New quantitative approach reveals heterogeneity in mitochondrial structure–function relations in tumor-initiating cells

The steady-state structure or morphology of mitochondria within a given cell depends on the cell type and its physiological state (Labbé et al., 2014; Miettinen and Björklund, 2017). Mitochondrial morphology is primarily determined by a balance of opposing fission and fusion events between individual mitochondria, collectively referred to as mitochondrial dynamics (Nunnari and Suomalainen, 2012). Mitochondrial dynamics (opposing fission and fusion) and mitochondrial energetics (linked ATP and redox states) (Murphy, 2009) impact each other to regulate cellular bioenergetics (Mishra and Chan, 2016; Willems et al., 2015). It has been proposed that, in nutrient excess, the balance of mitochondrial dynamics lies towards fission, decreasing mitochondrial energetic efficiency (Liesa and Shirihai, 2013; Schrepfer and Scorrano, 2016). Conversely, in nutrient deprivation, mitochondria are maintained in hyperfused networks, enhancing mitochondrial energetic efficiency.

In-depth understanding of the relationship between mitochondrial dynamics and energetics, or its underlying mechanisms, requires integrated quantitative analyses of these mitochondrial properties. Various approaches to quantify mitochondrial energetics exist in the field. Quantification of mitochondrial morphology reflecting mitochondrial dynamics status is less developed, but has been attempted by various laboratories (Harwig et al., 2018; Iannetti et al., 2016; Karbowski et al., 2004; Mitra et al., 2009; Ouellet et al., 2017; Parker et al., 2017). To address the critical need for integrated analyses, we first designed and validated metrics for quantifying mitochondrial dynamics status, and thereafter developed a multivariate approach towards identifying and quantifying mitochondrial [energetics] versus [dynamics] relationships in single cells. We named the microscopy-based high-resolution approach mito-SinCe² (mito-SinCe-SQuAReD: single cell simultaneous quantification of ATP or redox with dynamics of mitochondria). After validating and providing proof of principle of the mito-SinCe² approach, we used it in an example application on stem cell regulation.

The importance of mitochondrial dynamics and energetics in stem cell regulation is becoming increasingly recognized (Chen and Chan, 2017). Previously, we proposed that fission activity is critical for ovarian cancer (Tanwar et al., 2016). Here, we find that mitochondria-dependent ovarian tumor-initiating cells (ovTICs), which harbor stem cell properties and contribute to ovarian cancer development (Ishiguro et al., 2016), regulate mitochondrial energetics differently from the bulk tumor cells. Mito-SinCe² analyses revealed complexities of mitochondrial dynamics and energetics in ovTIC self-renewal and proliferation.

Single-cell metrics for parsing the contribution and dynamic interaction of the opposing fission or fusion processes in a given mitochondrial morphology have not been reported. Here, we used outputs of MitoGraph v2.1 image analysis software (Viana et al., 2015) and a photoactivation approach to compute and validate the quantitative metrics for steady-state mitochondrial fission and fusion, and also average mitochondrial diameter in single cells.

MitoGraph v2.1 application on high-resolution 3D confocal micrographs yields the following parameters for mitochondrial elements: total length and number, individual length and volume. First, we confirmed that the length and diameter (derived from volume) of mitochondrial elements correspond with qualitative differences in mitochondrial morphology in representative Mitotracker Green (MTG)-stained cells 1–4 (Fig. 1A). In particular, we found that (1) cells 1 and 2 have less than 50, whereas cells 3 and 4 have greater than 100 individual mitochondrial components (Fig. S1A, x-axis); (2) cells 1 and 2 have their longest element constituting greater than, and cells 3 and 4 less than, 20% of the total mitochondrial length (Fig. S1A, y-axis); and (3) cell 3 has the maximum abundance of mitochondria with diameter greater than 1 arbitrary unit (we refrain from using the unit ‘µm’ due to the limitation described in the Materials and Methods) (Fig. S1B).

Mitochondrial fission (along with biogenesis and clearance) primarily determines the number of mitochondrial elements in a cell (Liesa and Shirihai, 2013). Thus, we obtained a quantitative fission metric, denoted [Fission], by normalizing the total number of mitochondrial elements (N) within a cell by the total length of its mitochondrial network (ΣL) (to account for variations in total mitochondrial content). On the other hand, mitochondrial fusion may increase the length of individual mitochondrial elements (Mitra et al., 2009). Thus, we obtained quantitative fusion metrics, denoted [Fusion(1–10)], from the percentage coverage of the top(1–10) longest elements. We obtained [Fusion(1,3,5,10)] by normalizing the longest (L₁), or the summed length of the three (L₃), five (L₅) or ten (L₅) longest components by the total length of the mitochondrial network (ΣL). Therefore, [Fission]_(N/ΣL) relies on mitochondrial number and [Fusion]_{(L1–10/ΣL)} relies on mitochondrial length, and thus these two metrics report distinct structural properties of a mitochondrial network. We also calculated the mean mitochondrial diameter in each cell, denoted [Diameter], by weighting the diameter of each element by its volume (details in the Materials and Methods). We found that cells 1 and 2 have higher [Fusion(1–10)] and lower [Fission] in comparison to cells 3 and 4, consistent with the qualitative evaluation of the images (Fig. 1A,B). The higher [Diameter] of cell 3 is due to several small and thick mitochondria, whereas the non-uniformly thick mitochondrial tubules of cell 1 do not increase [Diameter] (Fig. 1A,B).

The first level of validation of our metrics for fission or fusion involved 2 h exposure to the carbon source galactose or the protonophore carbonyl cyanide m-chlorophenylhydrazone (CCCP). CCCP prevents and galactose promotes fusion, thus altering the steady-state mitochondrial morphology (Mishra and Chan, 2016). We used immortalized human cells (HaCaT cells) stably expressing our mitochondrial matrix-targeted photoswitchable mOrange2 (mito-PSmO2) probe, the basal fluorescence of which can be photoswitched (Subach et al., 2011). Using the basal mito-PSmO2 fluorescence, we detected expected differences in fission and fusion when comparing CCCP with dimethyl sulfoxide (DMSO), and galactose with glucose. We detected higher [Fission] and lower [Fusion(1–10)] in CCCP-treated cells, with opposite trends in cells in galactose (Fig. S1C). Also, CCCP-treated cells have higher [Diameter] (Fig. S1C). The dynamic range (value_max/value_min) is maximal for [Fusion1] and progressively decreases with [Fusion(1–10)] (Fig. S1D), indicating that inclusion of fewer elements, as in [Fusion1], increases metric sensitivity. More importantly, with linear regression analyses of the single-cell data (assessed by R² and P-value), we detected an inverse [Fission] versus [Fusion(1–10)] relationship, i.e. [Fission] increases linearly with decrease in [Fusion(1–10)] (Fig. 1C). These data are consistent with the consensus that fission and fusion counter each other (Hoppins, 2014), and reveal the quantitative relationship between opposing fission and fusion processes contributing to steady-state mitochondrial morphology. The optimal linearity between [Fission] and [Fusion] metrics is achieved with [Fusion5] or above (R² and P-value from Fig. 1C in Fig. 1D), indicating that inclusion of a greater number of long elements (five or ten) better represents the overall fusion status of the cell. We also find that [Diameter] decreases linearly with increase in [Fusion(1–10)] [Fig. 1E shows only Fusion(5)], the significance of which remains unknown. Next, we reasoned that enhancement or suppression of any biological process would respectively increase or decrease the variability in the metric measuring the process. Indeed, the variation (expressed as variance) in [Fusion(1–10)], and not in [Fission], is markedly reduced with CCCP treatment, which suppresses fusion, and increased in galactose, which enhances fusion (Fig. 1F).

