Summary
Cell competition is emerging as a quality control mechanism that eliminates unfit cells in a wide range of settings from development to the adult. However, the nature of the cells normally eliminated by cell competition and what triggers their elimination remains poorly understood. Here we have performed single cell transcriptional profiling of early mouse embryos and find that the cells eliminated show the hallmarks of cell competition, are mis-patterned and have mitochondrial defects. We demonstrate that mitochondrial defects are common to a range of different loser cell types and that manipulating mitochondrial function is sufficient to trigger competition. Importantly, we show that in the embryo loser epiblast cells display mitochondrial DNA mutations and that even small changes in mitochondrial DNA sequence can influence the competitive ability of the cell. Our results therefore suggest that cell competition is a purifying selection that optimises metabolic output prior to gastrulation.
Introduction
During the early stages of mammalian development, the cellular and molecular landscape is profoundly remodelled. As embryonic cells approach gastrulation, when the precursors of all embryonic tissues are specified, they need to rewire the transcriptional, epigenetic, metabolic and signalling networks that govern cell identity (Kojima et al., 2014). These changes are accompanied by a marked acceleration in the proliferation rate (Snow, 1977) and need to be orchestrated with the different morphogenetic processes that re-shape the embryo (reviewed in (Stower and Srinivas, 2018). The scale of this remodelling creates the potential for the emergence of abnormal cells that need to be removed to prevent them from contributing to the soma or germline during development. This requirement implies that there must be stringent cell fitness quality control mechanisms acting around the time of gastrulation. One such control has been postulated to be cell competition, a fitness sensing mechanism eliminating cells that, although viable, are less fit than their neighbours (reviewed in (Bowling et al., 2019; Diaz-Diaz and Torres, 2019; Madan et al., 2018). During cell competition, the cells that are eliminated are generically termed losers, while the fitter cells that survive are referred to as winners.
Cell competition has been primarily studied in Drosophila, where it was first described in the imaginal wing disc (Morata and Ripoll, 1975). Since then, it has also been found to be conserved in mammals. For example, in the mouse embryo between E5.5 and E6.5 cell competition has been proposed to eliminate less-fit cells from the epiblast, the pluripotent tissue that generates the three primary germ-layers (Claveria et al., 2013; Sancho et al., 2013). Importantly, in the time-window when cell competition acts, about one third of epiblast cells are eliminated by apoptosis (Bowling et al., 2018). These cells are marked by a loss of mTOR signalling, a read-out of loser status during cell competition in vitro (Bowling et al., 2018). This suggests that cell competition is the primary cause of this cell elimination. Other markers of those cells eliminated in the early post-implantation embryo have been identified as relative low c-MYC expression, high P53 expression or elevated levels of ERK phosphorylation (Bowling et al., 2018; Claveria et al., 2013; Diaz-Diaz et al., 2017; Sancho et al., 2013), that together could be considered as a cell competition signature.
In spite of the advance that identifying these markers signifies, we still do not know what overarching cellular and molecular features define the cells eliminated by cell competition in mouse. Using embryonic stem cells (ESCs) and embryo chimeras we have shown that mis-patterned cells, autophagy deficient cells and karyotypically abnormal cells are all eliminated upon differentiation by cell competition (Bowling et al., 2018; Sancho et al., 2013). Cell competition has also been shown to eliminate pluripotent cells that differentiate precociously (Diaz-Diaz et al., 2017) or that are not properly specified (Hashimoto and Sasaki, 2019). Therefore, a variety of different defective cell types can be eliminated by cell competition, but which is the underlying cause for their elimination remains poorly understood.
Mitochondria, with their diverse cellular functions ranging from determining the bioenergetic output of the cell to regulating its apoptotic response, are strong candidates for determining competitive cell fitness. During early mouse development mitochondria undergo profound changes in their shape and activity (reviewed in (Lima et al., 2018). In the pre-implantation embryo mitochondria are rounded, fragmented and contain sparse cristae, but upon implantation they fuse to form complex networks with mature cristae (Zhou et al., 2012). The mode of replication of the mitochondrial genome (mtDNA), that encodes for vital components of the bioenergetic machinery, also changes during early mouse development. After fertilization, mtDNA replication ceases and its copy number per cell decreases with every division until post-implantation stages, when mtDNA replication resumes (reviewed in (Lima et al., 2018). However, as the mutation rate of mtDNA is significantly higher than that of nuclear DNA (Allio et al., 2017; Khrapko et al., 1997) this increased replication most likely leads to an increased mutation load. A number of mechanisms have been proposed to reduce this mutation load, such as the bottleneck effect, purifying selection or biased segregation of mtDNA haplotypes (Burgstaller et al., 2014; Johnston et al., 2015; Latorre-Pellicer et al., 2019; Lee et al., 2012; Sharpley et al., 2012; Zhang et al., 2018). However, despite these mechanisms, inheritable mtDNA based diseases are reported with a prevalence of 5-15 cases per 100,000 individuals (Burgstaller et al., 2015; Gorman et al., 2016), highlighting both the importance and limitations of these selection mechanisms.
To understand the nature of the cells eliminated during early mouse post-implantation development we have analysed their transcriptional profile by single-cell RNA sequencing. Importantly, we have found that these cells share a cell competition signature. Analysis of the pathways mis-regulated in the cells eliminated identified mitochondrial dysfunction as a common feature. Furthermore, we demonstrate that manipulating mitochondrial activity either by disrupting mitochondrial dynamics or by introducing small mtDNA changes is sufficient to trigger cell competition during early mouse development. These results therefore pinpoint mitochondrial performance as a key cellular feature that determines the competitive ability of embryonic cells.
Results
Cells eliminated in the early mouse embryo have a distinct transcriptional profile
We have previously shown that in the early post-implantation mouse embryo about 35% of epiblast cells are eliminated and that these cells are marked by low mTOR signalling (Bowling et al., 2018). However, we currently do not understand the characteristics of these cells or what triggers their elimination. To answer these questions, we have analysed their transcriptional profile by single cell RNA sequencing (scRNA-seq). To ensure we can capture the eliminated cells, as we have done before (Bowling et al., 2018), we isolated embryos at E5.5 and cultured them for 16 hours in the presence of a caspase inhibitors (CI) or vehicle (DMSO) (Figure 1A and Figure S1A).
Unsupervised clustering of the scRNA-seq data revealed five clusters: two corresponding to extra-embryonic tissues (visceral endoderm and extra-embryonic ectoderm) and three that expressed epiblast marker genes (Figure 1B-C and Figure S1B-D). Interestingly, cells from CI- and DMSO-treated embryos are unequally distributed across the three epiblast clusters. In particular, one of these clusters (cluster 4) is only composed of cells from CI-treated embryos (Figure 1D-E). To establish the relationship between these epiblast clusters we computed a diffusion map (Angerer et al., 2016). For this, we selected only cells captured from CI-treated embryos, to eliminate possible confounding effects due to the caspase inhibitor (Figure 2A). However, when all epiblast cells are considered, the results remain unchanged (Figure S2A-C). This analysis identified a trajectory between the three epiblast clusters, with those cells unique to CI-treated embryos falling at one extreme end of the trajectory (corresponding to cluster 4; Figure 2A) and with those cells present in both DMSO and CI-treated embryos at the other (corresponding to cluster 1; Figure 2A and Figure S2A-C).
