Regulation of Mitoflash Biogenesis and Signaling by Mitochondrial Dynamics

Mitochondria are highly dynamic organelles undergoing constant network reorganization and exhibiting stochastic signaling events in the form of mitochondrial flashes (mitoflashes). Here we investigate whether and how mitochondrial network dynamics regulate mitoflash biogenesis and signaling. We found that mitoflash frequency was largely invariant when network fragmentized or redistributed in the absence of mitofusin (Mfn) 1, Mfn2, or Kif5b. However, Opa1 deficiency decreased spontaneous mitoflash frequency due to superimposing changes in respiratory function, whereas mitoflash response to non-metabolic stimulation was unchanged despite network fragmentation. In Drp1- or Mff-deficient cells whose mitochondria hyperfused into a single whole-cell reticulum, the frequency of mitoflashes of regular amplitude and duration was again unaltered, although brief and low-amplitude “miniflashes” emerged because of improved detection ability. As the network reorganized, however, the signal mass of mitoflash signaling was dynamically regulated in accordance with the degree of network connectivity. These findings demonstrate a novel functional role of mitochondrial network dynamics and uncover a magnitude- rather than frequency-modulatory mechanism in the regulation of mitoflash signaling. In addition, our data support a stochastic trigger model for the ignition of mitoflashes.

(FM) manner, with activators including mitochondrial Ca 2+ , oxidants such as reactive oxygen species (ROS), and hyperosmotic stress; and depressors including ROS scavengers and mPTP inhibitors 16,24,27,28 . However, an important yet unanswered question is whether and how network dynamics interact with the newly-discovered functional dynamism, mitoflashes, which represents fundamental mitochondrial signaling events.
Here we investigate possible roles of mitochondrial dynamics in the regulation of mitoflash biogenesis and signaling by manipulating key dynamic proteins that mediate mitochondrial fission, fusion, and motility in a variety of cells. We performed parallel respirometric measurement and examined mitoflash responses to non-metabolic stimulations to appraise and separate contributions from altered respiratory function, if any. Our results indicate that, when mitochondrial network reorganizes or redistributes, mitoflash frequency remains constant at the cellular level provided that respiratory function is intact; at the same time, mitoflash signal mass is enhanced or decreased when mitochondrial network hyperfuses or fragmentizes, respectively. These data are unified in a stochastic trigger model of mitoflash biogenesis and uncover that changing mitochondrial network affords a magnitude-modulatory rather than FM mechanism in the regulation of mitoflash signaling.

Results
Mitofusin deficiency does not alter mitoflash frequency. We first examined mitoflash activity in response to genetic manipulations of Mfn1 and Mfn2, both are crucial for promoting the OMM fusion. In mouse embryonic fibroblasts (MEFs) of wild-type, Mfn1 −/− , Mfn2 −/− , and Mfn1/2 double-knockout (DKO) background, mitochondrial targeted, circularly permuted yellow fluorescent protein (mt-cpYFP) was expressed as the mitoflash biosensor, the fluorescence of which increases upon bursting superoxide production and transient alkalization 16,29 . As expected, deficiency of either or both mitofusins transformed mitochondrial morphology from the mostly tubular and reticular network in wild-type cells to a severely fragmented network consisting exclusively of punctiform mitochondria of round or oval appearance (Fig. 1A). Time-lapse imaging of wild-type MEFs revealed spontaneous and quantal mitoflashes, and individual events exhibited punctiform, tubular or reticular appearance, with the tubular ones being predominant (Fig. 1B). Mitoflash activity persisted in all three types of Mfndeficient cells, but they were all confined to single punctiform mitochondria. The rate of mitoflash occurrence was similar in wild-type and Mfn-deficient cells, at a level of about 0.4 events/cell/100 s (Fig. 1D). Parametric measurement showed that their spatial area and signal mass (amplitude × duration × area) were decreased by 55-70% in the absence of either or both mitofusins, while their amplitude and duration were unchanged ( Fig. 1C and Supplementary information, Figure S1). As mitoflash area is related to mitochondrial functional connectivity 23 , the composite parameter "signal mass" is preferred in the context of mitoflash signaling. When cells were challenged with hyperosmotic stress (270 mM glucose or sorbitol) to elevate its mitoflash activity via a non-metabolic mechanism 28 , nearly identical mitoflash frequency was attained, reaching ~3.0 events/cell/100 s, regardless of mitofusin deficiency. By measuring the oxygen consumption rate (OCR) using the Seahorse XF24 analyzer, we found that basal metabolism and maximal oxidative capacity remained unchanged with deficiency of either mitofusin, but exhibited mild decreases (14-20%) in Mfn1/2 DKO cells (Fig. 1E), the latter in agreement with previous findings 30 . These results indicate that the rate of occurrence of spontaneous or stimulated mitoflashes at the cellular level is largely independent of OMM fusion in the present experimental conditions. Mitochondrial motility is dispensable for mitoflash biogenesis. Mitochondrial movement along microtubules, which is predominantly mediated by Kif5b, is essential for mitochondrial fusion and intracellular mitochondrial distribution 14 . Next, we examined mitoflash activity in normal rat kidney (NRK) cells in which mitochondrial motility and consequentially mitochondrial fusion are completely abolished with Kif5b knockout 15 . Mitoflashes displayed a range of morphological shapes in wild-type NRK cells, punctiform, tubular, or reticular (Supplementary information, Figure S2A), with their average spatial areas of 0.68, 2.41 and 4.10 μ m 2 and their relative percentages of 33, 44 and 23%, respectively ( Fig. 2D and Supplementary information, Figure S2B). In Kif5b −/− cells with densely packed perinuclear mitochondrial network ( Fig. 2A), the same three types of mitoflashes occurred with their respective proportions and spatial areas all similar as those in wild-type cells ( Fig. 2D and Supplementary information, Figure S2B), and their combined mitoflash signal mass and frequency was also unchanged (Fig. 2B,C). Furthermore, loss of Kif5b did not alter mitoflash response to either hyperosmotic stimulation by 270 mM glucose or oxidative stimulation by 80 μ M menadione (Fig. 2C). The distribution of stress-induced mitoflashes among three morphological categories was not changed, either (Fig. 2D). Parallel experiments revealed that respiratory function remained intact in Kif5b-null cells (Fig. 2E). Thus, mitochondrial motility as well as mitochondrial fusion does not play any major role in the regulation of mitoflash biogenesis and its stress responses.

Opa1 deficiency suppresses mitoflashes and mitochondrial respiration.
Recently, Santo-Domingo et al. have shown that knockout of Opa1, an IMM fusion protein, completely abolished mitoflashes (named "pH flashes" that were indicated by the probe mitoSypHer in their report) in MEFs; and, based on this finding, they proposed a model in which mitoflash biogenesis is causatively linked to the events of IMM fusion 31 . We therefore revisited Opa1 effect on mitoflash activity in MEFs. As expected, Opa1 −/− MEFs displayed aberrant mitochondrial morphology: they were all dispersed as punctiform mitochondria (Fig. 3A). Nonetheless, we still recorded discrete punctiform mitoflashes occurring in these cells, albeit at a markedly decreased frequency. The mitoflash frequencies in the presence and absence of Opa1 were 0.43 and 0.05 events/cell/100 s, respectively (Fig. 3C). Individual mitoflashes in wild-type and Opa1 −/− cells were indistinguishable in terms of average amplitude and kinetics ( Fig. 3B and Supplementary information, Figure S3), indicating that the machinery of mitoflash biogenesis is still functional in the absence of Opa1. Using the ratiometric pH indicator mito-SypHer as reported in the work of Santo-Domingo et al. 31  Opa1-deficient cells (Supplementary information, Figure S4A,B). Moreover, hyperosmotic stress also elevated the mitoSypHer-reported mitoflash activity (Supplementary information, Figure S4C).
