Prompted by enormous progress in single-cell technologies, this Editorial introduces the special issue ‘Single-cell analysis in development’ which aims to critically appraise the biological problems that may justify a single-cell approach (Fig. 1). Why just in development? This is because the early embryo is the only phase of the life cycle during which a totipotent programme operates (totipotency is the ability to give rise to the whole organism). Further, life has a hierarchical order in which each level of a biological structure builds on the level below it. We can consider the embryo as the elemental unit at the root of this order since we do not reproduce by half or quarter embryos, let alone by isolated blastomeres (the cells generated by the initial mitotic divisions of the zygote)—at least not naturally. Thus, the usefulness of a single-cell approach to embryos may not seem obvious. We will show that it is useful, provided some conditions are fulfilled.
According to the traditional way of thinking, the embryo is like the atom of classic (i.e. Newtonian) mechanics. Just as classical physicists considered particles interacting with each other by means of forces to be sufficient to describe the macroscopic world, biologists studying development viewed embryos with varying degrees of developmental competence to be sufficient to account for the reproductive success or failure of an organism. The discovery of subatomic particles confronted physicists with the fact that these particles operated by rules that radically differed from those they had known—namely, the rules of quantum mechanics. Seventy-two years ago, Nobel prize-winning physicist Erwin Schrödinger, in his famous book, ‘What is Life?’ (Schroedinger, 1944), envisioned that some aspects of embryology may bear similarity to the rules of quantum mechanics. For example, examining the subunits of the embryo may reveal biological evidence that was kept hidden from us by the low-resolution tools, we were using to study the intact embryo. For example, at the single-cell level, we can now capture the moment when a gene locus switches from transcribing one allele to transcribing the companion allele, resulting in a time gap in which mRNA synthesis is lacking. If we wanted to use this particular transcript to normalize the levels of gene expression, we would soon realize that dividing a number (e.g. mRNA amount) by zero is futile. Further, at the single-cell level, the presence of chromosomal aneuploidy (abnormal number of chromosomes in deficit or in excess of the complete set of chromosomes) can directly influence the gene expression measure because euploid companion cells are not there to compensate. If we wanted to discern whether a particular gene expression level is due to epigenetic regulation or template dosage, we would soon realize that we cannot give the answer unless we determine both the transcript amount and the karyotype of the single cell—a very challenging undertaking. These hypothetical experimental scenarios speak to the usefulness of single-cell approaches. However, single cells may not always be representative of the whole entity. For example, an embryo may still be capable of development even if some of its cells are aneuploid. Striking are the cases of healthy babies born from human embryos that would have been discarded based on the screening of single-cell biopsies for aneuploidy (Greco et al., 2015). These cases are related to the issue of mosaicism in early development. For a comprehensive appraisal of the issues associated with preimplantation genetic screening (PGS) and mosaicism, we refer the reader to the upcoming special issue of Molecular Human Reproduction on PGS (editors: Karen Sermon and Joep Geraedts). These issues point to the need for defining the questions worthy of single-cell approaches and developing new statistical algorithms tailored to single-cell analysis (Fig. 1).
We have committed this entire issue of Molecular Human Reproduction to the topic of ‘single-cell analysis in development’. We address in the following sections the field of single-cell analysis in general; and the open questions that are amenable to single-embryo or single-cell approaches in early development; we discuss the benefits, limitations and challenges of the single-cell approach; and we provide synopses of the articles. We are grateful to the five teams of authors, experts in this field, who have provided the articles that made this special issue possible. These teams—in alphabetical order, the Araúzo-Bravo team (Araúzo-Bravo and Gerovska), the Dovichi team (Dovichi, Sun, Champion and Huber), the Otu team (Otu, Yalcin and Hakguder), the Pantazis team (Pantazis, Welling and Ponti) and the Wessel team (Wessel and Brayboy)—provided articles that describe the single-cell analyses of blastomeres in early mammalian development and of oocytes in Xenopus. We are also grateful to the many colleagues who welcomed this initiative although they could not contribute. We are indebted to Chris Barratt for his encouragement and for trusting us to assemble this special issue.
The field of single-cell analysis: past and present
In spite of its seemingly recent development, single-cell analysis has been around for a long time. Early in the twentieth century, when pathologists stained cells of a tissue with dyes in order to identify the malignant cells, they were essentially trying to identify single cells by their morphological or biochemical properties. Solving this heterogeneity has been a major driver of single-cell analyses of adult tissues under pathological conditions; but in the twenty-first century, there is more to single-cell analysis than solving tissue heterogeneity, and there is also more to tissues than just somatic cells.
Unlike an adult organism, which comprises billions of cells with 200–300 different types, the distinct cells in an early embryo should be easier to tackle; they are relatively hard to miss among the few total cells, e.g. ∼100 embryonic cells in mammals and 1000 embryonic cells in Xenopus. Yet, scientists found their efforts to study embryos using single-cell approaches frustrated by limitations in the amount and accessibility of embryonic cellular material as well as by the lack of adequate tools and technologies available for the comprehensive investigation of the small numbers of cells obtainable from early mammalian embryos.
