To identify additional mechanisms of germline mtDNA control, we investigated how mtDNA quantity and quality are regulated in Caenorhabditis elegans PGCs. The entire C. elegans germ line descends from two PGCs, which are born early in embryogenesis and remain quiescent until early larval stages (Fukuyama et al., 2006). Although embryonic PGCs do not divide, they undergo a non-mitotic cellular remodeling process, discarding much of their cell mass and content. Remodeling occurs when PGCs form organelle-filled lobe-like protrusions, which adjacent endodermal cells cannibalize and digest (Figure 1A; Abdu et al., 2016; Sulston et al., 1983). Previously, we showed that PGCs lose much of their mitochondrial mass in the process of lobe cannibalism, suggesting that one role of this remodeling event could be to eliminate PGC mitochondria in bulk (Abdu et al., 2016). As such, lobe cannibalism might provide a novel mechanism for PGCs to adjust their mtDNA quantity and/or quality at the initial stages of germline development.

To begin to test this hypothesis, we used PGC-specific markers of the plasma membrane (PH_PLC1∂1::mCherry, ‘Mem-mCh^PGC’) and mitochondrial outer membrane (TOMM-20^1-54::Dendra2, ‘Mito-Dendra^PGC’) to follow the distribution of PGC mitochondria during lobe formation and cannibalization in living embryos. Most PGC mitochondria moved into lobes shortly after they formed (Figure 1B–C) but a subset returned to the cell body prior to lobe digestion (Figure 1D and F, Figure 1—figure supplement 1A-B). Cell body mitochondria that are retained in L1 PGCs (Figure 1E) represent the founding population present at the onset of larval germline expansion.

The PGC lobe fragments present within endodermal cells colocalize with the lysosomal marker LAMP-1, suggesting that mitochondria within lobes are targeted for destruction and digested (Abdu et al., 2016). To test this hypothesis more directly, we visualized the mitochondrial outer membrane marker Mito-Dendra^PGC, which is pH-sensitive (Chudakov et al., 2007) and should be quenched when mitochondria are present within lysosomes. In L1 larvae, Mito-Dendra^PGC fluorescence was greatly reduced in cannibalized lobe mitochondria compared to pH-insensitive Mito-mCh^PGC (Shaner et al., 2004; arrowheads, Figure 1G–H), whereas both markers labeled PGC cell body mitochondria robustly (dashed outline, Figure 1G–H). We conclude that PGC lobe mitochondria are digested by endodermal cells shortly after lobes are cannibalized, permanently removing them from the mitochondrial pool passed on to L1 larval PGCs.

To determine how elimination of mitochondria by lobe cannibalism affects the pool of germline mtDNAs, we first examined PGC mtDNAs visually. Mitochondrial transcription factor-A (TFAM), a component of the mtDNA nucleoid, is a well-characterized marker of mtDNA (Garrido et al., 2003; Lewis et al., 2016; Rajala et al., 2014). In human cells, individual TFAM nucleoids appear as puncta within the mitochondrial matrix and contain single, or at most a few, mtDNA genomes (Brown et al., 2011; Kukat et al., 2011). We tagged the C. elegans TFAM homolog (hmg-5) endogenously with green fluorescent protein (GFP). The TFAM-GFP protein was expressed ubiquitously and formed puncta that localized to mitochondria, consistent with its known binding to mtDNA in C. elegans (Figure 1—figure supplement 2A-B; Yang et al., 2022). Within PGCs, TFAM-GFP puncta were present in both cell body and lobe mitochondria, including those that had been recently cannibalized (arrowhead, Figure 1I). The number of TFAM-GFP foci decreased more than twofold between embryogenesis and the L1 larval stage (Figure 1J–L), suggesting that lobe cannibalism results in a substantial loss of PGC mtDNAs.

To quantify the number of mtDNAs within PGCs, we developed a fluorescence activated cell sorting (FACS) protocol to purify GFP-labeled PGCs from dissociated embryos (before lobe cannibalism), late embryos (after lobe cannibalism), and L1 larvae, which we paired with droplet digital PCR (ddPCR) to count mtDNA molecules per cell (Figure 2A, Figure 2—figure supplement 1, and Figure 2—figure supplement 2). We were able to isolate nearly pure populations of PGCs as determined by live imaging (Figure 2—figure supplement 2A) and post-sort analysis (see Methods). Additionally, PGCs isolated from late embryos and L1 larvae were less than half the volume of embryonic PGCs (Figure 2—figure supplement 2A, B), indicating that lobe cannibalism had not yet initiated in most of the sorted embryonic PGCs and was complete in late embryonic and L1 PGCs, as expected (Abdu et al., 2016).

We determined that each embryonic PGC contained 401 ± 11 mtDNAs (Figure 2B), which is 1.2% of the number of mtDNAs we detected in whole early embryos (33,840 ± 1784) (Figure 1—figure supplement 2C). The volume of each embryonic PGC in vivo (275 ± 7.2 µm³) is 1.2% of the volume of whole embryos (23,949 ± 175 µm³) (Figure 2—figure supplement 3), suggesting that embryonic PGCs inherit their mtDNAs from the pool present at fertilization through reductive embryonic cell divisions. By contrast, late embryonic PGCs, which lacked lobes, contained only 272 ± 8.1 mtDNAs (Figure 2B). Sorted L1 larval PGCs contained even fewer mtDNAs (220 ± 12; Figure 2B). Together, these data suggest that PGC lobe cannibalism could eliminate a third of the mtDNA molecules that each PGC inherits at its birth, and that an additional mechanism further reduces mtDNA numbers between late embryogenesis and the L1 larval stage. Thus, each PGC in the larval germ line contains roughly half the number that they inherit during embryogenesis.

To directly test whether lobe cannibalism contributes to the loss of mtDNAs in PGCs, we examined PGC mtDNA number in nop-1 mutants, in which most PGCs fail to form lobes (Figure 2C–D; Maniscalco et al., 2020). The nop-1 mutant L1 PGCs retained a significantly higher proportion of embryonic PGC mtDNAs compared to wild type (Figure 2E and G). This finding implicates lobe cannibalism in the reduction in mtDNA that occurs as PGCs transition from embryogenesis to the L1 stage.

Autophagy is an additional mechanism by which cells can remove cellular components and organelles, including mitochondria. During autophagy, an autophagosome membrane encapsulates organelles and cytoplasm, subsequently fusing with a lysosome to degrade its contents (Dikic and Elazar, 2018). To test if autophagy could be the source of lobe-independent mtDNA reduction in PGCs, we used the pH-discriminating Mito-mCh^PGC and Mito-Dendra^PGC reporters to observe whether any PGC mitochondria become acidified before lobe cannibalism occurs. We observed one or more large, distinct foci of acidified mitochondria [mCherry(+) Dendra(-)] within many PGCs prior to lobe cannibalism (Figure 2—figure supplement 4), and foci were absent in autophagy-defective atg-18/WD repeat domain phosphoinositide interacting 2 (WIPI2) mutants (see Figure 5F–G; Palmisano and Meléndez, 2019), suggesting that autophagy may be responsible for lobe-independent degradation of mtDNA in PGCs.

To directly test whether autophagy and lobe cannibalism fully account for mtDNA reduction in PGCs, we sorted embryonic and L1 PGCs from nop-1; atg-18 double mutants, which are deficient in both lobe cannibalism and autophagy. Notably, mtDNA reduction was entirely prevented in nop-1; atg-18 L1 PGCs, which inherited a larger proportion of embryonic PGC mtDNAs than nop-1 single mutants (Figure 2E and G). This is consistent with our finding that late embryonic PGCs contain slightly more mtDNAs than L1 PGCs (Figure 2B). Together, these results (and data below; see Figure 5A–B) suggest that lobe cannibalism and autophagy are both required for complete mtDNA copy number reduction in PGCs, and indicate that if any mtDNA replication occurs in late embryonic or L1 PGCs, it is insufficient to outpace autophagy-mediated mitochondrial destruction at this stage.

Lobe cannibalism and autophagy could reduce the number of mtDNAs to a fixed number, or alternatively, eliminate a fixed proportion of the mtDNAs present within PGCs regardless of how many are present. To distinguish between these possibilities, we took advantage of the fact that changing TFAM activity can alter mtDNA copy number (Larsson et al., 1998; Sumitani et al., 2011). Indeed, we found that whole embryos from the TFAM-GFP knock-in strain contained significantly fewer mtDNAs (8450 ± 768) than wild type (Figure 1—figure supplement 2C), indicating that the GFP tag partially interferes with TFAM function. Using the TFAM-GFP strain, we asked how many mtDNAs PGCs eliminate if they are born with a reduced number. The TFAM-GFP embryonic PGCs contained 89 ± 3 mtDNAs and TFAM-GFP L1 PGCs contained 54 ± 2 mtDNAs (Figure 2F). Thus, despite the presence of markedly fewer mtDNAs in the TFAM-GFP strain, L1 PGCs still inherit a comparable percentage of the mtDNAs contained within embryonic PGCs (wild-type: 55% and TFAM-GFP: 60%; Figure 2F–G). Conversely, when embryonic PGCs contained ~25% excess mtDNAs (in the mptDf2 mtDNA mutant strain), we still observed a twofold reduction in mtDNAs by the L1 stage (see Figure 4—figure supplement 1D-E). Together, these data indicate that lobe cannibalism and autophagy do not subtract the number of PGC mtDNAs to a defined number, but rather divide the population of PGC mtDNAs present by a fixed proportion.

Our results so far suggest that lobe cannibalism and autophagy could contribute to a germline mtDNA bottleneck by halving the number of PGC mtDNAs. However, if the initial cycles of larval germline proliferation proceed in the absence of bulk mtDNA replication, the number of mtDNAs per germ cell would continue to drop and an mtDNA low point (per cell) would occur at a later stage of germline development. When L1 larvae first encounter food, PGCs exit from quiescence and begin to proliferate, forming a population of undifferentiated germline stem cells (GSCs) (Fukuyama et al., 2006; Hubbard and Schedl, 2019). It is not known whether germline mtDNA replication has begun at this stage. Previous quantitative PCR (qPCR) experiments on whole worms first revealed a significant expansion of germline mtDNAs after the L3 larval stage (Bratic et al., 2009; Tsang and Lemire, 2002a). However, these experiments might have lacked the resolution to detect an increase in mtDNAs were it to occur within the relatively small number of GSCs present in whole L1 larvae.

To determine if mtDNAs replicate as L1 PGCs exit quiescence and divide to produce GSCs, we quantified TFAM foci as PGCs in fed L1s began to proliferate as GSCs. To circumvent the mtDNA replication defects that we noted in TFAM-GFP worms, we utilized split GFP, a form of bimolecular fluorescence complementation that brings together the GFP_1-10 and GFP₁₁ fragments of GFP, which are non-fluorescent until reunited (Cabantous et al., 2005; Kamiyama et al., 2016); this approach allowed us to tag TFAM (hmg-5) endogenously with the much smaller GFP₁₁ tag. To visualize TFAM-GFP₁₁, we expressed a mitochondrial matrix-targeted, PGC-specific, GFP_1-10 [Mito-GFP_1-10^(PGC)]. The GFP_1-10 alone was minimally fluorescent, but upon binding to GFP₁₁ formed a functional fluorophore (Figure 3—figure supplement 1A-B). Within PGCs, TFAM-GFP₁₁ detected with Mito-GFP_1-10^(PGC) showed an identical localization pattern to TFAM-GFP (Figure 3A–B), and did not cause significant defects in mtDNA copy number (Figure 3—figure supplement 1C). Larvae fed beginning at the L1 stage showed a progressive increase in the number of TFAM-GFP₁₁ foci per germ line (Figure 3C–F, compare with Figure 3B). The TFAM-GFP₁₁ foci numbers began to increase even before the first division of the PGCs was complete (early-L1, Figure 3C and G), and continued to expand through the L2 stage (an average of 22 GSCs), when we stopped our analysis (Figure 3D–G). In contrast to the increasing number of TFAM-GFP₁₁ foci per germ line over this period, the number of foci per germ cell remained constant after a transient spike in early-L1s (two cells), and stabilized at a number of foci similar to that of L1 that had not been fed (Figure 3H).