The very presence of mitoflashes of normal amplitude and time course in Opa1 −/− MEFs strongly suggests that Opa1 regulates mitoflash biogenesis via a mechanism other than regulating IMM fusion. In this regard, it has been shown that Opa1 plays a key role in regulating mitochondrial respiratory function because it alters the IMM crista structure, the respiratory supercomplex assembly, and hence the metabolic efficiency 32,33 . Direct respirometric measurement confirmed that loss of Opa1 significantly attenuated mitochondrial basal respiration and maximal oxidative capacity by 22.8 and 48.7%, respectively, in MEFs (Fig. 3D). This result lends support to the hypothesis that Opa1 deficiency secondarily impairs the biogenesis of spontaneous mitoflashes via a direct suppression of respiratory activity. To further test this possibility, we treated the wild-type cells with the respiratory chain inhibitors, including NaN3, oligomycin, and rotenone. These metabolic inhibitors attenuated the mitoflash frequency by ~90% while depressing the metabolic rate by ~50% (Supplementary information, Figure S5A,B). Concomitantly, mitochondrial membrane potential was similarly depolarized by ~50% in Opa1-mutant cells and in the presence of these metabolic inhibitors (Supplementary information, Figure S5C,D). These results indicate that inhibition of spontaneous mitoflash activity in Opa1-null cells can be largely accounted for by the depressed OCR, oxygen consumption rate. n = 3 independent experiments. *p < 0.05; **p < 0.01. metabolic rate and membrane potential. Furthermore, if this is the case, mitoflash response to stimuli bypassing metabolism might remain intact even in the absence of Opa1. Strikingly, we found that mitoflash frequency was greatly elevated by hyperosmotic stimulations in Opa1-deficient cells, to a level similar to that in wild-type cells under the same conditions (Fig. 3C). In other words, the lack of the pro-fusion protein Opa1 does not compromise the capacity of mitoflash response to hyperosmotic stress, excluding IMM fusion as a prerequisite of mitoflash biogenesis. Spatial synchrony in hyperfused mitochondria. Next, we sought to address whether and how mitochondrial hyperfusion regulates mitoflash frequency and signal mass. For this purpose, we knocked-down the pro-fission protein Drp1 using RNA interference (RNAi) or transiently expressed its dominant negative Drp1 K38A mutant 34 in HeLa cells stably expressing mt-cpYFP. Both Drp1 depletion and Drp1 K38A expression 31 promoted hyperfusion of mitochondria into extended tubular networks (Fig. 4A,B). Remarkably, mitoflash imaging revealed that, in > 78% cells, the entire intracellular mitochondrial network underwent synchronous activities ( Fig. 4A and Supplementary information, Movie S1), with average area of 166.4 ± 12.0 μ m 2 (n = 124 cells), indicating that all mitochondria were fused into a single functional unit, similar to large mitochondrial reticulum revealed by mitoflash imaging in mouse skeletal muscles in vivo 23 and in vitro 18 .
To determine spatial synchrony of whole-cell mitoflash activation, we performed linescan at 200 Hz and two consecutive events of different amplitudes are shown in Fig. 4C. It is clear that both events were activated abruptly and synchronously across the entire dimension of the cell. Even by linescan imaging at 763 μ s/line (Supplementary information, Figure S6), we failed to resolve any local delays in the rising phase of mitoflashes over a distance of tens of micrometers. This result argues against the possibility of diffusible chemicals as the primary synchronizer, and provides compelling evidence that synchronization within an extended mitochondrial reticulum is controlled by the spreading of electrophysiological signals (e.g., membrane depolarization) 23,35 . Using the cable theory 36 , we estimated that the electrical length constant is ~82 μ m for a contrived tubular mitochondrion (Supplementary information, Table S1), a value that is consistent with an instantaneous and near uniform distribution of electrical signals over the whole-cell reticulum. Mitoflashes and miniflashes in mitochondria-hyperfused cells. Visual examination and quantitative measurement of whole-cell mitoflashes revealed another striking finding -the appearance of a subpopulation of events of tiny amplitude and brief duration, or "miniflashes", in hyperfused mitochondrial reticulum (Fig. 4A). For the purpose of bisecting all events into two categories, we classified miniflashes as those with Δ F/F 0 < 0.15 and FDHM (full-duration at half maximum) < 4 s. Miniflashes, which were rare in control cells