Real-time assessment of mitochondrial DNA heteroplasmy dynamics at the single-cell level

To investigate the dynamics of heteroplasmy, we aimed to develop a system, which would allow real-time assessment of the segregation of two different mtDNA genotypes across multiple generations. For this purpose, we opted for two recently established yeast strains with opposing mating types, each homoplasmic for a different mtDNA variant. Specifically, these strains harbor mtDNA encoding Atp6 tagged C-terminally either with NeonGreen (mtDNA^Atp6-NG) or mKate2 (mtDNA^Atp6-mKate2) and hence exhibit green or red fluorescence within mitochondria, respectively, depending on the presence and expression of the respective mtDNA haplotype. According to our rationale, the mating of both strains and subsequent mitochondrial fusion would result in heteroplasmic zygotes, carrying ∼50% mtDNA^Atp6-NG and ∼50% mtDNA^Atp6-mKate2 (Jakubke et al, 2021; Nunnari et al, 1997; Okamoto et al, 1998). Upon growth, each heteroplasmic zygote would give rise to progeny, whose heteroplasmic state would be reflected by ongoing expression and fluorescence of Atp6-NG and/or Atp6-mKate2 (Fig. 1A). In the following analyses, the heteroplasmy state of each cell will be characterized by an “h-value,” calculated by dividing the amount of mtDNA^Atp6-mKate2 by the sum of mtDNA^Atp6-NG and mtDNA^Atp6-mKate2, inferred by the respective fluorescent signals (Fig. 1B). Unless otherwise stated, for all experiments, cells were pre-grown in glucose-containing rich medium, because of higher mating efficiency, while imaging was performed in minimal medium containing glucose, due to lower background fluorescence.

In order to continuously monitor heteroplasmy changes on a single-cell level, we used a microfluidic system, that allowed tracking of progeny derived from single zygotes for up to six generations. Segmentation and tracking of individual cells as well as lineage tracing enabled the construction of genealogical trees for every single population (Fig. 1C,D).

Before generating heteroplasmic zygotes, it was important to compare growth and mtDNA maintenance between strains harboring mtDNA^Atp6-NG or mtDNA^Atp6-mKate2 to determine potential influences of growth properties on segregational dynamics of both mtDNA variants in the following heteroplasmic segregation experiments. In line with previous analyses (Jakubke et al, 2021), both strains exhibited virtually wildtype-like growth under fermentable and non-fermentable carbon sources at 30 and 37 °C in plate growth assays (Appendix Fig. S1A,B). We furthermore quantified doubling times of mtDNA^Atp6-NG or mtDNA^Atp6-mKate2 strains in our microfluidic setup by segmenting and counting cells over a time course of 8 h. In line with our growth assay on plates, both strains exhibited very similar doubling times (Appendix Fig. S1C). Similarly, petite levels, indicative of loss or dysfunctional mtDNA, were equivalent between mtDNA^Atp6-NG, mtDNA^Atp6-mKate2, and wildtype strains (Appendix Fig. S1D).

Next, we aimed to assess the validity of using Atp6-NG or Atp6-mKate2 fluorescence intensities as proxies for mtDNA presence within single cells. First, we examined mtDNA^Atp6-NG or mtDNA^Atp6-mKate2 strains in our microfluidic setup and monitored fluorescent signals across multiple generations for 8 h. In every population derived from single haploid cells, we observed homogeneous fluorescent signal intensities in the mitochondrial networks across all cells after 8 h (Fig. 2A,B; Movies EV1, EV2). Next, we assessed Atp6-NG and Atp6-mKate2 fluorescence in cell populations lacking the mitochondrial HMG-box protein Abf2. ∆abf2 cells are known to gradually lose mtDNA if grown in a fermentable carbon source (YPD) (Schrott and Osman, 2023; Sia et al, 2009; Zelenaya-Troitskaya et al, 1998). We cultured ∆abf2 cells in a non-fermentable carbon source (YPG) to prevent mtDNA loss and monitored Atp6-NG and Atp6-mKate2 fluorescence upon transition to a minimal medium. Consistent with the occasional loss of mtDNA, ∆abf2 populations exhibited indeed a higher signal variability across cells, which is quantitatively revealed by a higher coefficient of variation for either mtDNA^Atp6-NG or mtDNA^Atp6-mKate2 in ∆abf2, compared to the respective WT cells (Fig. 2C,D; Appendix Fig. S2; Movies EV3, EV4). Some cells exhibited fluorescence, indicative of mtDNA presence, while others appeared to have fully lost mtDNA, as they were only visible in the DIC channel. We interpret these findings to reflect that all wildtype cells maintain mtDNA and continue to express the fluorescent Atp6 variants with relatively little cell-to-cell variability, while mtDNA maintenance deficits associated with ∆abf2 cells result in heterogeneity of Atp6-NG or Atp6-mKate2 fluorescence signals.