To further validate [Fusion(1–10)] obtained from MitoGraph v2.1 analysis of diffraction-limited micrographs, we assessed matrix continuity arising due to fusion of the double membranes between contiguous mitochondria. We refined our previously reported matrix continuity assay (Mitra et al., 2009) by photoswitching mito-PSmO2 and simultaneously measuring the dilution of the photoswitched fluorescence (FL₆₃₃) and the recovery of the bleached basal fluorescence (FL₅₅₅) (Fig. 1G). Our assay measures recovery of fluorescence over 2 min and controls for physical movement of mitochondria in and out of the regions of interest (ROIs), indicated by decrease in both basal and photoswitched pools (Fig. S1E, dashed lines). Therefore, our assay for quantification of [Matrix-continuity] is distinct from the reported assays of fusion performed over 30 min (Karbowski et al., 2004; Mishra et al., 2014; Xie et al., 2015). In our assay, we denoted the linear coefficient of the fitted curve for decay of FL_Ex-633/555 in an ROI as its [Matrix-continuity] (Fig. 1H). As expected, [Matrix-continuity] is less in the CCCP-treated cells than in the DMSO-treated cells (Fig. S1F). Importantly, linear regression analyses of the CCCP and DMSO groups combined indicate that matrix continuity increases linearly with fusion and thus validates [Fusion(1–10)] (Fig. 1I,J). Notably, [Matrix-continuity] decreases linearly with [Fission], but only modestly (Fig. 1J, right). The dynamic range of [Fusion5] is comparable to that of [Matrix-continuity], while that of [Fusion1] remains maximum (Fig. S1G).

Finally, we validated [Fission] and [Fusion(1,5)] using Mitotracker-633-stained mouse embryonic fibroblasts (MEFs) lacking the key fusion proteins mitofusin 1/2 (MFN1/2) (Chen et al., 2003) or the key fission protein dynamin-related protein 1 (Drp1) (Ishihara et al., 2009). As expected, MFN1/2-double knockout (DKO) cells exhibit higher [Fission] and lower [Fusion5] compared to the corresponding wild-type (WT), WTm, cells (Fig. 2A; Fig. S2A). Also, MFN1/2-DKO cells exhibit a robust increase in [Diameter], as validated by direct manual measurements (in μm) (Fig. S2B). As expected from loss of fusion, the inverse [Fission or Diameter] versus [Fusion5] linear relationship present in WTm cells is lost in MFN1/2-DKO cells (Fig. 2B). Also, consistent with the loss of fusion in the absence of MFN1/2, the variance of [Fusion(1,5)] is markedly reduced in MFN1/2-DKO in comparison to WTm cells (Fig. 2C); increased variance in Fission in MFN1/2-DKO cells may indicate unopposed fission of variable degree between cells. Similar comparison of Drp1-knockout (KO) and corresponding wild-type, WTd, MEFs shows significant increase only in [Diameter] (Fig. S2C), as noted by visual analyses of the confocal micrographs (Fig. 2D). The inverse [Fission] versus [Fusion5] linear relationship is modestly modified in Drp1-KO MEFs (Fig. 2E, left; reflected in the modest change in the slope of the regression lines). The inverse [Diameter] versus [Fusion5] relationship is not affected, although [Diameter] is markedly elevated in Drp1-KO MEFs (Fig. 2E, right). However, maximal reduction of the variance of [Fission] in Drp1-KO in comparison to WTd MEFs (Fig. 2F) is consistent with loss of regulation of fission by Drp1.

In summary, using various mitochondrial structural probes, we identified metrics for quantifying the contribution of the opposing fission and fusion processes to the mitochondrial dynamics state in single cells. We validated the metrics in metabolically distinct cells and in fission/fusion mutants. Notably, the steady-state [Fission] and [Fusion] metrics do not quantify the kinetics of fission and fusion that may be influenced by mitochondrial motility and other factors. Additionally, because [Fusion(1–10)] are derived from mitochondrial length, they do not reflect transient fusion events that do not result in increase in mitochondrial length.

To identify and quantify mitochondrial [dynamics] versus [energetics] relationships, we designed the high-resolution confocal microscopy-based approach, mito-SinCe². This involves single-cell quantitative analyses of the state of mitochondrial dynamics (using metrics for fission, fusion, matrix continuity and diameter as in Figs 1 and 2) and energetics (metrics for ATP or redox, using reported genetically encoded fluorescent ratiometric probes to rule out influence of mitochondrial mass). The mito-SinCe² workflow involves two independent arms: Dynamics-ATP and Dynamics-Redox (Fig. 3A; Fig. S3A,B). The ATP arm uses the mito-GFP- and OFP-based ATP (mito-GO-ATeam2) Förster resonance energy transfer (FRET) probe (FL_Em-555/488) (Nakano et al., 2011) to quantify the basal steady-state ATP levels in the mitochondrial matrix, denoted [ATP]. We used the direct excitation of the FRET acceptor (FL_Ex-555) to measure dynamics. The redox arm uses the mito-reduced-oxidized GFP (mito-roGFP) probe (FL_Ex-405/488) to quantify the oxidation status of the mitochondrial matrix (Hanson et al., 2004), denoted [Oxidation]. The dynamics metrics obtained with the functional mito-roGFP probe are not different between the CCCP- and galactose-treated cells (Fig. S3C), indicating that mito-roGFP cannot be used as a structural probe. Therefore, we used the fluorescently compatible mito-PSmO2 probe to quantify dynamics, in combination with mito-roGFP (as validated in Fig. 1C–F).