Large-scale molecular analyses of the mouse oocyte at the single-cell level occurred in the 1970s (Schultz and Wassarman, 1977). These pioneering studies examined proteins, a notable accomplishment because they predated the discovery of the PCR for nucleic acids. Since the advent of PCR, scores of researchers have applied the technique to single genes or small sets of genes as well as to single oocytes, single blastomeres and single embryos of mammalian species such as mouse and human (Gentile et al., 2004; May et al., 2009; Wong et al., 2010). The next advance was the ability to perform high-throughput single-cell DNA and RNA sequencing (RNA-seq), which was crowned ‘Method of the Year, 2013’ (2014). These technologies allow the dissection of heterogeneity at single-cell resolution instead of averaging out gene signals from a bulk of heterogeneous samples, which enables a better understanding of complex phenomena (see Fig. 1 for cases relevant to the embryo). Single-cell analyses can affect multiple areas of biomedicine; for example, next generation sequencing (NGS) in cancer research helps quantify changes in mutant allele frequencies during the clonal evolution of a cancer, allowing the detection and profiling of rare cancerous cells in the body of a patient. As of May 2015, the technology reached such a degree of penetration that thousands of single embryonic stem cells (ESCs) could be analysed for the NGS transcriptome (Klein et al., 2015; Macosko et al., 2015). In principle, these sequencing methods can also apply, pending adaptations, to the polar bodies (side products of meiosis) of oocytes and to the blastomeres of embryos (Xue et al., 2013; Yan et al., 2013; Deng et al., 2014), opening up the possibility of prioritizing oocytes for IVF or for transfer to the uterus following fertilization (article of Brayboy and Wessel).
Single-cell analysis does not stand alone but comes with microtools for picking individual cells and with dedicated algorithms to analyse the gene expression results. It becomes critical to efficiently select and manipulate single cells to achieve single-cell ‘omics’ (this neologism refers to biological fields of study ending in -omics, including genomics, transcriptomics, proteomics and metabolomics). This ability has boosted the use of microfluidics, as documented by the 2010 study of cell fate decisions in mouse development at single-cell resolution (Guo et al., 2010). The next step involves extracting the information from a tiny amount of biomass without compromising the raw material's integrity. While microarray tools and data have been around for about a decade, they can only detect sequences homologous to the probes that correspond to a predefined gene set, making the technology unsuitable for discovering new genes. In addition, isoforms are difficult to resolve using microarrays. In contrast, NGS high-throughput datasets are neither limited to a predefined set of probes nor blind to isoforms. It may be noted that the former limitation also can be overcome by other approaches, e.g. analysis of expressed sequence tags (Evsikov et al., 2006).
The combination of single-cell analysis with sequencing methods can generate staggering amounts of data, requiring more sophisticated bioinformatics methods than those used for microarrays, for example. As a consequence, the balance between technical possibilities and biological questions, which previously leaned to the side of the questions, has now shifted. Sometimes technological innovation is deployed and made available to scientists at such a rate that publications resemble the outcome of having tools available without necessarily a problem to solve. Current techniques keep increasing their degree of resolution to a point where scientific progress is only restricted by the capacity of each researcher to interrogate the data in the most meaningful way.
Technical challenges remain when trying to analyse some ‘omics’ at the single-cell level. For example, revealing the presence of phenotypically relevant gene products—the proteins, not the mRNAs—is still challenging at the single-cell level. Single-cell western blots and single-cell liquid chromatography–mass spectrometry (LC-MS/MS) remain exceptions (Hughes et al., 2014; Smits et al., 2014; Sun et al., 2014). Further, although tedious at the single-cell level and unaccomplished to date on the whole-genome scale, we can analyse epigenetic modifications to chromatin such as DNA methylation and hydroxymethylation, histone acetylation, methylation and phosphorylation (Qiao et al., 2010; Chavez et al., 2014). For example, the isolated blastomeres of 8-cell mouse embryos were probed for DNA-methylation errors at six loci—H19 (imprinted non-protein coding maternally expressed transcript), IG-DMR (intergenic differentially methylated region), Igf2r (insulin-like growth factor 2 receptor), Snrpn (small nuclear ribonucleoprotein N), Peg3 (paternally expressed gene 3), Nnat (neuronatin) (Lorthongpanich et al., 2013); the results showed that the embryos were ‘epigenetic chimeras’ and that this state was exacerbated in Trim28 (tripartite motif-containing 28) maternal mutants. This is just one example of how technical advances expose exciting biological questions. Rather than merely describing heterogeneity in embryos, we may seek to accurately define the moment when the sister blastomeres start to diversify and elucidate how this happens at different levels of the gene expression cascade (e.g. DNA, mRNA and protein). Irrespective of excitement, the answers will require validation with independent approaches so as to improve reproducibility—an issue that not many studies have mentioned, with few laudable exceptions (Blakeley et al., 2015).
Open questions amenable to a single-embryo or single-cell approach
We have already remarked on the technologically driven nature of the single-cell analysis field. In fact, several open issues that remain in developmental biology can further guide this technological progress.