To further support that Atp6-NG or Atp6-mKate2 fluorescence reflects the expression of the respective mtDNA, we measured the fluorescent decay rates by inhibiting mitochondrial translation with Chloramphenicol (CAP). Of note, this experiment was performed in strains lacking PDR5, a multidrug transporter localized in the plasma membrane, to prevent CAP export from the cell and allow efficient translational inhibition (Leonard et al, 1994). PDR5 deletion did not lead to a growth phenotype (Appendix Fig. S1A,B). Upon addition of CAP, we observed a decrease in both Atp6-mKate2 and Atp6-NG intensities, where within 5 h, signal intensities appeared to reach a bottom plateau, likely representing background fluorescence of cells and absence of the respective fluorophore (Fig. 2E; Appendix Fig. S3A–D; Movies EV5–EV8). By fitting exponential decay curves (Alber et al, 2018), we assessed the fluorescence decay rates (k) for both fusion proteins, which were virtually identical (k_mKate2 = 0.31, SEM:±0.01), k_NG = 0.31, SEM^: ± 0.01) (Fig. 2F). Of note, in the absence of CAP, the wildtype mtDNA^Atp6-NG strain exhibited an increase of fluorescence over time, while the mtDNA^Atp6-mKate2 strain exhibited a slight decrease over time. We also assessed the reappearance of fluorescent signals upon removal of CAP after a 6-h treatment. While both fluorescent signals reappeared rapidly, reflecting the maintenance of mtDNA, the Atp6-NG signal reemerged slightly faster compared to Atp6-mKate2 (Figs. 2F and EV1; Movie EV9). We speculate the difference in fluorescence levels over time between Atp6-NG and Atp6-mKate2 to be the result of a combination of various factors, including the switch from rich to minimal medium, which may affect mtDNA expression, the biophysical properties of the fluorescent proteins, as well as differences in the maturation time between NG and mKate2. To account for the difference, in all following experiments, NG and mKate2 fluorescence is always normalized to the median fluorescence per timeframe in the respective channel.

Based on the aforementioned experiments, we conclude that the presence of mtDNA^Atp6-NG or mtDNA^Atp6-mKate2 is unlikely to substantially alter cellular fitness, mtDNA maintenance, or influence the outcome of segregation dynamics within our heteroplasmic model. Furthermore, our findings support that Atp6-NG and Atp6-mKate2 serve as valid proxies for the presence of the mtDNA by which they are encoded.

Next, we followed the segregation of mtDNA^Atp6-NG and mtDNA^Atp6-mKate2 in populations derived from heteroplasmic zygotes, generated by mating parental cells harboring either mtDNA variant (Fig. 1A; Movie EV10). As expected, all newly formed zygotes displayed a fused mitochondrial network that exhibited signals in both fluorescent channels, indicative of carrying both mitochondrial genomes. We proceeded to generate time-lapse videos for a total duration of 8 h to track the mtDNA segregation across multiple generations. Remarkably, cell-to-cell heterogeneity became rapidly evident in all populations, reflecting different heteroplasmy levels (Fig. 3A; Appendix Fig. S4A). Most interestingly, patches of cells emerged in all populations that had transitioned to exhibiting fluorescence only in a single channel during the time course, indicating that these cells had retained predominantly or exclusively one of both mtDNA variants.

To quantitatively assess the heteroplasmy levels within each cell and population, we applied the heteroplasmy formula h = m1/(m1 + m2), (Aryaman et al, 2019; Johnston et al, 2015), where m1 and m2 are the fluorescent intensities of Atp6-NG and Atp6-mKate2 per cell that were normalized to cell volume as well as median fluorescence per timeframe. Furthermore, the background signal present in cell-free areas of images was subtracted from cellular fluorescence. As a result, an h-value of 0.5 is indicative of an equal ratio of both mtDNA variants within cells, equidistant to the homoplasmy extremes, which tend towards 1 or 0 when there is a shift towards mtDNA^Atp6-NG or mtDNA^Atp6-mKate2, respectively. By plotting the quantities of cells with specific h-values over time, we observed a shift in heteroplasmic states (Fig. 3B). Initially, the heteroplasmy distribution was narrowly centered around the average value of 0.5, where most cells exhibited fluorescence for both Atp6 variants. As time progressed, the middle peak flattened and the overall heteroplasmy distribution broadened (Fig. 3B). Focusing on the final timepoint, two peaks at both ends of the spectrum became apparent, closer to h-values of either 1 or 0 (Fig. 3C; Appendix Fig. S4B), likely representing homoplasmic cells.

