Intrinsic oxidative phosphorylation limitations underlie cellular bioenergetics in leukemia

Margaret Nelson East Carolina University Kelsey McLaughlin East Carolina University James Hagen East Carolina University Hannah Coalson East Carolina University Cameron Schmidt East Carolina University Kimberly Kew East Carolina University Joseph McClung East Carolina University P. Neufer East Carolina University https://orcid.org/0000-0002-2256-6051 Patricia Brophy East Carolina University Nasreen Vohra East Carolina University Darla Liles East Carolina University Myles Cabot East Carolina University Kelsey Fisher-Wellman (  sherwellmank17@ecu.edu ) East Carolina University https://orcid.org/0000-0002-0300-829X


Introduction
Although all mitochondria make ATP, the e ciency of this process varies widely across the human body's >200 distinct cell types [1][2][3][4] . Tissue-speci c differences in mitochondrial function (i.e., mitochondrial specialization) are operationally de ned through differences in mitochondrial protein expression, as well as an ever-growing list of post-translational modi cations (PTMs) [5][6][7][8] . Regardless of the mechanism(s), the biological reward of such specialization is the alignment of bioenergetic e ciency with organ physiology (i.e., establishment of bioenergetic delity). In the case of leukemia, as with most cancers, bioenergetic delity becomes misaligned from the host, resulting in uncontrolled proliferation of neoplastic progenitors. While this misalignment is in part associated with increased glucose uptake/glycolytic ux, a growing body of evidence is emerging that links multiple aspects of leukemia biology (e.g., tumorigenesis, chemoresistance) to altered mitochondrial quantity and quality [9][10][11][12][13][14] .
In comparison to normal hematopoietic cells, various human leukemias present with increased mitochondrial mass and higher basal respiration rates [10][11][12]15,16 , the latter of which appears to sensitize them to respiratory inhibition 9,13,17,18 . Although these studies have ignited interest in mitochondrialtargeted chemotherapeutics 19,20 , experimental rationale for targeting oxidative phosphorylation (OXPHOS) in leukemia is largely based on the assumption that heightened respiration is representative of the cancerous mitochondrial network's attempt to accommodate an increased ATP demand. However, identical increases in mitochondrial respiration can derive from any number of physiological stimuli, ranging from increased demand for ATP resynthesis to decreased OXPHOS e ciency. Distinguishing between these potential outcomes is critical, as such insight demarcates the difference between e cacious targeted drug delivery and undesirable systemic toxicity. For example, it is currently unclear how targeting 'increased OXPHOS reliance' in leukemia can speci cally disrupt leukemic oxidative metabolism without impacting OXPHOS in other highly metabolic organs (e.g., brain, heart, muscle).
Given the ubiquitous necessity of OXPHOS for healthy cellular metabolism, a major barrier to mitochondrial-targeted drugs in leukemia relates to the need for cancer-cell selectivity 21,22 . The current project was based on the premise that establishing cause and effect between mitochondrial bioenergetics and cancer is one of the keys to developing targeted and more effective therapies. To do this requires advanced technical approaches capable of quantifying the interplay among the major mitochondrial thermodynamic free energy driving forces to distinguish between changes in bioenergetic demand versus e ciency. To this end, our group recently developed a diagnostic biochemical work ow that quanti es the changes in free energy driving forces over the entire range of respiratory demand, thus providing a comprehensive pro le of mitochondrial bioenergetic e ciency and capacity, relative to the underlying proteome [23][24][25] . Herein, we leveraged this work ow across several acute leukemia cell lines and primary human leukemias.
Compared to healthy controls, intrinsic limitations to mitochondrial OXPHOS were found to characterize the mitochondrial network in leukemia. Parallel assessment of the underlying mitochondrial proteome linked leukemia-speci c OXPHOS de ciency to shifts in adenine nucleotide translocase (ANT) isoform expression and increased abundance of TNF receptor associated protein 1 (TRAP1). Speci cally, decreased ANT1 and increased ANT2 in leukemia facilitated matrix ATP uptake, rather than export, that, in the presence of functional TRAP1, directly constrained the ability of mitochondrial OXPHOS to contribute to the cellular ATP free energy (ΔG ATP ) charge. These ndings are consistent with recent evidence demonstrating that mitochondrial OXPHOS is dispensable for tumor growth 26 , and raise the intriguing possibility that the requirement for mitochondria in leukemia may have little to do with oxidative ATP production, but instead re ect a universal requirement for continuous mitochondrial ux to support other cellular functions (e.g., metabolite export, nucleotide synthesis). Taken together, the ndings provide proof-of-principle that pharmaceutical intervention designed to restore, rather than disrupt, OXPHOS may impart therapeutic e cacy across various hematological malignancies. Given that increased OXPHOS e ciency is advantageous across non-cancerous tissues, the obvious bene t to this novel treatment paradigm is the elimination of secondary toxicity (i.e., a wide therapeutic window).

Results
Mitochondrial bioenergetic pro ling of acute leukemia reveals respiratory ux limitations. To begin to characterize the mitochondrial network in leukemia, we selected three commercially available acute leukemia cell lines -HL-60, KG-1, MV-4-11 -and comprehensively evaluated their bioenergetic pro les.
These cells arise from unique precursors along the hematopoietic lineage, express a diverse array of cell surface markers, and have distinct underlying genetics [27][28][29] . Results were compared to peripheral blood mononuclear cells (PBMC) isolated from healthy volunteers. The utilization of distinct leukemic cell lines, as well as multiple PBMC controls, was intended to provide the experimental design contrast needed to identify intrinsic mitochondrial bioenergetic signatures potentially necessary for leukemia survival.
Using intact PBMC and leukemia cell lines, respiratory ux (JO 2 ) was assessed under basal conditions, as well as in response to ATP synthase inhibition (oligomycin), and FCCP titration (i.e., mitochondrial uncoupler). Following FCCP titration, respiration was inhibited with a combination of rotenone (inhibits complex I) and antimycin A (inhibits complex III). Consistent with prior work in human leukemia 10,30 , basal respiration normalized to cell count was elevated above PBMC across all leukemia lines and maximal respiratory ux was higher in KG-1 and MV-4-11 (Fig. 1A). When normalized to basal respiration, oligomycin similarly inhibited respiration across groups and the fold change induced by FCCP was consistently blunted in leukemia (Fig. 1B).
Given the large differences in cell size between PBMC and leukemia (Supp. Fig. 1A), we reasoned that normalization to total protein would likely provide the most accurate index of absolute respiratory kinetics across groups. Interestingly, upon normalization to total protein, although basal respiration remained higher in KG-1 and MV-4-11, differences in maximal respiratory ux were eliminated, particularly at higher FCCP concentrations (Fig. 1C). In fact, maximal FCCP-supported JO 2 was nearly two-fold lower in HL-60 compared to PBMC when normalized to total protein (Fig. 1C). Relative to PBMC, maximal respiration induced by FCCP occurred at much lower concentrations in leukemia (i.e., lower Km; Fig. 1D To determine if ux differences in leukemia could be explained by differences in mitochondrial content, nuclear and mitochondrial volumes were assessed independently by tetramethylrhodamine methyl ester (TMRM) or MitoTracker uorescence and confocal microscopy ( Fig. 1E-I, Supp. Fig 1D). Absolute nuclear and mitochondrial volumes were higher in all leukemia lines ( Fig. 1E-H), consistent with leukemia's larger cell size (Supp. Fig. 1A). However, when normalized to nuclear volume, mitochondrial content was elevated above PBMC only in HL-60 and MV-4-11 (Fig. 1I). Interestingly, across all cell types, considerable discrepancies were apparent when protein-normalized maximal respiratory ux (Fig. 1C) was compared to mitochondrial content (Fig. 1I). This was particularly evident in HL-60 cells where mean maximal respiration, relative to the size of the underlying mitochondrial network, was ~5-fold lower compared to PBMC (Fig. 1J). Together, these data suggested that respiratory ux may be partially constrained across the mitochondrial network in human leukemia, potentially indicative of overt bioenergetic ine ciency.
Intrinsic limitations to OXPHOS kinetics characterize the mitochondrial network in leukemia cell lines. To directly test OXPHOS kinetics in leukemia, two complementary assays were designed, both of which used digitonin-permeabilized cells. In the rst assay, the maximal capacity of the electron transport system (ETS) was assessed by energizing permeabilized cells with saturating carbon substrates (Pyr/M/Oct/Glut/Succ; 'MULTI') and titrating in FCCP ( Fig. 2A-B). The use of multiple substrates was intended to fully saturate the 'fuel' node such that maximal ETS ux could be quanti ed. Using this approach, absolute respiration in substrate-replete permeabilized cells was comparable to that observed using intact cells treated with FCCP ( Fig. 2C), con rming maximal ETS ux in the permeabilized system.
Note, maximum FCCP-supported ux under these conditions is indicated throughout as JH + Total ( Fig. 2A). Relative to PBMC and similar to that observed in intact cells, JH + Total was lower in HL-60, unchanged in KG-1, and higher in MV-4-11 (Fig. 2B).
In mammalian cells, the vast majority of the adenylate pool is represented by ATP (i.e., ΔG ATP ), with typical values for ATP free energy ranging from -56 to -64 kJ/mol [31][32][33] . Thus, to evaluate OXPHOS kinetic e ciency in leukemia mitochondria across a physiological range of ATP resynthesis demands, we utilized the creatine kinase (CK) energetic clamp 23,34,35 . This technique leverages the enzymatic activity of CK, which couples the interconversion of ATP and ADP to that of phosphocreatine (PCr) and free creatine (Cr) such that extramitochondrial ATP free energy (i.e., ΔG ATP ) can be empirically titrated using PCr. Using this approach, permeabilized cells were energized with the same carbon substrate mix used for the ETS capacity assay and respiration was stimulated at minimal ATP free energy. Note, ΔG ATP equal to -54.16 kJ/mol re ects an ATP/ADP ratio in vivo that would be expected to induce 'maximal' OXPHOS ux and is thus referred to throughout as 'JH + OXPHOS ' (Fig. 2D). Cytochrome C (Cyt C) was added to assess the integrity of the mitochondrial outer-membrane, and ΔG ATP was then titrated via sequential additions of PCr. With respect to the ETS capacity assay, respiration stimulated by ΔG ATP partially normalized JO 2 between MV-4-11 and PBMC and revealed decreased respiratory kinetics in both HL-60 and KG-1 (Fig. 2E), indicating substantial OXPHOS limitations in leukemia.
In both assays, the utilization of identical substrates ('Pyr/M/Oct/Glut/Succ') allowed us to directly quantitate absolute OXPHOS kinetics ('JH + OXPHOS '), relative to the maximal capacity of the electron transport system, ('JH + Total '). Together, JH + OXPHOS and JH + Total provide a quantitative index of fractional OXPHOS capacity as the ratio of the two re ects the proportion of the entire respiratory system that can be used for OXPHOS (Fig. 2F). A ratio of '1' re ects maximal OXPHOS reliance, whereas a ratio of '0' indicates that the mitochondrial proton current cannot be utilized for ATP synthesis. Strikingly, calculated fractional OXPHOS in leukemic mitochondria was consistently decreased compared to PBMC, corresponding to a factor of ~0.5 (Fig. 2G), indicating that only half of the available ETS capacity in leukemia can be dedicated to OXPHOS under physiological ATP-free energy constraints. Given that the OXPHOS network is responsible for driving ATP/ADP disequilibrium to establish cellular ΔG ATP , low fractional OXPHOS was interpreted to re ect reduced bioenergetic e ciency in leukemia. Moreover, such ndings indicate that traditional measurements of 'OXPHOS' capacity using intact cells woefully underestimate true OXPHOS kinetics.
To differentiate between bioenergetic signatures inherent to proliferating cells and those which are unique to leukemia, experiments were repeated in primary human muscle precursor cells (human myoblasts -'HMB' Supp. Fig. 2A-B). These cells were cultured from muscle biopsies uniformly collected from the gastrocnemius muscle (10 cm distal to the tibial tuberosity) of healthy human subjects and were intended to serve as a non-cancerous human progenitor control. Importantly, fractional OXPHOS was elevated above leukemia in human muscle progenitor cells (Fig. 2G: 'HMB'), indicating that decreased bioenergetic e ciency is not an absolute requirement of cellular proliferation, but rather a unique bioenergetic feature of leukemic mitochondria.
Exposure to physiological ΔG ATP reveals direct inhibition of ETS ux by ATP in leukemic mitochondria. To gain insight into the mechanism of OXPHOS limitations in leukemia, at the end of the ΔG ATP titration, oligomycin was added to inhibit ATP synthesis and maximal uncoupled respiration was stimulated with FCCP titration. Maximal FCCP-supported ux under these conditions is denoted as 'FCCP ΔGATP ' (Fig. 2D).
In the intact cell assay (Fig. 1C), increased glycolytic ux is presumed to maintain cellular ΔG ATP during FCCP titration. Thus, the continued presence of extra-mitochondrial ΔG ATP in our permeabilized cell system was intended to model the adenylate constraints present in intact cells. By comparing maximal OXPHOS ux ('JH + OXPHOS ') to maximum FCCP-stimulated respiration in the presence of ΔG ATP ('FCCP ΔGATP '), it becomes possible to quantitate any ux limitations imposed by physiological ATP/ADP. Importantly, ATP synthase is not functional during the assay, thus any ux limitations imposed by ΔG ATP would be interpreted to re ect direct ETS regulation. In PBMC and human muscle progenitor cells, the addition of FCCP at the end of the ΔG ATP titration restored respiration to levels obtained under low (-54.16 kJ/mol) ATP free energy (Fig. 2E, Supp. Fig. 2A-B), indicating minimal ETS ux inhibition by ΔG ATP .
Surprisingly, relative to JH + OXPHOS , FCCP-stimulated respiration was substantially blunted in the presence of high ATP free energy across all three leukemia cell lines (Fig. 2E), resulting in a 2-fold difference in the FCCP ΔGATP /JH + OXPHOS ratio, termed 'FCCP Effect' throughout ( Fig. 2H). Importantly, in the absence of ATP, the addition of CK and PCr up to 21mM did not impact FCCP-supported ux in permeabilized MV-4-11 cells (Supp. Fig. 2C), con rming that ATP free energy was required to induce ETS ux inhibition in leukemia. To determine the sensitivity of ETS inhibition by ΔG ATP , FCCP-supported ux in MV-4-11 cells was assessed at de ned ATP free energies. In these experiments, extramitochondrial ΔG ATP was administered after CV inhibition with oligomycin, followed by FCCP titration. Results revealed a dosedependent decrease in uncoupled respiration in response to increasing ΔG ATP (Fig. 3A).
Inhibition of respiratory ux mediated by ΔG ATP could re ect a number of potential mechanisms ranging from cytoskeletal alterations, direct inhibition of the matrix dehydrogenase network (e.g., inhibitory phosphorylation of pyruvate dehydrogenase), and/or ETS inhibition 23 . To differentiate between these potential outcomes, mitochondria were isolated from PBMC and each of the three leukemia cell lines and similarly assessed for OXPHOS kinetics. In mitochondria energized with saturating carbon, increasing ΔG ATP led to a more pronounced decrease in respiration in mitochondria of all three leukemia cell lines.
The ability of FCCP to restore maximal respiratory ux was also once again blunted in leukemia mitochondria ( Fig. 3B-C), consistent with lower fractional OXPHOS ( Fig 2G). This inhibitory effect of ATP free energy on JH + Total was present in both isolated leukemia cell mitochondria and permeabilized leukemia cells, ruling out any involvement of the cytoskeleton. Together these ndings demonstrate that mitochondrial ux inhibition by ATP free energy is an intrinsic bioenergetic feature of leukemic mitochondria.