To complement these experiments, we sorted GSCs (Figure 3—figure supplement 2) from fed mid-L1 larvae (containing an average of 4 GSCs) and L2 larvae (containing an average of 18 GSCs), and counted the number of mtDNA molecules per GSC. Consistent with our TFAM-GFP₁₁ observations, the number of mtDNAs per germ line increased over this period (Figure 3I), although the number of mtDNAs per GSC remained constant and similar to that in starved L1s (~200; Figure 3J). Together, these results indicate that mtDNAs replicate in bulk as L1 PGCs begin to divide to form GSCs, and thereafter balance mtDNA replication with cell division to maintain a constant number of mtDNAs per GSC, through at least the L2 stage.

The observation that GSCs contain the same number of mtDNAs as L1 PGCs suggests that lobe cannibalism and autophagy might function to reduce PGC mtDNA numbers to an optimal level. To explore this hypothesis, we examined mtDNA number in GSCs of nop-1 mutants, since we found that nop-1 L1 PGCs contain excess mtDNAs. Remarkably, nop-1 GSCs isolated from mid-L1 and -L2 larvae contained a similar number of mtDNAs as did wild-type GSCs (~200; Figure 3K–L). These results suggest that GSCs actively coordinate mtDNA replication with cell division to maintain ~200 mtDNA per cell, even if excess mtDNAs are present at the onset of germline expansion.

Our experiments so far have not addressed whether PGCs eliminate mitochondria indiscriminately, or alternatively, if poorly functioning mitochondria, containing high levels of mutant mtDNA, are preferentially targeted for destruction. To examine this question, we investigated PGCs containing the uaDf5 mtDNA deletion. The uaDf5 deletion removes 3.1 kb of the mitochondrial genome, including several essential genes (Figure 4A), and therefore can exist only when in heteroplasmy with wild-type mtDNA (Tsang and Lemire, 2002b). However, uaDf5 persists stably because it is preferentially replicated compared with wild-type mtDNA (Yang et al., 2022; Gitschlag et al., 2016; Gitschlag et al., 2020; Lin et al., 2016). Our experiments above suggest that bulk mtDNA expansion does not occur until PGCs differentiate and divide in L1 larvae, potentially providing an opportunity for purifying selection to reduce uaDf5 levels before larval germline growth begins.

First, we measured uaDf5 heteroplasmy in whole embryos, embryonic PGCs, and L1 PGCs, as well as in GSCs of fed larvae (Figure 4B, Figure 4—figure supplement 1A-C). The uaDf5 deletion occurred at 48% ± 0.3 heteroplasmy in embryonic PGCs, which was nearly identical to its heteroplasmy in whole embryos (Figure 4B), suggesting that there is no strong selection against uaDf5 during embryogenesis prior to PGC birth. However, in L1 PGCs, uaDf5 heteroplasmy dropped by 4.5% (Figure 4B). This effect was not specific to uaDf5, as PGC heteroplasmy of the 1.5 kb mptDf2 deletion was also reduced between embryogenesis and the L1 stage (Figure 4—figure supplement 1F). Within starved L1 PGCs, uaDf5 was present at 43.5% ± 0.5 heteroplasmy and was maintained at a similar level after the PGCs divided once to form four GSCs (mid-L1 stage). However, by the L2 stage (average of 20 GSCs), uaDf5 heteroplasmy increased to 53% ± 1.7 a level nearly identical to that of whole adult worms (Figure 4B). These findings suggest that PGCs utilize purifying selection to reduce levels of mutant mtDNAs at a stage when uaDf5 cannot take advantage of bulk mtDNA replication to expand selfishly within the germ line. However, once the number of mtDNAs expands in larval GSCs, the percentage of uaDf5 mutant mtDNAs can once again rise.

To test whether lobe cannibalism is responsible for purifying selection against uaDf5 in PGCs, we examined uaDf5 heteroplasmy in nop-1 mutant PGCs. Similar to wild type, uaDf5 PGCs reduced their total mtDNA content ~twofold between embryogenesis and L1, and as expected, nop-1; uaDf5 PGCs failed to reduce their mtDNA significantly (Figure 4C). Surprisingly, we found that in nop-1; uaDf5 mutants, uaDf5 heteroplasmy still decreased from embryonic PGCs to L1 PGCs (Figure 4D–E). We conclude that lobe cannibalism is not responsible for the reduction in uaDf5 heteroplasmy within PGCs, implicating an alternative pathway in PGC mtDNA purifying selection.