Of note, for all segregation experiments, zygotes were chosen that gave rise to medial buds, which inherit mitochondria from both parental cells and, therefore, end up being heteroplasmic (Strausberg and Perlman, 1978). The inheritance of a mixed content derived from both parental cells is apparent from the green and red fluorescence within the mitochondria of medial buds (Appendix Fig. S5A). Second-daughter cells from zygotes often bud off of lateral positions and predominantly or exclusively inherit mitochondrial content from the parental cell from which they originate. Importantly, the segregation pattern did not significantly change, when we excluded cells derived from second-generation daughter cells that had appeared on lateral zygote positions and, therefore, had started out with predominantly only one of both mtDNA variants (Appendix Fig. S5B).

To estimate the proportions of homo- and heteroplasmic cells, we fit three Gaussian curves to the observed trimodal heteroplasmy distribution at the final timepoint. These curves characterize cells primarily exhibiting homoplasmy for either mtDNA^Atp6-NG (on the left) or mtDNA^Atp6-mKate2 (on the right), as well as heteroplasmic cells (in the middle). While definitively distinguishing between homo- and heteroplasmy is not feasible, we approximated their quantities by establishing thresholds at the intersections of individual Gaussian curves. Cells falling below an h-value of 0.35 or above 0.7 were classified as homoplasmic for mtDNA^Atp6-NG or mtDNA^Atp6-mKate2, respectively (Fig. 3C). This approximation is further supported by the fact that residual non-mated haploid cells homoplasmic for either mtDNA^Atp6-NG or mtDNA^Atp6-mKate2 exhibited h-values comparable to diploid cells classified as homoplasmic (Appendix Fig. S6). We posit that the homoplasmic peaks not being centered around 0 and 1 is primarily attributable to autofluorescence in the “absent” channel. To confirm that diploid homoplasmic cells indeed fall below or above these thresholds, we assessed heteroplasmic states of 20 independent populations derived from heteroplasmic zygotes after 24 h, where we expected virtually full segregation of mtDNA variants. Spatial limitations within the microfluidic system prevented its use for this approach. Therefore, we opted for a microdissection of heteroplasmic zygotes and observed mtDNA segregation in single cells after 24 h of growth on plates (Fig. EV2). Indeed, the 24 h cell distribution appeared bimodal, where the middle curve, representing heteroplasmic cells, was virtually absent (Fig. 3D). The two pronounced peaks fell below or above our homoplasmy thresholds, establishing them as good estimates to quantify homoplasmic cells. Based on these cutoffs, 97% of cells were characterized as homoplasmic, revealing, as expected, nearly complete segregation of two mtDNA variants within a growing yeast population after 24 h. Thus, our approach allows real-time imaging of mtDNA segregation dynamics at a single-cell level. Our analysis reveals rapid segregation, which can be observed already during the first 8 h and virtually completes within 24 h.

To better understand how the observed mtDNA segregation occurs, we developed a mathematical model that simulates the dynamics of segregation in a growing cell population arising from a single heteroplasmic cell. In vivo, mtDNA is distributed throughout a tubular mitochondrial network within cells and partitioning of mitochondria, including mtDNA, occurs via the transport of mitochondrial tubules into daughter cells during cell division (Osman et al, 2015). Hence, mtDNA copies present in the same segments of the tubular mitochondrial network are more likely to be segregated together into daughter or mother cells, respectively. Therefore, a simple random-pick model would not adequately capture the in vivo situation. To reflect this morphological aspect in the model, we used arrays, representing the tubular organization of mitochondria that contained two mtDNA variants either denoted as 0 or 1. Of note, we do not consider branching of the array. The length of this array represents the previously determined average of 32 mtDNA copies per diploid cell (Göke et al, 2020), neglecting the cell-to-cell variability in the copy numbers that may occur due to cell size differences. Importantly, we simulated zygotes with equal amounts of 0 and 1 s in an initial ’mixed’ structure sequence of 01010101… (Fig. 4A). We substantiated the model’s robustness through manipulation of the array, by mildly varying the mtDNA copy numbers between 26 and 38 (Appendix Fig. S7A), as well as by substituting the initial structure of the founder cell with a semi-mixed array (“0011100011…”) or a non-mixed array (“0000011111…”) (Appendix Fig. S7B,C).

To account for mitochondrial fission and fusion events, we introduced the nspl parameter, which represented the number of fragments an array could split into, followed by re-fusion in random order during growth. Upon completion of this shuffling process, the array was allowed to split into two parts, which partitioned to mother and daughter cells. The daughter cell received the shorter of both parts containing a specific amount of mtDNA copies, defined by the ndau parameter (Fig. 4A). Depending on the inherited mtDNA variants, each daughter cell consisted of either all 0 s, all 1 s, or a mixture of both. Upon division, mtDNA copies were allowed to replicate until the total number reached 32. To reflect relaxed mtDNA replication, we randomly chose mtDNA molecules for replication, allowing multiple rounds of replication for the same mtDNA molecule during the span of one cell cycle. Notably, newly replicated mtDNA copies were always placed next to the template mtDNA. We neglected the possibility of spontaneous cell death and mtDNA degradation.