First, we validated the mito-roGFP and mito-GoATeam2 probes. For the widely used mito-roGFP probe, we confirmed that the oxidizing agent tertiary-butyl hydroperoxide (t-BH) increases [Oxidation] and the reducing agent dithiothreitol (DTT) causes no reduction beyond the basal redox levels (Fig. S3D). It is noteworthy that mito-roGFP detects oxidation status but does not detect reactive oxygen species directly. We extensively validated the newer mito-GO-ATeam2 probe for quantification of both ATP and dynamics. Mitochondrial ATP synthesis is coupled to oxygen consumption via the mitochondrial transmembrane potential (ΔΨ) maintained by electron transport chain (ETC) complexes (Hill et al., 2012). To validate mito-GO-ATeam2, we used mitochondrial inhibitors that reduce mitochondrial ATP levels. The inhibitors used were Oligomycin (inhibits ATP synthase/ATPase), Antimycin (inhibits complex III) and CCCP (uncouples ATP synthesis and oxygen consumption by dissipating ΔΨ) (Brand and Nicholls, 2011). The cell lines used were paired WTd and Drp1-KO MEFs, paired WTm and MFN1/2-DKO MEFs, normal human HaCaT line, and paired chemosensitive-A2780-IP and chemoresistant-A2780-CP ovarian cancer lines. We noted that mito-GO-ATeam2 fluorescence is enriched in certain specks in ∼10% of the cells. These specks colocalize with Mitotracker (not shown), are bioenergetically less active (Fig. S3E), are continuous with the surrounding mitochondrial elements (Fig. S3F), and their abundance can be reduced in the presence of galactose (Fig. S3G). We excluded cells containing these specks from our analyses as they are of unknown significance and they interfere with MitoGraph-based quantification, which requires uniform signal intensity in individual cells. For validating mito-GO-ATeam2 as an ATP probe, we exposed cells to the inhibitors (10 min) and computed the pre-inhibitor/post-inhibitor ratio of [ATP] to signify the [Inhibitor-sensitive-ATP], with the ‘pre’ value signifying the [Basal-ATP] (Fig. 3B); [Inhibitor-sensitive-ATP] value >1 indicates inhibition, whereas 1 indicates no change. Using mito-GO-ATeam2, we detected a 1- to >4-fold reduction in [Basal-ATP] caused by the mitochondrial inhibitors. Under the same conditions, these inhibitors impact oxygen consumption rate (OCR) measured by Seahorse metabolic flux analyses (Fig. S3H). [Oligomycin-sensitive-ATP] or [Antimycin-sensitive-ATP] increases with increase in [Basal-ATP], which is not observed for [CCCP-sensitive-ATP] (Fig. 3B). Importantly, as expected, the mean [CCCP-sensitive-ATP] for each cell line, obtained using mito-GO-ATeam2, increases with increase in the [ATP-linked-OCR] of that cell line, obtained using metabolic flux analyses (Fig. 3C). Thus, we could successfully validate the use of mito-GO-ATeam2 as an ATP probe in mito-SinCe². Next, the use of mito-GO-ATeam2 for measuring the metrics for dynamics was validated by obtaining the following results: (1) inverse [Fission] versus [Fusion5] relationship (Fig. S3I); (2) CCCP induced a decrease in [Fusion (1,5)] in the majority of the cells in six lines but not in MFN1/2-DKO cells (Fig. S3J); (3) CCCP induced reduction of variance of [Fusion(1,5)] combined from all cell lines (Fig. S3K), similar to that of fusion-defective MFN1/2-DKO cells (Fig. 2C); (4) expected reduction in [Fission] in the Drp1-KO and decrease in [Fusion (1,5)] in the MFN1/2-DKO MEFs in comparison to their respective WT MEFs (not shown); notably, Mito-GO-ATeam2 appears to be more sensitive than Mitotracker-633 at detecting differences between WT and KO MEFs.

Next, we provide proof of principle of the ATP arm of mito-SinCe². Mitochondrial ATP output can influence mitochondrial fusion (Mishra and Chan, 2016). Therefore, we used the ATP arm of the mito-SinCe² to analyze the impact of Oligomycin-induced inhibition of mitochondrial ATP synthase on the dynamics metrics, in the presence and absence of the key fission and fusion proteins Drp1 and MFN1/2, respectively. Thus, we performed linear regression analyses of Oligomycin-sensitive [ATP] and [Fusion] (experimental plan as in Fig. 3B); >1 value indicates inhibition, 1 indicates no change and <1 indicates induction (Fig. 3D). [Oligomycin-sensitive-Fusion1] increases linearly with [Oligomycin-sensitive-ATP] in the Drp1-KO MEFs, while this linear relationship was not significant in the WTd MEFs (Fig. 3D, left). A similar trend was observed with [Oligomycin-sensitive-Fusion5] (not shown). This indicates that fusion decreases linearly with inhibition of ATP synthesis in the absence of Drp1, while any non-linear nature of the relationship in the presence of Drp1 remains to be identified. Thus, our data indicate that Drp1 may modify the reported regulation of fusion by ATP (Mishra and Chan, 2016). No statistically significant linear relationship was observed in the MFN1/2-DKO/WTm pair (Fig. 3D, right). Note that the WTd MEFs, obtained at embryonic stage E10 (Ishihara et al., 2009), and WTm MEFs, obtained at embryonic stage E13 (Chen et al., 2003), have different spreads of data in the bivariate plots (compare left and right panels in Fig. 3D).

Next, we provide proof of principle of the redox arm of mito-SinCe². Here, we investigated the impact of CCCP-driven disruption of mitochondrial potential, given that CCCP can either induce or prevent mitochondrial oxidation (Izeradjene et al., 2005; Murphy, 2009). We found that [Oxidation] linearly increases with [Fusion5] and [Matrix-continuity] within the CCCP-treated population, while the linear relationship was not statistically significant in the DMSO-treated cell population (Fig. 3E, left and middle). Notably, CCCP increases [Oxidation] only in cells with [Fusion5] comparable to that in DMSO-treated cells (Fig. 3E, left). It remains to be tested whether the CCCP-driven oxidation in cells with higher mitochondrial fusion is due to reported elevated oxidative phosphorylation in these cells (Liesa and Shirihai, 2013; Schrepfer and Scorrano, 2016). No statistically significant linear relationship was detected between [Oxidation] and [Fission] (Fig. 3E, right).

Thus, we provide proof of principle of the mito-SinCe² approach by validating and refining reported findings. Hereafter, we will employ the approach to address biologically relevant questions (Figs 4–7), and to generate predictions for future experimental validations (Fig. 8). To investigate the various other aspects of quantitative relationships between the status of mitochondrial dynamics and ATP or redox, the data generated by mito-SinCe² can be meaningfully studied in various other ways, supported by appropriate statistics. Also, the current mito-SinCe² probes can be replaced and validated with other compatible ATP or redox probes as appropriate.

Given our interest in ovarian cancer (Tanwar et al., 2016), we compared chemosensitive A2780-IP with chemoresistant A2780-CP cells derived after carboplatin treatment of the A2780-IP cells (Landen et al., 2010). Chemoresistance acquisition is associated with enrichment of ovTICs (Ishiguro et al., 2016). Thus, compared to A2780-IPs, A2780-CPs are enriched in cells with enhanced activity of a functional ovTIC marker, aldehyde dehydrogenase (Aldh) (detected by ALDEFLUOR staining) (Landen et al., 2010) (Fig. S4A). Using mito-SinCe² on A2780-IPs and A2780-CPs, we did not find any difference in basal [ATP], [Fission], [Fusion(1,5)], [Diameter] or [Oxidation] between the cell lines (Fig. S4B). However, compared to the A2780-IPs, the A2780-CPs have significantly higher [Oligomycin-sensitive-ATP] associated with greater Oligomycin-induced decrease in [Fusion(1,5)], and increase in [Fission] and [Diameter]; the increase in [Oxidation] remained insignificant (Fig. 4A). These data demonstrate that steady-state mitochondrial ATP levels and dynamics status are more dependent on active ATP synthesis in the ovTIC^Aldh+-enriched A2780-CPs than in the A2780-IPs. In contrast, mitochondrial energetics and dynamics status are equally dependent on mitochondrial transmembrane potential in the A2780-IPs and A2780-CPs, as the cell lines are comparably impacted by disruption of mitochondrial potential (by CCCP) or inhibition of mitochondrial complex III (by Antimycin) (Fig. S4C,D). Bivariate analyses of [Oligomycin-sensitive-ATP] and [Oligomycin-sensitive-Fusion1] distinguishes the A2780-CP cells contributing to the elevated [Oligomycin-sensitive-ATP] (Fig. 4B). Interestingly, [Oligomycin-sensitive-ATP] increases with [Oligomycin-sensitive-Fusion1] up to a certain level and decreases thereafter (as indicated by a non-linear regression model) (Fig. 4B, left); no such relationship is observed in the A2780-IPs. Thus, specifically in the A2780-CPs, Oligomycin-driven inhibition of ATP synthesis inhibits fusion only up to a certain threshold (Fig. 4B, left, gray cells), beyond which this relationship does not hold true (Fig. 4B, left, red cells). These red cells are identified as distinct outliers [by interquartile range (IQR) analyses] in the bivariate analyses of [Oligomycin-sensitive-ATP] and [Oligomycin-sensitive-Fusion5], while these metrics linearly increase with each other within the gray cell population (Fig. 4B, right). Bivariate analyses of [Oligomycin-sensitive-ATP] and [Oligomycin-sensitive-Fission] demonstrate that Oligomycin induces fission in all A2780-CPs, while Oligomycin can induce or suppress fission in A2780-IPs (P-value obtained from Fisher's exact test) (Fig. 4C).