To further define the identity of the epiblast cells of CI-treated embryos we analysed the genes differentially expressed along the trajectory (see Methods and Figure S2D) using Ingenuity Pathway Analysis (IPA) to characterize gene signatures (Kramer et al., 2014). Importantly, we found that these differentially expressed genes fell under molecular and cellular function categories associated with cell death and survival, protein synthesis and nucleic acids (Figure 2B). Analysis of the factors with enriched targets within the genes differentially expressed along the trajectory revealed RICTOR (an mTOR component), TLE3, MYC, MYCN, P53 and IGFR (that is upstream of mTOR) as the top upstream regulators (Figure 2C). Breaking down the differentially expressed genes into those down-regulated or up-regulated along the winner-to loser trajectory revealed that the targets of RICTOR, MYC, MYCN and IGFR primarily fell within the down-regulated genes (Supplementary Tables 1 and 2). P53 activated targets were preferentially up-regulated and P53 repressed targets were preferentially down-regulated (Figure S2E-F). Moreover, genes related to protein synthesis were primarily found to be downregulated.
The first identified trigger of cell competition were differences in protein synthesis (Morata and Ripoll, 1975), therefore the observation that the genes differentially expressed along the trajectory fall into cell death and protein synthesis categories, as well as being mTOR, MYC and P53 targets strongly suggests that cells at each end of the trajectory are the winners and losers of cell competition. For this reason, we hereafter refer to those epiblast cells unique to CI-treated embryos as “loser” epiblast cells and to those at the opposite end of the trajectory as the “winner” epiblast cells. Those cells lying between these two populations on the trajectory are considered “intermediate”. Using this knowledge we can define a diffusion pseudotime (dpt) coordinate (Haghverdi et al., 2016) originating in the “winner” cluster that tracks the position of cells along the trajectory and that can be interpreted as a “losing score”, i.e., it quantifies how strong the signature of the “losing” state is in the transcriptome of a cell (see Figure 2D-E).
Loser cells show a signature of mis-patterning
The most notable change that embryonic cells undergo during early development is the onset of differentiation. During this process cells transit from naïve to primed pluripotency and then undergo germ layer specification (Nichols and Smith, 2009). For this reason, we analysed the expression of pluripotency and differentiation markers as a function of the losing score of epiblast cells (Figure 2F). We found that loser cells (with high losing scores) have lower levels of expression of genes such as Fgf5 and Tdgf1, which are characteristic of the primed pluripotency state. In contrast to this, they displayed high levels of expression of some naïve pluripotency markers such as Klf4 and Klf5, but not of others, such as Rex1 (Zfp42). Analysis of lineage specific markers revealed a complex pattern. For example, some loser cells showed high expression of the endoderm marker Sox17, but not of Gata6. Something similar was observed for neuroectoderm markers, whereby loser cells showed significantly higher expression of Neurod1 than winner cells, but comparable levels of Sox1.
Plotting the levels of markers of naïve pluripotency and germ layers against each other for single cells revealed that some loser epiblast cells expressed higher levels of both naïve pluripotency and germ-layer markers than normal epiblast cells (Figure 2G). Together, these data suggest that loser epiblast cells represent a generic mis-patterned state, rather than precocious or delayed differentiation.
Loser cells are characterised by defects in mitochondrial function
We next analysed using IPA the cellular pathways mis-regulated in loser epiblast cells and found that the top two pathways (mitochondrial dysfunction and oxidative phosphorylation) are related to mitochondrial function (Figure 3A). Detailed inspection of the oxidative phosphorylation signature in loser versus normal epiblast cells indicated that 92.6% of the genes in this pathway were mis-regulated (Figure 3A), including the majority of genes encoding proteins of the five complexes (Complexes I to V) of the electron transport chain (ETC)(Figure 3B), that were down-regulated along the winner-to-loser trajectory (Supplementary Table 1). For example, we found a down-regulation along the winner to loser trajectory of the mitochondrial DNA encoded mt-Nd3 and mt-Atp6 (Figure 3C), of regulators of mitochondrial dynamics such as Opa1, as well as of genes involved in mitochondrial membrane and cristae organisation such as Samm50, (Figure 3C).
A recent body of evidence has revealed that stress responses, such as the integrated stress response (ISR) or the closely related unfolded protein response (UPR), when triggered in cells with impaired mitochondrial function prompt a transcriptional program to restore cellular homeostasis (Melber and Haynes, 2018; Munch, 2018; Topf et al., 2016). We observed that loser epiblast cells displayed a characteristic UPR-ISR signature (Figure S3A-C) (Mouchiroud et al., 2013; Nargund et al., 2012; Quiros et al., 2016; Zhao et al., 2002) and key regulators of this response, such as Atf4, Ddit3, Nrf2 and Foxo3 were all up-regulated in these cells (Figure S3D). Similarly, Sesn2, a target of p53 that controls mTOR activity (Saveljeva et al., 2016), was also up-regulated in loser cells (Figure S3D). These findings support the possibility that loser epiblast cells present mitochondrial defects, leading to the activation of a stress response in an attempt to restore cellular homeostasis (Yun and Finkel, 2014).
To validate the significance of the mitochondrial defects observed, we did two things. First, we asked if the changes observed at the mRNA level are also reflected at the protein level. We observed that in CI-treated embryos, loser cells that persist and are marked by low mTOR activity (Bowling et al., 2018), also show significantly lower OPA1 levels (Figure 3D-F). This finding is in agreement with the observation that OPA1 levels are decreased during mitochondrial stress (Quiros et al., 2017). We also found that DMSO-treated embryos showed strong DDIT3 staining (an UPR-ISR marker) in the dying cells that accumulate in the proamniotic cavity, and that in CI-treated embryos, DDIT3 expression was up-regulated in a proportion of epiblast cells (Figure S3E-G). The second thing we did to validate the importance of the mitochondrial defects observed was to study in loser epiblast cells the mitochondrial membrane potential (Δψm), an indication of mitochondrial health. We observed that while the cells of DMSO-treated embryos showed a high Δψm that fell within a narrow range, in CI-treated embryos the proportion of cells with a low Δψm significantly increased (Figure 3D and 3G-H). Together, these results suggest that loser epiblast cells have impaired mitochondrial activity that triggers a stress response.
Mitochondrial dysfunction is common to different types of loser cells
The above data indicate that loser epiblast cells are mis-patterned and show mitochondrial defects. To address if mitochondrial defects are a common feature of mis-patterned cells we analysed ESCs that are defective for BMP signalling (Bmpr1a-/-), as these are abnormally patterned (Di-Gregorio et al., 2007) and eliminated by cell competition (Sancho et al., 2013). In parallel, to determine if mitochondrial defects are present in other loser cells eliminated by cell competition, we also studied tetraploid cells (4n) (Sancho et al., 2013). We first carried out a mass spectrometry analysis using the Metabolon platform and found that metabolites and intermediates of the TCA cycle, such as malate, fumarate, glutamate and α-ketoglutarate are depleted in both Bmpr1a-/- and 4n ESCs in differentiation culture conditions (Figure 4A). Next, we performed an extracellular flux Seahorse analysis of Bmpr1a-/- ESCs to measure their glycolytic and oxidative phosphorylation (OXPHOS) rates. We observed that when these cells are maintained in pluripotency culture conditions, that are not permissive for cell competition (Sancho et al., 2013), they showed a similar glycolytic activity, but a higher OXPHOS rate than control cells (Figure S4A-D). In contrast, when Bmpr1a-/- cells are induced to differentiate, this phenotype is reversed, with mutant cells showing lower ATP generated through OXPHOS and a higher glycolytic capacity than controls (Figure 4B-E; Figure S4E-F). This suggests that upon differentiation Bmpr1a-/- cells are unable to sustain proper OXPHOS activity.
To further test the possibility that defective ESCs have impaired mitochondrial function, we assessed their Δψm. We found that whilst Bmpr1a-/- and 4n cells had a similar Δψm to control cells in pluripotency conditions (Figure S4G-H), upon differentiation both these cell types presented a loss of Δψm, irrespective of whether they were separate or co-cultured with wild-type cells (Figure 4F-G). This reduction in Δψm is unlikely to be due to excessive mitochondrial reactive oxygen species (ROS) production or to a lower mitochondrial mass within mutant cells since, as for example, Bmpr1a-/- cells have lower ROS levels and similar TOMM20 and mt-CO1 expression as control cells (Figure 4H-J; Figure S4I). The fact that the loss of Δψm and lower OXPHOS activity can be observed even when loser cells are cultured separately, suggests that the mitochondrial dysfunction phenotype is an inherent property of loser cells and not a response to them being out-competed. These results also indicate that the mitochondrial defects are directly linked to the emergence of the loser status: in conditions that are not permissive for cell competition (pluripotency) mutant cells do not show defective mitochondrial function, but when they are switched to differentiation conditions that allow for cell competition, they display impaired mitochondrial function.
To further explore the relationship between mitochondrial activity and the competitive ability of the cell, we analysed the Δψm of BMP defective cells that are null for p53 (Bmpr1a-/-; p53-/- ESCs), as these are not eliminated by wild-type cells (Bowling et al., 2018). Remarkably, we observed that mutating p53 in Bmpr1a-/- cells not only rescues the loss of Δψm of these cells, but also causes hyperpolarisation of their mitochondria (Figure 4K). These results strongly support a pivotal role for mitochondrial activity in cell competition.
Impaired mitochondrial function is sufficient to trigger cell competition
The mitochondrial defects observed in loser cells led us to ask if disrupting mitochondrial activity alone is sufficient to trigger cell competition. During the onset of differentiation, mitochondria go from having a fragmented shape to fusing and forming complex networks (reviewed in (Lima et al., 2018). We therefore tested if disrupting mitochondrial dynamics induces cell competition. MFN1 and MFN2 regulate mitochondrial fusion and DRP1 controls their fission (Chen et al., 2003; Prudent and McBride, 2017; Smirnova et al., 2001). We generated Drp1-/- ESCs, that show hyper-elongated mitochondria, and Mfn2-/- ESCs, that have enlarged globular mitochondria (Figure 5A). Analysis of the Drp1 mutant cells showed that although they did not grow significantly slower than wild-type cells when cultured separately in differentiation inducing conditions, they were out-competed by wild-type cells in co-culture (Figure 5B). In contrast to this, we observed that Mfn2-/- ESCs displayed very poor growth upon differentiation (data not shown). For this reason, we tested their competitive ability in pluripotency conditions, that we have previously found not to induce the out-competition of Bmpr1a-/- or 4n cells (Sancho et al., 2013). Interestingly, we found that although Mfn2-/- cells grow similarly to wild-type cells in separate cultures, they were out-competed in co-culture (Figure 5C). The observation that disrupting mitochondrial dynamics can induce cell competition even in pluripotency culture conditions, suggests that mitochondrial activity is a dominant parameter determining the competitive ability of the cell.
Loser epiblast cells accumulate mtDNA mutations
There is strong evidence for selection against aberrant mitochondrial function induced by deleterious mtDNA mutations in mammals (Fan et al., 2008; Freyer et al., 2012; Kauppila et al., 2016; Sharpley et al., 2012; Stewart et al., 2008). Given that we observe that cell competition selects against cells with impaired mitochondrial function, we asked if cell competition could be reducing mtDNA heteroplasmy (frequency of different mtDNA variants) during mouse development. It has been recently shown that scRNA-seq can be used to reliably identify mtDNA variants, although with a lower statistical power compared to more direct approaches, like mtDNA sequencing (Ludwig et al., 2019). We therefore tested if mtDNA heteroplasmy is present in our scRNA-seq data and whether this is associated with the losing score of a cell. Our analysis revealed that the frequency of specific mtDNA polymorphisms increased with the losing score of epiblast cells (Figure 6A), and such mtDNA changes occurred within mt-Rnr1 and mt-Rnr2 (Figure 6B-H and Figure S5A-E). Moreover, these changes were not dependant on the litter from which the embryo came from (Figure S5F-K). The mutations we detected in mt-Rnr1 and mt-Rnr2 strongly co-occurred in the same cell, with those closest together having the highest probability of co-existing (Figure 6I and Figure S5L). This is suggestive of mtDNA replication errors that could be ‘scarring’ the mtDNA, disrupting the function of mt-Rnr1 (12S rRNA) and mt-Rnr2 (16S rRNA) and causing the loser phenotype. Importantly, the presence of these specific mtDNA mutations in the loser cells suggests that cell competition is contributing to the elimination of deleterious mtDNA mutations during early mouse development.
Changes in mtDNA sequence can determine the competitive ability of a cell
To explore this possibility further we analysed if alterations in mtDNA can induce cell competition by testing the competitive ability of ESCs with non-pathological differences in mtDNA sequence. For this we compared the relative competitive ability of ESCs that shared the same nuclear genome background but differed in their mitochondrial genomes by a small number of non-pathological sequence changes. We derived ESCs from hybrid mouse strains that we had previously engineered to have a common nuclear C57BL/6N background, but mtDNAs from different wild-caught mice (Burgstaller et al., 2014). Each wild-derived mtDNA variant (or haplotype) contains a specific number of single nucleotide polymorphisms (SNPs) that lead to a small number of amino acid changes when compared to the C57BL/6N mtDNA haplotype. Furthermore, these haplotypes (BG, HB and ST) can be ranked according to their genetic distance from the C57BL/6N mtDNA (Figures 7A and S6A). Characterization of the isolated ESCs revealed that they have a range of heteroplasmy (mix of wild-derived and C57BL/6N mtDNAs) that is stable over several passages (Figure S6B). Importantly, these different mtDNA haplotypes and different levels of heteroplasmy do not alter cell size, cell granularity, mitochondrial mass or mitochondrial dynamics, nor do they substantially impact the cell’s Δψm (Figure S6C-F).
When we tested the competitive ability of these mtDNA ESCs we observed that cells carrying the mtDNAs that are most distant from the C57BL/6N mtDNA, such as the HB(100%), the HB(24%) and the ST(46%) ESCs could all out-compete the C57BL/6N line (Figure 7B-C and S6G). Similarly, when we tested the HB(24%) line against the BG(99%) or the BG(95%) lines (that have mtDNAs more closely related to the C57BL/6N mtDNA), we found that cells with the HB haplotype could also out-compete these ESCs (Figure 7D and S6H). In contrast, we observed that the HB(24%) ESCs were unable to out-compete either their homoplasmic counterparts, HB(100%), or the ST(46%) cells that carry the most distant mtDNA from C57BL/6N (Figure 7E and S6I). These results tell us three things. First, that non-pathological differences in mtDNA sequence can trigger cell competition. Second, that a competitive advantage can be conferred by only a small proportion of mtDNA content, as indicated by our finding that HB(24%) behave as winners. Finally, these findings suggest that the phylogenetic proximity between mtDNA variants can potentially determine their competitive cell fitness.
To characterise the mode of competition between different mtDNA cells we focussed on the HB(24%) and the BG(95%) ESCs. Analysis of these cell lines revealed that specifically when co-cultured, the BG(95%) cells display high levels of apoptosis (Figure 7F), indicating that their out-competition is through their elimination. To gain further insight we performed bulk RNA-seq of these cells in separate and co-culture conditions (Figure S6J) and analysed the differentially expressed genes by gene-set enrichment analysis (GSEA). We found that in separate culture the most notable features that distinguish BG(95%) from HB(24%) cells were a down-regulation of genes involved in oxidative phosphorylation and an up-regulation of those associated with cytokine activity (Figure 7G). Interestingly, in the co-culture condition, in addition to these signatures, BG(95%) cells revealed a down-regulation in signature markers of MYC activity and mTOR signalling (Figure 7H), whose downregulation are known read-outs of a loser status during cell competition in the embryo (Bowling et al., 2018; Claveria et al., 2013; Sancho et al., 2013)(Figure 2C). These results suggest that the elimination of loser mtDNA cells occurs through the same mechanism as the out-competition of defective cells in the embryo (Figure 7I).
The finding that the genes down-regulated in BG(95%) cells when co-cultured with HB(24%) cells fell under functional categories relating to mitochondrial function (Figure S7A) led us to analyse the degree of overlap between these genes and the genes differentially expressed along the winner-to-loser trajectory in the embryo. We observed a significant overlap in down-regulated genes (Figure S7B), as well as in the functional components that these genes can be categorised into (Figure S7C). This further highlights the importance of relative mitochondrial activity for determining the competitive ability of embryonic cells.
Discussion
The emerging role of cell competition as a regulator of cell fitness in a wide range of cellular contexts, from the developing embryo to the ageing tissue (reviewed in (Bowling et al., 2019; Diaz-Diaz and Torres, 2019; Madan et al., 2018), has highlighted the importance of understanding what cell types are normally eliminated by this process. With the aim of understanding this question, we have analysed the transcriptional identity of the cells eliminated in the early mouse embryo. We have found not only that they present a cell competition signature, but also that they are mis-patterned and marked by impaired mitochondrial function. Starting from these results, we leveraged in vitro models of cell competition to show that: (i) mitochondrial function is impaired in loser cells eliminated by cell competition, and (ii) that differences in mitochondrial activity are sufficient to trigger cell competition in ESCs. Overall, this points to mitochondrial performance as a key determinant of the competitive ability of cells during early mammalian embryonic development. One implication of our findings is that a range of different types of defect, such as mis-patterning, karyotypic abnormalities or mtDNA mutations, all lead to dysfunctional mitochondria at the onset of differentiation and that ultimately it is their impaired mitochondrial function that triggers cell competition, inducing their elimination (Figure 7I).
It is well known that the successful development of the embryo can be influenced by the quality of its mitochondrial pool (reviewed in (Lima et al., 2018). Moreover, divergence from normal mitochondrial function during embryogenesis is either lethal or can lead to the development of mitochondrial disorders (Chinnery and Hudson, 2013). Deleterious mtDNA mutations are a common cause of mitochondrial diseases and during development selection against mutant mtDNA has been described to occur through at least through two mechanisms, the bottleneck effect and intra-cellular purifying selection. The bottleneck effect is associated specifically with the unequal segregation of mtDNAs during primordial germ cell specification, for example as seen in the human embryo (Floros et al., 2018). In contrast to this, purifying selection, as the name implies, allows for selection against deleterious mtDNAs and has been proposed to take place both during development and post-natal life (Burr et al., 2018). Importantly, purifying selection has been found to occur not only at the organelle level, but also at the cellular level (Rajasimha et al., 2008). Our findings indicate that purifying selection can occur not only at the intra-cellular level but also inter-cellularly (cell non-autonomously). We show that epiblast cells are able to sense their relative mitochondrial activity and that those cells with mtDNA mutations, lower or aberrant mitochondrial function are eliminated. By selecting those cells with the most optimised mitochondrial performance, cell competition would not only prevent cells with mitochondrial defects from contributing to the germline or future embryo, but also ensure optimization of the bioenergetic performance of the epiblast, contributing to the synchronization of growth during early development.
Cell competition has been studied in a variety of organisms, from Drosophila to mammals, and it is likely that multiple different mechanisms fall under its broad umbrella (reviewed in (Bowling et al., 2019; Diaz-Diaz and Torres, 2019; Madan et al., 2018). In spite of this, there is considerable interest in understanding if there could be any common feature in at least some of the contexts where cell competition has been described. The first demonstration of cell competition in Drosophila was made by inducing clones carrying mutations in the ribosomal gene Minute (Morata and Ripoll, 1975) and this has become one of the primary models to study this process. Our finding that that during normal early mouse development cell competition eliminates cells carrying mutations in mt-Rnr1 and mt-Rnr2, demonstrates that in the physiological context mutations in ribosomal genes also trigger cell competition. Furthermore, our observation that mis-patterned and karyotypically abnormal cells show impaired mitochondrial activity indicates that during early mouse development different types of defects impair mitochondrial function and trigger cell competition. Interestingly, mtDNA genes are amongst the top mis-regulated factors identified during cell competition in the mouse skin (Ellis et al., 2019). In the Drosophila wing disc oxidative stress, a general consequence of dysfunctional mitochondria, underlies the out-competition of Minute and Mah-jong mutant cells (Kucinski et al., 2017). Similarly, in Madin-Darby Canine Kidney (MDCK) cells, a loss of Δψm occurs during the out-competition of RasV12 mutant cells and is key for their extrusion (Kon et al., 2017). These observations raise that possibility that differences in mitochondrial activity may be a key determinant of competitive cell fitness in a wide range of systems. Unravelling what mitochondrial features can lead to cellular differences that can be read between cells during cell competition will be key not only for understanding this process, but also to open up the possibility for future therapeutic avenues in the diagnosis or prevention of mitochondrial diseases.
Author Contributions
A.L. performed most of the experimental wet lab work. J.B. and A.L. derived heteroplasmic mESC lines. J.B. performed heteroplasmy measurements in heteroplasmic mESCs. B.P. generated Mfn2-/- and Drp1-/- mESCs and J.M.S did characterisation of mitochondria shape and pluripotency status. D.H. performed embryo dissections, treatments and cell dissociation prior to scRNA-seq experiments. G.L. did the bioinformatic analysis of scRNA-seq data. E.M., N.J. and A.G. participated in the analysis of mitochondrial DNA heteroplasmy. A.D.G. performed the metabolomic studies using Metabolon platform and participated in embryo dissections and immunohistochemistry stainings for validation of results obtained by scRNA-seq. M.D, and M.K. performed the bioinformatic analysis of bulk RNA-seq experiments. N.J., S.S. and D.C. participated in the design of experimental work and analysis of results. A.L., G.L., A.S and T.R. interpreted results and wrote the paper. T.R. and A.S. directed and designed the research.
