHS) ester–based dye for pan-staining of ultrastructures, as well as a validated antibody targeting the HA epitope (Panluminate Inc.). Through a series of 0.2-μm optical slices (>25 slice stacks), we observed consistent overlap of stained ^Flag-HASLC25A51 with mitochondrial cristae throughout the cell (Fig. 7D and movie S1). ^Flag-HASLC25A51 was not associated with any other cellular structure. Its staining pattern was distinct from that observed with outer membrane proteins such as translocase of the outer mitochondrial membrane 20 (TOM20) (42) and consistent with known crista proteins (42, 43). We saw a similar localization pattern for expressed ^Flag-HASLC25A51 R278L variant (fig. S7B and movie S2), indicating that its activity would also be subject to ΔΨm.

DISCUSSION

Our work provides insight into how cells maintain different NAD⁺ concentrations between the cytosol and mitochondrial matrix, and we found that maintenance of this gradient is essential for mitochondrial function. MCF transporters control much of the trafficking of molecules across the inner mitochondrial membrane that are used as fuel, signaling intermediates, and cofactors in the matrix, as well as the export of products and signaling intermediates that regulate and get used by other parts of the cell (5, 6). Hence, the activities of MCF transporters have tremendous influence and can dictate the precise role of mitochondria and dependent pathways in real time. Most cargo are composed of charged molecules, and so, transport must also consider both relative electrical potential and relative concentrations of the molecules on either side of the membrane (44–47).

With transporters SLC25A51, Ndt1p, and our observations with mitochondrial ATP gradient, this work provides evidence in cells for a charge compensation mechanism and electrogenic transport by mitochondrial carriers (34–37)_. The concepts of charge compensation and electrogenic transport in the MCF family were first proposed for AACs based on proteoliposomes and in vitro capacitive current measurements (34–36). In this model, the net charge in the binding site of AAC (approximately +3.3) was proposed to shield the (−3) charge of ADP to enable its import against the membrane potential into mitochondria. In contrast, an ATP molecule with its (−4) charge would retain a negative charge in its bound state and be exported. Mavridou et al. (37) advanced this model by identifying residues in AAC that governed electrogenic transport and determined that residues in the substrate binding site were the primary contributors. In this work, we expand the model of charge compensation and electrogenic transport beyond AAC to the NAD⁺ carriers and identify that contributing charges in the cases of NAD⁺ carriers include residues both in the binding pocket and in the plane of the matrix gate (Fig. 3). Furthermore, by extending the model to cells, this work was able to test the functional consequence of this mechanism. We showed that a specific array of positively charged residues in the pores of both SLC25A51 and Ndt1p transporters was needed to sustain ΔΨm-driven NAD⁺ import, which, in turn, was required for creation of the mitochondrial NAD⁺ gradient and sustainment of mitochondrial function (Fig. 7C). In the NAD⁺ import model, membrane potential is used by the carriers to sustain a sustained directionality of their transport cycle, and this overcomes the electrochemical barriers opposing sustained import of NAD⁺ into the matrix. This occurs through charge movement and requires an interior net charge in the transporter of approximately +2. Together, the data support at least two MCF members that depend on this mechanism in cells, with the likelihood that others may as well due to shared structural and mechanistic features within the family (39, 48–51).

An implication is that any physiological fluctuations in ΔΨm could be a homeostatic mechanism to regulate these transporters and the import of mitochondrial NAD⁺. SLC25A51 is positioned such that it would be responsive to such fluctuations (Fig. 7D). Local changes in ΔΨm can occur, for example, during mitochondrial remodeling (52–55). During fission or fusion, an acute loss of ΔΨm may serve to rapidly and reversibly pause NAD⁺ import (53–56), and this could further occur in microdomains within the organelle (57–59). Conversely, an accumulation of mitochondrial NADH can drive flux through the electron transport chain, and this may transiently and locally increase ΔΨm to enable increased import of mitochondrial NAD⁺.