Several adaptor proteins function upstream of the autophagy pathway to specifically eliminate mitochondria – a process called mitophagy. The mitophagy receptor BCL2 interacting protein 3 like (BNIP3L) is required for autophagy-driven mitochondrial clearance in erythrocytes (Meiklejohn et al., 2007; Sandoval et al., 2008; Schweers et al., 2007; Zhang et al., 2009) and for mtDNA purifying selection in the Drosophila ovary (Lieber et al., 2019). To test whether mitophagy or autophagy preferentially removes uaDf5 mtDNAs in PGCs, we sorted embryonic and L1 PGCs in uaDf5 mutants containing putative null mutations in dct-1/BNIP3L, as well as in atg-18 and atg-13, which block autophagy at the elongation and initiation steps, respectively (Palmisano and Meléndez, 2019; Tian et al., 2009). Surprisingly, we found no defect in either PGC mtDNA copy number reduction or purifying selection in dct-1/BNIP3L; uaDf5 mutants (Figure 5—figure supplement 1). The L1 PGCs in atg-18 and atg-13 mutants had reduced numbers of total mtDNAs compared with embryonic PGCs, although a smaller percentage (atg-18: 25% and atg-13: 20%) of mtDNAs were eliminated compared to uaDf5 alone (52%; Figure 5A–B). Consistent with our findings above (see Figure 2), these data suggest that the autophagy pathway, but not the mitophagy receptor DCT-1/BNIP3L, acts in parallel with lobe cannibalism and is partially responsible for the reduction of mtDNAs in PGCs. Unexpectedly, in both atg-13; uaDf5 and atg-18; uaDf5 mutants, uaDf5 heteroplasmy was still reduced in L1 PGCs compared to embryonic PGCs (Figure 5C–D). We conclude that autophagy likely eliminates a subset of mitochondria and mtDNAs within PGCs non-selectively, but is not responsible for purifying selection against uaDf5 mutant mtDNA. Consistent with this interpretation, we observed acidified mitochondria [mCherry(+) Dendra(-)] in uaDf5 PGCs at comparable frequencies to wild-type PGCs (Figure 5E and G, Figure 2—figure supplement 4), and acidified foci were completely absent in atg-18; uaDf5 null mutant embryos (Figure 5F–G).

The PINK1/Parkin signaling pathway, which consists of the mitochondrial kinase PINK1 and its effector ubiquitin ligase Parkin, can recognize and mark defective mitochondria for destruction either via autophagy or through autophagy-independent pathways (Zhang et al., 2019; Hammerling et al., 2017; McLelland et al., 2014). C. elegans contains single orthologs of PINK1 (encoded by pink-1) and Parkin (encoded by pdr-1) (Hamamichi et al., 2008; Springer et al., 2005). To address whether PINK1 or Parkin are required for purifying selection of uaDf5, we examined uaDf5 heteroplasmy in PGCs with putative null mutations in pink-1, pdr-1, and pink-1; pdr-1 double mutants. As expected, single and double mutants had reduced mtDNA content in L1 PGCs compared to embryonic PGCs, although to a lesser extent than uaDf5 controls (Figure 6A–B). However, even though uaDf5 heteroplasmy was markedly higher in all three backgrounds compared to uaDf5 controls (see Discussion), only pink-1, and pink-1; pdr-1 double mutants, but not pdr-1 single mutants, abrogated the reduction in uaDf5 heteroplasmy between embryonic and L1 PGCs (Figure 6C–D). Taken together, these findings indicate that PINK-1 alone, acting independently of PDR-1/Parkin, is required for autophagy-independent purifying selection against mutant mtDNAs within C. elegans PGCs.

Unless otherwise stated, all strains were maintained at 20°C on nematode growth medium plates seeded with Escherichia coli strain OP50 according to standard methods (Brenner, 1974). For egg isolation and L1 synchronization, semi-synchronized L1 larvae were outgrown on 10 cm enriched peptone plates seeded with E. coli strain NA22. Gravid adults were then washed off and early-stage embryos were isolated via worm bleaching. Isolated eggs were broken into two populations: one for immediate embryo dissociation and another which was allowed to hatch and starved overnight in M9 for L1 synchronization/dissociation. For late embryo dissociations, early embryos were isolated as above and incubated in M9 at 25°C for 6 hr. For L1 feeding experiments, synchronized L1 larvae were plated onto enriched peptone plates and grown for 12 and 24 hr at 20°C (for cell sorting), or for 6, 9, 12, and 24 hr at 23°C (for live imaging). A list of all strains used/generated in the study is available in the Key resources table.

