The monoplastidic bottleneck in algae and plant evolution

Plastids (chloroplasts) define the cytosol of algae and plants like no other compartment. The origin of plastids traces back to the endosymbiotic integration of a cyanobacterium into the cytosol and biochemical pathways of a heterotrophic eukaryote. We know close to nothing about the nature of the protist host, and from which cyanobacterial phyla the primal plastid evolved is also uncertain (Deschamps et al., 2008; Deusch et al., 2008; Criscuolo and Gribaldo, 2011). Based on the overall sequence similarity, the plastid donor appears to belong to a group of cyanobacteria able to fix atmospheric nitrogen (Dagan et al., 2013). Another phylogenetic analysis points to the recently discovered freshwater-dwelling Gloeomargarita clade (Ponce-Toledo et al., 2017). What is commonly accepted is that the three archaeplastidal lineages – Glaucophyta, Rhodophyta and Chloroplastida (Zimorski et al., 2014; Archibald, 2015) (Fig. 1A) – arose monophyletically (Rodríguez-Ezpeleta et al., 2005; Jackson and Reyes-Prieto, 2014; Burki, 2014) (see Glossary in Box 1 for specialized terms used throughout). The Archaeplastida emerged early during eukaryotic evolution (He et al., 2014), probably sometime between the ‘upper limit’ for the origin of eukaryotes ∼1.9 billion years ago (Gya) (Eme et al., 2014) and the fossilization of early red algae some 1.6 to 1.2 Gya ago (Butterfield, 2000; Bengtson et al. 2017). These fossil records display multicellular organisms, in the case of Bangiomorpha with branched filaments, indicating that they had already evolved some level of morphological complexity and that single-celled algae – maybe comparable to extant glaucophytes – are older.

The consummated endosymbiotic integration of a prokaryote into a eukaryote is a rare event. There are examples of endosymbiotic integration of prokaryotes in several clades of eukaryotes (Kneip et al., 2008; Nakayama et al., 2014; Nowack, 2014; Bennett and Moran, 2015; Husnik and McCutcheon, 2016), but it was the endosymbiotic integration of the mitochondrion and later the plastid that were key to the evolution of macroscopic life. Among the reasons for why such an integration is a rare event are the challenges of: (1) fusing a prokaryote and an eukaryote genome (Timmis et al., 2004); (2) establishing new transport routes for proteins now encoded in the host nucleus as a consequence of endosymbiotic gene transfer (EGT) (Soll and Schleiff, 2004; Leister, 2016; Garg and Gould, 2016); (3) developing a means of communication such as retrograde signalling (Woodson and Chory, 2008; Chandel, 2015; Singh et al., 2015); and (4) synchronizing prokaryotic fission with the cell cycle of the eukaryotic host (Miyagishima, 2011). The co-evolution of organelle and host depends on the simultaneous and successful implementation of these events. However, some of these events compete, such as EGT and the number of endosymbionts per cell. These conflicts add an additional layer of complexity to the transition from endosymbiont to organelle.

In this Review, we discuss the evolutionary ancestry of plastid division, from the successful integration of the cyanobacterial plastid progenitor into the host, to the additional layers of division control that land plants evolved. We collate information on the mechanisms underlying plastid division and the number of plastids per cell, and examine the implications for organelle evolution and inheritance, before aiming to provide an answer as to why most algae bear only a single plastid per cell.

EGT was a key component of organelle integration during the early stages of algae evolution and is an ongoing process in plants (Matsuo et al., 2005; Cullis et al., 2009). EGT typically occurs in the form of DNA chunks, rather than individual genes (Henze and Martin, 2001; Yuan et al., 2002; Michalovova et al., 2013). This is thought to occur through the uncontrolled lysis of endosymbionts, thus releasing their DNA, of which some can then be randomly incorporated into the nuclear genome (Martin, 2003). Many nuclear genomes of eukaryotes carry evidence for recent gene transfers that are known as nuclear mitochondrial and nuclear plastid DNA sequences (Richly and Leister, 2004; Hazkani-Covo et al., 2010). The number of plastids (or mitochondria) that are present when EGT happens is crucial: if an alga carries only one plastid, then its lysis results in the death of the host cell and hence, no progeny will emerge. That is why EGT preferably occurs in the presence of multiple donors per cell, which is true for the cyanobionts that evolved into modern plastids (Barbrook et al., 2006) – both mitochondria and plastids in their current form (Smith 2011a; