A reduction in ATP demand and mitochondrial activity with neural differentiation of human embryonic stem cells

Human embryonic stem cells (hESCs) are long-term self-renewing cells that are able to differentiate into any cell type in the body (Thomson et al., 1998). The potential of hESCs to act as an unlimited source for somatic cell generation is of major interest for downstream applications, including drug testing, disease modelling and cell replacement therapy (Braam et al., 2009; Desbordes et al., 2008; Han et al., 2009; Kiskinis and Eggan, 2010; Passier et al., 2008; Zeng and Rao, 2008). Additionally, these cells might yield insights into the processes underlying cellular senescence, cancer and ageing (Zeng, 2007). Obtaining a more complete understanding of the biology of pluripotent cells is important for maximising the benefits from these areas.

The cells of the inner cell mass of the blastocyst, from which hESCs are derived, divide rapidly. This property of aggressive growth is reflected in cultured ESCs by their ability to form teratomas and their short-term growth-factor-autonomous cell cycling (Becker et al., 2010; Blum and Benvenisty, 2009), although there are differences in the division rate of different pluripotent cell types (Hanna et al., 2010). Progenitor cells in the adult organism, such as multipotent neural stem cells (NSCs), must possess the unique properties of long-term viability and self-maintenance, while remaining in what is typically a quiescent state. Multipotent NSCs can be derived in pure populations from cultured hESCs (Reubinoff et al., 2001; Swistowski et al., 2009; Zhang et al., 2001), but little is known about their bioenergetic requirements or how closely their energy metabolism is governed by the constraints of the genetic developmental programme. The transition from a pluripotent stem cell through progressive stages of differentiation probably involves dynamic changes in the energy demand for cellular processes, depending on the needs of the individual cell types. ATP generation in aerobic cells is divided between two main pathways: anaerobic glycolysis and mitochondrial oxidative phosphorylation. On the basis of changing energy demands, and the need to limit the production of mitochondrial reactive oxygen species, it might be expected that the relative contribution of these two pathways would change during differentiation.

Some models suggest that an expansion of the mitochondrial population regardless of the differentiation lineage might be a universal phenomenon from the onset of ESC differentiation (Facucho-Oliveira and St John, 2009). However, many signalling and energy-consuming processes might be decreased in the early transition to a differentiated state, and it might not be expected that this would be accompanied by a mitochondrial expansion (Armstrong et al., 2006; Becker et al., 2006; Efroni et al., 2008; Meshorer and Misteli, 2006). By contrast, the evidence that mitochondrial biogenesis occurs in some instances of somatic stem-cell differentiation could be consistent with the acquisition of specialised energy-requiring processes, as reported for mesenchymal stem-cell differentiation (Chen et al., 2008; Uldry et al., 2006). Changes in mitochondrial regulation might also be driven by genetic developmental programmes, reflecting the typical energy requirements of these cell types in vivo as opposed to the energy requirements in culture. This provides a rationale for following the genetic regulators of energy metabolism in concert with the bioenergetics to illuminate the different forces driving changes in mitochondrial function.

In the present study, we characterise mitochondrial function in hESCs and in several stages of neuronal differentiation using advanced bioenergetic technologies. We conclude that hESCs have a high mitochondrial activity that declines disproportionately, alongside several related parameters of mitochondrial biogenesis, with differentiation into NSCs and that this is coupled to a decrease in overall ATP turnover, a decrease in macromolecule biosynthesis and a decrease in protein secretion.

To study the bioenergetics of hESC neural differentiation, we examined hESCs (Fig. 1Ai), hESC-derived NSCs (Fig. 1Aii) and NSCs differentiated for 35 days, which consist of >80% β-III-tubulin-positive neurons (termed D-35; Fig. 1Aiii), as shown previously (Swistowska et al., 2010; Swistowski et al., 2009). It has been demonstrated previously that these NSCs are nestin- and Sox1-positive and have a wide differentiation capacity (Swistowski et al., 2009). For comparisons with another somatic cell type, we chose to examine a well-characterised primary human fetal-foreskin-derived fibroblast line (BJ; termed HDF).

The oxygen consumption rate and extracellular acidification rate were determined in adherent cultures using a Seahorse XF-24 extracellular flux analyzer, as described previously (Gerencser et al., 2009). Basal respiration rate in hESC colonies was 1.8-fold higher than in NSCs and 2.8-fold higher than in HDFs, when normalised to the cell protein level (Fig. 1B). The coupled rate (the total basal rate minus the rate with oligomycin, where oligomycin is an inhibitor of ATP synthesis) was 111 pmol of O₂ per minute per 10 μg of cell protein in hESCs and dropped to 54 pmol of O₂ per minute per 10 μg of cell protein in NSCs, indicating that the mitochondrial ATP output decreases in the transition from hESCs to NSCs. Data were combined from the I6 and H9 lines for hESCs and NSCs to increase the reliability of our final determinations (no significant difference was observed between these lines). Differentiation of NSCs into D-35 cells caused little change in respiration or oxidative phosphorylation compared with the parental NSCs. Furthermore, the maximal uncoupler-evoked oxygen consumption rate (indicating maximal respiratory capacity) was significantly decreased in NSCs when compared with hESCs. The extracellular acidification rate is mainly due to lactate and bicarbonate production and, when calibrated as the proton production rate, indicates glycolytic rate (Wu et al., 2007). In contrast with respiration, the glycolytic rate was slightly increased in NSCs but declined significantly in D-35 cells (Fig. 1C). An absolute quantification of both the oligomycin-sensitive oxygen consumption rate and glycolytic rate can be converted into the ATP production rate, to assess the relative contribution in each cell type. The oligomycin-sensitive oxygen consumption was converted into the ATP production rate using a P/O ratio of 2.3 (Brand, 2005), and there is a one-to-one relationship between the lactate production rate and the ATP production rate from glycolysis. Remarkably, hESCs made at least 77% of their ATP using oxidative phosphorylation, whereas this value was significantly lower in NSCs (55%) and in HDFs (59%) (Fig. 1D, both P<0.01 compared with hESCs), but was similar to that of hESCs in D-35 cells (76%). The results in Fig. 1D show that hESCs had a faster turnover of ATP than NSCs, D-35 cells or HDFs and that this demand is met primarily by higher oxidative phosphorylation in hESCs.

To address the apparent bias hESCs show towards high oxidative phosphorylation for ATP production, we assessed various parameters of mitochondrial content in each cell type. Western blots showed that although the levels of voltage-dependent anion-selective channel (VDAC) remained unchanged upon neural differentiation, cytochrome c levels were markedly decreased in NSCs but were upregulated again upon differentiation into D-35 cells (Fig. 2A,B). The important antioxidant proteins manganese superoxide dismutase (MnSOD) (mitochondrial) and copper/zinc superoxide dismutase (CuZnSOD) (cytosolic) showed little change (MnSOD was slightly reduced in NSCs), indicating that the cellular antioxidant system remains fairly stable between these states.

Quantification of absolute mitochondrial DNA (mtDNA) copy number revealed that hESCs are rich in mtDNA, with around 1200–1500 copies per cell, whereas NSCs, D-35 cells and HDFs contain 600–1000 copies per cell (Fig. 2C; values are the means of the H9 and I6 cell lines at each differentiation stage). A third hESC line BG01 was found to have approximately 1200 copies per cell. No significant difference was found in mtDNA content between hESCs maintained on feeder cells and those maintained on matrix in mouse embryonic fibroblast (MEF)-conditioned medium (data not shown). When normalised for cell size (see supplementary material Fig. S1), these values equate to an approximately 1.5- to 2-fold greater mtDNA density in hESCs. Differentiation of hESCs, as embryoid bodies or as a monolayer in inactivated serum-replacement-containing medium, showed that this progressive transition in mtDNA copy number occurred during the differentiation process itself and that by day 12 the amount was close to the content of the NSCs (Fig. 2D). This pattern was mimicked in B6 mouse ESC differentiation (supplementary material Fig. S2).

We also measured mitochondrial organelle content by employing live-cell confocal imaging stereology in cells labelled with MitoTracker Red and calcein-AM. In agreement with differentiation-based changes in respiration and mtDNA content, stereology revealed a small but significant reduction in mitochondrial volume density from 5.4±0.3% in hESCs to 4.0±0.3% in NSCs (Fig. 2E,F; P<0.01), which was reversed upon further differentiation (D-35: 5.9±0.4%). The mitochondrial volume density of the HDFs was similar to that of the NSCs (HDF: 4.0±0.2%; P<0.01 compared with hESCs).

To investigate whether the high levels of mitochondrial activity and changes in mitochondrial biogenesis upon differentiation had underlying nuclear genetic regulation, we examined the expression of the ‘master regulators’ of mitochondrial gene expression – that is transcriptional coactivators and corepressors thought to control the expression of multiple downstream targets (Feige and Auwerx, 2007; Handschin and Spiegelman, 2006). We measured the expression levels of the genes encoding the two transcriptional coactivators peroxisome-proliferator-activated receptor γ coactivator (PGC)-1α and PGC-1β and the transcriptional corepressor receptor-interacting protein 140 (RIP140). We first compared hESCs with several somatically derived primary cell types and two cancer cell lines. Consistent with our functional observations, the mRNAs encoding PGC-1α and PGC-1β were both expressed to a greater extent in hESCs than in most of the somatic cell types (Fig. 3A,B). The cell types most similar to hESCs were melanocytes (HEMN-LP) and the cancer cell line MCF7, which showed high expression of the mRNAs encoding both PGC-1α and PGC-1β. RIP140, a repressor of genes involved in oxidative metabolism, was weakly expressed in hESCs compared with all the other cell types (Fig. 3C). PGC-1α and PGC-1β expression decreased and RIP140 expression increased during hESC differentiation into NSCs (Fig. 3D,E). Similar changes in PGC-1α and PGC-1β expression with hESC embryoid body differentiation have also been reported recently (Prigione et al., 2010). A large induction in PGC-1α upon differentiation into D-35 cells is consistent with our observation of increased mitochondrial biogenesis and a return to a reliance on mitochondrial ATP production in these cells.

To confirm further the expression pattern of these genes in human pluripotent cells, we assessed the expression of these genes in an induced pluripotent stem cell (iPSC) line derived from IMR90 fibroblasts (Mali et al., 2010). There was a strong induction of the mRNA encoding PGC-1β, whereas that encoding RIP140 was repressed upon reprogramming (Fig. 3F). PGC-1α was not induced but already had elevated expression in these fibroblasts compared with the other fibroblast strains we measured. iPSCs that were induced to differentiate into NSCs and D-35 cells, as described previously (Swistowski et al., 2010), showed similar mRNA expression changes to the hESC differentiations (Fig. 3F).

Finally, we attempted to control the energetic supply in differentiated cells by stable overexpression of PGC-1β or short hairpin RNA (shRNA) depletion of RIP140 in I6 NSCs or HDFs. Surprisingly, neither manipulation substantially boosted mitochondrial respiration or respiratory capacity (supplementary material Fig. S3). RIP140 knockdown did subtly increase maximal respiratory capacity in both cell types and stimulated glycolysis in NSCs. These findings suggest that manipulation of these genes individually is insufficient to alter mitochondrial bioenergetics in these cell types.

The higher basal ATP turnover and maximal ATP-generating capacity of hESCs cannot be explained solely by the abundance of mitochondria in these cells. Mitochondrial membrane potential (Δψ_m) accounts for the majority of the proton-motive force used for driving ATP synthesis and therefore has a significant effect on the maximal ATP-generating capacity (Nicholls, 2006). The resting Δψ_m reflects the balance between supply (substrate and terminal oxidation) and demand (ATP consumption). A high ATP demand in hESCs might outstrip supply and thus result in a lower Δψ_m than in differentiated cells such as NSCs and HDFs, where ATP demand is smaller. Conversely, a high ATP supply originating from Δψ_m might keep ATP consumption running faster in hESCs.

Δψ_m in adherent cultures of hESCs and related lines was determined by fluorescence microscopy. All known Δψ_m measurement techniques (fluorescence, voltammetric or isotopic) rely on the potential-dependent distribution of lipophilic cations, such as the fluorescent indicator tetramethylrhodamine methyl ester (TMRM). The mitochondrial accumulation of these cations is a function of Δψ_m, the plasma membrane potential (Δψ_p) and the mitochondrial-matrix-to-cell-volume ratio (Nicholls, 2006). These parameters are expected to differ between cell types. To unveil Δψ_m, Δψ_p was followed with a lipophilic anionic plasma membrane potential indicator (PMPI) (Nicholls, 2006), and volume ratios were determined separately. TMRM and PMPI fluorescence of whole microscopic view fields were monitored in time, and Δψ_p and Δψ_m were determined in millivolts by temporally deconvoluting the fluorescence intensity time lapses, taking into account the slow potential-dependent redistribution and nernstian behaviour of these probes (A.A.G. and M.D.B., unpublished).

Δψ_p was calibrated by establishing a K⁺-equilibrium potential {a potential governed by the intra- and extra-cellular (EC) [K⁺]} using a cocktail of K⁺-channel openers and making stepwise increments in [K⁺]_EC (Fig. 4A–C). The resting Δψ_p pooled from the two hESC lines (I6 and H9), as well as from HDFs is shown in Fig. 4D. The mitochondrial-matrix-to-cell-volume ratio was obtained as the product of the mitochondria-to-cell-volume ratio (see above) and the matrix-to-mitochondria-volume ratio. The latter was measured as the volume density of matrix by transmission electron microscopic stereology and was 0.84 for hESCs and 0.76 for NSCs. For HDFs, we assumed the same ratio as for NSCs. Interestingly, mitochondria in hESCs were mostly in the ‘orthodox’ conformation (Hackenbrock, 1966), whereas NSCs showed a ‘condensed’ conformation, with higher electron density in the matrix and more dilated cristae (Fig. 4G). The resting Δψ_m was calculated from the efflux kinetics of TMRM following an acute and complete mitochondrial depolarisation (Fig. 4E,F). The decay of TMRM fluorescence under these conditions was determined by the previous resting Δψ_m, volume ratios and the present Δψ_p (see the supplementary methods).

The mean determinations are shown in Fig. 4H. We conclude that there is no significant change in Δψ_m upon differentiation of hESCs into NSCs. Therefore the decrease in respiration in NSCs when compared with hESCs is not caused by a large decrease in ATP supply or demand; instead supply and demand are kept in balance by concerted regulation.

Our data suggest that hESCs turnover ATP more rapidly than differentiated cells. To pinpoint the process(es) responsible for the increased ATP demand, we compiled a complete budget of the ATP demand in hESCs and NSCs. Individual ATP-consuming processes were acutely inhibited, causing a consequential drop in ATP demand and triggering a proportional drop in respiration. Dose–response curves for the protein translation inhibitor cycloheximide, the DNA and/or RNA synthesis inhibitor actinomycin D and the proteasome inhibitor MG132 are shown in supplementary material Fig. S4. In addition, the tubulin and actin turnover and the activities of Na⁺/K⁺-ATPase, the endoplasmic reticulum (ER) and plasma membrane Ca²⁺-ATPase and xanthine oxidase were investigated. Using the minimum saturating concentrations of each inhibitor, the relative contributions to the total respiration of these cells were assigned cumulatively for each of these processes (Table 1; Fig. 5). Inhibition of these processes caused a measurable drop in the oxygen consumption rate for both cell types, with the exception of the Ca²⁺-ATPases and xanthine oxidase. The relative demand from each of these processes was largely maintained, with protein synthesis and DNA and/or RNA synthesis still accounting for the majority of mitochondrial respiration in both cell types. However, the absolute oxygen consumption rates that were sensitive to inhibition of protein synthesis, DNA and/or RNA synthesis and Na⁺/K⁺-ATPase were significantly higher in hESCs, indicating that these processes consume more ATP in hESCs than in NSCs.

We considered the possibility that differences in cell division rates between hESCs and NSCs could account for some of the observed differences in ATP turnover. Hence, generation times were determined through time-lapse imaging. Surprisingly, the generation time of NSCs was significantly shorter than that of hESCs (H9 hESCs at 20.3 hours compared with NSCs at 14.2 hours; I6 hESCs at 24.2 hours compared with NSCs at 14.1 hours) (Fig. 6A–C; supplementary material Movie 1). No significant differences in cell death rates were observed between hESCs and NSCs (Fig. 6D). In conclusion, the higher ATP turnover rate in hESCs cannot be explained by a higher growth rate. Instead, processes that consume ATP in the same proportion as at lower growth rates must be increased; these might include autophagy or secretion of macromolecules.

Electron microscopy showed that there were hallmarks of macromolecule secretion in hESCs but not in NSCs (Fig. 7A,B). hESCs typically contained clusters of single-membrane-bound vesicles with homogeneous content and a less electron-dense core (Fig. 7A–D). Although no definite exocytotic events were observed, electron micrographs suggest a goblet-cell-like secretion mechanism (Fig. 7C), which is supported by previous results showing the secretion of cytosolic and cytoskeletal proteins in the proteome of hESC-conditioned medium (LaFramboise et al., 2010). By contrast, evidence of autophagy was uncommon, but both lines contained large multivesicular bodies signifying heterophagy. To assess the relative rates of protein secretion directly, we analysed medium conditioned for 10 hours by hESCs and NSCs and found a significantly higher concentration of secreted proteins in medium conditioned by hESCs (Fig. 8).

Thus, the high ATP production rate in hESCs might be required to support RNA synthesis, protein synthesis, ion pumping and cytoskeletal rearrangements during macromolecule secretion by a goblet-cell-like mechanism, in which large parts of the cell, including cytosolic and cytoskeletal components, are secreted.

To our knowledge, this is the first report of the bioenergetics of hESC differentiation. Our principal finding is that hESCs maintain a high rate of mitochondrial oxidative phosphorylation, which funds their high ATP demand for processes other than growth. We report a discrepancy wherein decreasing energy turnover is concomitant with increasing cell division rate during the transition from hESCs into NSCs, which might be explained by macromolecule secretion in hESCs.

ESCs have a fundamentally different in vivo biological role compared with that of somatic progenitor cells, which can remain quiescent for long periods of time. It seems sensible that these cell types would be bioenergetically tuned towards their normal developmental roles. Using a novel experimental platform enabling measurement of oxygen consumption rates and glycolytic rates in normal adherent cultures, we discovered that hESCs had an elevated rate of oxidative phosphorylation compared with hESC-derived NSCs, but without increased glycolysis. The hESC lines we investigated also had a relatively high mitochondrial biogenesis compared with NSCs and HDFs (Fig. 2), and overall our data support a model whereby mitochondrial biogenesis is decreased upon the transition from hESCs into early progenitors but increased again on differentiation into more specialised cell types, such as the neuronal population examined in the present study. The changes in mitochondrial biogenesis correlated with the changing contribution of oxidative phosphorylation to ATP output during differentiation. The higher maximal respiratory capacity of hESCs could be caused by the higher matrix-to-mitochondria and mitochondria-to-cell-volume ratios and the greater amounts of respiratory complexes, as indicated by the abundance of mtDNA and cytochrome c. Other studies have presented data suggesting that some indicators of mitochondrial content might increase overall during hESC differentiation in a mixed cell population differentiating in the presence of serum (Armstrong et al., 2010; Saretzki et al., 2008). This could reflect the preferential induction of other cell types, such as cardiomyocytes, in the presence of serum, but conclusions are difficult to draw on the basis of whole-cell fluorescent measurements because changes in cell size, and both mitochondrial and plasma membrane potentials, will affect these readings (see Fig. 4D). We have shown in the present study that, in a pure population of hESC-derived NSCs, mitochondrial content and activity are reduced compared with those of the parent cells. Further study will be needed to assess the mitochondrial and bioenergetic changes in other cell lineages. However, preliminary data from a cardiac differentiation model suggest a unique mitochondrial development for beating compared with non-beating cells (M.J.B., unpublished), highlighting further the importance of studying pure populations of cells.