First, we simulated the influence of nspl and ndau parameters on the segregation rate by modeling the amount of homoplasmic cells after 7.5 h (Fig. 4B). We observed that higher nspl or ndau parameters, corresponding to more shuffling or more mtDNA copies partitioned to daughter cells, predict a relatively low percentage of homoplasmic cells after 7.5 h. Conversely, lower nspl and/or ndau values lead to an acceleration of mtDNA segregation. Our simulations show that distinct ndau-nspl combinations can result in similar levels of homoplasmic cells after 7.5 h. We next asked, which of these combinations aligns with our experimental data. To this end, we applied the previously defined homoplasmy thresholds (Fig. 3C) to all timepoints within our 8-hr in vivo experiment and quantified amounts of homoplasmic cells (Fig. 4C, black curve). We compared the fractions of homoplasmic cells between the experimental and the simulated data at timepoints 1.5, 3, 4.5, 6, and 7.5 h by applying the least-squares criterion. We found the best-fitting parameter values to be ndau = 14 and nspl = 4 (Fig. 4C,D). However, we identified various combinations of ndau and nspl approximating the experimental results. In particular, there was an inverse relationship between the values for nspl and ndau that fit the data well, where higher ndau values required lower nspl ones, and vice versa. Specifically, the combinations of relatively higher ndau values (ranging from 10-16) and lower nspl values (within the range of 3–6) came closest to the empirical data (Fig. 4C, light green lines). In contrast, we observed a poor fit to the data, when ndau and nspl were simultaneously very low or very high (Fig. 4D). Thus, these results suggest that fusion and fission frequencies, representing shuffling events, as well as the number of mtDNA copies transferred to daughter cells are contributing to the segregation kinetics of mtDNA variants in a yeast population, under our neutral experimental setup.

Next, we used our model to examine how increased mtDNA copy numbers within cells may affect mtDNA variant segregation. To this end, we simulated segregation in cells with 56 (observed in cells with increased mtDNA copy number lacking the gene MRX6 (Göke et al, 2020)) or 90 mtDNA copies (as an extreme example) (Fig. EV4A,B) and compared segregation kinetics to the kinetics from the best-fitting simulations for cells with 32 mtDNA copy numbers, where ndau and nspl equaled 14 and 4, respectively (Fig. EV4C–E). We found that the segregation speed is very similar if the nspl value stays the same (nspl=4) and the same percentage of total copies of mtDNA are passed on to the daughter cell (e.g., ∼44% would mean 14 copies in a cell with 32 copies, but ∼25 or ∼39 copies in a cell with 56 or 90 copies, respectively) (Fig. EV4C). However, it is also possible that increased mtDNA copy number affects the ndau and/or nspl parameters. Similar to our simulations for mtDNA copy number equal to 32 (Fig. 4D), our simulations suggest that also in scenarios with increased mtDNA copy number, a higher percentage of mtDNA copies transmitted to daughter cells and increased fusion-fission frequencies could result in slower segregation of mtDNA variants. On the contrary, a lower percentage of transmitted copies and decreased fission-fusion frequencies would result in faster segregation (Fig. EV4D,E).

Altogether, these results suggest that fusion and fission frequencies, representing shuffling events, as well as the number of mtDNA copies transferred to daughter cells are contributing to the segregation kinetics of mtDNA variants in a yeast population, under our neutral experimental setup.

Our mathematical model suggests that multiple different combinations of nspl and ndau parameters could explain the observed segregation of mtDNA variants. With respect to the ndau parameter, the values range from seven to fifteen, representing about a fifth or half of available mtDNA copies passed on to the daughter cell, respectively. Therefore, we asked which of the possible model predictions reflects the in vivo situation. First, we examined our experimental heteroplasmy dataset to evaluate if equal or unequal amounts of mtDNA copies are partitioned to mother and daughter cells during cell division. If unequal amounts of mtDNA copies are transferred per division to the progeny, we would expect a lower correlation between heteroplasmy values between mothers (M) and daughter cells (D) compared to a correlation of h-values between mothers to themselves at a later timepoint (M_D). In a scenario where mtDNA copies are equally split between mother and daughter, the M-D and M-M_D h-value correlations are expected to be similar. Hence, we compared heteroplasmy levels of mother cells (M) at timepoint t_M to those of their daughters (D) at t_D or granddaughters (GD) at t_GD, as well as to those of their aging selves M_D and M_GD at t_D or t_GD, respectively (Fig. 5A). The two timepoints, t_D and t_GD, were chosen based on the same growth stage, where cells had small buds (1/5 bud-to-mother volume ratio). In line with an unequal partitioning of mtDNA copies, we observed a high correlation between M and M_D in contrast to a statistically significant lower correlation coefficient between M and D (Fig. 5B). Comparison of mothers to their granddaughters further supported this hypothesis, as the M-GD correlation further decreased compared to the M-D correlation. In contrast, aged mothers (M_GD) still displayed a stronger similarity to their original heteroplasmic state (M). These results underscore that cell division, most likely through the transmission of a limited number of mtDNA copies, is a major driver for the progressive divergence of heteroplasmic states in a proliferating yeast population.