In summary, mito-SinCe² analyses reveal that the mitochondrial dynamics status is more dependent on mitochondrial energetics in the ovTIC^Aldh+-enriched A2780-CPs in comparison to the parental A2780-IPs. It remains to be tested whether and how mitochondrial dynamics modulates energetics in this system, which needs specific experimental design.

Mito-SinCe² analyses identified uniqueness in mitochondrial energetics in the ovTIC^Aldh+-enriched A2780-CPs in comparison to the A2780-IPs (Fig. 4A–C). Thus, we compared mitochondrial energetics in conventional ways between the A2780-IP and A2780-CPs in high-density cultures that further enhance the expression of the ovTIC marker, Aldh1A (Fig. S4E). First, bioenergetic profiling from mitochondrial OCR (details in the Materials and Methods) showed that the A2780-CPs have higher reserve respiratory capacity due to their elevated maximal respiration (Fig. 4D). These data suggest that the wiring of mitochondrial energetics in A2780-CPs may allow them to handle mitochondria-related stress situations more efficiently than A2780-IPs. Second, flow cytometric analyses of the A2780-CPs stained with the potentiometric dye tetramethylrhodamine ethyl ester (TMRE) reveal a population with markedly low TMRE uptake, thus defining distinct ΔΨ^lo and ΔΨ^hi populations (Fig. 4E), which can be visibly confirmed in confocal micrographs (boxes in Fig. 4F). The ΔΨ^lo peak is also detected in our newly derived paclitaxel-resistant A2780-PX line and a naturally carboplatin-resistant ovarian cancer line, SKOV-3 (Landen et al., 2010) (Fig. S4F,G). The non-potentiometric MTG stain does not detect distinct ΔΨ^lo and ΔΨ^hi populations (Fig. S4F), confirming that the ΔΨ^lo population is not due to reduced mitochondrial mass. Consistently, the dynamic range of signal from single cells (in confocal micrographs) stained with MTG or mitochondrial markers is 5-fold lower than that of TMRE where the detected ΔΨ^lo cells markedly increase the dynamic range (Fig. S4H).

To determine whether ΔΨ is linked to ovTIC self-renewal/proliferation, we took advantage of the previously reported TMRE-based cell sorting (Schieke et al., 2008; Sukumar et al., 2016) and combined it with ALDEFLUOR-based ovTIC sorting. Using this approach, we isolated four populations; namely, Aldh⁺ΔΨ^hi, Aldh⁺ΔΨ^lo, Aldh⁻ΔΨ^hi and Aldh⁻ΔΨ^lo (Fig. 4G). We used the standard extreme limiting dilution assay (ELDA) (Hu and Smyth, 2009) to determine ovTIC frequencies of each sorted population. Towards this end, we quantified the frequency of tumorspheres formed in tumor-initiating cell (TIC) medium and low-attachment conditions that allow ovTICs to self-renew and proliferate in vitro (Schultz et al., 2016) (details in the Materials and Methods). Expectedly, the ovTIC frequency is ∼200-fold higher in the Aldh⁺ populations than in the Aldh⁻ populations, irrespective of ΔΨ (Fig. 4H; Fig. S4I,J). The Aldh⁻ population with attenuated tumorsphere-forming ability still maintains cell proliferation (not shown). Interestingly, the ovTIC frequency in the Aldh⁺ΔΨ^hi population is ∼10-fold higher than that in the Aldh⁺ΔΨ^lo populations (Fig. 4H; Fig. S4I,J). Importantly, the Aldh⁺ΔΨ^hi and Aldh⁺ΔΨ^lo populations equilibrate to form a population with an intermediate ΔΨ in the tumorspheres, while Aldh⁻ΔΨ^hi and Aldh⁻ΔΨ^lo maintain their original ΔΨ^hi or ΔΨ^lo status (Fig. 4I); the Aldh^+/− populations harbor a proportionally comparable number of ΔΨ^hi/lo cells (Fig. 4G) and the Aldh⁺ status does not change during ΔΨ equilibration (not shown). Consistent with ΔΨ driving ATP synthesis, we investigated whether the ovTICs^Aldh+, which equilibrate between ΔΨ^lo and ΔΨ^hi states, are dependent on mitochondrial ATP synthesis as observed with certain other TICs (Viale et al., 2015). Indeed, 3-day exposure to a picomolar dose of Oligomycin significantly reduced survival of Aldh⁺ cells, not affecting the Aldh⁻ cells (Fig. 4J).

Thus, we further employed mito-SinCe² to investigate any existing linear relationships between the status of mitochondrial dynamics and energetics in the ΔΨ^hi/lo populations that equilibrate in mitochondrial energy-dependent ovTICs^Aldh+ of the A2780-CPs.

Here, we describe the differences in mitochondrial dynamics status in the identified ΔΨ^hi/lo and the sorted Aldh^+/− A2780-CPs, as detected by the validated [Fission], [Fusion5] and [Diameter] metrics; [Fusion5] is particularly optimal and thus chosen (Fig. 1D,I).

Since the ΔΨ^hi/lo cells can be distinguished in the confocal micrographs (Fig. 4F), we quantified the dynamics metrics using the MTG signal in TMRE and MTG co-stained cells (given that the functional TMRE signal cannot be used for this purpose). We compared cells in glucose and galactose, because galactose elevates OCR (Fig. S5A) and stimulates mitochondrial fusion (Fig. 1C,F) (Mishra and Chan, 2016). Replacement of glucose with galactose reverts ΔΨ in both the ΔΨ^hi/lo groups within 2 h (Fig. 5A), reducing the markedly high dynamic range of [ΔΨ] within the ΔΨ^lo group (Fig. 5B). We noted that the ΔΨ^hi group has higher [Fusion5], and lower [Fission], than the ΔΨ^lo group in the presence of glucose or galactose (Fig. S5C). Bivariate analyses of [Fission] and [Fusion] identified the subset of ΔΨ^hi cells having higher [Fusion5], and lower [Fission], than the ΔΨ^lo group (Fig. 5C, left, arrow pointing to the abundance of ΔΨ^hi cells). Linear regression analyses confirmed that [Fission] decreases linearly with increase in [Fusion5] within both ΔΨ^hi/lo groups in the presence of glucose or galactose (Fig. 5C). [Diameter] is modestly higher in the ΔΨ^hi group and is further increased by galactose (Fig. S5C), while a linear decrease in [Diameter] with increase in [Fusion5] is detected only within the ΔΨ^lo group in galactose (Fig. 5D). The significance of these findings related to [Diameter] is yet to be elucidated.