dominated over mitoflashes in both Drp1-knockdown (KD) and Drp1 K38A -expressing cells (Fig. 5A). When mitoflashes and miniflashes were combined, the frequency of all events was markedly elevated in these mitochondria-hyperfused cells (Fig. 4D). However, when comparison was confined to the mitoflash subgroup, we found that regular mitoflash frequency was unaffected in cells with hyperfused mitochondria (Fig. 5A). Notably, robust miniflash activity was also present when mitochondrial hyperfusion was induced with a small molecule S-3 that enhances mitofusin activity 37 or by depletion of Mff (Supplementary information, Figure S7), another essential factor for mitochondrial fission 10 . At the first glance, the above results suggest that network hyperfusion promotes miniflash production without altering regular mitoflash activity as well as respiration (Fig. 4F). However, a caveat is that, before we reach that conclusion, we should determine whether mitoflashes and miniflashes are of the same or different origin. In this regard, we showed that both mitoflashes and miniflashes were accompanied by transient depolarization of the mitochondrial membrane potential (Δ Ψ m ) (Fig. 4A). They were equally sensitive to inhibition by ROS scavengers (e.g., mitoTEMPO and SS31), the mPTP inhibitor (CsA), and the respiratory chain inhibitor (rotenone) (Fig. 5B). Moreover, we found that histograms for the amplitude and duration of all events were well-fitted to single-exponential functions, with decay constants of 0.22 (Δ F/F 0 ) and 5.57 s (FDHM), respectively (Fig. 5C,D). Together with the shared feature of spatially synchronous activation, mitoflashes and miniflashes appear to belong to a single population of events, the parameters of which displaying continuous distributions. If miniflashes and mitoflashes are mechanistically identical, then the emergence of miniflashes in Drp1-inhibited cells could be only apparent, an epiphenomenal associated with hyperfusion. To this end, we noticed that, basal mt-cpYFP signals (F 0 ), spatially averaged over the entire mitoflash area, displayed similar absolute values in both control and Drp1-inhibited groups, but an 81% smaller standard deviation (SD), equivalent to a 5.2-fold higher signal-to-noise ratio (SNR) in cells harboring whole-cell mitochondrial reticulum (Fig. 5E,F). As such, miniflashes would mostly go undetected in control group but would be readily resolved in Drp1-inhibtied group; nevertheless, mitoflashes of regular size and duration would be faithfully detected in both groups. Since regular mitoflashes should display similar detectability and their frequency was unchanged in wild-type versus Drp1-defective cells, we interpret that mitochondrial hyperfusion does not alter the rate of occurrence of mitoflashes in a cell, and the apparent increase in all-event frequency is attributable to greatly improved detectability for small and brief events. It is noteworthy that the mitoflash signal mass and the frequency per mitochondrion were both dramatically increased when all mitochondria fused into one single reticulum in a cell. Furthermore, we found that the Drp1-KD and control cells exhibited similar levels of mitochondrial respiratory function (Fig. 4F), in general agreement with previous reports 38 .

Local swelling at preferred sites in mitoflashes of hyperfused mitochondria. Another interest-
ing observation is that mitochondrial reticulum exhibited reversible local swellings during a mitoflash (Fig. 6A  and Supplementary information, Movie S2), similar to those previously reported by us 24 and by others (termed "mitochondrial contraction") 38 . These local swellings have been thought to be triggered by local transient openings of mPTP and thus provide an opportunity to characterize kinetics of mPTP gating in intact cells 24 . To this end, Fig. 6 shows a sequence of repetitive mitoflashes of progressively increasing amplitude in a whole-cell mitochondrial reticulum in a Drp1-KD cell. Local swelling sites were conspicuous at the second event and became more vigorous in the third event. Nearly all swelling sites in the second event re-appeared, while new sites were recruited in the third event (Fig. 6B,E). The nearest intervals between swelling sites range from 0.68 to 3.00 μ m. Similar results were obtained in cells expressing the Drp1 K38A mutant (data not shown). These data indicate that mitoflash-associated local swelling is gradable and is associated with preferred locations of the reticular mitochondria. This result is consistent with the idea that transient mPTP opening plays an important role in the biogenesis of mitoflashes.