Declaration of Interests
The authors declare no competing interests.
Figure titles and legends
List of Tables
Supplementary Table 1. List of genes down-regulated along the winner-to-loser trajectory in the embryo.
Supplementary Table 2. List of genes up-regulated along the winner-to-loser trajectory in the embryo.
Supplementary Table 3. Genes related to the unfolded protein response and integrated protein response pathways (UPR_ISR) that were analysed in the genes differentially expressed along the winner-to-loser trajectory.
Supplementary Table 4. List of background genes for the winner-to-loser trajectory in the embryo.
Supplementary Table 5. List of genes down-regulated in BG(95%) cells when co-cultured with HB(24%) cells.
Supplementary Table 6. List of genes up-regulated in BG(95%) cells when co-cultured with HB(24%) cells.
Supplementary Table 7. List of background genes used for the analysis of genes differentially expressed between co-cultured BG(95%) and HB(24%) cells.
STAR Methods
KEY RESOURCES TABLE
Presented in a separate file
LEAD CONTACT AND MATERIALS AVAILABILITY
Tristan Rodriguez: tristan.rodriguez{at}imperial.ac.uk and Antonio Scialdone antonio.scialdone{at}helmholtz-muenchen.de
EXPERIMENTAL MODEL AND SUBJECT DETAILS
Cell lines and cell culture routine
E14, kindly provided by Prof A. Smith, from Cambridge University, were used as wild-type control cells tdTomato-labelled or unlabelled. GFP-labelled or unlabelled cells defective for BMP signalling (Bmpr1a-/-), tetraploid cells (4n) and Bmp1a-/- null for p53 (Bmpr1a-/-;p53-/-) are described elsewhere (Bowling et al., 2018; Sancho et al., 2013). Cells null for Dynamin-related protein 1 (Drp1-/-) or Mitofusin 2 (Mfn2-/-) were generated by CRISPR mutagenesis. Cells with different mitochondrial DNA (mtDNA) content in the same nuclear background were derived from embryos of hybrid mice, generated elsewhere (Burgstaller et al., 2014).
Cells were maintained pluripotent and cultured at 37°C in 5% CO2 in 25 cm2 flasks (Nunc) coated with 0.1% gelatin (Sigma) in DPBS. Growth media (ES media) consisted of GMEM supplemented with 10% FCS, 1mM sodium pyruvate, 2 mM L-glutamine, 1X minimum essential media non- essential amino-acids, 0.1 mM β-mercaptoethanol (all from Gibco) and 0.1% leukemia inhibitory factor (LIF, produced and tested in the lab). Cells derived from hybrid mice (C57BL/6N nuclear background) were maintained on 0.2% LIF. The growth media was changed daily, and cells were split every 3 days.
CRISPR mutagenesis
Drp1 and Mfn2 knockout ESCs were generated by CRISPR-Cas9 mediated deletion of Drp1 exon 2 and Mfn2 exon 3 respectively. sgRNA guides flanking Drp1 exon 2 or Mfn2 exon 3 were cloned into the PX459 vector (Addgene)(Ran et al., 2013): Drp1 exon 2 upstream sgRNA: 5’ TGGAACGGTCACAGCTGCAC 3’; Drp1 exon 2 downstream sgRNA: 5’ TGGTCGCTGAGTTTGAGGCC 3’; Mfn2 upstream sgRNA: 5’ GTGGTATGACCAATCCCAGA 3’; Mfn2 downstream sgRNA: 5’ GGCCGGCCACTCTGCACCTT 3’. E14 ESCs were co-transfected with 1ug of each sgRNA expression using Lipofectamine 2000 (Invitrogen) according to manufacturer’s instructions. As control E14 ESCs were transfected in parallel with equal amount of empty PX459 plasmid. Following 6 days of Puromycin selection, single colonies were picked from both Drp1 sgRNA and empty vector transfected ESCs and screened for mutations. Drp1 exon 2 deletion was confirmed by PCR genotyping using the following primers: Drp1_genot F: 5’ GGATACCCCAAGATTTCTGGA 3’; Drp1_genot R: 5’ AGTCAGGTAATCGGGAGGAAA 3’, followed by Sanger Sequencing. Mfn2 exon 3 deletion was confirmed by PCR genotyping using the following primers: Mfn2_genot F: 5’ CAGCCCAGACATTGTTGCTTA 3’; Mfn2_genot R: 5’ AGCTGCCTCTCAGGAAATGAG 3’, followed by Sanger Sequencing.
Derivation of mESCs from hybrid mouse strains
The derivation of new mESC lines was adapted from (Czechanski et al., 2014). Heteroplasmic mESCs were derived from embryos of hybrid mouse strains BG, HB and ST. These contain the mtDNA of C57BL/6N (Bl6) lab mouse and mtDNA variants from wild-caught mice (Burgstaller et al., 2014).
Embryos were isolated at E2.5 (morula stage) and cultured in 4-well plates (Nunc, Thermo Scientific) containing KSOM media (Millipore) plus two inhibitors (KSOM+2i): 1 μM MEK inhibitor PDO325901 (Sigma-Aldrich) and 3 μM GSK-3 inhibitor CHIR9902 (Cayman Chemicals) for 2 days at 37°C in 5% CO2 incubator. To reduce evaporation, the area surrounding the wells was filled with DPBS. Embryos were further cultured in a fresh 4-well plates containing, N2B27+2i+LIF media: N2B27 media supplemented with 1 μM MEK inhibitor PDO325901 and 3 μM GSK-3 inhibitor and 0.1% LIF for up to 3 days until reaching the blastocyst stage. Each embryo was then transferred to a well of a 96-well plate coated with 0.1% gelatin in DPBS and containing 150 μL of N2B27+2i+LIF media per well. In these conditions, the embryos should attach to the wells allowing the epiblast to form an outgrowth. This plate was then incubated at 37°C in 5% CO2 incubator for 3 to 7 days until ES-like colonies start to develop from the epiblast outgrowth. Cells were passaged by dissociation with Accutase (Sigma) and seeded in gradual increasing surface area of growth (48-well, 24-well, 12-well plate, T12.5 and T25 flask), until new cell lines were established. At this stage cells were weaned from N2B27+2i+LIF media and then routinely cultured in ES media.
Animals
Mice were maintained and treated in accordance with the Home Office’s Animals (Scientific Procedures) Act 1986. All mice were housed on a 10 hr-14 hr light-dark cycle with access to water and food ad libitum. Mattings were generally set up in the afternoon. Noon of the day of finding a vaginal plug was designated embryonic day 0.5 (E0.5). Embryo dissection was performed at appropriate timepoints in M2 media (Sigma), using Dumont No.5 forceps (11251-10, FST). No distinction was made between male and female embryos during the analysis.