Open questions about the zygote and the initiation of totipotent development
The chromosomes of oocyte and sperm arrive in the zygote in totally different chromatin states. They undergo epigenetic restructuring before embryonic gene transcription commences. Two fundamental biological questions are front and centre: first, how are early epigenetic and transcriptional events initiated and molecularly coupled and, secondly, how do these events influence the course of cleavage, which leads to the founder cell lineages of the blastocyst? The influence of oocytoplasm on embryo development is largely (but not completely) lost with embryonic genome activation (EGA), which marks the transition into a new gene expression pattern without the meiotic features of the oocyte (Schultz, 2002). Even within the same species, oocytes can vary substantially in their ability to support EGA (article of Brayboy and Wessel). The preparation for EGA requires exchanging and remodelling the chromatin components of the gametes and making massive histone changes and DNA modifications on both sets of parental chromosomes to shape the chromosomal landscape for totipotent development (Posfai et al., 2012). Chromatin mobility surely plays an important role in defining the properties of cells in general, including the propensity of blastomeres to embark (and stay) on trajectories of differentiation (article of Welling, Ponti and Pantazis). While conserved developmental principles (such as the accelerated degradation of maternal transcripts upon oocyte fertilization) are becoming increasingly clear, the transition to an embryonic pattern of gene expression spans multiple cell cycles whose number varies across species, as shown in mice (Hamatani et al., 2004; Evsikov et al., 2006) and in humans (Bermudez et al., 2004; Dobson et al., 2004; Wong et al., 2010). The molecular nature and complexity of upstream regulatory factors remain unresolved. They certainly include protein complexes, such as Trithorax group proteins, present during late oogenesis (Andreu-Vieyra et al., 2010) and riboprotein complexes, known as maternal effect structures (Yurttas et al., 2010), which store maternal effect proteins or protein-encoding mRNAs until they need to be released for action (Li et al., 2010).
In addition to maternal factors, it has become increasingly apparent that the de novo synthesis of factors not initially present in the oocyte, such as TET1 and TET2 (Ten-eleven translocation methylcytosine dioxygenases 1 and 2), NANOG and CDX2 (caudal type homeobox 2) proteins, contributes to general epigenomic and specific transcriptomic programming. These factors operate in close association with maternally provided factors, such as DPPA3 (developmental pluripotency-associated 3; also known as PGC7/STELLA), TET3 (Ten-eleven translocation methylcytosine dioxygenase 3), and the transcription factor OCT4 (OCTAMER 4), the product of the Pou5f1 gene (Ovitt and Schoeler, 1998; Pesce et al., 1998). OCT4, central in controlling stem cell pluripotency (ability to give rise to endodermal, mesodermal and ectodermal derivatives but not extraembryonic lineages), is present throughout the female germ line (primordial germ cells, gonia, oocytes and preimplantation embryos). It has long been a question how OCT4 exerts its role in early embryonic development. OCT4 probably is not the most upstream key regulator of pluri/totipotency, and certainly it does not work alone. Biologists thought that mouse embryos lacking the Oct4 locus could not initiate the formation of the founder cell lineages (Foygel et al., 2008), but newer evidence challenges such a notion (Frum et al., 2013; Wu et al., 2013). Also quite unexpected is the discovery that the expression of Oct4 varies significantly between species, whereby the lesson taught by mice may or may not apply to other vertebrates, e.g. humans (Cauffman et al., 2009). Like that of Oct4, the expression of Cdx2 also varies; CDX2 was detected in the mouse morula prior to cavitation while in the human it appeared after cavitation (Niakan and Eggan, 2013). Regardless of how OCT4 may operate, it may be mediated by a more or less dynamic and receptive state of chromatin (article of Welling, Ponti and Pantazis). The ability to profile, on a genome-wide scale, the expression of genes in blastomeres during the early stages of differentiation will help to unravel the network of molecular regulators that drive the activation of early embryonic transcriptional programmes (article of Gerovska and Araúzo-Bravo). Deep mRNA sequencing of single oocytes, zygotes and blastomeres has begun to unravel the contribution and developmental timing of further gene products in the coordination of EGA (Tang et al., 2011; Tan et al., 2013).
Open questions about zygotic cleavage and earliest signs of daughter cell diversification
We can use a very limited repertoire of markers to spot differences as they start to emerge among the blastomeres of an embryo (Fig. 1). We can also use a very limited repertoire of markers to unambiguously distinguish an embryo from an oocyte. One proposal suggests that the axis of cell cleavage (meridian or equatorial, as opposed to symmetric or asymmetric) correlates strongly with the fate of blastomere progeny in the blastocyst, but other researchers still seek upstream regulators. The point at which the sperm enters the oocyte has been suggested to determine the position of the first division plane in the mouse zygote (Piotrowska and Zernicka-Goetz, 2001). Not all scientists share this deterministic view, which suggests similarities to developmental principles in other vertebrates, such as Xenopus (Hiiragi and Solter, 2004). Observations of the heterogeneous expression of crucial transcription factor (e.g. Oct4, Nanog) and epigenetic regulator genes (e.g. Carm1, coactivator-associated histone-3-specific arginine methyltransferase 1; see below) within the cleavage-stage embryo have raised the intriguing possibility that a stochastic component is involved in switching from symmetric to asymmetric cell divisions, setting the stage for lineage divergence at later time points in developing blastomeres. Research has suggested that the subcellular protein dynamics of OCT4 play a role in the symmetric or asymmetric division of the early blastomeres (Plachta et al., 2011). However, this role has been challenged by the recent observation that mouse zygotes lacking the maternal supply of the long A isoform of OCT4 can still form blastocysts that comprise an inner cell mass (ICM) and trophectoderm (TE) (Wu et al., 2013).