Subsequently, we aimed to directly determine the number of mtDNA copies passed from mother to daughter per cell division. Therefore, we employed the mtLacO-LacI system, which allows the detection of single mtDNA copies by fluorescence live-cell microscopy (Osman et al, 2015). First, we quantified the number of fluorescent foci, representing mtDNA copies, migrating across the bud neck from a mother to its bud in 5-minute windows. Cells were in different cell cycle stages, however only mother-bud pairs whose mitochondrial network appeared interconnected or where mitochondrial fragments traveled from mother to daughter were considered. We observed that, on average, 1.18 mtDNA foci were transmitted to daughter cells within these 5 min (Fig. 5C,D; see Methods; Appendix Fig. S8A,B; Movie EV11). Next, we examined the duration for which the exchange of mitochondrial content occurred between mother-daughter pairs, by live-cell microscopy using the nuclear-encoded matrix-targeted mKate2 (see Methods). We did not observe mitochondrial content exchange, neither through continuous mitochondrial tubules spanning the mother-bud neck, nor through transport of mitochondrial fragments, after an average time period of ∆t = 47.5 min (Fig. 5E; Appendix Fig. S8C; Movie EV12). This observation aligns well with a previous study that similarly observed that daughter cells gain mitochondrial volume at the expense of the mother during the first half of the cell cycle (Rafelski et al, 2012). Taken together, we estimate that, on average, 11.2 nucleoids migrate from mother to daughter in a single-cell division (Fig. 5F).

To further examine mtDNA transmission to daughter cells, we performed live-cell microscopy of diploid cells containing the mtLacO-LacI system over the duration of an entire cell cycle and assessed dynamic changes of mtDNA content in virgin mother and emerging daughter cells. In line with our observation that the exchange of mitochondrial content occurs between mother and daughter cells for about 45 min, we observed a reduction of mtDNA foci from about 35 to below 30 in the mother cells during this time period, whereas the number of foci increased to 15 in daughter cells. After 45 min, the number of foci remained constant in the mother cell, while the foci number further increased to about 30 in the daughter cell (Fig. EV4). These results suggest that mtDNA copy number is replenished in mother cells, while mtDNA copies are being passed on to daughters, and that after 45 min mtDNA replication continues in daughter cells. Of note, we did not observe that the starting foci number of 35 was reestablished in the mother cell or the daughter cell. We assume that this observation is due to continued imaging causing phototoxicity or bleaching.

Besides asymmetric apportioning of mtDNA copies to daughter cells, our mathematical model also predicts a critical role for mitochondrial fusion and fission cycles in determining the rate of mtDNA variant segregation by shuffling mtDNA copies in the mitochondrial network. To experimentally test this prediction, we performed mtDNA variant segregation experiments in cells lacking the gene DNM1, which is essential for mitochondrial fission. Specifically, we mated ∆dnm1 cells containing mtDNA^Atp6-NG or mtDNA^Atp6-mKate2, pre-grown in YPG medium to prevent mtDNA loss, and microscopically evaluated the percentage of heteroplasmic cells in colonies formed from individual zygotes after 18 h of growth on YPD plates. In line with our model, the absence of mitochondrial fission resulted, on average, in a lower amount of heteroplasmic cells compared to mating between mtDNA^Atp6-NG or mtDNA^Atp6-mKate2 WT cells, indicating faster segregation kinetics in the absence of mitochondrial fission (Fig. 5G). This conclusion is further supported by the observation that we detected exclusively homoplasmic cells in 57% of colonies derived from heteroplasmic ∆dnm1 cells compared to 14% in WT cells (Appendix Fig. S9B).

We additionally experimentally tested the effect of increased mtDNA copy number on mtDNA segregation dynamics. To this end, we performed segregation experiments in ∆mrx6 cells, that have previously been shown to have a twofold increase of mtDNA (Göke et al, 2020). We assessed the number of heteroplasmic cells in populations after growth for 18 h derived from ∆mrx6 zygotes containing mtDNA^Atp6-NG and mtDNA^Atp6-mKate2. Interestingly, mtDNA variant segregation was delayed in ∆mrx6 cells and a higher percentage of cells remained heteroplasmic after 18 h of growth (Fig. 5G), with none of the populations having entirely segregated (Appendix Fig. S9B). In conjunction with predictions from our model (Fig. EV3), the experimentally determined delayed segregation in ∆mrx6 cells suggests that these cells have higher fission-fusion frequencies and/or transmit a higher percentage of mtDNA copies to daughter cells.