Discussion
In this study, we have systematically examined possible roles of mitochondrial network dynamics in the regulation of mitoflash biogenesis and signaling. Based on data from a range of cell types with genetic manipulation of multiple dynamic proteins, including Mfn1, Mfn2, Opa1, Drp1 and Kif5b, our overall conclusion is that the rate of mitoflash occurrence in a given cell is independent of network reorganization by fission, fusion and motility, when metabolic contributions as well as detection bias are appraised and separated. In contrast, the magnitude of mitoflash signaling (signal mass) in a cell is dynamically regulated by network fragmentation and hyperfusion. These findings provides compelling evidence for the stochastic trigger model of mitoflash biogenesis 39 and reveals a novel role for network dynamics in the regulation of mitochondrial signaling (see below).

Constancy of mitoflash biogenesis in changing mitochondrial network. Our conclusion is based
on critical synthesis of multiple lines of evidence. Experimental results from Mfn1, Mfn2 and Kif5b are straightforward: dramatic fragmentation or redistribution of mitochondrial network causes no detectable change in mitoflash frequency while exerting mild or no effect on mitochondrial respiration. However, cases with Opa1 and Drp1 appear to be counterintuitive and merit further analysis. In the absence of Opa1, we observed that spontaneous mitoflash activity was decimated, but not completely abolished, in fragmented network, dismissing Opa1 as an obligatory factor for mitoflash biogenesis 31 . However, this effect is unlikely attributable to altered mitochondrial morphology because similar network fragmentation in mitofusin-knockout cells caused no significant changes in mitoflash frequency. Instead, parallel experiments revealed that Opa1 deficiency markedly impairs mitochondrial respiratory function, in agreement with previous reports 32, 33 . This superimposed metabolic change could have accounted for the lowered mitoflash activity because of the coupling of mitoflash biogenesis to metabolism 18,19,22,23 . To directly test this possibility, we verified that wild-type and Opa1 −/− MEFs exhibited similar mitoflash response to hyperosmotic stress. This result supports that spontaneous mitoflash activity in Opa1 −/− MEFs is altered by depressed respiration rather than network fragmentation.
In the situation with Drp1, there emerged a new population of events (miniflashes) that even outnumbered regular mitoflashes, such that the combined event frequency was much higher in Drp1-KD or Drp1 k38A cells. The emergence of miniflashes in hyperfused mitochondrial network was independently verified with different experimental approaches involving activation of mitofusin by S-3 and depletion of the fission factor Mff. Does this mean that hyperfused mitochondria generate new types of events? The answer is no. First, similar to regular mitoflashes, miniflashes were synchronous over fused mitochondrial network, and were also associated with transient loss of Δ Ψ m . Second, miniflashes were sensitive to inhibition by ROS scavengers, mPTP inhibitor, and respiratory complex inhibitor, as regular mitoflashes did. Third, the amplitude and duration histograms of all events were fitted to single-exponential decaying functions, revealing no subpopulation of events. It is also noteworthy that these miniflashes closely resemble "precursor events" occurring at the rising feet of regular mitoflashes in cardiomyocytes 39 . Thus, we conclude that both miniflashes and regular mitoflashes are manifestations of the same stochastic process.
Then, do the Drp1-deficient cells exhibit hyperactive mitoflashes? The answer is unlikely when biases in detectability are taken into account. We showed that the noise level, quantified as SNR of basal mt-cpYFP fluorescence, was 5.2-fold higher in a whole-cell network (on average, 166 μ m 2 per mitoflash) than in a regular mitochondrial network (7.6 μ m 2 per mitoflash), as expected for the signal averaging effect. All else equal, greater SNR confers greater ability to identify small and brief events amidst noise. A fair and direct comparison should be limited to those events that are of larger amplitudes and duration and thus relatively insensitive to changes in detectability. In this regard, we found that the frequency of regular mitoflashes was unchanged in cells with hyperfused mitochondrial reticulum. Together with the constancy of respiration in these cells, this finding provides additional evidence supporting the conclusion that mitochondrial hyperfusion, while enhancing detectability and increasing observable mitoflash frequency, does not affect mitoflash biogenesis per se.