A second implication of this mechanism arises from the requirement of a net positive pore charge in the carriers of approximately +2. In both SLC25A51 and Ndt1p, lysine residues contribute to the net charge, and we showed that modification of lysine residue charges can change the extent of uptake in presence of membrane potential. The possibility therefore exists that posttranslational acylation on these accessible lysines could regulate the activities of these transporters. Posttranslational modifications of lysine residues in the binding site and matrix gate have been previously identified for other MCF carriers (60, 61).

The findings in this study were enabled by genetically encoded fluorescent biosensors that monitored relative matrix NAD⁺ and ATP levels in intact cells. Advantages of these measurements include being able to characterize the effects of the variants in situ amid physiological metabolite concentrations, a native lipid environment, with retention of gradients and voltage, and in cells that have retained respiring and oxidative states. That is, by retaining the environment surrounding the transporter, we identified a mechanism for how its microenvironment regulated activity.

Throughout this work, we found that ectopic expression of SLC25A51 variants with only a +1 charged pore resulted in lowered mitochondrial NAD⁺ levels in wild-type cells. We showed that these same variants retained uptake activity in SLC25A51 KO cells. We conclude that these variants were unable to sustain continuous uptake. Collectively, the data further suggest that the variants enabled an equilibration of NAD⁺ across the membrane (Fig. 6H). Previous reports showed that loss of SLC25A51 did not alter ΔΨm (9, 10), and, instead, one prominent difference between wild-type and SLC25A51 KO cells would be inverted NAD⁺ concentration gradients between the cytosol and matrix, which may explain the inverse effects of the variants in these different backgrounds. Equilibration may be occurring through SLC25A51, as suggested by some studies (62–64), but this will need further investigation as there could also be independent mechanisms. A reverse transport activity has been observed for AAC in cells lacking a functional electron transport chain (65–68). The drop in membrane potential enabled ATP import for hydrolysis by complex V to reinstate ΔΨm (65–68).

Last, we examined other MCF members for similar regulation. The ATP-Mg/Pi carriers (APCs) coimport ATP⁻⁴ and Mg⁺² to supplement the adenine nucleotide matrix pool in exchange for negatively charged phosphate (69). APC has a net pore charge of +3 (binding site⁺² and matrix gate⁺¹) and thus would be predicted to depend on ΔΨm to aid import of ATP⁻⁴ and Mg⁺² while exporting phosphate (39, 48, 49). A different example may be the mitochondrial oxoglutarate carrier (OGC/SLC25A11) in the malate-aspartate shuttle, which may use charge compensation to provide flexibility for the directionality of its transported cargo. OGC antiports similarly charged malate⁻² and α-ketoglutarate⁻² (70), which both approximate the net charge of ~2 in the pore [binding site⁺³, matrix gate⁻¹, and a putative fractional positive c-state–specific charge (37)]. Thus, the activity of OGC would be expected to be less dependent on ΔΨm and more driven by mass action and the relative concentrations or gradients of each ligand. Future studies are needed to determine how this mechanism may apply across other MCF carriers.

MATERIALS AND METHODS

Cells

Cell lines were regularly tested for mycoplasma contamination. HeLa [American Type Culture Collection (ATCC), CCL-2] and HEK293T (ATCC, CRL-3216) cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM) with glucose (4.5 g/liter), sodium pyruvate (1 mM), l-glutamine (4 mM), 10% fetal bovine serum, penicillin-streptomycin (1×), and 25 mM Hepes buffer (pH 7.4). HEK293 and HEK293 SLC25A51 KO cells were previously published (15). Sensors, SLC25A51, Ndt1p, and their variants were either transiently transfected or stably expressed using lentiviral transduction and selection with puromycin (2 μg/ml; 4 days) or with fluorescence-activated cell sorting enrichment for fluorescent cells.