Cell dissociation of embryos and larvae was performed as described previously (Lee et al., 2017; Strange et al., 2007) with slight modifications described in detail below.

Embryos, adults, and larvae were mounted on 5 and 10% agarose pads, respectively. Larvae were immobilized prior to and during image acquisition using 1.25 mM levamisole in M9 buffer. Animals were imaged on a Leica SP8 laser-scanning confocal microscope, using a 63 × 1.4 NA oil-immersion objective with 488 nm and 594 nm lasers and HyD detectors; or on a Zeiss AxioImager A2, using a 40 × 1.3 NA oil-immersion objective and a charge-coupled device (CCD) camera (model C10600-10B-H, S. 160522; Hamamatsu). For sorted PGC imaging, 5 µL of sorted embryonic and larval PGCs in conditioned L-15 (see ‘FACS and PGC isolation’ above) were mounted on custom depression slides to avoid crushing the cells. Sorted PGCs were then imaged on a Zeiss AxioImager A2 as above. Images were analyzed and processed in ImageJ (NIH), and Adobe Photoshop.

In all cases, CRISPR/Cas9 mediated genome editing was performed using pre-incubated Cas9 (Berkeley)::(crRNA +tracrRNA) (IDT) ribonucleoprotein, and injection quality was screened using the co-CRISPR dpy-10(cn64) mutation as previously described (Paix et al., 2017). DNA repair templates contained ~25–35 bps of homology on each arm, and varied depending on the size of insertion as either dsDNA PCR product (>150 bps), or ssDNA oligos (<150 bps) (IDT). The crRNAs and insertion sequences are listed in the Key resources table and Supplementary file 1. For the generation of putative null alleles [pdr-1(xn199), dct-1(xn192)] we used the ‘STOP-IN’ method (Wang et al., 2018) to insert an early stop and frame-shift into either the first or second, exon of the target gene. For the generation of xnSi73 [mex-5p::GFP_1-10::nos-2 3’UTR, unc-119(+)], pAS07 was used as a PCR template to amplify GFP_1-10 with ~35 bp of homology to replace tomm-20^1-54::Dendra2 by CRISPR at the xnSi67 locus. To generate xnSi85 [mex-5p::mito(matrix)GFP_1-10::nos-2 3’UTR, unc-119(+)], an oligo repair template (see Key resource table) was used to introduce an N-terminal mitochondrial-matrix localization sequence to xnSi73. To generate hmg-5(xn107[hmg-5-GFP]) and hmg-5(xn168[hmg-5-GFP₁₁]), full length GFP with ~35 bp homology arms or an oligo-containing sequence for GFP₁₁ were used to generate C-terminal tags at the endogenous hmg-5 locus. All strains generated by CRISPR are included in the Key resources table and relevant sequences are in Supplementary file 1.

Statistical analysis was performed using GraphPad Prism 9 software. For categorical data, such as scoring acidified mitochondria in PGCs, contingency tables were made and Fisher’s exact test was used to calculate p-values. For all other data, one-tailed or two-tailed Student’s t-tests were performed, as applicable. For mtDNA copy number comparisons, unpaired t-tests were used since embryos and L1s could come from the same or different adult populations; for heteroplasmy experiments, paired tests were used since embryos and L1 PGCs always came from the same adult population. Data in graphs are shown as Superplots (Lord et al., 2020), with individual data points from three independent color-coded biological replicates (except for ddPCR experiments where small dots are technical replicates of the ddPCR analysis) shown as small dots, the mean from each experiment shown as a larger circle, the mean of means as a horizontal line, and the SEM as error bars. Sample size, t-test type, and p-value ranges are reported in figure legends. Where applicable, no corrections for multiple comparisons were made to avoid type II errors (Armstrong, 2014). For live imaging, embryos and larvae were selected based on orientation on the slide and on health. For all datasets, at least three biologically independent experiments were performed and the arithmetic means of biological replicates were used for statistical analysis. Combined source data for all ddPCR experiments can be found in Supplementary file 2.