Chromatin changes may play a role in instructing blastomere identity. The level of histone 3 monomethylation on arginine 26 (H3R26me1) distinguishes mouse blastomeres at the 4-cell stage (Torres-Padilla et al., 2007). This observation correlates with the order of the second cell division (equatorial, meridian) of the 2-cell blastomeres. Along these lines, one group proposed that the cleavage order of 2-cell mouse blastomeres may predetermine the formation of cell progenies for the ICM or TE in the blastocyst. Consistent with this logic, the overexpression of Carm1 appears to promote the formation of ICM programmes in individual blastomeres (Torres-Padilla et al., 2007). The design of these experiments typically involves microinjecting one blastomere each of two distinct 2-cell embryos with either test (e.g. Carm1) or control mRNA. Hence, the experimental unit is the embryo (i.e. the experiment uses two distinct embryos), although one would like to learn about the individual sister blastomeres of the same embryo. It is desirable that future studies give more importance to the experimental design, and that both blastomeres of the same embryo are subjected to treatment. Failure to do so introduces a variable: the shape of the zona pellucida (extracellular glycoprotein coat) may vary from embryo to embryo (Motosugi et al., 2005).
In addition, paternal and maternal chromosomes of the mouse embryo remain differently marked (e.g. by dimethylation of lysine 9 on histone H3, H3H9me2) until the 4-cell stage, when they occupy distinct interphase nucleus compartments (Lepikhov et al., 2008). Whether such molecular differences truly cause or instead result from lineage commitment remains an open question. Various observations argue against an early commitment of cells in the developing embryo and rather favour a self-generated yet enigmatic rise of cell identity (Chazaud et al., 2006; Dietrich and Hiiragi, 2007). The aspect of different pronuclear marking according to the parent of origin also has implications for the science of somatic cell nuclear transfer, since the two pseudo-pronuclei originating from the somatic nucleus could not be marked differently, unlike the pronuclei that experienced spermatogenesis and oogenesis.
A deeper knowledge of the molecular dynamics that accompany the switch of cells in the early embryo from undetermined to determined states will be an asset for molecular embryology in the coming years. The last few years have seen the development of novel (e.g. less invasive) microscopic approaches that address such dynamic processes more systematically, as shown for the spindle formation changes in the early embryo (Courtois et al., 2012) and as described in this special issue (article of Welling, Ponti and Pantazis).
Open questions about the rise and fixation of diversity among blastomeres
How the developmental histories and trajectories of blastomeres result in a blastocyst with an epithelium-like TE that encases a clump-like ICM remains unclear. Understanding how and when interblastomere differences arise and reach a ‘point of no return’ has, however, become critically important (article of Gerovska and Araúzo-Bravo). Classic approaches aimed at exposing any developmental diversity were based on studying the cell progeny and the developmental potential of blastomeres engrafted in chimeric embryos or examined individually after embryo splitting. Chimeras are embryos that are composed of genetically distinct cells. In the mouse, classic chimera experiments showed that some blastomeres are, if not totipotent, at least broadly potent up to the 8- (Kelly, 1977) or even 16-cell stage (Tarkowski and Wroblewska, 1967; Rossant, 1976; Tarkowski et al., 2010). However, these blastomeres are not identical, since they express differentiation preferences (Piotrowska-Nitsche et al., 2005). In humans, Edwards has suggested that the blastomeres of the 4-cell embryo are distinguishable at the mRNA level (Edwards and Hansis, 2005), though Galan and colleagues could not confirm this finding (Galan et al., 2010).
When performing single-blastomere analysis, it is difficult to know if ‘departures’ from the normal pattern of gene expression are due to regulation as opposed to reflecting the presence of more or less gene template (aneuploidy) (Fig. 1). The incidence of aneuploidy in human embryos at the cleavage stage can be astounding. Up to 80% of embryos comprise at least some aneuploid cells; that is, 20% of the embryos are euploid in all cells, as suggested by different studies using different approaches, e.g. comparative genome hybridization, sequencing and karyotyping (Vanneste et al., 2009; Johnson et al., 2010; Chavez et al., 2012; Chow et al., 2014; Huang et al., 2014; Wang et al., 2014; Kung et al., 2015; Lukaszuk et al., 2015). It is remarkable that such embryos exist and that the human species ‘entrusts’ them with its evolutionary future; indeed, it has been suggested that the aneuploid embryos are purged at or prior to implantation, accounting for the low fertility of the human species (Macklon et al., 2002; McCoy et al., 2015). The aetiology of aneuploidy, at least in humans, deserves further investigation. Recent molecular studies are already going in this direction; for instance, as long as the fertilized oocyte has not cleaved (first cell cycle after fertilization), the use of transcriptomic signatures can predict embryo ploidy (Vera-Rodriguez et al., 2015). However, it is not clear how embryo (aneu)ploidy contributes to the interblastomere differences of gene expression revealed by single-cell studies. The analysis of transcriptomic signatures may also shed light on the outcomes of classical embryo-splitting studies. In bovine animals, embryo-splitting experiments resulted in one set of monozygotic quadruplets produced from a split 4-cell embryo (Johnson et al., 1995). Embryo bisection yielded full-term twins from the 2-cell stage in sheep (Willadsen, 1979) and in mice (Tsunoda and McLaren, 1983; Togashi et al., 1987; Wang et al., 1997; Morris et al., 2012). Among primates, ‘Tetra’, the Rhesus monkey reproduced by embryo splitting (Chan et al., 2000), proved that one blastomere of the donor's 4-cell embryo was totipotent but could not offer any information about the other three blastomeres. In addition to classical tools, such as chimeras, modern tools have become available to investigate the similarities and differences among blastomeres. By performing RNA-seq, Tang et al. (2011) showed that allele-specific expression could distinguish two blastomeres, although the imbalances are minute. Regardless of how minute the imbalances are, the prevalence of this phenomenon seems to be quite high (Xue et al., 2013; Deng et al., 2014).