METHOD DETAILS
Embryo experiments
Early mouse embryos were isolated at E5.5 (from pregnant CD1 females, purchased from Charles River, UK). Following dissection from the decidua, embryos were cultured overnight in N2B27 “poor” media (same formulation as N2B27 media but supplemented with 0.5xB27 supplement and 0.5xN2 supplement) with pan-caspase inhibitors (100 μM, Z-VAD-FMK, FMK001, R&D Systems, USA) or equal volume of vehicle (DMSO) as control. On the next morning, embryos were processed for single cell RNA-Seq (scRNA-seq) or functional validation (Δψm analysis and immunohistochemistry for markers of loser cells).
For the scRNA-seq and Δψm analysis embryos were dissociated into singe-cells. Briefly, up to 12 embryos were dissociated in 600 μL Acccutase (A6964, Sigma, UK) during 12 min at 37°C, tapping the tube every two minutes. Accutase was then neutralised with equal volume of FCS, cells span down and stained with TMRM, for Δψm analysis, or directly re-suspended in 300 μL DPBS with 1% FCS, for single cell sorting and RNA-seq. Sytox blue (1:1000, S34857, ThermoFisher Scientific, UK), was used as viability staining.
Cell competition assays
Cell competition assays between wild-type and Bmpr1a-/-, 4n or Drp1-/- cells were performed in differentiating conditions. Cells were seeded onto fibronectin-coated plates (1:100, Merck) in DPBS during 1h at 37°C and grown in N2B27 media - to promote the differentiation of mESCs into a stage resembling the post-implantation epiblast, as cell competition was previously shown to occur in these conditions (Sancho et al., 2013). N2B27 media consisted of 1:1 Dulbecco’s modified eagle medium nutrient mixture (DMEM/F12) and Neurobasal supplemented with N2 (1x) and B27 (1x) supplements, 2 mM L-glutamine and 0.1 mM β-mercaptoethanol - all from Gibco. Cell competition assays between wild-type and Mfn2-/- and between mESCs with different mtDNA content were performed in pluripotency maintenance conditions (ES media).
Cells were seeded either separately or mixed for co-cultures at a 50:50 ratio, onto 12 well plates, at a density of 8E04 cells per well, except for assays between wild-type and Mfn2-/- mESCs, where 3.2E05 cells were seeded per well. The growth of cells was followed daily and compared between separate or co-culture, to control for cell intrinsic growth differences, until the fourth day of culture. Viable cells were counted daily using Vi-CELL XR Analyser (Beckman Coulter, USA), and proportions of each cell type in co-cultures were determined using LSR II Flow Cytometer (BD Bioscience), based on the fluorescent tag of the ubiquitously expressed GFP or TdTomato in one of the cell populations.
Metabolomic Analysis
The metabolic profile was obtained using the Metabolon Platform (Metabolon, Inc). Each sample consisted of 5 biological replicates. For each replicate, 1E07 cells were spun down and snap frozen in liquid nitrogen. Pellets from 5 independent experiments for each condition were analysed by Metabolon Inc by a combination of Ultrahigh Performance Liquid Chromatography-Tandem Mass Spectroscopy (UPLC-MS/MS) and Gas Chromatography-Mass Spectroscopy (GC-MS). Compounds were identified by comparison to library entries of purified standards based on the retention time/index (RI), mass to charge ratio (m/z), and chromatographic data (including MS/MS spectral data) on all molecules present in the library. Samples were normalized to protein content measured by Bradford assay. Statistical analysis was done using Welch’s two-sample t-test and statistical significance defined as p ≤0.05.
Analysis of mitochondrial membrane potential (Δψm)
Quantitative analysis of Δψm was performed by flow cytometry. Cells were grown in pluripotency or differentiating conditions, as described above. Cells were dissociated and pelleted to obtain 2E05 cells per sample for the staining procedure.
For TMRM staining in single cells from early mouse epiblasts, embryos were dissected at E5.5 and cultured overnight in the presence or absence of caspase inhibitors. On the following morning, to avoid misleading readings, epiblasts were isolated initially by an enzymatic treatment with of 2.5% pancreatin, 0.5% trypsin and 0.5% polyvinylpyrrolidone (PVP40) - all from Sigma-Aldrich- to remove the visceral endoderm (VE). Embryos were treated during 8 min at 4°C, followed by 2 min at RT. The VE was then pealed with the forceps and the extraembryonic ectoderm removed to isolate the epiblasts. Up to 16 epiblasts were pooled per 600µL of Accutase (Sigma-Aldrich) for dissociation into single cells prior to staining. Reaction was stopped with equal volume of FCS and cells subjected to TMRM staining.
Cells were loaded with 10 nM of the Nernstian probe tetramethylrhodamin methyl ester perchlorate (TMRM, Sigma), prepared in N2B27 media. After incubating for 15 min at 37°C, cells were pelleted again and re-suspended in flow cytometry (FC) buffer (3% FCS in DPBS). Sytox blue (1:1000, Invitrogen, UK) was used as viability staining. Stained cell suspensions were analysed in BD LSRII flow cytometer operated through FACSDiva software (Becton Dickinson Biosciences, UK). For TMRM fluorescence detection the yellow laser was adjusted for excitation at λ=562 nm, capturing the emission light at λ=585 nm for TMRM (Floryk and Houštěk, 1999; Scaduto and Grotyohann, 1999). In the case of GFP-labelled cell lines, for GFP fluorescence detection the blue laser was adjusted for excitation at λ=488 nm, capturing the emission light at λ=525 nm. Results were analysed in FlowJo vX10.0.7r2.
Qualitative analysis of Δψm was performed by confocal microscopy. Wild-type and Bmpr1a-/- cells were grown in fibronectin-coated glass coverslips. At the third day of differentiation, cells were loaded with 200 nM MitoTracker Red probe (Life Technologies), prepared in N2B27 media, for 15 min at 37°C. Cells were then washed with DPBS and fixed with 3.7% formaldehyde for subsequent immunocytochemical staining of total mitochondria mass, with TOMM20 antibody.
Immunofluorescence
Cells were washed with DPBS and fixed with 3.7% formaldehyde (Sigma, UK) in N2B27, for 15 min at 37°C. Permeabilization of the cell membranes was done with 0.4% Triton X-100 in DPBS (DPBS-Tx), at RT with agitation. Blocking step with 5% BSA in DPBS-Tx 0.1% was performed for 30 min, at RT with agitation. Mitochondria were labelled with TOMM20 antibody (1:100, Santa Cruz Biotechnologies). Dead cells were labelled with cleaved caspase-3 antibody (1:400, CST) and NANOG antibody was used to mark pluripotent cells (1:100, eBioscience). Secondary antibodies were Alexa Fluor (1:600, Invitrogen). Primary antibody incubation was performed overnight at 4°C and secondary antibody incubation during 45 min, together with Hoechst to stain nuclei (1:1000, ThermoScientific), at RT and protected from light. In both cases antibodies were diluted in blocking solution. Three 10 min washes with DPBS-Tx 0.1% were performed between each critical step and before mounting with Vectashield medium (Vector Laboratories).
Samples were imaged with a Zeiss LSM780 confocal microscope (Zeiss, UK) and processed with Fiji software (Schindelin et al., 2012). Mitochondria stainings were imaged with a 63x/1.4 Oil objective. For samples stained with TOMM20 antibody and MitoTracker Red, Z-stacks were acquired and processed for deconvolution using Huygens software (Scientific Volume Imaging, https://svi.nl/). Samples stained with cleaved caspase-3 were imaged with 20x/0.8 magnification objective. Imaging and deconvolution analysis were performed with the support and advice from Mr. Stephen Rothery from the Facility for Imaging by Light Microscopy (FILM) at Imperial College London.