To differentiate between respiratory ux inhibition localized to the matrix dehydrogenases or the ETS, NADH/NAD + redox poise was measured in parallel in substrate-replete isolated mitochondria exposed to an identical ΔG ATP span. Results are depicted as a percentage of complete reduction, where 0% reduction re ects isolated mitochondria at 37°C without added substrates and 100% reduction is recorded at the end of the assay with the addition of cyanide. Except for a slight hyper-reduction in HL-60 mitochondria, NADH/NAD + redox was similar across groups (Supp. Fig. 2D), indicating that ATP-mediated respiratory ux inhibition in leukemia is not due to a generalized impairment in dehydrogenase ux. Having eliminated the cytoskeleton and the dehydrogenase network as potential sites of inhibition, we next turned our attention to the ETS. To determine if ux inhibition induced by ΔG ATP was speci c to a given respiratory complex, OXPHOS kinetics was assessed in isolated mitochondria energized with either complex I (CI)-or CII-linked substrate combinations. Note, the presence of saturating malate in the CI substrate mix results in CII inhibition via malate-fumarate equilibration (Fig. 3D). Likewise, the addition of rotenone in the presence of succinate eliminates residual CI-supported ux by downstream products of succinate oxidation (Fig. 3D). Using either CI-or CII-linked substrates, we once again observed a more pronounced decrease in respiration in response to ΔG ATP titration in leukemia mitochondria, as well as a stark inability of FCCP to restore maximal respiratory ux ( Fig. 3E-H). Taken together, these ndings indicate that ΔG ATP -mediated respiratory ux limitations are speci c to leukemia mitochondria, independent of carbon substrate source, and thus most likely attributable to ETS inhibition, potentially downstream of CI or CII.
Subcellular proteomics reveals unique isoform expression of the adenine nucleotide translocase (ANT) in leukemia. To identify potential protein mediators responsible for reduced bioenergetic e ciency in leukemia, we conducted a proteomics screen using TMT-labeled peptides prepared from the same isolated mitochondria samples used for functional characterization. To control for group differences in percent mitochondrial enrichment, nLC-MS/MS raw data were searched using the MitoCarta 2.0 database, as previously described 36 . Using this approach, total mitochondrial protein abundance was similar between groups (Supp. Table 1), thus allowing for intrinsic mitochondrial signatures to be identi ed across leukemia. In total, 135 differentially expressed mitochondrial proteins (adjusted P value < 0.01) were identi ed comparing PBMC to each of the three leukemia lines (Fig. 4A). For pair-wise comparisons of mitochondrial protein expression between PBMC and each of the three leukemia cell lines, see Supp. Table 1. With respect to the shared differentially expressed proteins, several of these proteins have previously been implicated in cancer biology, such as decreased MAOB 37 and HK1 38 , and increased MTHFD1L 39 , and COX17 40 (Fig. 4A).
Focusing on the OXPHOS proteome, we assessed the abundance of the individual protein subunits that comprise CI, CII, CIII, and CIV, as well as the protein components of the phosphorylation system which include ATP synthase (CV), the phosphate carrier (SLC25A3), and ANT (Fig. 4B). Although considerable heterogeneity was present across groups, comparing protein expression pro les of the individual subunits that comprise CI-CV and SLC25A3 (Fig. 4C, Supp. Fig. 3A-D) revealed that only 6 of the 110 subunits were similarly altered in leukemia (Supp. Table 1). With the exception of COX6A1, all protein subunits were involved in the assembly of CI (NDUFB10), CIV (COA4, COA7, COX17) or CV (ATPAF2). In stark contrast, the expression pro les of the three main ANT isoforms were entirely distinct between PBMC and leukemia mitochondria, highlighted by reduced ANT1 (SLC25A4) and increased ANT2 (SLC25A5) and ANT3 (SLC25A6) in leukemia (Fig. 4D).
Inhibition of ETS ux by ΔG ATP requires ANT. Given that ATP free energy was required to induce ETS ux inhibition, we hypothesized that this effect may be mediated by ATP transport into the matrix, facilitated by enhanced ANT2 expression in leukemia 41 . To test this hypothesis, FCCP-supported respiration was assessed in energized, permeabilized MV-4-11 and HL-60 cells exposed to ΔG ATP of -61.49 kJ/mol in the absence and presence of the ANT inhibitor carboxyatractyloside (CAT) 42 . Consistent with our prior ndings, the addition of FCCP in the presence of ΔG ATP was incapable of restoring ux to levels obtained with minimal ATP free energy in leukemia ( Fig. 5A-B; 'ΔG ATP (-61.49 kJ/mol)'). However, relative to no adenylates, as well as minimal ΔG ATP (e.g., -54.16 kJ/mol), the addition of CAT restored maximal FCCPsupported ETS ux in the presence of high ATP free energy ( Fig. 5A-C). Similar experiments performed in MV-4-11 isolated mitochondria (Fig. 5D), as well as with the ANT inhibitor bongkrekic acid (Supp. Fig. 4C) revealed nearly identical results, con rming that ETS ux inhibition by ΔG ATP requires functional ANT.
Together, these ndings reveal an intrinsic bioenergetic phenotype, speci c to leukemia, whereby extramitochondrial ATP gains access to the matrix space, presumably via increased ANT2, and directly constrains ETS ux across a physiological ΔG ATP span.
Reduced mitochondrial bioenergetic e ciency is a common feature of human leukemia. To determine if the bioenergetic phenotypes present in leukemia cell lines translated to the clinic, we recruited patients diagnosed with leukemia and comprehensively evaluated mitochondrial bioenergetic function in mononuclear cells isolated from bone marrow aspirates. Although all patients had con rmed leukemia at the time of sample acquisition, the type of leukemia ranged from acute myeloid leukemia (N=5), chronic myeloid leukemia (CML, N=2), and granular lymphocytic leukemia (N=1). Biochemical results were compared to PBMC isolated from age-matched participants without a prior history of leukemia. Mitochondrial respiration rates in substrate-replete permeabilized cells in the absence of adenylates were identical between groups (Fig. 6A). In contrast, respiration stimulated by ΔG ATP revealed decreased OXPHOS kinetics in primary leukemic cells (Fig. 6B). In agreement with impaired OXPHOS ux in leukemia, calculated fractional OXPHOS was reduced in primary leukemia (Fig. 6C), entirely consistent with results observed across the leukemia cell lines (Fig. 2G). To determine if ΔG ATP was capable of directly limiting ETS ux via ANT-mediated ATP uptake, FCCP titration was performed using substratereplete permeabilized cells in the absence (Fig. 6B) and presence (Fig. 6D) of carboxyatractyloside. In the absence of ANT inhibition, the presence of ΔG ATP restricted FCCP-supported ux speci cally in leukemic cells (Fig. 6B, quanti ed in Fig 6E; '-CAT'). Despite the continued presence of ΔG ATP , the addition of CAT restored FCCP-supported ux in leukemic cells to that of PBMC (Fig. 6D, quanti ed in Fig 6E), con rming direct ETS ux inhibition by ATP in primary leukemia. Taken together, these ndings from primary human leukemia cells corroborate the results across the three acute leukemia cell lines and suggest that speci c impairments in OXPHOS kinetics is a hallmark characteristic of the mitochondrial network in human leukemia.
Inhibition of ETS ux via extramitochondrial ΔG ATP requires functional TRAP1. Having established that extramitochondrial ΔG ATP must gain access to the matrix space to inhibit ETS ux in leukemia, we next set out to elucidate the potential protein mediator(s) of this effect. To do this, we searched our proteomics dataset for mitochondrial proteins with known kinase and/or ATPase function that were substantially upregulated across all three leukemia lines and identi ed mitochondrial TRAP1 (Fig. 7A). TRAP1 is the mitochondrial paralog of the heat shock protein 90 (HSP90) family and is widely recognized as a potential anti-cancer drug target across multiple human malignancies, including leukemia [43][44][45][46][47][48][49] . Given that ATPase activity is required for TRAP1 function 50 , we hypothesized that ETS inhibition in leukemic mitochondria exposed to ΔG ATP may be driven by acute activation of TRAP1. To test this hypothesis, OXPHOS kinetics were assessed in permeabilized MV-4-11 cells in the absence and presence of the TRAP1 inhibitor 17-AAG 51 . In substrate-replete permeabilized MV-4-11 cells, the presence of 17-AAG had no impact on maximal FCCP-supported respiration (Supp. Fig. 4A), indicating that TRAP1 does not impinge on ETS ux in the absence of physiological adenylates. Remarkably, the presence of 17-AAG increased JH + OXPHOS and calculated fractional OXPHOS relative to vehicle control and completely restored FCCP-supported respiration in the presence of ATP free energy (Fig. 7B-D). Similar results were observed using permeabilized HL-60 cells (Supp. Fig. 4B, Fig. 7C-D), as well as using the mitochondrialtargeted TRAP1 inhibitor Gamitrinib TPP hexa uorophosphate (Supp. Fig. 4C).
Given that functional TRAP1 was apparently required for ATP-mediated respiratory inhibition in leukemia, we next sought to determine the sensitivity of TRAP1 to physiological ΔG ATP . Related to this, although ATP is widely understood to be the universal energy currency in cells, it is critical to consider that ATP alone has minimal bio-synthetic power; rather, its utilization as a common energy currency is solely a function of the remarkable displacement of the molecule from equilibrium (~10 orders of magnitude) 52 . This means that biological processes driven by ATP hydrolysis, such as those carried out by TRAP1, are presumably fueled by ATP free energy, rather than ATP levels per se. The primary advantage of the CK clamp technique is that it allows for mitochondrial bioenergetics to be evaluated across a physiological ΔG ATP span without appreciable changes in ATP concentration (Supp. Fig. 4D). Thus, we reasoned that the CK clamp could be utilized to assess TRAP1 sensitivity to ΔG ATP under conditions in which free [ATP] is not rate-limiting. To do this, we assessed the ability of 17-AAG to impact FCCP-supported respiration across a physiological ΔG ATP span. For contrast, we compared the respiratory impact of 17-AAG to that of the commonly used ETC inhibitor antimycin A. Although antimycin A decreased respiratory ux in permeabilized MV-4-11 cells, percent inhibition by the compound was largely insensitive to ΔG ATP , consistent with direct respiratory complex inhibition (Supp. Fig. 4E). In contrast, across all leukemia cell lines, respiratory inhibition induced by functional TRAP1 was exquisitely sensitive to ΔG ATP (Fig. 7E-G), indicating that the ability of TRAP1 to impinge on ETS ux depends entirely upon the prevailing matrix ATP free energy. Given that we observed similar maximal respiratory capacities using both intact and permeabilized cells (Fig. 2C), these data inform a model whereby intrinsic OXPHOS limitations in leukemia serve to chronically constrain the mitochondria's ability to drive ATP/ADP disequilibrium, thereby lowering cellular ΔG ATP charge and stimulating compensatory rapid and continuous metabolic ux (Fig. 7H).

Discussion
Increased mitochondrial oxidative metabolism, an established metabolic hallmark of leukemia 10,11,13,17,53,54 , has been historically interpreted to re ect an increased reliance on mitochondrial ATP production. However, direct evaluation of fractional OXPHOS kinetics had not been empirically evaluated in leukemia at the onset of this project. Thus, it remained to be determined whether higher basal respiration in leukemia re ected accelerated demand for ATP regeneration or intrinsic bioenergetic ine ciency. Both conditions would be expected to similarly restrict cellular ATP/ADP equilibrium displacement (i.e., ΔG ATP charge) and thus potentially result in identical respiratory pro les in intact cells.
Applying a diagnostic biochemical work ow, we determined that intrinsic limitations in fractional OXPHOS characterize the mitochondrial network across human leukemias. In fact, approximately half of the leukemic mitochondrial network is incapable of contributing to oxidative ATP production across a physiological ΔG ATP span. Subsequent experiments linked intrinsic OXPHOS limitations in leukemia to a unique biochemical mechanism in which extra-mitochondrial ATP gained access to the matrix space, where it then directly inhibited electron transport ux in a ΔG ATP -dependent manner. Restoration of mitochondrial bioenergetic e ciency was observed upon administration of 17-AAG, unveiling a critical role for the pro-neoplastic protein TRAP1 in driving reduced fractional OXPHOS in leukemia.
Although prior work has implicated ANT2 42,55,56 and TRAP1 50 as potential anti-cancer targets in leukemia, we present for the rst time here a potential mechanism for their coordinated regulation of leukemic cell metabolism. Speci cally, our ndings inform a model of leukemia bioenergetics in which decreased ANT1 and increased ANT2 favors the uptake of extra-mitochondrial ATP into the matrix space. The transfer of ΔG ATP from the cytosol to the matrix in turn activates matrix-localized TRAP1 by providing substrate for its ATPase activity. While TRAP1 has been shown to interact with CII and CIV of the ETS 48 , the present ndings challenge the conclusion that activated TRAP1 directly inhibits ETS ux as a means of upregulating glycolytic metabolism. In fact, the exquisite sensitivity of TRAP1 to ΔG ATP in vitro indicates that TRAP1 restricts the ability of the mitochondrial network to contribute to the cellular ΔG ATP charge. In this way, chronic maintenance of low ΔG ATP maximizes both glycolytic and mitochondrial metabolism, entirely consistent with the known metabolic phenotype of human leukemia 11,12,15 . Such conditions are likely advantageous to proliferating leukemic blasts, as the increased ATP demand commensurate with pro-growth signaling 57,58 likely synergizes with OXPHOS limitations to chronically constrain cellular ΔG ATP , thereby establishing the thermodynamic metabolic 'pull' needed for rapid and continuous nutrient ow across the plasma membrane (Fig. 7H). Although ΔG ATP in proliferating blasts was not directly quanti ed here, the ATP free energy-dependent ETS inhibition by TRAP1 (Fig. 7E-G) strongly suggests a low (i.e., more positive) ΔG ATP is maintained in leukemia.
Currently, there is great interest in developing novel pharmacotherapies that target mitochondrial OXPHOS in leukemia 13,16,17,19,20 . Yet, a large caveat of targeting mitochondria is the ubiquitous necessity of OXPHOS for healthy cellular metabolism. If indeed the proliferative potential of leukemia depends upon maintenance of low fractional OXPHOS, it is tempting to speculate that a pharmaceutical intervention designed to restore normal ΔG ATP could effectively halt cell proliferation, in turn allowing for these blasts to succumb to apoptosis. Such a targeted approach would minimize secondary toxicity as increased OXPHOS e ciency is advantageous across non-cancerous, highly metabolic tissues (e.g., brain, heart, muscle), as well as in the context of adaptive cellular immunity 59,60 . Based upon our proposed model, targeting the acute ETS regulating capabilities of the pro-neoplastic protein, TRAP1, provides an appealing, leukemia speci c target for pharmaceutical intervention, as acute TRAP1 inhibition in the matrix restored OXPHOS kinetics in leukemic mitochondria. Moreover, given that TRAP1 has previously been shown to interact with matrix localized c-Src 47 , it is possible that TRAP1 activation indirectly regulates respiratory ux in leukemia by facilitating post-translation OXPHOS inhibition (i.e., inhibitory phosphorylation).
Taken together, the present ndings provide a novel mechanism to explain the enhanced glycolytic ux and altered mitochondrial metabolism previously observed in leukemia 1,2,5-7 . In particular, we provide direct evidence that leukemic blasts display an inherent limitation in OXPHOS kinetics that is mediated by a dynamic interplay between TRAP1 ATPase activity and the cellular ATP free energy. Lastly, this study establishes the utility of comprehensive mitochondrial diagnostics to inform therapeutic strategies targeting OXPHOS in leukemia and sets the stage for an entirely new research area focused on understanding and leveraging the unique biochemical characteristics intrinsic to cancerous mitochondria.

Methods
Unless otherwise stated, all reagents and chemicals were purchased from Sigma-Aldrich.

Blood collection and isolation of PBMCs
All procedures involving human subjects were approved by the Institutional Review Board of the Brody School of Medicine. Venous blood from the brachial region of the upper arm was collected from 30 healthy volunteers, ranging from 18-70 years. Whole blood was collected in 20 sodium-heparinized Cell Preparation Tubes (CPT) (BD Biosciences, Franklin Lakes, NJ) and centrifuged at 1,800 x g for 15 min.
Fractions containing peripheral blood mononuclear cells (PBMC) were collected from 4 CPT tubes and used for intact and permeabilized cell experiments while the remaining PBMC fractions were used to isolate mitochondria.