Next, we performed similar analyses on sorted Aldh^+/− cells maintained in Roswell Park Memorial Institute (RPMI) growth medium or TIC medium to promote self-renewal and proliferation of ovTICs (see Materials and Methods). Analyses of Mitotracker-633-stained cells showed higher [Fusion5], and lower [Fission], in the Aldh⁺ group than in the Aldh⁻ group maintained in RPMI growth medium (Fig. S5D). Bivariate analyses of [Fission] and [Fusion] identified the subset of Aldh⁺ cells having higher [Fusion5] and lower [Fission] than the Aldh⁻ group in RPMI growth medium (Fig. 5E, left, arrow pointing to the abundance of Aldh⁺ cells). However, TIC medium maintains markedly higher [Fission] and lower [Fusion5] in comparison to the RPMI medium, particularly in the Aldh⁺ group (Fig. 5E; Fig. S5D). Also, TIC medium dramatically increased [Diameter] in both Aldh^+/− groups (Fig. 5F; Fig. S5D). Linear regression analyses confirmed that [Fission] or [Diameter] increase linearly with decrease in [Fusion5] within both Aldh^+/− groups in RPMI and TIC media (Fig. 5E,F).

Therefore, ΔΨ^hi and Aldh⁺ groups harbor cells with unopposed fusion with >80 [Fusion5]. This implies that the Aldh⁺ΔΨ^hi group, which harbors maximum ovTIC frequency (Fig. 4H), has unopposed fusion state in RPMI medium that shifts to an unopposed fission state after exposure to TIC medium.

Mitochondrial fusion has been bidirectionally linked with mitochondrial ATP output (Liesa and Shirihai, 2013; Mishra and Chan, 2016; Schrepfer and Scorrano, 2016). To verify and quantify such a relationship in the ΔΨ^lo/hi groups, we employed the ATP arm of mito-SinCe² (as in Fig. 3). Specifically, we tested whether ΔΨ^lo/hi groups of the A2780-CPs have differential [Oligomycin-sensitive-ATP] in relation to the differential basal [Fission] or [Fusion5]. We identified ΔΨ^hi/lo groups in the TMRE-stained mito-GO-ATeam2-expressing cells by generating [ΔΨ] histograms (see Materials and Methods) (Fig. S6A,B).

K-means clustering analyses identified two distinct clusters in the ΔΨ^lo group (predicted by gap statistics, see Materials and Methods), which can be separated by [Fusion5] values into ΔΨ^lo-Fusion5^hi and ΔΨ^lo-Fusion5^lo clusters (Fig. 6A). ΔΨ^hi cells can be assigned to either of the two ΔΨ^lo clusters by their minimum Euclidean distance from the clustered ΔΨ^lo cells. As expected, similar assignment puts the CCCP-treated cells in the Fusion5^lo and the galactose-utilizing cells in the Fusion5^hi clusters (Fig. 6A). Importantly, the ΔΨ^lo-Fusion5^lo cells have markedly higher [Oligomycin-sensitive-ATP] in comparison to the ΔΨ^lo-Fusion5^hi and ΔΨ^hi cells (Fig. 6B). Next, we performed linear regression analyses between parameters (Fig. 6C; ‘+’ or ‘−’ signify direct or inverse linear relationships, respectively; INS signifies no statistically significant linear relationship due to P>0.05). Surprisingly, we detected no linear relationship between [Oligomycin-sensitive-ATP] and [Fusion5] within the ΔΨ^lo-Fusion5^lo group that had maximum [Oligomycin-sensitive-ATP] (Fig. 6C, left). Only within the ΔΨ^lo-Fusion5^hi group do [Oligomycin-sensitive-ATP] and [Fusion5] increase linearly with each other (Fig. 6C, right); [Fission] has opposite relationships to [Fusion5]. The ΔΨ^hi group also did not harbor any statistically significant relationship between [Oligomycin-sensitive-ATP] and [Fusion5] or [Fission] (Fig. S6C). Notably, [ΔΨ] and [Fusion5] increase linearly with each other within the ΔΨ^hi cells falling in the ΔΨ^lo-Fusion5^hi cluster (Fig. 6D), while no such relationship is observed within the ΔΨ^lo group, in which [ΔΨ] decreases linearly with [Diameter] (Fig. S6D). The significance of such relationships of [ΔΨ] with mitochondrial [Fusion] or [Diameter] remains to be investigated further. [Diameter] does not have any significant relationship with [Oligomycin-sensitive-ATP] in ΔΨ^hi/lo groups (not shown).

[Oligomycin-sensitive-ATP] is proportional to ATP synthesis, which is proportional to ATP consumption, and modulated by ADP levels in steady state (Brand and Nicholls, 2011). Notably, enhancement of ATP synthesis reduces ΔΨ, provided the ETC activity is constant. Therefore, elevated [Oligomycin-sensitive-ATP] in the ΔΨ^lo group (Fig. 6B) can potentially reduce ΔΨ to maintain the ΔΨ^lo state. Indeed, ΔΨ linearly decreases with increase in [Oligomycin-sensitive-ATP] in the ΔΨ^lo group, but not in the ΔΨ^hi group (Fig. 6E). More importantly, inhibition of ATP synthesis by Oligomycin (10 µM for 15 min) increases the [ΔΨ] of the ΔΨ^lo group, while slightly decreasing the ΔΨ^hi peak, which converts the ΔΨ^lo-ΔΨ^hi groups to a single ΔΨ group (Fig. 6F). These results indicate that differential [Oligomycin-sensitive-ATP] may reflect differential ATP synthase activity between the ΔΨ^hi/lo group. Since Oligomycin also inhibits the reverse ATPase activity, the [Oligomycin-sensitive-ATP] reflects a smaller decrease than expected from ATP synthase inhibition.

In summary, quantitative mito-SinCe² analyses reveal that a functional link between fusion and ATP synthesis is established beyond the value of 80 [Fusion5] in the ΔΨ^lo-Fusion5^hi state. Such a level of fusion may represent a ‘hyperfused’ state that is linked to enhanced ATP production (Liesa and Shirihai, 2013; Schrepfer and Scorrano, 2016). Importantly, dynamics is not related to ATP in the ΔΨ^lo-Fusion5^lo state, although it maintains higher Oligomycin-sensitive mitochondrial ATP. Also, the ΔΨ^hi state associates with an unopposed fusion state, while the ΔΨ^lo state may be maintained by elevated ATP synthesis (and minimum ETC regulation).

Mitochondrial redox states can be modulated by mitochondrial ATP synthesis (Murphy, 2009), and mitochondrial dynamics and redox states can influence each other (Willems et al., 2015). Here, we performed the redox arm of mito-SinCe² (as in Fig. 3) to investigate whether the identified [ATP] versus [dynamics] linear relationships (Fig. 6C) can be linked to [redox] versus [dynamics] linear relationships in the ΔΨ^lo/hi groups of A2780-CPs. Like in the HaCaT cells (Fig. 3E, DMSO), regression analyses did not detect any linear relationship between [Oxidation] versus [Fusion5] or [Matrix-continuity] in the A2780-CPs (Fig. S6E). However, unlike HaCaT cells (Fig. 1J), [Matrix-continuity] modestly decreases linearly with increase in [Fusion5] in the A2780-CPs (Fig. S6F) (this can arise due to discoordination between fusion of outer and inner mitochondrial membranes in the cancer line).