Embryo immunofluorescent staining for p-rpS6, OPA1 and DDIT3 (CHOP) markers was performed as follows. Cultured embryos were fixed in 4% PFA in DPBS containing 0.01% Triton and 0.1% Tween 20 during 20 min at RT. Permeabilization of the membranes was done during 10 min in DPBS with 0.5% Triton. Embryos were blocked in 5% BSA in DPBS with 0.25% Triton during 45 min. Incubation with primary antibodies - CHOP (1:500, CST), OPA1 (1:100, BD Biosciences) and p-rpS6 (CST, UK) - was done overnight at 4°C in 2.5% BSA in DPBS with 0.125% Triton. On the following morning, hybridisation with secondary antibodies Alexa Fluor 568 and Alexa Fluor 488 (diluted 1:600 in DPBS with 2.5% BSA and 0.125% Triton) was done next during 1h at RT. Hoechst was also added to this mixture to stain nuclei (1:1000). Three 10 min washes with filtered DPBS-Tx 0.1% were performed between each critical step. All steps were done with gentle agitation.
Embryos were imaged in embryo dishes (Nunc) in a drop of Vectashield using Zeiss LSM780 confocal microscope at 40x magnification.
Western Blotting
Cells were washed in DPBS and lysed with Laemmli lysis buffer (0.05 M Tris-HCl at pH 6.8, 1% SDS, 10% glycerol, 0.1% β-mercaptoethanol in distilled water). Total protein quantification was done using BCA assay (Thermo Scientific, UK) and samples (15μg of protein per lane) were loaded into 12% Bis-Tris protein gels (BioRad). Resolved proteins were transferred into nitrocellulose membranes (GE Healthcare). The following primary antibodies were incubated overnight at 4°C: rabbit anti-TOMM20 (1:1000, CST-42406), rabbit anti-α-Tubulin (1:1000, CST-2144), mouse anti-mt-CO1 (1:2000, ab14705), rabbit anti-DRP1 (1:1000, CST-8570), mouse anti-MFN1 (ab57602), mouse anti-MFN2 (ab56889) and mouse anti-Vinculin (1:1000, Sigma V9131). On the following morning, HRP-conjugated secondary antibodies (Santa Cruz) were incubated for 1h at RT. Membranes were developed with ECL reagents (Promega) and mounted in cassette for time-time-controlled exposure to film (GE Healthcare).
Bulk RNA-Seq and Single cell RNA-Seq
For bulk RNA Seq in the competitive scenario between cells with different mtDNA, HB(24%) and BG(95%) mESCs were grown separately or in co-culture. On the third day of culture cells were dissociated and subjected to fluorescence activated cell sorting (FACS) to separate the cell populations in co-culture. To control for eventual transcriptional changes due to the FACS process, a mixture of the two separate populations was subjected to the same procedure as the co-cultured samples. Total RNA isolation was then carried out using RNA extraction Kit (RNeasy Mini Kit, Qiagen). PolyA selection/enrichment was the method adopted for library preparation, using the NEB Ultra II RNA Prep Kit. Single end 50bp libraries were sequenced on Illumina Hiseq 2500. Raw basecall files were converted to fastq files using Illumina’s bcl2fastq (version 2.1.7). Reads were aligned to mouse genome (mm9) using Tophat2 version 2.0.11 (Kim et al., 2013) with default parameters. Mapped reads that fell on genes were counted using featureCounts from Rsubread package (Liao et al., 2019). Generated count data were then used to identify differentially expressed genes using DESeq2 (Love et al., 2014). Genes with very low read counts were excluded. Finally, Gene Set Enrichment Analysis was performed using GSEA software (Mootha et al., 2003; Subramanian et al., 2005) on pre-ranked list generated by DESeq2.
To investigate the nature of cells eliminated by cell competition during early mouse embryogenesis by means of Single Cell RNA-Sequencing (scRNA-seq), early mouse embryos were dissected at E5.5 and cultured overnight in the presence or absence of caspase inhibitors. On the following morning, embryos were dissociated with Accutase and subjected to single-cell sorting into 384- well plate. Total RNA isolation was then carried out using a RNA extraction Kit (RNeasy Mini Kit, Qiagen). scRNA-seq was performed using the Smart-Seq2 illumina method. PolyA selection/enrichment with Ultra II Kit (NEB) was the method adopted for library preparation.
Data processing, quality control and normalization
We performed transcript quantification in our scRNA-seq data by running Salmon v0.8.2 (Patro et al., 2017) in the quasi-mapping-based mode. First, a transcriptome index was created from the mouse reference (version GRCm38.p4) and ERCC spike-in sequences. Then, the quantification step was carried out with the “quant” function, correcting for the sequence-specific biases (“-- seqBias” flag) and the fragment-level GC biases (“--gcBias” flag). Finally, the transcript level abundances were aggregated to gene level counts. On the resulting raw count matrix including 1,495 cells, we apply a quality control to exclude poor quality cells from downstream analyses.
For the quality control we used the following criteria: we identified the cells that have a log10 total number of reads equal to or greater than 4, a fraction of mapped reads equal to or greater than 0.8, a number of genes with expression level above 10 reads per million equal to or greater than 3000 and a fraction of reads mapped to endogenous genes equal to or greater than 0.5. This resulted in the selection of 723 cells, which were kept for downstream analyses. Transcripts per million (TPM) normalization (as estimated by Salmon) was used.
Identification of highly variable genes and dimensionality reduction
To identify highly variable genes (HVG), first we fitted a mean-total variance trend using the R function “trendVar” and then the variance was decomposed into biological and technical components with the R function “decomposeVar”; both functions are included in the package “scran” (version 1.6.9 (Lun et al., 2016)).
We considered HVGs those that have a biological component that is significantly greater than zero at a false discovery rate (Benjamini-Hochberg method) of 0.05. Then, we applied further filtering steps by keeping only genes that have an average expression greater to or equal than 10 TPM and are significantly correlated with one another (function “correlatePairs” in “scran” package, FDR<0.05). This yielded 1921 genes, which were used to calculate a distance matrix between cells defined as , where ρ is the Spearman’s correlation coefficient between cells. A 2D representation of the data was obtained with UMAP package (version 0.2.0.0 https://cran.r-project.org/web/packages/umap/index.html) using the distance matrix as input.
Cell clustering
To classify cells into different clusters, we ran hierarchical clustering on the distance matrix (see above; “hclust” function in R with ward.D2 aggregation method) followed by the dynamic hybrid cut algorithm (“cutreeDynamic” function in R package “dynamicTreeCut” (https://CRAN.R-project.org/package=dynamicTreeCut) version 1.63.1, with the hybrid method, a minimum cluster size of 35 cells and a “deepSplit” parameter equal to 0), which identified five clusters. Cells from different batches were well mixed across these five clusters (see Figure S1), suggesting that the batch effect was negligible.