Plasmids

We used the 2× HiFi builder assembly master mix [New England Biolabs (NEB), #E2621S] to create pLentiCMV-4×(cox8)-Flag-HA-ChemoG-NAD⁺-IRES-puro, pLentiCMV-Flag-HA-ChemoG-NAD⁺-IRES-puro, pLentiCMV-4×(cox8)-Flag-HA-mCardinal-cpSFGFP, pLentiCMV-4×(cox8)-Flag-HA-mCardinal-iATPSnFR2, pLentiCMV-NES-Flag-HA-mCardinal-cpSFGFP, pLentiCMV-NES-Flag-HA-mCardinal-iATPSnFR2 plasmid, and pMW7-Flag-HA-Ndt1. Q5 site–directed mutagenesis kit (NEB, #E0554S) or 2× HiFi builder assembly master mix (NEB, #E2621S) was used to generate mutants in pMW7–SLC25A51 (amino acids 29 to 297), pLentiCMV–Flag–HA–SLC25A51 (amino acids 1 to 297)–IRES–puro, pLentiCMV-Flag-HA-Ndt1p-IRES-puro (9, 15), and pLentiCMV-4×(cox8)-Flag-HA-ChemoG-NAD⁺-IRES-puro.

Western blotting

A total of 1 million mammalian cells were lysed in 200 μl of 2× Laemmli sample buffer containing dithiothreitol. Lysate was incubated at 95°C for 5 min, and 40 or 15 μl was resolved by SDS–polyacrylamide gel electrophoresis (PAGE) using NuPAGE 10% or 4 to 12% bis-tris protein gels (Invitrogen). Protein was transferred to Bio-Rad 0.45-μm nitrocellulose membrane. The nitrocellulose membrane was blocked with 5% bovine serum albumin (BSA) or 5% nonfat milk in tris-buffered saline with 0.1% (v/v) Tween 20 (TBST) (pH 7.4). The antibodies used were prepared in TBST with 1% BSA or nonfat milk. The primary antibody was incubated overnight at 4°C, followed by three washes of 5 min each with 1× TBST. The secondary antibody was hybridized for 1 hour at room temperature, followed by three washes of 5 min each with 1× TBST. We used LI-COR Odyssey CLx for imaging. The antibodies and their dilutions are as follows: anti-Flag M2 (Sigma-Aldrich, F1804, RRID: AB_262044; 1:3000), anti–heat shock protein 60 (HSP60) monoclonal antibody (Invitrogen, MA3-012, RRID: AB_2121466; 1:3000), anti-SLC25A51 (MyBiosource, MBS1496255; 2 μg/ml), anti-mouse immunoglobulin G (IgG) (H+L), Alexa Fluor Plus 800 (Invitrogen, A32730, RRID: AB_2633279; 1:10,000), anti-rabbit IgG (H+L) IRDye 800CW (Abcam, ab216773, RRID: AB_2925189; 1:10,000), and anti-mouse IgG (H+L) Alexa Fluor 680 (Invitrogen, A10038, RRID: AB_11180593; 1:10,000).

NAD⁺ measurement using single fluorescent protein-based NAD⁺ biosensor

We seeded HeLa cells stably expressing ^mitocpVenus and ^mitosensor at a density of 200,000 cells per well in a six-well plate (9). Cells were transfected the next day with plasmids expressing transporter variants using polyethylenimine (PEI; 1 mg/ml) at a PEI:DNA ratio of 5:1. Cells were collected 24 hours posttransfection or after indicated treatments in cold DMEM and kept on ice until flow cytometry analysis using NovoCyte 3000 VYB. Standard gating was applied to exclude debris and cell doublets. Steady-state NAD⁺ levels were calculated as the geometric mean of 10,000 fluorescent cells as described earlier using the formula (15, 16, 71)

$$\text{Normalized fluorescence}={\left(\frac{{\text{Sensor}}_{\frac{488}{405}}}{{\text{cpVenus}}_{\frac{488}{405}}}\right)}^{-1}$$

The fluorescence was measured for both ^mitosensor- and ^mitocpVenus-expressing lines in parallel at excitations of 488 nm and 405 nm with emission of 530 nm ± 30 nm. Measurements resulting from expression of transporter variants were presented relative to measurements obtained with empty vector control.