We identified the ΔΨ^hi/lo groups in mito-roGFP and mito-PSmO2 co-expressing cells modifying the strategy described in Fig. S6A (Fig. S6G). Interestingly, [Oxidation] is maximum in the ΔΨ^hi group (Fig. 7A), without a statistically significant linear relationship between [Oxidation] and ΔΨ in this group. Notably, in the ΔΨ^hi group, the low [Oligomycin-sensitive-ATP] does not linearly change with [dynamics] (Fig. S6C). Although, Oligomycin (10 µM for 10 min) oxidizes cells in both the ΔΨ^lo/hi groups (Fig. S6H), [Oligomycin-induced-Oxidation] linearly increases with decrease in [ΔΨ] only within the ΔΨ^lo group (Fig. 7B), similar to increase in [Oligomycin-sensitive-ATP] (Fig. 6E). This raises the possibility that Oligomycin-induced inhibition of ATP synthesis in the ΔΨ^lo cells dissipates the ΔΨ consumed for mitochondrial ATP synthesis to cause matrix oxidation.

Next, to analyze [Oxidation] versus [dynamics] linear relationships, we partitioned the cells into Fusion5^hi/lo, which differ in [Oligomycin-sensitive-ATP] and its relationship with dynamics (Fig. 6A–C). Importantly, regression analyses show that [Oxidation] and [Fission] linearly increase with each other within both ΔΨ^hi-Fusion5^lo and ΔΨ^lo-Fusion5^lo groups (Fig. 7C,D; Fig. S6I); no linear relationship was detected between [ATP] and [dynamics] (Fig. 6C; Fig. S6C). Notably, [Fusion5] linearly decreases with [Oxidation] only in the ΔΨ^hi-Fusion5^lo group (Fig. 7D; Fig. S6I). It remains to be seen whether reactive oxygen species induce fission in the Fusion5^lo groups, similar to what has been reported in some systems (Willems et al., 2015). We noted that [Diameter] linearly decreases with increase in [Oxidation] within the ΔΨ^hi-Fusion5^hi group (Fig. 7D; Fig. S6I), in which [ΔΨ] and [Fusion5] linearly increase with each other (Fig. 6D). This interesting connection between [ΔΨ], [Fusion5], [Diameter] and [Oxidation] in the ΔΨ^hi-Fusion5^hi group can be further explored in pathological situations associated with swelling of mitochondria. Such a conundrum does not exist in the in the ΔΨ^lo group, in which [Diameter] linearly increases with decrease in [Matrix-continuity] (Fig. 7E).

Thus, comparison of the Mito-SinCe² relationships in the ATP (Fig. 6) and redox (Fig. 7) arms suggests that mitochondrial dynamics status changes linearly either with ATP or with redox, but not simultaneously with both.

To test whether mito-SinCe² analyses can have any potential predicting abilities, we further analyzed the quantitative mito-SinCe² metrics obtained from A2780-CPs described above. [Fusion] and [Fission] may not always hold opposite relationship with energetics metrics (Figs 1J, 3E and 4B,C). This highlights the distinction between the [Fusion] and [Fission] metrics and indicates they may not be interchangeably used, although an inverse linear relationship exists between them. However, the inverse [Fission] versus [Fusion5] linear relationship is not uniform between Fusion5^lo/hi cells, as reflected in the higher spread of data points from the regression line in the Fusion5^lo range (Fig. 6A). Such non-uniformity in inverse [Fission] versus [Fusion5] linear relationship was also observed in pooled data from various cell types (Fig. S3I). To probe further, we pooled data from Figs 5–7 to analyze the [Fission] residuals from linear regression analyses (indicating deviation from the regression line) separately in the Fusion5^lo and Fusion5^hi range. The [Fission] residuals progressively increase with decrease in [Fusion5] (Fig. 8A), with cells in CCCP and galactose occupying zones with maximal and minimal residuals, respectively (Fig. S7A). Further analyses of standard deviation of [Fission] in windows of five [Fusion5] units revealed that the variation in [Fission] increases at a constant rate with decrease in [Fusion5] (Fig. 8B). Thus, these data predict the existence of one or more factor(s) influencing the inverse [Fission] versus [Fusion5] relationship.

TIC and RPMI media maintain distinctly different statuses of dynamics between the Aldh^+/− cells (Fig. 5E,F). TIC medium maintains higher variance in [Fission] specifically in the Aldh⁺ cells, lower variance in [Fusion5] and higher variance in [Diameter] in both Aldh^+/− cells (Fig. 8C). This may predict that the TIC medium represses fusion in both Aldh^+/− cells, while maintaining a wider range of fission activity in the Aldh⁺ cells, similar to MFN1/2-DKO cells (Fig. 2C).

Next, we performed multivariate analyses to determine whether combination of the mito-SinCe² metrics can classify the cells in distinct ΔΨ^lo/hi clusters. Thus, we performed partial least square discriminant analyses (PLSDA) using [Fission], [Fusion5], [Diameter], [Oligomycin-sensitive-ATP-fraction] and [ΔΨ]. The ΔΨ^lo/hi cells form statistically significant overlapping clusters, with the majority of the cells in the non-overlapping zones and the overlapping zone consisting of the ΔΨ^lo-Fusion5^hi cells (Fig. 8D). Whereas [ΔΨ] contributes maximally to principal component 1, as expected, [Fusion5] contributes maximally to principal component 2 (Fig. S7B). This confirms that the ΔΨ^lo/hi clusters are indeed distinct, and the overlapping zone may ‘potentially’ signify the path of equilibration of the two clusters as the ovTICs^Aldh+ form tumorspheres (Fig. 4I). Thus, PLSDA of the mito-SinCe² metrics of any unknown cell can be used to potentially classify the cell and predict its TIC and energetic state.

In summary, mito-SinCe² can potentially make the following experimentally testable predictions: (1) new features that can influence the fission–fusion relationship; (2) whether any physiological or pathological condition particularly impacts fission and/or fusion; and (3) classification of cells into functional classes.

The microscopy-based high-resolution mito-SinCe² approach is the first step using single cells towards quantitative analyses of the interplay between mitochondrial dynamics (fission, fusion, matrix continuity and diameter) and energetics (ATP or redox) status. We designed and validated individual mito-SinCe² metrics (Figs 1 and 2), provided proof of principle of the mito-SinCe² approach by refining published findings (Fig. 3), employed the approach to address a biologically relevant question (Figs 4–7) and generated predictions to be tested experimentally in the future (Fig. 8). The mito-SinCe² approach can be used in specific experimental design to investigate a cause-and-effect relationship between dynamics and energetics (Fig. 4 as an example). Also, data generated from mito-SinCe² analyses can be analyzed in various other statistical ways to refine the linear relationships reported here, or to identify meaningful non-linear relationships between the mito-SinCe² metrics. Currently, mito-SinCe² remains a low-throughput approach and can be made high throughput with the development of high-resolution automated confocal microscopes. It is noteworthy that absolute values of the mito-SinCe² metrics may be impacted by the fluorescent probes in use, and the mito-SinCe² approach does not assess fission–fusion kinetics.

We employed mito-SinCe² to investigate the distinct ΔΨ^lo and ΔΨ^hi cell groups that equilibrate in ovTICs^Aldh+ (Fig. 4I) and likely maintain differential ATP synthesis (Fig. 6B). Based on mito-SinCe² analyses, we hypothesize that mitochondria-dependent self-renewing/proliferating ovTICs^Aldh+ interconvert between three mito-SinCe² states; namely, state 1 (ΔΨ^hi), state 2a (ΔΨ^lo-Fusion5^lo) and state 2b (ΔΨ^lo-Fusion5^hi) (Fig. 8D,E). We speculate that the reducing equivalents from carbon sources could build up ΔΨ in state 1, ATP synthesis in Fusion5^lo mitochondria with weakened energetic efficiency (Liesa and Shirihai, 2013) could dissipate ΔΨ in state 2a, and establishment of a direct relationship between ATP synthesis and fusion in state 2b could aid the transition from ΔΨ^lo to a ΔΨ^hi state. Importantly, dynamics relates to redox (but not to ATP) in states 1 and 2a, and to ATP (but not to redox) in state 2b.