From these observations, we inferred that hESCs might have a genetic programme that facilitates their high mitochondrial activity and identified expression patterns consistent with this for the ‘master regulators’ PGC-1α, PGC-1β and RIP140. The PGC-1 coactivators have been widely studied in other models and shown to be potent regulators of mitochondrial gene expression (Arany et al., 2007; Fritah et al., 2010; Lelliott et al., 2006; Lin et al., 2003; Meirhaeghe et al., 2003; Powelka et al., 2006; Seth et al., 2007). We observed high levels of expression of the mRNA encoding coactivators PGC-1α and PGC-1β in hESCs and, interestingly, also in the cancer cell line MCF7, pointing to potential commonalities in energy metabolism. MCF7 cells have been reported to obtain 80% of their ATP from oxidative phosphorylation (Guppy et al., 2002). Nearly all the other primary somatic cell types studied showed reduced expression of the mRNA encoding PGC-1α and PGC-1β, together with increased expression of that encoding the corepressor RIP140. Recent publications support these observations by showing that many other genes involved in mitochondrial biogenesis, including those encoding the mtDNA transcription factor TFAM and the mitochondrial RNA polymerase POLRMT, are also downregulated upon hESC and iPSC differentiation (Armstrong et al., 2010; Prigione et al., 2010). Whether these expression changes occur as a programme of differentiation or whether they change in response to reduced energy demand remains an unanswered question. Increasing energy demand in fibroblasts through mitochondrial uncoupling has been reported to induce the PGC-1 coactivators (Rohas et al., 2007). The expression levels of the PGC-1 coactivator could decrease if energy demand fell in the early stages of differentiation. However, as we see a strong induction of PGC-1α upon neuronal differentiation from NSCs, without any increase in ATP demand, and a return towards reliance on oxidative phosphorylation-generated ATP, a developmental genetic programme might also be a driving force in both of these transitions. Manipulation of PGC-1β or RIP140 in NSCs and HDFs was insufficient to drive a change in the bioenergetics, and so other genes might oppose or limit their effects in these cells.

Measurements of mitochondrial membrane potential showed that hESCs and NSCs had a similar Δψ_m of approximately −130 mV. The higher rate of oligomycin-sensitive respiration in hESCs in the absence of a significantly different mitochondrial membrane potential suggests a parallel increase in the rate of ATP-supplying and -demanding processes. Concerted regulation of demand and supply could be in favour of avoiding increases in Δψ_m and an associated increase in the production of reactive oxygen species (Brand, 2010; Murphy, 2009). A breakdown of ATP budgets revealed that in both hESCs and NSCs ATP was consumed through largely the same pathways. Protein and nucleic acid synthesis together accounted for the largest portion of the energy budget (hESCs, 41%; NSCs, 52% of mitochondrial respiration), indicating that most of the ATP drives macromolecule synthesis. As the total energy budget was greater in hESCs than in NSCs, this indicated a faster macromolecule synthesis rate in hESCs; however, in the absence of increased cell division, these extra macromolecules must ultimately be degraded or secreted. hESCs have been shown to secrete a diverse range of chemokines, cytokines, growth factors and matrix remodelling enzymes, along with cytosolic and cytoskeletal components (Bendall et al., 2009; LaFramboise et al., 2010). We found that hESCs had morphological indications of goblet-cell-like secretion and secreted significantly more protein into the extracellular environment than did NSCs. Further study is warranted to investigate the importance of this secretory activity. Secretion could be a driving force necessitating an increased rate of biosynthesis; however, an elevated biosynthesis rate in the absence of faster division might also drive secretion instead of increasing cell size. Goblet-cell-like hESCs have also been observed by Sathananthan and colleagues (Sathananthan et al., 2002). hESCs are known to divide relatively slowly – much slower than mouse ESCs – so it is plausible that in hESCs the division rate might not keep pace with the programme driving biosynthesis, but this hypothesis will require further study. The expression of potent oncoproteins such as c-Myc is a characteristic of ESCs and could be an example of such a programme. Myc itself is an important regulator of ribosome biogenesis and protein synthesis, and also a regulator of mitochondrial activity and biogenesis (Fan et al., 2010; Kim et al., 2008; Li et al., 2005; van Riggelen et al., 2010). The importance of biosynthesis and energy metabolism in ESCs has also been highlighted in recent studies (Dejosez et al., 2010; Wang et al., 2009).

In summary, we provide what we believe to be the first evidence that the energy demand for many cell processes in hESCs is elevated relative to their division rate and funded primarily by ATP generated through mitochondrial oxidative phosphorylation. Developing embryos of several species studied in vitro show a large increase in oxygen consumption at the blastocyst stage, during which protein synthesis is at its maximal rate, and it has been estimated that the majority of ATP comes from oxidative phosphorylation during this phase (Houghton et al., 1996; Sturmey and Leese, 2003; Thompson et al., 1996). In this respect, hESCs might mirror the metabolic characteristics of cells of the developing blastocyst during this important phase of expansion.

hESC lines I6, H9 and BG01 were maintained as described previously (Zeng et al., 2004) or were grown on Geltrex-coated dishes (1:50) (Invitrogen, Carlsbad, CA) in the same medium previously conditioned for 24 hours on inactivated mouse embryonic fibroblasts (MEFs). B6 mouse ESCs were grown in the same unconditioned medium but supplemented with 1000 units/ml leukaemia inhibitory factor (LIF) (Millipore) instead of fibroblast growth factor 2 (FGF2). For hESC differentiation, colonies were detached with collagenase IV (1 mg/ml), triturated and suspended as embryoid bodies for up to 16 days or re-plated on Geltrex-coated dishes for up to 12 days, both in standard ES medium without FGF2. mESCs were differentiated as embryoid bodies in standard ES medium without LIF.

NSC derivation, culture and differentiation were performed as described previously (Swistowska et al., 2010; Swistowski et al., 2009). H9 hESC-derived NSCs were from Millipore. BJ fibroblasts were from the ATCC and grown in high-glucose Dulbecco's modified Eagle's medium (DMEM) containing sodium pyruvate and L-glutamine (Invitrogen) plus 10% foetal bovine serum (FBS) and penicillin and streptomycin. All other cell types used for RNA expression analysis were from Invitrogen and cultured briefly using the medium recommended by the manufacturer before RNA isolation (for details, see supplementary material Table S1).

To calculate the cell generation times, NSCs were imaged at 24 hours after seeding, using phase-contrast microscopy in a humidified chamber with 5% CO₂, at 4-minute intervals for 8 hours. At 24 hours after seeding, hESCs were labelled with 2 μg/ml Hoechst 33342 (Molecular Probes H3570) and fluorescent images were recorded at 8-minute intervals with minimal excitation intensity for 24 hours. Cells were randomly selected from several fields, tracked by eye and division or death events were recorded. The mean generation time and frequency of death were calculated.

For the analysis of conditioned media, I6 hESCs or NSCs were grown to high density on Geltrex-coated 6-cm-diameter dishes in normal growth medium as above, rinsed twice and then incubated in DMEM with Ham's F12 plus Glutamax for 10 hours. After collection, the medium was centrifuged to remove debris, and the protein component was precipitated using a 1:8:1 ratio of medium, acetone and trichloroacetic acid, respectively, at −80°C for 1 hour. Precipitates were centrifuged at 18,000 g for 15 minutes at 4°C and resuspended in a volume of SDS-PAGE loading buffer proportional to the relative amounts of cellular protein harvested from the dish. Total cell lysates were made from the monolayers by using mammalian protein extraction reagent (Pierce) containing protease inhibitor cocktail (Roche). The cell lysate protein levels were quantified according to the Biuret method and confirmed on test gels with GelCode Coomassie Brilliant Blue staining (Thermo Scientific). Precipitated samples were subjected to SDS-PAGE alongside the protein lysates and stained as above. Gels were scanned and lane intensities were quantified using densitometry in ImageJ software (NIH).

Respiration and acidification rates were measured on adherent cells using a Seahorse XF-24 analyser (Seahorse Bioscience, North Billerica, MA). For hESCs and NSCs, the XF24 V7 assay plates were coated with Geltrex (1:50). Cells were seeded at approximately 24 hours before measurement in their normal growth medium; NSCs were seeded at 7×10⁴ cells per well and hESCs as colonies totalling ~3×10⁴ cells per well. The assay was performed in bicarbonate-free DMEM (Sigma D-5030) supplemented with 15 mM glucose, 2 mM L-glutamine, 0.5 mM sodium pyruvate and 0.4% BSA. Cells were washed twice and preincubated in this medium for 1 hour before measurement. Oligomycin was used at 0.2 μg/ml; carbonyl cyanide p-trifluoromethoxyphenylhydrazone (FCCP) was titrated in four injections to 1.5 μM for hESCs, NSCs and D-35s, and 2 μM for HDFs; and rotenone and antimycin A were added at 1 μM and 2 μM, respectively. Oxygen consumption values were calculated from 3-minute measurement cycles by the XFReader software Version 1.4 updated with a recent correction (Gerencser et al., 2009). Acidification rates were taken as the mean rate from the second and third baseline readings. After the assay, a standard protein assay was performed. Alternatively, with the D-35 cells, some wells were fixed with 4% paraformaldehyde for immunocytochemical staining, as described below.