Thus, our experimental data suggest that unequal amounts of mtDNA copies are partitioned to daughter cells during the cell cycle, which appears to be a major driving force for mtDNA segregation. Additionally, we find that fission deficiency leads to faster mtDNA variant segregation, while increased mtDNA copy number results in slower homoplasmy establishment.

Finally, we asked if we could apply our experimental setup to observe purifying selection of intact over mutant mtDNA. Currently, no yeast strains are available that harbor a mutant mtDNA encoding a fluorescent protein. Therefore, we conducted experiments in which cells containing intact intron-containing mtDNA^Atp6-NG were mated with cells containing “dark” intronless mtDNA^il-∆cob, lacking the open reading frame encoding Cytochrome b, which is essential for respiratory growth. Notably, deletion of COB was generated in mtDNA lacking introns (Gruschke et al, 2011). Therefore, we conducted control experiments where cells containing mtDNA^Atp6-NG were mated with cells harboring intact intron-containing (mtDNA^ic) or intact intronless mtDNA (mtDNA^il), both lacking genes encoding fluorescent proteins. Upon growth for 24 h, we assessed the percentage of “dark” cells, lacking mtDNA^Atp6-NG, in populations arising from heteroplasmic zygotes (Fig. EV5). Our data show that in matings between cells harboring mtDNA^Atp6-NG and cells harboring mtDNA^ic, no preference for either mtDNA variant was present as equal segregation into ’dark’ and fluorescent cells was observed (Fig. 5H). Matings between cells harboring mtDNA^Atp6-NG and cells harboring mtDNA^il revealed a slight preference for the intron-containing mtDNA as less “dark” cells were detected. Most strikingly, the number of “dark” cells was significantly lower after the growth of zygotes containing mtDNA^Atp6-NG and mtDNA^il-∆cob, indicating selection against the mutant mtDNA. This result confirms our previous finding that yeast cells favor generation of progeny with functional mtDNA copies (Jakubke et al, 2021), and that our experimental pipeline can be used to study the purifying selection of mtDNA.

All yeast strains employed in this study were generated in the W303 background. Detailed genotypes are listed in Appendix Table S1. Construction of single deletion mutants followed the procedure described in (Janke et al, 2004). Primers and plasmid information are listed in Appendix Tables S2 and S3, respectively. For the deletion of ABF2, transformed cells were plated and kept on non-fermentable (YPG) carbon source, to avoid loss of mtDNA prior to the microfluidic experiment.

Prior to all experiments, cells were freshly streaked from −80 °C on YPD plates, left to grow at 30 °C overnight, and then inoculated in fermentable (YPD) liquid culture. The next day, cells were diluted to 0.1 OD₆₀₀ and further incubated to reach the log phase. Of note, the ∆abf2 cells, for either variant, were maintained on YPG, to prevent spontaneous mtDNA loss.

For the growth tests, log-phase cells were serially diluted and 3 µl of each dilution were spotted onto fermentable (YPD) and non-fermentable (YPG) plates. Plates were incubated at 30 and 37 °C, and were imaged after 24 and 48 h of growth. Appendix Fig. S1A shows the data after 48 h of growth. For the petite analyses to assess the loss of mtDNA in cells, 200 cells growing in log-phase were plated on YPG plates containing 0.1% glucose. Plates were imaged and colonies were counted after 3 days at 30 °C and classified into grande (big) or petite (small) colonies. Cells unable to properly respire show a petite phenotype.

Time-lapse data was acquired using the microfluidics CellAsic ONIX system (Millipore) attached to a Nikon Eclipse Ti inverted microscope, equipped with a CFI Apochromat TIRF 100XC oil objective with a numerical aperture of 1.49 and a working distance of 0.12 mm (Nikon MRD01991). A TwinCam LS dual-camera splitter, coupled with two Photometrics Prime 95B 25-mm cameras, facilitated simultaneous imaging of red and green fluorophores. Every experiment was performed under the same settings for both fluorophores, with 40 ms exposure time, and 40% LED power. NeonGreen was excited at 506 nm wavelength, while mKate2 at 588 nm. Z-stacks were captured with increments of 0.3 µm over a total of 6 µm. Information on filters and dichroics can be made available upon request.