NAD⁺ measurement using semisynthetic ChemoG-NAD⁺ FRET biosensor

Cells expressing the ^mitoChemoG-NAD⁺ or ^cytoChemoG-NAD⁺ FRET biosensor were seeded at 100,000 per well cell density in a 24-well plate. The medium was replaced the next day with fresh DMEM containing 167 nM JF635 dye (Promega, #HT1050). Forty-eight hours postseeding, flow cytometry analysis was performed using NovoCyte 3000 VYB directly or after indicated treatments. The relative NAD⁺ levels were calculated using the geometric mean of FRET/EGFP for 10000 fluorescent cells analyzed using FlowJo. FRET was calculated with an excitation of 488 nm and an emission of 660 nm ± 20 nm, and EGFP fluorescence was measured at an excitation of 488 nm and an emission of 530 nm ± 30 nm.

ATP measurements using iATPSnFR2

Cells expressing the ^mitomCardinal-iATPSnFR2 or ^cytomCardinal-iATPSnFR2 biosensor were seeded at 200,000/ml of cell density in a 12- or 24-well plate. Twenty-four hours postseeding, flow cytometry analysis was performed using NovoCyte 3000 VYB directly or after indicated treatments. The relative ATP levels were calculated using the geometric mean of sensor/mCardinal for 10,000 fluorescent cells analyzed using FlowJo. The sensor fluorescence was normalized to measurements for cpSFGFP/mCardinal conducted in parallel. Sensor fluorescence was measured with an excitation of 488 nm and an emission of 530 nm ± 30 nm, and mCardinal fluorescence was measured at an excitation of 561 nm and an emission 660 nm ± 20 nm.

Fluorometry for iATPSnFR2 in HEK293T lysates

A total of 500,000 HEK293T cells seeded in six-well plate were transfected 24 hours later with NES-mCardinal-iATPSnFR2 or Mito-mCardinal-iATPSnFR2. Transfection mixture was prepared by incubating 2 μg of DNA with 10 μg of PEI MAX (molecular weight, 40,000; catalog #24765-1) diluted in Opti-MEM. Forty-eight hours posttransfection, cells were collected and lysed in 250 μl of lysis buffer [1% NP-40, 100 mM tris, 150 mM NaCl, and 1× protease inhibitor (pH 7.4)]. Lysate was centrifuged at 13,000g for 30 min at 4°C. Cleared lysate was diluted in assay buffer [100 mM tris, 150 mM NaCl, and 1× protease inhibitor (pH 7.4)]. Horiba fluorimeter was used to perform excitation scans of the cell lysate mixed with indicated ATP concentrations. The fluorescence shown is the average of three neighboring wavelengths at excitation peak for both sensor and mCardinal.

ΔΨm estimations

HEK293 cells stably expressing different SLC25A51 variants were collected in prewarmed medium and incubated at 37°C and 5% CO₂ with DMEM containing 200 nM MitoTracker Red CMXRos (Thermo Fisher Scientific, #M7512) for 15 or 30 min. To measure the changes in membrane potential with different mitochondrial modulators, MitoTracker Red CMXRos was added to the cells for the last 30 min of the treatment accounting for the drug concentration. Mean fluorescence intensity was measured for 10,000 cells using flow cytometry with NovoCyte 3000 VYB (excitation, 561 nm; emission, 615 nm ± 20 nm).

Structural modeling

The c-state structural models for human SLC25A51 and Ndt1p were generated using AlphaFold2, and the m-state model for SLC25A51 was generated using SWISS-MODEL with ttAAC (Protein Data Bank ID: 6gci) as template (26, 72). The model for ChemoG-NAD⁺ in NAD⁺-bound form was generated using AlphaFold3 (41). PyMOL was used for the visualization, analysis, and generation of structural graphics.