ATP demand was measured using the Seahorse XF-24 analyser as above with the addition of 10 mM HEPES to the assay medium. The basal respiration rate was recorded and then the selective inhibitors were injected and respiration recorded for the following 2–5 minutes. Small changes from vehicle-only controls, which were included on each plate, were subtracted to yield the specific effects of the inhibitors. The inhibitors used for each cellular process, and the concentrations tested, include: protein synthesis (cycloheximide, 0.01–50 μg/ml); DNA and/or RNA synthesis (actinomycin D, 0.1–10 μM); proteasome (MG132, 0.1–50 μM); tubulin dynamics (nocodazole, 0.1–50 μM); actin dynamics (latrunculin A, 0.1–3 μM); Na⁺/K⁺-ATPase (ouabain, 1 mM); sarcoplasmic/endoplasmic reticulum Ca²⁺-ATPase (SERCA) (thapsigargin, 1–1000 nM); plasma-membrane Ca²⁺-ATPase (PMCA) (LaCl₃, 1–4 mM); and xanthine oxidase (allopurinol, 10–1000 μM). Mitochondrial respiration was subsequently fully inhibited with rotenone and antimycin A, as described above. All compounds were from Sigma, except MG132 (Calbiochem).

Imaging was conducted in eight-well LabTek I (Nunc, Rochester, NY) coverglass chambers. For hESCs, the glass was coated with 0.1% gelatin and the colonies were grown on a feeder layer of MEFs. For NSCs the glass was coated with poly-ornithine (20 μg/ml) and then Geltrex (1:100). Tetramethylrhodamine methyl ester (TMRM) (7.5 nM; Invitrogen), plasma membrane potential indicator (PMPI) (1:200; no. R8042 FLIPR membrane potential assay kit from Molecular Devices; Sunnyvale, CA) and Na-tetraphenylborate (1 μM) were continuously present in the experimental buffer (120 mM NaCl, 1.75 mM KCl, 0.2 mM KH₂PO₄, 10 mM TES, 2.5 mM NaHCO₃, 0.6 mM Na₂SO₄, 1.3 mM CaCl₂, 2 mM MgCl₂, 15 mM glucose, 2 mM L-glutamine and 0.5 mM Na-pyruvate, pH 7.4) at 37°C. Imaging was performed on a Nikon Eclipse Ti-PFS inverted fluorescence microscope controlled by NIS Elements 3.1 (Nikon, Melville, NY) and equipped with a S-Fluor 20×/0.75 NA lens, a Lambda LB-LS17 light source (Sutter Instruments, Novato, CA) a Cascade 512B camera (Photometrics, Tucson, AZ) and a MS-2000 motorised stage (ASI; Eugene, OR). The filter sets given as excitation–dichroic mirror–emission in nm were: for PMPI 500/24–520LP (long pass)–542/27 and for TMRM: 582–593LP (all from Semrock, Rochester, NY)–610LP (Omega Optical, Brattleboro, VT). Images were taken at 10-second intervals, typically cycling in the 2–12 stage positions. To establish the K⁺ equilibrium potential, a cocktail of drugs (KEC) was added including K⁺-channel activators: flupirtine (10 μM), NS309 (10 μM) and diazoxide (100 μM); chloride channel co-transport inhibitors: R-(+)-IAA94 (100 μM), R-(+)-DIOA (10 μM), bumetanide (80 μM); and the sodium channel inhibitor tetrodotoxin (1 μM). For calculation of Δψ_m mitochondria were completely depolarised with a cocktail (MDC) of valinomycin (1 μM), oligomycin (2 μM), FCCP (1 μM) and myxothiazol (1 μM). Complete cell depolarisation was induced with a cocktail (CDC) of ionophores: gramicidin (10 μM), nigericin (10 μM), monensin (10 μM) and the glycolysis inhibitor iodoacetate (500 μM). This cocktail also included the components of the MDC. Calculation of Δψ_p and Δψ_m from the measured fluorescence intensities were performed as described in the supplementary methods.