Both haploid and diploid cells were visualized for 8 up to 12 h, depending on the experiment, and images were acquired at 30 °C in 15 min intervals throughout the duration of the time-lapse using a temperature-controlled chamber around the microscope. Cells were continuously replenished with a fresh minimal medium at a flow rate of 15 kPa. For all microfluidic experiments, the CellAsic ONIX plates (Millipore) for haploid yeast cells were used. To obtain the diploid cells used for diploid live cell imaging, log-phase haploid cells harboring Atp6-NG were mated with haploid Atp6-mKate2 cells, in a 1:1 ratio on a YPD plate. After 2 h at 30 °C, once zygotes were formed, cells were resuspended in minimal medium, and single zygotes were picked in individual field of views to be monitored overnight using the CellAsic ONIX system.

Cell segmentation and tracking was performed frame by frame based on the DIC channel using the Cell-ACDC software and the implemented YeastMate cell segmentation algorithm (Bunk et al, 2022; Padovani et al, 2022). Cell cycle annotations, mask correction, and lineage tracking errors, as well as cell volume calculation, were also performed through the Cell-ACDC software. All cell masks and traces were also manually verified. Fluorescent signals were quantified from the mean projection of the z-stack per timepoint from non-deconvolved video recordings. Fluorescent intensities per cell were first corrected by subtraction of background signal, measured in cell-free areas, and normalized to cell volume. Subsequently, for comparability between the different cell populations and to account for photobleaching, the cell signal intensity for the respective fluorophore was additionally normalized to the median intensity of all segmented cells of the given timepoint and technical replicate. Microscopy images in all figures were deconvolved with the Huygens software (Scientific Volume Imaging).

For assessing the proportion of homoplasmic Atp6-mKate2 and Atp6-NG cells after 24 h, the respective strains were grown to log phase and mated in a 1:1 ratio for 90 min on YPD plates. Once zygotes were formed, they were micro-dissected and placed in separate areas on a YPD plate, where they grew into single colonies overnight (Fig. EV2). The next morning, each distinct colony was picked and diluted into 200 µl and imaged on an Ibidi glass slide, previously coated with the immobilizing agent Concanavalin A (Jakubke et al, 2021). To get an estimate of heteroplasmy within the population, a minimum of 200 cells were imaged and evaluated from each individual colony. First, the h-value of all segmented cells was calculated, using the heteroplasmy equation (Johnston et al, 2015), and subsequently, after binning the data, cells were characterized as homoplasmic for Atp6-NG, Atp6-mKate2 or heteroplasmic if they fell within the middle range of heteroplasmy, defined by the intersection points of the single Gaussian curves (Fig. 3C).

For assessing the level of homoplasmy in cells competing between intact and mutant mtDNA, we mated WT strains harboring mtDNA^Atp6-NG with cells containing intronless mtDNA^∆cob (Gruschke et al, 2011). As a control, we also mated cells harboring mtDNA^Atp6-NG with WT “dark” mtDNA^ic, or WT “dark” mtDNA^il. To assess the percentage of homoplasmic cells, we manually determined a threshold for ’signal’ or ’no signal’ by thresholding the fluorescent intensity of Atp6-NG based on the distribution of cells across each replicate as follows. Across all experiments, two distributions, exhibiting or virtually lacking fluorescent signals, were observed. The lowest value at the valley between the two peaks was set as the threshold for signal per replicate.

For assessing the heteroplasmy levels of cells with increased copy number (∆mrx6) or fission-deficient ones (∆dnm1), the zygote dissection pipeline was performed as described above, only that colonies were imaged after 18 h of growth on YPD plates. The rationale for earlier imaging was to capture differences in the percentage of remaining heteroplasmic cells more clearly compared to later imaging, where mtDNA variant segregation would have been virtually complete even in matings with a decreased rate of segregation.

To estimate the number of mtDNA copies crossing the mother-bud neck, we used a diploid strain developed in the lab, containing mtDNA with integrated LacO repeats and expressing the nuclear-encoded matrix-targeted 3xNG-LacI protein as well as a matrix-targeted mKate2 to monitor the mitochondrial network (3xNG-LacI-PGK1-su9-mKate2) (Osman et al, 2015). Log-phase cells, pre-grown in YPD, were visualized in 10-second intervals for 5 min in total, since longer exposure (with such frequent intervals) induces significant photobleaching.

By focusing exclusively on the neck region where the two cells are joined, we quantified the migration of mtDNA fluorescent foci from the maternal to the daughter side. Notably, we only analyzed cells where the mitochondrial networks between mother and bud were connected. Given that cells were at different stages of the cell cycle, we calculated the average amount of foci transmitted in 5 min, regardless of bud size. We cannot rule out that if two mtDNA foci are in close proximity, they might be counted as a single molecule. Of note, all aforementioned analysis was done using all z-slices and 3D modeling of each cell, and no maximum projection of z-stacks, to avoid loss or misinterpretation of the number of foci per cell pair.