Mito-SinCe² analyses indicate how TIC self-renewal may involve both unopposed fusion and fission states that have been individually linked to stemness (Chen and Chan, 2017). We propose that, during the cycling of the Aldh⁺ cells between ΔΨ^hi and ΔΨ^lo states (Fig. 8E), the Aldh⁺ΔΨ^hi cells with unopposed fusion state are ‘primed’ towards shifting to an unopposed fission state supported by the TIC medium (Fig. 5E,F). This explains the maximal ovTIC enrichment in the Aldh⁺ΔΨ^hi group (Fig. 4H); reduced ovTIC frequency of the Aldh⁺ΔΨ^lo group could be due to lag time for conversion to Aldh⁺ΔΨ^hi state through cell cycle transitions (not shown).

Regulation of TICs by mitochondrial energetics has been demonstrated in certain cases, but remains controversial (De Francesco et al., 2018). Our data indicate that interconversion between mitochondrial ATP^hi/lo states occurs during self-renewal and proliferation of ovTICs (Fig. 8E). Our data support a concept where the ATP^hiΔΨ^lo state is metastable (likely due to elevated ATP synthesis with minimal ETC regulation), and the stabilization of this state is achieved by conversion to an ATP^loΔΨ^hi state. The same phenomenon can be tested in other normal or neoplastic stem cells.

Currently, mito-SinCe² can be used on immortalized/primary cells, including patient cells. Expression of relevant probes in transgenic animals will allow in/ex vivo mito-SinCe² analyses. Appropriately targeted probe sets to other mitochondrial compartments would expand mito-SinCe² abilities, which currently focus on the matrix. Finally, mito-SinCe² analyses can be expanded to include other functional parameters, such as mitochondrial calcium, DNA, mitochondrial antiviral signaling proteins, moonlighting proteins on mitochondria, etc.

Cell lines were either purchased from American Type Culture Collection or provided as gifts [A2780-IPs and A2780-CPs by Dr C. Landen (University of Virginia, Charlottesville, USA); Drp1-KO and WTd MEFs by Dr K. Mihara (Kyushu University, Fukuoka, Japan); and MFN1/2-DKO and WTm MEFs by Dr D. Chan (California Institute of Technology, Pasadena, USA)]. The paclitaxel-resistant A2780 cell line was generated by culturing parental A2780-IPs in 3 nM paclitaxel, while changing the medium every alternate day. The resistant colonies were maintained in 3 nM paclitaxel and experiments were performed without paclitaxel. Cells lines were treated with BM-cyclin as required to eliminate any detected mycoplasma.

Media and biochemical reagents were obtained from Gibco or Sigma-Aldrich; ALDEFLUOR reagent was from Stem Cell Technologies. We subcloned the mito-roGFP (plasmid #49437 from Addgene) into the pCDH lentiviral vector with hygromycin selection. We constructed the mito-PSmOrange in the pCDH lentiviral vector with puromycin selection, by tagging mitochondrial targeting sequence of the human cytochrome oxidase VIII subunit to the previously reported PSmO2 [gift from Dr George Patterson (National Institutes of Health, Bethesda, USA)]. Mito-GO-Ateam2 was a gift from Dr H. Imamura (Osaka University, Osaka, Japan). Transfections, transductions and immunoblotting were carried out as previously described (Parker et al., 2015).

ALDEFLUOR staining was performed according to the manufacturer's protocol (Stem Cell Technologies) with modifications as required. We used 2 million cells/ml with 5 µl ALDEFLUOR reagent/ml. Stained cells were excited using a 488 nm laser with a 525/50 band-pass (BP) [505 long-pass (LP)] filter on an LSR II flow cytometer (BD Biosciences) and a 530/30 BP (505 LP) filter on the FACS Aria II (BD Biosciences). TMRE was detected using a 532 nm laser and 582/15 BP filter on the LSR II and using a 561 nm laser and the 576/26 BP filter on the FACS Aria II. Cell sorting was performed after applying appropriate gating and compensations. A recovery period (∼12 h) is necessary after ALDEFLUOR sorting before performing assays to investigate mitochondrial properties because both FACS and ALDEFLUOR staining impact mitochondrial morphology (data not shown).

Sorted cell groups were maintained in TIC medium, i.e. RPMI supplemented with 1× N1 supplement, 500 mg/ml insulin, 20 ng/ml human epidermal growth factor and 10 ng/ml basic fibroblast growth factor to a volume of 200 ml or 500 ml RPMI with 1% sodium pyruvate, 1% L-glutamine and 1% penicillin–streptomycin (Schultz et al., 2016). For using TIC medium in the limiting-dilution tumorsphere assay, we sorted cells into this medium and seeded 1, 10, 100 and 1000 cells in a 96-well ultra-low attachment plate, with TIC medium supplementation (10% of the medium volume in each well) performed every alternate day. After 15 days, spheres counts were analyzed using ELDA statistics, or the tumorspheres were dissociated by pipetting to analyze the ALDEFLUOR/TMRE profile.

For using TICs in the microscopy and survival/proliferation assays, we handled sorted cells as above and then allowed them to recover overnight on Geltrex in the presence of growth medium (RPMI) or TIC medium, because Geltrex maintains Aldh activity similar to suspension (data not shown; see below for microscopy). For survival/proliferation assays, we seeded sorted cells on Geltrex in TIC medium and allowed them to attain ∼100% confluence before treating them with Oligomycin (250 pM) or DMSO (250 pM) for 3 days. Cell survival/proliferation was determined by Crystal Violet staining, as previously described (Parker et al., 2015).

Cells were seeded in Labtek chambers (on Geltrex in case of sorted cells) at least 24 h before experimentation, and fresh medium was added 1 h before microscopy.

Confocal microscopy of live cells was carried out on a laser scanning confocal microscope (LSM 700, Zeiss Microscopy) using a 40× plan-apochromat/1.4 NA oil objective, in a temperature- and CO₂-controlled chamber. The following lasers were used for excitation: 488 nm for mito-GO-ATeam2, 405 nm and 488 nm for mito-roGFP, 555 nm for basal, 488 nm for photoswitching and 639 nm for photoswitched mito-PSmO2. Appropriate detectors were used in each case. Multichannel image acquisition was designed to rule out crosstalk and cross excitations, with automated switching lasers between channels after each scanned line. The 3D confocal images were acquired at optical zoom 3 with 1 Airy unit pinhole and 0.5 µm Z interval. Mean background corrected fluorescence intensity from single-cell ROIs were obtained from maximum-intensity projections of the optical slices, and appropriate ratios were obtained for the ratiometric analyses.

Staining with TMRE (or with Mitotrackers) was carried out as previously described (Mitra and Lippincott-Schwartz, 2010). Higher concentrations of TMRE and MTG, and increased laser power, were used for the ΔΨ^lo cells (for Fig. 5). We noted a higher abundance of cells with [Fission] >0.7 in the TMRE/MTG doubly stained A2780-CPs, which may indicate adverse effects of TMRE/MTG double staining because these cells are rare when other probes are used in this cell line. Thus, we excluded those cells from our analyses (Fig. S5B), although the overall conclusions remain unchanged with the inclusion of the cells.