Cells were grown in LabTek I chambers as described above and were loaded with calcein-AM (1 μM; Invitrogen) and MitoTracker Red (hESC, 50 nM; NSC, D-35 and HDF, 25 nM; Invitrogen) in the above experimental buffer for 20 minutes, then imaged in experimental buffer with a Zeiss LSM 510 laser scanning confocal microscope. Single planes of 1024×1024 pixel images at 44 nm/pixel resolution were recorded using a Plan-Apochromat 100×/1.4 NA oil lens. Calcein and MitoTracker Red were excited simultaneously using the argon 488 nm and helium–neon 543 nm laser lines at high intensity, and fluorescence emissions were collected in the range of 500–530 nm and above 560 nm, respectively. To record serial optical sections without recording photo-damaged areas, random images were acquired at different x- and y-locations, and the z-plane was advanced from the bottom of the culture to the highest focal plane in steps of 1–1.5 μm. Images were binarised by highpass filtering and locally adaptive thresholding using Image Analyst MKII software (Image Analyst Software, Novato, CA). To calculate mitochondria-to-cell-volume ratios, the total number of mitochondrial pixels was divided by the total number of cytosolic pixels in the complete image set and multiplied by 0.67 – a stereologic correction factor (Weibel and Paumgartner, 1978) that accounts for the projection and clipping of mitochondria by the optical section. See the supplementary methods.

Sample preparation and processing are described in the supplementary methods. For determination of matrix volume ratios, mitochondria were selected in a uniform random approach and imaged at ×105,000 magnification. Matrix-to-whole-mitochondrial-volume ratios were calculated by the Cavaileri estimator using Stereo-Investigator (MBF Bioscience; Williston, VT).

Absolute mtDNA copy number was assessed by comparing PCR amplification of a mitochondrial amplicon [human, NADH-ubiquinone oxidoreductase chain 1 (ND1); mouse, cytochrome b] with a nuclear amplicon (human, Δ- and/or γ-globin; mouse, glucagon) from DNA isolated using a Qiagen DNA mini kit. Primer sequences are in supplementary material Tables S2 and S3. DNA templates were made of regions spanning each amplicon by PCR amplification. Dilutions from 1×10⁸ down to 1×10⁴ copies of the isolated DNA were used to generate a standard curve for quantification.

Total cell lysates were made using the mammalian protein extraction reagent (Pierce) containing protease inhibitor cocktail (Roche). Gels were run and proteins transferred using standard techniques. Antibodies were: anti-cytochrome c (no. 556433, BD Pharmingen); anti-VDAC (no. MSA03, MitoSciences); anti-MnSOD (no. sc-30080, Santa Cruz Biotechnology); anti-CuZnSOD (no. sc-11407, Santa Cruz Biotechnology). Immunocytochemistry and staining procedures were performed as described previously (Zeng et al., 2003).

RNA was isolated using TRIzol (Invitrogen) and was treated with DNase-I (Sigma). cDNA was synthesised using the Superscript III reverse transcriptase system (Invitrogen). Real-time PCR was performed using the PerfeCTa SYBR Green 2× mix (Quanta Bioscience) on an Applied Biosystems 7900HT fast real-time PCR machine. Primer details and sequences are presented in supplementary material Tables S4 and S5.

cDNA encoding PGC-1β was obtained from Open Biosystems (clone ID: 40146993) and cloned into a pLenti CMV/TO Neo vector (Addgene: 17292). The shRNA targeting RIP140 was constructed using sense: 5′-gcagcagtattctcgagaaGTGTGCTGTCCttctcgagaatactgctgc-3′ and antisense 5′-gcagcagtattctcgagaaGGACAGCACACttctcgagaatactgctgc-3′ sequences and expressed in a pLenti X2 Puro vector (Addgene number 17296). The uppercase nucleotides indicate the target sequence. The empty pLenti CMV/TO Neo vector or an shRNA against luciferase (Addgene: 21472) were used as controls. This lentiviral system has been described previously (Campeau et al., 2009). All vector inserts were sequenced to confirm identity. For packaging lentiviruses, human embryonic kidney HEK-293FT cells were used (Invitrogen). Lentiviral supernatants were concentrated using ultracentrifugation and titers were used to infect ~80–90% of the cells, as judged by a control virus with a cytomegalovirus promoter driving the expression of green fluorescent protein (GFP). Cells were selected with neomycin or puromycin for 1 week before analysis. Two separate derivations were made for I6 NSCs and HDFs for bioenergetic analyses.

This work was supported by the California Institute for Regenerative Medicine Grant CL1-00501-1 (X.Z.) and the Larry L. Hillblom Foundation (X.Z.). RS1-00163-1 (D.T.M., laboratory of Dale E. Bredesen), National Institutes of Health grants PL1 AG032118 (A.L.O., A.A.G., C.V. and M.D.B.), P01 AG025901, P30 AG025708 and R01 AG033542 (M.D.B.), the W. M. Keck Foundation (A.A.G. and M.D.B.), and The Ellison Medical Foundation (AG-SS-2288-09) (A.L.O. and M.D.B.). We thank Linzhao Cheng at the Johns Hopkins University for providing the iPSC line. We thank Eric Campeau at the University of Massachusetts for providing the lentiviral vectors. Deposited in PMC for release after 12 months.