Multiple sequence alignment and analysis of SLC25A51 homologs

We retrieved the top 500 results from a search of SLC25A51 homologs on UniProt (release January 2025) (73) and eliminated sequences of less than 269 amino acids. The cutoff was selected because the shortest human MCF carrier is 275 amino acids and the smallest functional length tested for the human SLC25A51 (29 to 297 amino acid) is 269 amino acids. We manually reviewed each sequence and removed results that were identified to be AACs or yeast Ndt1 and Ndt2 carriers. We were left with 331 distinct SLC25A51 homologs of length indicating a functional protein. Clustal Omega was used to perform the multiple sequence alignment, which was analyzed using Unipro UGENE (74, 75). We used ggseqlogo for visualization of the amino acid conservation of binding site plane residues in SLC25A51 homologs (76).

Phylogenetic analysis for SLC25A51

We selected 20 SLC25A51 homologs and human SLC25A52 covering major phyla from the animalia kingdom including phyla of Porifera, Chordata, Nematoda, Arthropoda, Placozoa, Ctenophora, Rotifera, Mollusca, and Cnidaria. The organism and the SLC25A51 homolog identifier included in the analysis are as follows: Paramacrobiotus metropolitanus (UniRef #UPI002445FEA5), Mnemiopsis leidyi (UniParc #UPI003B49CAF3), Rotaria sordida (UniProt #A0A818THJ4), T. zimbabwensis (UniProt #A0A0V1HNT6), Pinworm (UniProt #A0A0N4VJM6), Steinernema carpocapsae (UniProt #A0A4U5NDW0), Drosophila (UniProt #Q7K483), Caerostris extrusa (UniProt #A0AAV4VCX6), Geodia barretti (UniProt #A0AA35TZ29), Toxolasma texasiensis (UniParc #UPI00297BE9D6), Trichoplax adhaerens (UniProt #B3S9X7), Stylophora pistillata (UniProt #A0A2B4RDE6), zebrafish (UniProt #E7F6T6), bovine (UniProt #A5PJY3), human (UniProt #Q9H1U9), human SLC25A52 (UniProt #Q3SY17), rat (UniProt #Q52KK3), mouse (UniPort #Q5HZI9), Xenopus (UniProt #A9UMI0), chicken (UniProt #A0A8V0X9Q0), and snake (UniProt #A0A6P9B6W2). The multiple sequence alignment for these 21 sequences was performed using Clustal Omega (74). The sequence alignment was used to generate a phylogenetic tree using MEGA12 software using maximum likelihood method with Jones-Taylor-Thornton substitution model. The tree estimation was done using bootstrapping with 500 replicates (77).

Cell proliferation assay

Cells were seeded in 12-well plates at a density of 10,000 per well for glucose or 20,000 per well for galactose-supplemented DMEM [glucose or galactose (4.5 g/liter) with 1 mM sodium pyruvate, 4 mM l-glutamine, 10% fetal bovine serum, and 25 mM Hepes buffer (pH 7.4)]. Cell count was determined daily using NovoCyte 3000 VYB with gating excluding the cell debris. The proliferation rate was calculated using the formula

$$\text{Proliferation rate}=\left[{\text{log}}_2(\text{count}\text{on}\text{day}4/\text{count} \text{on}\text{day}0)\right]/4$$

Expansion microscopy

HEK293 SLC25A51 KO cells stably expressing Flag-HA-SLC25A51 variants were fixed for 15 min at room temperature with 3% formaldehyde and 0.1% glutaraldehyde (Electron Microscopy Sciences, catalog nos. 15710 and 16019, respectively) in 1× phosphate-buffered saline (PBS). Coverslips were rinsed three times with 1× PBS for cell expansion by Panluminate Inc. as previously described (42). Briefly, cell coverslips were incubated in a solution of acrylamide and formaldehyde. After fixation, the cells were embedded in the expansion gel solution and placed in Milli-Q water for expansion. Gels were then reembedded, and the process was repeated until cells reached a linear expansion factor of approximately 19-fold. Expanded cells were pan-stained to visualize proteins indiscriminately (Atto 488 NHS ester, Sigma-Aldrich, catalog no. 41698) using a 488-nm excitation wavelength. Anti-HA antibody (Millipore, catalog no. H6908) was used to stain Flag-HA-SLC25A51 variants. Anti-HA antibody was visualized with goat anti-rabbit IgG (H+L) CF640R (Biotium, catalog no. 20202) with excitation at 638 nm. Images were acquired using an Andor BC43 spinning disk confocal microscope with a CFI Apochromat LWD Lambda S 40×/1.15 water immersion objective.