To identify the ΔΨ^hi/lo populations of the TMRE-stained mito-GO-ATeam2- or mito-PSmO2-expressing cells, we designed the following strategy (Fig. S6A,G). To obtain the contribution from TMRE from the overlapping of the TMRE and mito-GO-ATeam2 or PSmO2 signals, we performed microscopy on the same cells before and after TMRE staining, and subtracted the pre-TMRE FL₅₅₅ value of an individual cell from its post-TMRE FL₅₅₅ value. We plotted the histogram profile of the TMRE fluorescence (bin width of 200) and identified the inflection point between ΔΨ^hi and ΔΨ^lo. We then used the inflection point and the difference between the lowest and highest values to normalize the TMRE fluorescence, allowing comparison of values from different experiments. The [ATP] and [Oxidation] were obtained from the pre-TMRE images, to avoid any impact of TMRE on those mitochondrial functions. Because we found that Oligomycin ablates the ΔΨ^lo population, pre- and post-Oligomycin images were acquired after measurement of ΔΨ with TMRE. This precluded measuring Oligomycin-driven changes in dynamics in these cells. In the ΔΨ^hi group identified in the redox arm, TMRE signal was saturated due to the higher laser power required to detect mito-PSmO2, precluding any regression analyses in this group.

To perform the photoswitching-based assay for matrix continuity assay, we first confirmed colocalization of mito-PSmO2 with MTG (not shown). Optimal conversion (>3-fold) of basal FL₅₅₅ signal of PSmO2 to FL₆₃₃ signal was obtained with 40–90% 488-nm laser power with two iterations in 50×50-pixel ROIs. Thereafter, the photoswitched pool was followed by time-lapse microscopy (2 µm optical slice) within 2 min at 15-s intervals. The ratio of the increase in basal fluorescence and concomitant decrease in photoconverted fluorescence was quantified, with appropriate background corrections, by determining the linear coefficient of the parabolic fit to the time lapse (Fig. 1H). Five cells, in which the whole mitochondrial population in each cell was photoconverted, were used to obtain the average bleaching rate in each experiment. Because the photoconverting 488-nm laser also bleaches mito-roGFP signal, our analyses of redox states are consistently from pre-photoconverted images. Neither mito-PsmO2 nor mito-roGFP can be used with mito-GO-ATeam2 due to overlapping fluorescence spectra.

The MitoGraph v2.1 software (https://github.com/vianamp/MitoGraph) involves an R script that uses the iGraph package to calculate mitochondrial network properties per mito-component. MitoGraph v2.1 was run on TIFF images of individual cells with a single relevant fluorescent channel by entering the following command string in the terminal: ./MitoGraph –path ∼/Desktop/FolderName –xy 0.104 –z 0.5 –analyze. The *.mitograph output files were combined in Microsoft Excel using a custom-built macro. After an initial quality control step (see below), the calculations of [Fission], [Fusion(1–10)] and [Diameter] were performed in Excel. We derived the [Diameter] metric using a weighting approach analogous to the calculation of atomic weight by isotope abundance. We weighted the diameter of each mitochondrial element (d_n, derived from volume and length, assuming the elements are cylindrical) with its percentage of the total mitochondrial volume (V_n/V_T) and calculated the weighted mean diameter {[Diameter]} of the mitochondria for each cell ⁠. In MitoGraph v2.1, mitochondrial components reduced to a single pixel during skeletonizing are arbitrarily given at least one neighbor on the xy plane. Therefore, [Diameter] is an overestimation in cells with many small components. Sometimes, this leads to aberrant values for component volume that could not be verified by manual measurements. These components are excluded from the [Diameter] but not the [Fission] calculations. If the sum of the component volumes was more than 0.5% off the total volume reported by MitoGraph, the reported total volume was replaced by the sum of the component volumes.

For quality control, the binary MitoGraph outputs were compared to the TIFFs to determine whether coverage was acceptable in randomly chosen cells. Binary images were definitively checked if the MitoGraph output indicated that the cell (1) had a total length of less than 100 µm; (2) had a total component number of less than 30 or greater than 100; or (3) had a longest fragment of less than 20 µm. After acquiring our metrics, the binary images of outliers were re-examined. The expected inverse relationship between [Fission] and [Fusion] parameters was confirmed with every data set. Although the newer MitoGraph v3.0 is now available, the primary differences between these versions of the software relate to post facto utility. However, use of MitoGraph v3.0 may allow additional parameters to be considered in the overall mito-SinCe² analyses.

This was performed on Seahorse XF24 platform (Agilent). Cells (25,000 for Fig. 3 or 100,000 for Fig. 4) were seeded and analyzed after 24 h. We included 10% fetal bovine serum in the assay medium, given that growth factor components in the serum can impact mitochondrial substrate utilization and thus mitochondrial energetics. After experimentation, cells were immediately fixed by formalin and stained with Crystal Violet to obtain total cell content, which was used to normalize OCR values. The bioenergetic profiling was performed with normalized OCR values, as in Brand and Nicholls (2011). Non-mitochondrial respiration was obtained by measuring OCR after Antimycin addition. Basal respiration was obtained by subtracting non-mitochondrial respiration from OCR. Proton leak was obtained by subtracting non-mitochondrial respiration from OCR after Oligomycin addition. Maximal respiration was obtained by subtracting non-mitochondrial respiration from the OCR achieved after CCCP addition as titrated. Reserve capacity was obtained as the difference between the maximal and basal respiration. We additionally computed the fold change in OCR in response to inhibitors (Oligomycin, CCCP and Antimycin) in order to compare with the mito-SinCe² results.

Data presented are either pooled from multiple replicates or presented as representative of multiple replicates, as appropriate. Student's t-tests (parametric) or Kruskal–Wallis tests (non-parametric), and linear regression analyses were performed with Microsoft Excel or SPSS.

K-means clustering was carried out using the ‘cluster’ package in R (https://www.r-project.org/) using default parameters. The number of clusters (k) was determined using the clusGap function of the R package ‘cluster’ and with visual examination using the fviz_nbclust function from factoextra package. Briefly, the ‘gap statistics’ of a cluster is the change in variance of a random set of points within the cluster boundaries and variance of the actual points, after clustering the points into k clusters. This statistic is calculated for different values of k. The final number of cluster was determined using the default method of ‘firstSEMax’. In this method, the optimum number of k is the smallest value of k for which the gap statistic value is within 1 standard error of the lowest local maximum value. The other points were assigned a cluster based on smallest average Euclidean distance of each point from the points in each cluster.

PLSDA is used to classify points based using partial least square regression. The analyses were carried out using R MixOmics package.

We used generalized additive model (GAM) to identify the best-fit curve for data showing non-linear relationship. We fit a cubic spline to the dataset using the ‘gam’ function from the R package mgcv. The R² estimate of each model was determined using the proportion of the variance explained by the fit as implemented in mgcv package.

For detection of outliers in a dataset, we used the standard IQR. In each dataset, we subtracted the lower quartile from the upper quartile to calculate the IQR. Points above the upper quartile or below the lower quartile beyond 1.5 times the IQR were considered as outliers.

To rule out false-positive discovery, our approach involved multiple testing correction for ten or more P-value comparisons between two given parameters or relationships on any data set. Statistical significance was considered with a P-value <0.05.

We acknowledge Drs G. Patterson and H. Imamura for sharing constructs; Drs K. Mihara, D. Chan and C. Landen for sharing cell lines; Drs S. Rafelski and M. Vienna for help with accessing and understanding MitoGraph capabilities; and J. Wirth for the custom-built Excel macro.