Stochastic trigger model for mitoflash biogenesis. In a previous study we have proposed that, at the single-mitochondrion level, stochastic excitation of discrete trigger site (e.g., mPTP) signifies the ignition of mitoflash 39 . The constancy of mitoflash frequency at the single-cell level with altered mitochondrial dynamics leads us to further extend this model for the biogenesis of mitoflashes (Fig. 7A). Specifically, we propose that each cell contains a certain number of discrete trigger sites (N) that are randomly distributed throughout mitochondrial network. Different mitochondria or reticular networks operate independently and stochastically. When all intracellular mitochondria are considered as a whole, cell-level mitoflash frequency is proportional to the product NP o , where P o is the probability of excitation of a trigger site. Whether network hyperfuses or fragmentizes, cell-level mitoflash frequency will be constant as long as N and respiration, which is an important determinant of P o , remain unchanged.
From the viewpoint of a single mitochondrion or a fused network, however, its number of trigger sites (n) increases additively with fusion, and hence its propensity of mitoflash production (nP o ) is size-dependent. That is, a hyperfused mitochondrial reticulum should exhibit higher rate of mitoflash ignition; conversely, smaller mitochondria in pro-fission conditions should display reduced mitoflash frequency per mitochondrion. Indeed, both predictions were verified with our data from hyperfused or fragmented networks.
It should be noted that our results argue against an alternative model in which mitoflashes arise from intrinsic oscillation of the mitochondria 40,41 . This model predicts an invariant oscillation frequency when mitochondria hyperfuse or fragmentize (Fig. 7B). An exemplary intrinsic oscillator system can be found with cardiac pacemaking. An intact sinoatrial node consists of hundreds of pacemaker cells that are electrically coupled into a functional syncytium. The rate of action potentials of the whole system is similar to that of its constituent pacemaker cells in isolation 42 , in sharp contrast to the additive behavior of mitoflash biogenesis when mitochondria hyperfuse or fragmentize.
The hypothesis on the existence of discrete trigger sites is also in agreement with possible role of mPTP in mitoflash ignition 16,17 . In this study, data shown in Fig. 6 provide vivid details that match with known characteristics of mPTP: multiple states of increasing conductance of mPTP may be progressively recruited during sustained and repetitive activities 43 . Given that open probability of mPTP is modulated by diverse endogenous and exogenous factors, this stochastic trigger model also provides a unifying framework for understanding the FM regulation of mitoflashes.

Regulation of mitoflash signaling by network dynamics. While previous studies have established
various FM mechanisms of mitoflash regulation 17 , the present study represents the first report of a primarily magnitude-modulatory mechanism whereby dynamic reorganization of mitochondrial network regulates the total mass of mitoflash signaling. Specifically, mitoflash frequency is largely independent of network dynamics; however, the average area of individual mitoflashes, and hence the mitoflash signal mass, is directly controlled by network fragmentation and hyperfusion. That is, mitochondrial dynamics constitute a magnitude-modulatory mechanism to regulate the total mass of mitoflash signals, independently of mitoflash frequency. That mitoflash frequency remains constant when network reorganizes indicates that this magnitude-modulatory mechanism is orthogonal to FM mechanisms, such that mitoflash signaling can be dynamically and independently regulated in both FM and magnitude-modulatory modes. Taken together, these findings provide new insights into roles of mitochondrial dynamics in the regulation of mitochondrial function and signaling.
Plasmids and adenoviral vectors. The dominant-negative Drp1 K38A mutant was inserted into pcD-NA4-TO vector (Invitrogen). DNA fragment of mt-cpYFP or mitoSypHer was inserted into pENTR/ TEV/D-TOPO vector (Invitrogen) for adenovirus and the adenoviral expression system (Invitrogen) was used for infection.
Cell culture. NRK cell line was gifted by Professor Li Yu 15 . Stable expression of mt-cpYFP in HeLa cells was established as previously described 24 . MEFs, NRK cells, and HeLa cells were all maintained in Dulbecco Minimal In the stochastic trigger model, hyperfused mitochondrial reticulum exhibits additive rate of mitoflash production. In contrast, oscillatory trigger model predicts synchronous oscillations at invariant frequency when the size of mitochondrial reticulum varies.