In sum, these observations do not conclusively show whether sister blastomeres of the same early embryo are similar to or different from each other beyond very small variances, which are difficult to resolve from the influence of cell cycle asynchrony, aneuploidy and so on. It seems puzzling that a single blastomere may form a whole organism after embryo splitting, whereas removing a single blastomere, as occurs in PGS, can perturb the phenotypic traits in the adult (Yu et al., 2009). Perhaps the heterogeneity of changes in molecular markers observed for different mammalian species (Oron and Ivanova, 2012) speaks to the lack of a universal early point of no return in mammalian development. The molecular signatures that define such a point in the developmental trajectory of blastomeres remain cryptic, warranting more intense and focused research.
Open questions about the multiple levels of the gene expression cascade that controls cell identity
The above observations show that understanding the regulation of early embryogenesis requires an integrated view of the molecular mechanisms; by themselves, the transcriptome, the proteome and the epigenome cannot sufficiently explain the establishment of specific regulatory networks. While certainly very informative, transcriptional analyses (Robert, 2010) have a downside: changes occurring at the mRNA level of the gene expression cascade do not necessarily produce changes at the protein level. The transcript and protein changes in cells analysed to date have rather poor concordance: 50% on average (Ghazalpour et al., 2011; Schwanhäusser et al., 2011). This implies that blastomeres indistinguishable on the mRNA level may, nevertheless, be poised to suffer different fates (e.g. by protein-mediated epigenetic control, such as DNA methylation and hydroxymethylation of upstream gene control regions). Conversely, differences on the mRNA level may be offset during post-transcriptional events and might not reach the protein level.
Poor mRNA-protein concordance has multiple origins, including, but not limited to: Because of these variables, it is essential that we increase our understanding of the connections among transcriptional regulation, RNA storage, RNA processing, protein translation/processing and the concerted regulation of all these steps by oocytoplasmic and de novo synthesized proteins during early cleavage stages. Technologies to perform comparative proteomics on smaller and smaller sample sizes are on the horizon (see ‘Recent technological advances’ in Vogel and Marcotte, 2012) and can be applied to single oocytes and embryos in Xenopus (article of Sun, Champion, Huber and Dovichi), whereas mammalian oocytes still require cell numbers in the range of hundreds of oocytes and embryos (Schwarzer et al., 2014). Additional regulatory information will come from comparative epigenomic studies. Last but not least, the discovery of large amounts of novel oxidized forms of 5-methylcytosine during early embryogenesis demonstrates the importance of epigenetic changes during early embryogenesis (Wossidlo et al., 2011). The development of novel molecular profiling methods, such as hairpin-bisulfite sequencing (Arand et al., 2012) and oxidative-bisulfite sequencing (Booth et al., 2012), allow the study of molecular changes at the level of single DNA molecules during early embryogenesis to define the impact of these new and important DNA modification types on allelic methylation, gene expression, blastomere development and cell fate decisions during the first cleavages.
the influence of abundant non-protein-coding RNAs, such as small nuclear RNAs, small nucleolar RNAs and long non-coding RNAs, which regulate transcript conversion and export from the nuclear envelope, or of microRNAs for transcript processing by the RNA-induced silencing complex (Jeffries et al., 2011);
degree of recruitment of mRNAs to polyribosomes for translation (Potireddy et al., 2006);
the different intrinsic stability and degradation rates of proteins (Doherty et al., 2009);
the different subcellular localization of proteins, such as TEAD4 (TEA domain family member 4) (Home et al., 2012).
Open questions about blastocyst formation and preparation for implantation and post-implantation development
As described above, a pair of long-standing questions in embryology are, first, how do the blastomeres make the decision to become the ICM or the TE, and secondly, how do the cells of the ICM differentiate into either the primitive ectoderm (pEct, also known as the epiblast) or the primitive endoderm (pEnd, also known as the hypoblast), as opposed to the cells of the TE, which contribute to the placenta? Lineage commitment is commonly defined by the position of the cells within the embryo and by the expression of marker genes in time and space: for example, Oct4 in the ICM and Cdx2 in the TE. The gene expression profile of 8-cell stage mouse blastomeres is more similar to the TE than the ICM (Guo et al., 2010; Tang et al., 2011); in the absence of cell–cell contacts, the blastomeres seem to acquire a pattern of gene expression characteristic of neither the ICM nor TE, yet somewhat closer to the TE (Lorthongpanich et al., 2012). Taken together, these observations suggest that the embryo's default pathway is primarily devoted to the generation of extraembryonic tissues and secondarily to the generation of the embryonic ones. Up to the 32-cell stage, at least some blastomeres coexpress marker genes characteristic of multiple lineages; the identity of these blastomeres has become more defined, but has not yet become fixed. At the early blastocyst stage, in mice as well as in bovines, the ICM cells have a ‘salt and pepper’ distribution of NANOG and GATA6 (GATA binding protein 6) (Chazaud et al., 2006; Kuijk et al., 2008, 2012). Later on, the GATA6-positive cells relocate to the surface of the ICM to form the pEnd (Chazaud et al., 2006) and the NANOG-positive cells form the pEct (Ralston and Rossant, 2010). The mechanisms involved in these topological changes are currently unknown, although candidate genes exist (e.g. Dab2, disabled 2; Yang et al., 2002). Similar information is largely lacking for the human embryo. An RNA-seq study that analysed individual human embryos produced with the sperm of one father covered all preimplantation stages except the blastocyst (Xue et al., 2013).
As the ICM differentiates into pEct and pEnd, findings based on Nanog gene expression appear to contradict the paradigm accepted previously whereby the pEnd sends instructive signals to the pEct (Nichols et al., 2009; Messerschmidt and Kemler, 2010). Instead, these findings suggest that the relationship between pEct and pEnd remains an unsolved question of cell autonomy versus non-cell-autonomy in the commitment of blastomeres to cell lineages. One model suggests that the ICM contains a mix of naïve cells whose positions decide subsequent cell fate, while another proposes that an earlier cell divisional history autonomously causes pEct/pEnd lineage specification. In part, the situation is confounded by the use of different genetic strains of mice (Morris, 2011; Yamanaka, 2011). Increasingly, researchers have turned their attention to the possible role of monoallelic versus biallelic gene expression as an early regulatory event for cell fate determination. Reportedly, Nanog mRNA is produced monoallelically during mouse embryo cleavage, biallelically in the epiblast, and then monoallelically again in ESCs (Miyanari and Torres-Padilla, 2012). Although mouse embryos that have one instead of two Nanog alleles are viable and normal (Mitsui et al., 2003), they lead to a cell deficit in the blastocyst (Miyanari and Torres-Padilla, 2012). Two recent studies have called the monoallelic Nanog expression in ESCs into question (Faddah et al., 2013; Filipczyk et al., 2013), without, however, examining the embryo. These contradictory studies show that drawing any conclusion about the embryo based on ESC data requires an exercise in caution. Moreover, allelic bias or allele-specific expression of Nanog has been documented in mouse and human embryos using RNA-seq (Tang et al., 2011; Xue et al., 2013). The RNA-seq data suggest that monoallelic versus biallelic gene expression is a real feature of the early embryo, but its functional significance warrants further investigation. In particular, it remains to be seen if the allelic fluctuations of mRNA are coupled with protein fluctuations. In summary, we need further functional analyses of single blastomeres (e.g. combining single-cell transcriptomics with fluorescence marker protein tracing) to unravel the mechanisms and molecular decisions that define the commitment of cells into pEnd and pEct.
Benefits, limitations and challenges of the single-cell approach
Given the technical feasibility of single-cell analysis, the articles in this issue consider the value of analysing single cells in development, e.g. single oocytes or single blastomeres. Clearly, as these articles show, we do not perform single-cell analysis on embryos merely to characterize heterogeneity in a cell population. Altschuler and Wu remarked that ‘while ensemble measurements may be too simplistic, capturing all variation among cells may be unnecessary. […] A general challenge is to determine which details give rise to biologically-meaningful distinctions within and among populations, and which can be ignored (i.e. when an ensemble average is justified)’ (Altschuler and Wu, 2010). For one thing, ‘interesting’ heterogeneity resulting from regulative/epigenetic processes occurs alongside ‘uninteresting’ sources of heterogeneity, e.g. (un)equal partition during cytokinesis, cell cycle asynchrony and aneuploidy (Vera-Rodriguez et al., 2015; Fig. 1). The physicist Walter Elsasser (in his often-cited article ‘Outline of a theory of cellular heterogeneity’) argued that, given the immense number of atoms in a cell, the chance of their becoming equally partitioned upon cell division and then combining into exactly the same bonding possibilities is inconceivably small (Elsasser, 1984). Thus, some (quantitative) diversity is bound to happen. Even if the origin of diversity is trivial, diversity may nonetheless exert beneficial roles in the survival or adaptation of cells to changing environmental conditions.
One way to combat these ‘uninteresting’ sources of heterogeneity is to perform many more replicates than pooled-cell studies would require. In light of the ‘uninteresting’ heterogeneity, we propose the use of single-cell analysis to determine the earliest moment at which cells of the same embryo start to diversify. This information might point out embryo markers of clinical relevance; knowledge of this moment would probably mean that the substrate—the chromatin—is undergoing modification concurrently. As these (mostly) destructive analyses are restricted in human embryos, the first attempts will use animal models.
Whenever possible, single-cell data should be pursued using minimally invasive approaches, as opposed to physically removing cells from an embryo; this will facilitate knowledge transfer to the clinic. Among minimally invasive approaches, live-cell imaging takes into account the idea that cell decisions get made by communities of cells and not by the cells in isolation. Lorthongopanich and colleagues showed that blastomeres forced to remain single cells (by repeated splitting) do not form the gene expression patterns seen in cells belonging to a community (Lorthongpanich et al., 2012). Furthermore, since live-cell imaging can make use of proteins such as fluorochrome reporters (e.g. green fluorescent protein: GFP), thereby including the phase of post-mRNA regulation, it can, in part, rescue the poor correlation observed between the transcriptomes and proteomes of oocytes. Clearly, the more complex gene expression landscape of single cells adds to the difficulty of analysis, since effects that would be averaged out in pools of cells are, instead, revealed in many more (and more varied) facets. Thus, single-cell approaches coupled with the various ‘omics’ generate staggering amounts of data. These datasets require a special analytical effort to extract and interpret. Even in the face of computers becoming far faster and capable of handling far more data, we will face an inevitable trade-off among ever increasing volumes of data, analytical capabilities and an incremental gain of knowledge (article of Yalcin, Hakguder and Otu).