Mononuclear cell isolation from bone marrow aspirates
Bone marrow aspirates were collected from patients undergoing con rmatory diagnosis for a range of hematological malignancies. Patients with con rmed leukemia were enrolled in the study. Type of leukemia ranged from acute myeloid leukemia (AML, N=5), chronic myeloid leukemia (CML, N=2), and granular lymphocytic leukemia. Patient age ranged from 32-78 years (male/female, 3/5). Bone marrow aspirates were compared to PBMC isolated from age-matched participants without a prior history of any hematological malignancy. Peripheral blood and bone marrow aspirates were collected in sodiumheparinized Cell Preparation Tubes (CPT) (BD Biosciences, Franklin Lakes, NJ) and centrifuged at 1,800 x g for 15 min. Mononuclear cells were isolated and then washed in ammonium-chloride-potassium (ACK) lysis buffer to remove red blood cells.
Cell culture HL-60, KG-1, and MV-4-11 (ATCC, Manassas, VA) human leukemia cells were cultured in IMDM (Thermo Fisher Scienti c, Waltham, MA) supplemented with glutamine, 10% FBS, and 1% penicillin/streptomycin and incubated at 37°C in 5% CO 2 . Upon reaching an average cell density of 1.5x10 6 cell/mL the cells were harvested and used for whole cell and isolated mitochondria experiments. Primary human muscle progenitor cells (human myoblasts, 'HMB') were derived from fresh muscle biopsy samples, as described previously 61 . Cells were cultured on collagen-coated asks using HMB growth medium (GM: Ham's F10, supplemented with 20% FBS and 1% penicillin/streptomycin, and supplemented immediately prior to use with 5 ng/ml basic FGF).
Cells were held in place with a thin 1% agarose pad that was applied immediately prior to imaging in order to minimize rapid motion interference during imaging of live non-adherent cells 62 .
All imaging was performed using an Olympus FV1000 laser scanning confocal microscope (LSCM) with an onstage incubator at 37°C. Acquisition software was Olympus FluoView FSW (V4.2). The objective used was 60X oil immersion (NA=1.35, Olympus Plan Apochromat UPLSAPO60X(F)). Images were 800x800 pixel with 2μs/pixel dwell time, sequential scan mode, resulting in a 4X digital zoom. Hoechst 33342 was excited using the 405nm line of a multiline argon laser; emission was ltered using a 560nm dichroic mirror and 420-460nm barrier lter. MTG-FM was excited using the 488nm line of a multiline argon laser; emission was ltered using a 560nm dichroic mirror and 505-540nm barrier lter. TMRM was excited using a 559nm laser diode; emission was ltered using a 575-675nm barrier lter. Zero detector offset was used for all images and gain at the detectors was kept the same for all imaging. The pinhole aperture diameter was set to 105μm (1 Airy disc).
Images were analyzed using Fiji 63 . Spatial resolution was measured using sub-resolution uorescent beads (Thermo Fisher) and curve tting was performed using the MetroloJ plugin in Fiji. 16-bit images were made into a composite. Circular ROIs were manually selected using the ROI manager plugin. Images were then decomposed into separate 16-bit image stacks leaving the ROI positions intact. A Huang autothreshold was used for automated selection of signal for all three channels. Following threshold application, each signal was measured using the multi-measure feature. Only whole cells were analyzed (i.e. cells on edges of the FOV were excluded). Slices containing cells above the lowest monolayer were removed from stacks to avoid oversampling. The following calculations were performed to determine the relevant signal volumes.

Signal Volume (μm 3 ) = [A*Z]/N
Where A is the signal-positive area selected using a Huang auto-threshold (μm 2 ), Z is the optical section thickness (axial resolution; μm), and N is the number of steps within each optical section (i.e. axial resolution divided by the step size). The latter operation is necessary to correct for oversampling of the signal volumes.

Respiratory ux in intact and permeabilized cells
Approximately 3 x 10 6 cells were used for each intact and permeabilized cell experiment. High-resolution respirometry measurements were performed using the Oroboros Oxgraph-2k (O2k; Oroboros Instruments, Innsbruck, Austria) in a 1mL reaction volume at 37°C. At the conclusion of each experiment 1mL of cell suspension was collected from each chamber and centrifuged at 2,000 x g for 10 min at 4°C. Cells were lysed using low-percentage detergent buffer (CelLytic) followed by a freeze-thaw cycle, and protein concentration was determined using a BCA protein assay.

Isolation of mitochondria from PBMCs and leukemia cells
In order to pellet cells, PBMC fractions were washed with PBS and centrifuged at 3,000 x g for 10 min at 4°C and leukemia cells were centrifuged at 300 x g for 10 min followed by a PBS wash. Cell pellets were resuspended in Mitochondrial Isolation Buffer with BSA (100mM KCl, 50mM MOPS, 1mM EGTA, 5mM MgSO 4 , 0.2% BSA, pH 7.1) and homogenized using a borosilicate glass mortar and Te on pestle.
Homogenates were centrifuged at 800 x g for 10 min at 4°C. The supernatant was collected, and the remaining pellet was resuspended in Mitochondrial Isolation Buffer with BSA, then homogenized and centrifuged again. This process was repeated a total of 3 times. The collected supernatant was centrifuged at 10,000 x g for 10 min at 4°C to pellet the mitochondrial fraction. The fraction was resuspended in Mitochondrial Isolation Buffer without BSA, transferred to a microcentrifuge tube and subjected to a second spin at 10,000 x g. The mitochondrial pellet was resuspended in ~100µL of Mitochondria Isolation Buffer and protein concentration was calculated using the Pierce BCA assay. Respiration assays using isolated mitochondria were similar to that described for permeabilized cells.