To precisely determine the duration of content exchange between mothers and daughters, we employed the same strain but under different imaging conditions. Log-phase cells were visualized for 120 min, in 5-min intervals, for successful monitoring of complete division cycles and mitochondrial network association between mothers and daughters. We defined the initiation of budding, when cells were in the very early S phase, as timepoint zero. We tracked the exchange process until the daughter’s mitochondrial network had completely separated from that of the mother. At this point, just before the ultimate separation of networks, most daughter cells had developed into buds with an average cell volume equivalent to 55% of the mother’s size, most likely corresponding to mid- to late G2 phase. In cases where mothers had large buds, we observed no network connectivity, indicating that these buds had already entered the late G2 phase in preparation for cytokinesis, which is consistent with existing literature (Koren et al, 2010). By analyzing 33 cells from timepoint zero to the last timepoint, we calculated an average mitochondrial content exchange period of 47.5 min, which was subsequently used to calculate an estimate of nucleoids transmitted per division.

In order to assess the number of mtDNA foci during a complete cell cycle, we used the same diploid strain (3xNG-LacI-PGK1-su9-mKate2) used for mtDNA foci assessment during mother-bud partitioning. Cells were harvested at log phase and imaged for 2 h at 7 min intervals. mKate2 and NeonGreen were simultaneously captured using the dual-camera system (50 ms exposure time, 30 and 100% light intensity, respectively). Of note, all aforementioned analyses were done using all z-slices and 3D modeling of each cell, and no maximum projection of z-stacks, to avoid loss or misinterpretation of the number of foci per cell pair. Foci were counted with custom-built Fiji Macros using the “3D Maxima finder” command. For the analysis, only virgin mothers, that is, cells that had just budded off from their mother, were considered. Timepoint zero was determined based on the point of bud emergence from the virgin mothers, per time-lapse video.

In the model, we represent mitochondria as lists of 0 and 1 s, representing two different mtDNA genotypes. Mitochondria are modeled as one-dimensional objects without branches. When a mtDNA molecule is duplicated, the copy is placed right next to the original one. For the mtDNA copy number of diploid yeast cells, we apply actual measurements derived from (Göke et al, 2020), and set the average amount of mtDNA copies to 32 in each simulated cell. We additionally assume that one nucleoid contains one mtDNA copy and all mtDNA copies have identical replication rates. Importantly, the model neglects the possibility of a preferential replication of only a fixed subset of copies at any timepoint.

In order to accurately reflect the heteroplasmic state of founder cells, we define two different mtDNA variants in each zygote, mtDNA type 0 and mtDNA type 1. Hence, a cell can have two different states, homoplasmic for either 0 or 1, or a combination of both when heteroplasmic. In each simulation, founder cells were consistently characterized by virtually absolute heteroplasmy, featuring a distribution of mtDNA haplotypes according to the average of all nine biological replicates and a mitochondrion structure composed of alternating 0 and 1 sequences (that is 0101010101… in the case of 50% mtDNA haplotype 1). Given our experimental observations, where spontaneous mtDNA depletion is absent, we also postulate the absence of mtDNA depletion in each cell in silico. Furthermore, to simulate yeast lineages as close to the real data as possible, we allow each cell to divide only upon reaching the predefined mtDNA copy number. This way, when a daughter cell is created, it is allowed to first grow and replicate its content, before proceeding to division.

To account for the fission and fusion events in mitochondria, we establish the nspl parameter, defining the number of fragments every mitochondrial sequence splits into. This fragmentation leads to a reshuffling of the mtDNA copies within each mitochondrion, mimicking the fission-fusion dynamics. Upon this shuffling process, all fragments refuse and each cell is allowed to divide, by splitting into two parts. We also define a second variable ndau for the number of copies a mother cell gives to the daughter per division. We set the maximum ndau value to always be half of the total number of copies in the given founder cell. For example, in the simulations with increased copy number, where n = 56 or n = 90 copies/cell, each daughter cell could inherit a maximum of 28 or 45 copies, respectively. The ndau parameter, along with the nspl, are the only model parameters to be fitted to the data. Specifically, we first fit a sigmoid curve to the experimental heteroplasmy values at six timepoints (t = 1.5, 3, 4.5, 6, 7.5 h) and then identified the best-fitting pairs (ndau, nspl), using least-squares fitting.

Data analysis and statistics were performed in Python and the in silico mathematical modeling was constructed using R. A dependent two-sided t-test (scipy.stats.ttest_rel) was chosen for related lineage data (Fig. 5B). An independent two-sided t-test (scipy.stats.ttest_ind) was chosen for all data derived from independent samples (Fig. 5G,H). All scripts are provided in a GitHub repository under https://github.com/statgenlmu/yeast_mito_segregation.