The question, therefore, becomes whether more resolution is always better. Unlike technical noise, which single cells exhibit less of than pools of cells, biological systems have some stochasticity (e.g. the pulsed nature of transcription), which is magnified at the single-cell level. Thus, promoter transcription generates pulses of mRNA; at a particular instant, the cell may seem to have stopped transcribing the particular mRNA of interest. This kind of resolution seems to provide more trouble than benefit, because the mRNA would most likely start transcribing again a fraction of a second later, and this short hiatus would make no difference to the cell's phenotype. The issue of the pulsed nature of transcription might be less consequential for proteins because proteins have longer half-lives, on average, than mRNAs and can more easily bridge the hiatus. Like the pulsed nature of gene expression, the cell cycle dependency of gene expression becomes more of an issue in single-cell analyses than in pooled-cell analyses.
Last but not least, single-cell analysis in development faces confounding factors specific to oocytes and embryos that are not encountered in any other type of cell. The shape and curvature of the zona pellucida can influence the cleavage planes of sister blastomeres irrespective of upstream changes of gene expression. The mechanical constraints within the early embryo were studied in a 3D model by Honda et al. (2008) and by Krupinski et al. (2011), who added proliferation estimates and a ‘simple’ gene regulatory network in order to explain the mosaic expression pattern of some pluripotency factors observed earlier (Chazaud et al., 2006; Dietrich and Hiiragi, 2007). From a computational modelling perspective, an urgent need exists to reconcile and refine the existing models with quantitative data—generated, for example, by measuring mRNA and protein expression time series in blastomeres—based on the development and application of bioimage informatics tools such as 3D tracking, segmentation and quantification (Wennekamp et al., 2013). Buettner and Theis reanalysed the dataset of Guo and colleagues (2010) by means of a novel data reduction approach, which considers the temporal structure of expression datasets by identifying a subpopulation of blastomeres already in the 16-cell stage (Buettner and Theis, 2012). Determining the timing of even earlier distinctions leading to subpopulations and their determinants is a standing challenge of single-cell analysis.
Ultimately, the oocyte might prove to drive heterogeneity, as Brayboy and Wessel propose. This fits with the observation by Torres-Padilla that differences in histone methylation patterns become apparent as early as the 2-cell stage in mouse development and relate to the pattern of blastomere cleavage to the 4-cell stage (Torres-Padilla et al., 2007). If indeed the oocyte turns out to drive heterogeneity, then it could be meaningful to analyse the variance (of, for example, gene expression) to see if the part of variance (V) within an embryo is larger than the part of variance among embryos. If Vintra is >Vinter, then it would become useful to study how this diversity arises during mitosis. If Vintra is <Vinter, then it would mean that differences in oocyte composition carry more weight and that any differences we detect between blastomeres are probably less important than the differences between oocytes (Shaw et al., 2013). Gap junctions may be missing links that promote cell autonomy and, hence, diversification among the early blastomeres, as proposed by Brison and colleagues (Brison et al., 2014).
One future task will involve integrating at least subsets of data from various sources (e.g. transcriptome, proteome and epigenome) and/or from multiple species, provided that the levels of abstraction required for comprehensive interspecies comparison (gene ontology categories and pathways) become more efficient and public access to the datasets improves. Another task will involve determining a better definition of the experimental unit in single-cell studies. The experimental unit may be defined as the smallest division of the experimental material such that any two experimental units can receive different treatments. Currently, treating distinct cells within one embryo in different ways is difficult. This might be accomplished by triggering recombination events in one cell but not in its neighbour with two-photon microscopy (via e.g. photoactivatable recombinase; Schindler et al., 2015) or by producing primary chimeras in which some cells but not others in the same embryo can respond to the experimental cue (e.g. adding doxy and dexa to the culture medium). Thus, in most cases, the experimental unit is the undivided embryo, even if we can perform single-blastomere analysis. What develops is the embryo, not its blastomeres in isolation, as shown by the difficulty in achieving totipotency (full development) after embryo splitting. It remains worthwhile to resolve the heterogeneity within the embryo, being aware that blastomeres can become heterogeneous due to: epigenetic regulation mediated, at least in part, by maternal factors inherited from the oocyte; to cleavage asynchrony; or to accidents (e.g. chromosome malsegregation).
All in all, we propose that the real power of single-cell analysis comes from its ability to accurately date the initiation of developmental events in order to pursue the epigenetic/chromatin markers that prepare the ground for, or seal the transition from, a naïve or totipotent state to a state of more advanced differentiation. Whether or not these events come to completion once started is a question that single-cell analysis cannot answer unless the cell is tracked microscopically in a live-cell setting. In this regard, one future challenge involves developing fluorochromes and dyes that afford greater sensitivity with improved spatiotemporal resolution.
Synopses of the articles of this special issue
Article of Gerovska and Araúzo-Bravo: Does mouse primordial germ cell activation start before preimplantation as suggested by single-cell transcriptomics dynamics?