Identification of a single-cell trajectory in the epiblast
We calculated a diffusion map (“DiffusionMap” function in the R package “destiny” version 2.6.2 (Angerer et al., 2016) on the distance defined above on the epiblast cells from CI-treated embryos. The pseudotime coordinate was computed with the “DPT” function with the root cell in the winner epiblast cluster (identified by the function “tips” in the “destiny” package). Such pseudotime coordinate can be interpreted as a “losing score” for all the epiblast cells from the CI-treated embryos.
We estimated the losing scores of the epiblast cells from DMSO-treated embryos by projecting such data onto the diffusion map previously calculated (function “dm_predict” in the destiny package). Finally, for each of the projected cells, we assigned the losing score as the average of the losing scores of the 10 closest neighbours in the original diffusion map (detected with the function “projection-dist” in the destiny package).
Differential gene expression analysis along the trajectory
To identify the genes that are differentially expressed along the trajectory, first we kept only genes that have more than 15 TPM in more than 10 cells (this list of genes is provided in Supplementary Table 4); then, we obtained the log-transformed expression levels of these genes (adding 1 as a pseudo-count to avoid infinities) as a function of the losing score and we fitted a generalized additive model to them (R function “gam” from “GAM” package version 1.16.). We used the ANOVA test for parametric effect provided by the gam function to estimate a p-value for each tested gene. This yielded a list of 5,311 differentially expressed genes (FDR < 0.01).
Next, we looked for groups of differentially expressed genes that share similar expression patterns along the trajectory. To this aim, similarly to what we did when clustering cells, we calculated a correlation-based distance matrix between genes, defined as , where ρ is the Spearman’s correlation coefficient between genes. Hierarchical clustering was then applied to this matrix (hclust function in R, with ward.D2 method) followed by the dynamic hybrid cut algorithm (dynamicTreeCut package) to define clusters (“cutreeDynamic” function in R with the hybrid method and a minimum cluster size of 100 genes and a deepSplit parameter equal to 0). This resulted in the definition of four clusters, three of genes that decrease along the trajectory (merged together for the GO enrichment and the IPA analysis) and one of increasing genes (Figure S2D). IPA (QIAGEN Inc., https://www.qiagenbio-informatics.com/products/ingenuity-pathway-analysis), was run on all genes differentially expressed (FDR < 0.01) along the trajectory from winner to loser cells (see Figures 2A-D and Figures 3A-C), using all the tested genes as a background (see Supplementary Table 4). This software generated networks, canonical pathways and functional analysis. The list of decreasing/increasing genes is provided in Supplementary Tables 1 and 2.
Analysis of Mitochondrial DNA heteroplasmy
We used STAR (version 2.7 (Dobin et al., 2013)) to align the transcriptome of the epiblast cells from CI-treated embryos (274) to the mouse reference genome (mm10). Only reads that uniquely mapped to the mitochondrial DNA (mtDNA) were considered. From these, we obtained allele counts at each mtDNA position with a Phred Quality Score greater than 33 using the samtools mpileup function.
Next, we applied filters to remove cells and mtDNA positions with a low coverage. First, we removed cells with fewer than 2,000 mtDNA positions covered by more than 50 reads. Second, we removed positions having less than 50 reads in more than 50% of cells in each of the three epiblast clusters (winner, intermediate and loser). These two filters resulted in 259 cells and 5,192 mtDNA positions being considered for further analyses.
Starting from these cells and positions, we applied an additional filter to keep only positions with a sufficiently high level of heteroplasmy. To this aim, for each position with more than 50 reads in a cell, we estimated the heteroplasmy as:
where fmax is the frequency of the most common allele. We kept only positions with H>0.01 in at least 10 cells.
Finally, using generalized additive models (see above), we identified the positions whose heteroplasmy H changes as a function of the cells’ losing score in a statistically significant way. We found a total of eleven significant positions (FDR < 0.001), six of them in the mt-Rnr1 gene and five in the mt-Rnr2 gene. All of these positions have a higher level of heteroplasmy in loser cells (see Figure 6B-G and Figure S5F-K). The results remain substantially unaltered if the Spearman’s rank correlation test (in alternative to the generalized additive models) is used.
For the barplot shown in Figure 6H and the correlation heatmaps in Figure 6I and S5L, we took into account only cells that covered with more than 50 reads all the significant positions in the mt-Rnr1 gene (215 cells, Figures 6H and 6I) or in both the mt-Rnr1 and mt-Rnr2 genes (214 cells, Figure S5L).
As a negative control, we repeated the analysis described above using the ERCC spike-ins added to each cell. As expected, none of the positions was statistically significant, which suggested that our procedure is robust against sequence errors introduced during PCR amplification.
Common features of scRNA-seq and bulk RNA-seq datasets
Differential expression analysis between the co-cultured winner HB(24%) and loser cell line BG(95%) was performed using the package EdgeR version 3.20.9 (Robinson et al., 2010). Batches were specified in the argument of the function model.matrix. We fitted a quasi-likelihood negative binomial generalized log-linear model (with the function glmQLFit) to the genes that were filtered by the function filterByExpr (with default parameter). These genes were used as background for the gene enrichment analysis.
We set a FDR of 0.001 as a threshold for significance. The enrichment analysis for both the scRNA-seq and bulk RNA-seq datasets were performed using the tool g:Profiler (Reimand et al., 2011). The list of up-regulated, down-regulated and background genes related to the DE analysis for the bulk RNA-seq dataset are provided in the Supplementary Tables 5, 6 and 7.
Quantification and Statistical Analysis
Flow cytometry data was analysed with FlowJo Software.
Western blot quantification was performed using Image Studio Lite (LI-COR). Protein expression levels were normalised to loading controls vinculin or α-tubulin.
The quantification of the DDIT3 and OPA1 expression in embryos was done by two distinct methods. DDIT3 expression was quantified by counting the number of epiblast cells with positive staining in the embryos of each group. The expression of OPA1 was quantified on Fiji software as the mean fluorescence across a 10 pixel width line drawn on the basal cytoplasm of each cell with high or low p-rpS6 fluorescence intensity, as specified in (Bowling et al., 2018). A min of 8 cells were quantified per condition (high vs low mTOR activity) in each embryo. Six embryos treated with CI were analysed. Mean grey values of OPA1 fluorescence for each epiblast cell are pooled on the same graph.
The statistical analysis of the results was performed using GraphPad Prism v8 Software (GraphPad Software, United States of America). Data was tested for normality using Shapiro-Wilk normality test. Parametric or non-parametric statistical tests were applied accordingly. Details about the test used in each of the experiments are specified in figure legends. Statistical significance was considered with a confidence interval of 0.05%. n.s., non-significant; * p<0.05; ** p<0.01;*** p<0.001.
Data and Code Availability
Data were analysed with standard programs and packages, as detailed above. Code is available on request. Raw as well as processed data are available through ArrayExpress, accession numbers E-MTAB-8640, for scRNA-seq data, and E-MTAB-8692, for bulk RNA-seq data.
Acknowledgments
We would like to thank Stephen Rothery for guidance and advice with confocal microscopy. Gratitude also goes to James Elliot and Bhavik Patel for performing cell sorts. Research in Tristan Rodriguez lab was supported by the MRC project grant (MR/N009371/1) and by the British Heart Foundation centre for research excellence. Ana Lima was funded by a BHF centre of excellence PhD studentship. Shankar Srinivas was funded through Wellcome awards 103788/Z/14/Z and 108438/Z/15/Z.