Mitochondrial NADH/NAD + redox in isolated mitochondria Fluorescent determination of NADH/NAD + was performed using a QuantaMaster Spectro uorometer (QM-400, Horiba Scienti c, Kyoto, Japan). The NADH/NAD+ was detected at Ex/Em: 350/450. NADH/NAD+ was measured in mitochondria isolated from PBMC and leukemia cell lines using the CK clamp assay. Experiments were performed at 37°C in a 200µL reaction volume. To start, Respiratory Buffer supplemented with creatine (200µL), Cyt C (10µM), mitochondrial lysate (100µg) were added into a glass cuvette. Mitochondria were incubated at 37°C for ~ 5 minutes in the absence of substrate to induce 0% reduction of the NADH pool. Saturating carbon substrates were added (Pyr/Mal/ Oct/Glut/Succ, 'Multi'), and respiration was stimulated with the CK clamp. Titration of ΔG ATP was performed via PCr titration (6, 15, 21mM). Oligomycin (0.02µM) was added to inhibit ATP synthesis and cyanide (CN, 10mM) was added to induce 100% reduction of the matrix NADH pool. The NADH/NAD+ was expressed as a percentage reduction of the CN value (i.e. 100% reduction) based upon the formula % Reduction = (F-F 0% )/(F 100% -F 0% )*100.
Mitochondrial lysis, protein digestion, and peptide labeling for TMT quantitative proteomics Mitochondrial pellets from leukemia cells and PBMC (approximately 250 µg of protein) were lysed in icecold 8 M Urea Lysis Buffer (8 M urea in 50 mM Tris, pH 8.0, 40 mM NaCl, 2 mM CaCl 2 , 1x cOmplete ULTRA mini EDTA-free protease inhibitor tablet), as described previously 36 . The samples were frozen on dry ice and thawed for three freeze-thaw cycles and further disrupted by sonication with a probe sonicator in three 5s bursts set at an amplitude of 30 (Q Sonica, Newtown, CT). Samples were centrifuged at 10,000 × TMT labeling TMT labeling was performed as previously described 36 . The samples from isolated mitochondria were resuspended in 100 μL of 200 mM triethylammonium bicarbonate (TEAB), mixed with a unique 10-plex Tandem Mass Tag (TMT) reagent (0.8 mg re-suspended in 50 μL100% acetonitrile), and shaken for 4 hr at room temperature (Thermo Fisher). A total of 2 x 10-plex kits were used and one sample was TMT-labeled in both kits to control for quanti cation differences across multiplex preparations. Following quenching with 0.8 μL 50% hydroxylamine samples were frozen, and lyophilized. Samples were re-suspended in ~1 mL of 0.5% TFA and again subjected to solid phase extraction, but with a 100 mg tC18 SEP-PAK SPE column (Waters). The multiplexed peptide sample was subjected to high pH reversed phase fractionation according to the manufacturer's instructions (Thermo Fisher). In this protocol, peptides (100 µg) are loaded onto a pH-resistant resin and then desalted with water washing combined with low speed centrifugation. A step-gradient of increasing acetonitrile concentration in a high-pH elution solution is then applied to columns to elute bound peptides into 8 fractions. Following elution, fractions were frozen and lyophilized. Peptides were grouped to proteins using the rules of strict parsimony and proteins were ltered to 1% FDR using the Protein FDR Validator node of PD2.2. MS2 reporter ion intensities for all PSMs having coisolation interference below 0.5 (50% of the ion current in the isolation window) and an average S/N > 10 for reporter ions were summed together at the peptide and protein level. Imputation was performed via low abundance resampling.

Statistical analysis for TMT proteomic
The protein group tab in the PDv2.2 results was exported as tab delimited.txt. les, and analyzed based on a previously described work ow 36 . First, M2 reporter (TMT) intensities were summed together for each TMT channel, each channel's sum was divided by the average of all channels' sums, resulting in channelspeci c loading control normalization factors to correct for any deviation from equal protein input in the 10-plex experiments. Reporter intensities for proteins were divided by the loading control normalization factors for each respective TMT channel. All loading control-normalized reporter intensities were converted to log 2 space and the average value from the ten samples per kit was subtracted from each sample speci c measurement to normalize the relative measurements to the mean of each kit. Data from each kit were then combined for statistical comparisons. For comparison of PBMC to leukemia cell lines, condition average, standard deviation, p-value (p, two-tailed student's t-test, assuming equal variance), and adjusted p-value (P adjusted , Benjamini Hochberg FDR correction) were calculated 65,66 . For protein-level quanti cation, only Master Proteins-or the most statistically signi cant protein representing a group of parsimonious proteins containing common peptides identi ed at 1% FDR-were used for quantitative comparison.
Proteomics data availability and software All raw data for proteomics experiments is available online using accession number "PXD020715" for Proteome Xchange 67 and accession number "JPST000934" for jPOST Repository 68 .

Statistical Analysis and Software
Statistical analysis was performed using GraphPad Prism 8.4. Among groups, data were analyzed using one-way ANOVA and Tukey's multiple comparison tests. The assumption of equal variance was assessed using the Brown-Forsythe test. All data are represented as mean ± SEM and analysis were conducted with a signi cance level set at p<0.05.     Data are presented as mean ±SEM and analyzed by unpaired t-tests.   *p<0.05, **p<0.005, ***p<0.0005, ****p<0.0001

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.