Gerovska and Araúzo-Bravo show that genes expressed at very low levels warrant a single-cell analysis and that analysing time series data on a single-cell basis can time the start of some molecular events to an earlier point than pooled-cell analysis. The authors integrated microarray transcriptomic data available from public databases, comparing single-cell datasets with pooled-cell datasets among embryonic Days 3.25, 3.5 and 4.5. Results uncover new candidate markers of pEct, pEnd and TE, and suggest a regulation of Dnmt3l (DNA (cytosine-5-)-methyltransferase 3-like) by TCFAP2C (transcription factor AP-2 gamma) for pEnd and pEct specification in the mouse preimplantation embryo. Apparently, the most variably expressed genes in the context of the primitive endoderm-epiblast transitions are Fgf4 (fibroblast growth factor 4), Tcfpa2c and Dnmt3l. The expression of Fgf4 varies to such an extent that its signal can come across in a pooled-cell analysis, while pools of cells would mask the variability of the other two genes. It appears that primordial germ cell markers, such as Ifitm3 (interferon induced transmembrane protein 3, also known as Fragilis) and 4930432K21Rik mRNA, which go undetected in pooled-cell analysis, can be detected in single blastomeres, although primordial germ cells are supposed to arise after blastocyst implantation.
Article of Welling, Ponti and Pantazis: Symmetry breaking in the early mammalian embryo: the case for quantitative single-cell imaging analysis
Welling and colleagues demonstrate the use of transcription factor kinetics on live mouse embryos to identify very early differences between blastomeres, before morphological signs of differentiation become apparent. After a comprehensive review of the tools available to perform quantitative single-cell imaging analysis (e.g. Brainbow, PhOTO), the authors describe the properties of photoactivatable GFP reporter driven by the Oct4 gene promoter (OCT4-GFP). Mouse blastomeres can be sorted into two groups that differ by the rates of photoactivated OCT4-GFP kinetics: in one group, blastomeres are less able to retain OCT4 in the nucleus and tend to divide symmetrically (both daughter cells outside); in the other group, blastomeres can better retain OCT4 in the nucleus and tend to divide asymmetrically (one daughter cell inside, one daughter cell outside). One explanation for this behaviour is accessibility to binding sites, which is mediated by chromatin mobility, among other things. Thus, the authors point at one possible upstream event that could regulate blastomere potency, namely a receptive state of chromatin.
Article of Brayboy and Wessel: The double-edged sword of the mammalian oocyte: advantages, drawbacks, and approaches for basic and clinical single-cell analyses
The variability of the starting material—the oocytes—contributes, at least in part, to the differences among, and perhaps within, embryos. Brayboy and Wessel make a strong case for performing cell-by-cell analyses of human oocytes in vitro; unlike mouse oocytes, human oocytes show much more variance in the transcript pool. Since it faithfully reflects transcript prevalence in the oocyte, the polar body's composition allows us to infer inter-oocyte variability. Towards this and other applications aimed at prioritizing oocytes for assisted reproduction, the Wessel laboratory is currently developing a single-cell analytical device for oocytes, which they call a ‘simple perfusion apparatus’.
Article of Sun, Champion, Huber and Dovichi: Proteomics of Xenopus development
Unfortunately, the poor correlation between mRNA and protein abundance affects the predictive value of mRNA abundance for cell phenotypes. The Dovichi laboratory shows that mass spectrometry can overcome this problem, now in Xenopus oocytes and maybe soon in mammalian oocytes. The authors have shown that a single Xenopus oocyte or embryo can afford a proteomic depth of ∼6000 proteins. Approximately 40% of the proteins showed significant expression changes across the development stages, while replicates of the same stages, remarkably, exhibited greater similarity.
Article of Yalcin, Hakguder and Otu: Bioinformatics approaches to single-cell analysis in developmental biology
A main driver for single-cell studies was the heterogeneity of cell populations or communities. Some features can be extracted better from single cells than from pooled populations. However, in order to account for the subtle differences among cells, a single-cell-based study must typically include many more biological replicates than a pooled-cell-based study. Although well-established bioinformatics methods exist for analysing sequence data, a limited number of bioinformatics approaches address experimental design, sample size considerations, amplification bias, normalization, differential expression, coverage, clustering and classification issues specifically for application to single-cell studies. This, in turn, generates more costs and more demand for better analytical algorithms. The authors review the vagaries of single-cell analysis.
Upon reading the five articles, we hope you will appreciate—as we have tried to convey by this editorial commentary—just how much progress has been made in single-cell technologies and how this can be applied to the study of oocytes and embryos at the single-cell level. However, not all questions that are worth asking in early development are suited for a single-cell approach: some details are biologically meaningful and others can be ignored, while both require extended analytical power in order to be processed. Paradoxically, if the cells of the same embryo differ from each other and are not representative of the ensemble, comparisons between embryos will have to be made at the single-cell level analysing individually each and every cell; but then, there would be no embryo left.
M.B. and J.B.C. wrote this article together.
No dedicated funding was available for this manuscript.
Conflict of interest
M.B. and J.B.C. are not aware of any conflict of interest that may prejudice the impartiality of this article.
Authors are grateful to their host institutions (M.B. to Max Planck Institute, J.B.C. to Michigan State University) for infrastructural support.