Live-cell imaging

Imaging was done in an environmentally controlled OKO system using an inverted Olympus IX83 confocal microscope fitted with a spinning disk CSU-W1 confocal scanner, Hamamatsu ORCA-Fusion camera, and a UAPO 100× total internal reflection fluorescence objective, numerical aperture of 1.49, WD 0.1 mm, W/CC objective. Cells were incubated for 30 min with 200 nM MitoTracker Red CMXRos, which localizes specifically to mitochondria. Multichannel imaging was done using 100-mW excitation lasers with excitations of 488 nm and 561 nm and emissions of 525 nm ± 50 nm and 630 nm ± 75 nm respectively to simultaneously image sensors and MitoTracker CMXRos dye. Image analysis was done using ImageJ software.

Recombinant E. coli uptake analysis

We used a previously described uptake assay for SLC25A51 (amino acids 29 to 297) (15). In summary, pMW7 expression plasmid harboring yellow fluorescent protein (YFP), SLC25A51 (amino acids 29 to 297), or Flag-HA-ScNdt1gene was transformed into BL21 DE3 pLysS E. coli strain. A colony for each strain was picked and grown overnight in LB medium. The cultures were diluted to optical density at 600 nm (OD₆₀₀) of 0.5 and induced with 1 mM isopropyl-β-d-thiogalactopyranoside for 1 hour. Cells were pipetted and centrifuged to obtain a cell pellet equivalent to OD₆₀₀ of 10 in 25 μl. The cells were resuspended in 25 μl of uptake buffer [120 mM KCl, 5 mM KH₂PO₄, 2 mM Hepes-NaOH (pH 7.4), and 1 mM EGTA] containing either DMSO or 10 μM FCCP and incubated for 5 min. The uptake was initiated by adding 25 μl of uptake buffer with the final reaction concentration of 100 μM of unlabeled NAD⁺ traced with 3.33 nM of ³²P-NAD⁺ in presence of either DMSO or 10 μM FCCP. ³²P-NAD⁺ was synthesized using previously described method from ³²P-ATP (78). The uptake reaction was stopped by diluting the reaction 20× with ice-cold uptake buffer and spinning down the cells for 2 min at 4000g. The supernatant was discarded, and 1 ml of ice-cold uptake buffer was added to each tube. A vacuum pump with a filtration assembly was used to pass the samples through 0.22-μm mixed cellulose ester filter and washed two times with 4 ml of uptake buffer. The filters were organized on a support and exposed to Bio-Rad phosphor imaging screen for 18 hours. The phosphor imaging screen was imaged with Typhoon 9500, and the image was analyzed using ImageJ. The exposure dots for YFP were subtracted as background from the samples.

Quantification and statistical analysis

Statistical analyses were performed using GraphPad Prism v10.6.0. The data are shown either as means ± SD or box and whisker plot with hinges at 25th/75th percentile, whiskers are minimum and maximum, and the line represents median. The number of replicates are described in the figure legend. Unpaired two-sided Student’s t test was used to calculate statistical significance between two groups, one-way analysis of variance (ANOVA) with post hoc Dunnett’s or Tukey’s test for comparing three or more groups and two-way ANOVA with more than two groups and variables. P values under 0.05 were considered significant and *P < 0.05, **P < 0.01, and ***P < 0.001.

Supplementary Materials

The PDF file includes:

Figs. S1 to S7

Legend for data S1

Legends for movies S1 and S2

Other Supplementary Material for this manuscript includes the following:

Data S1

Movies S1 and S2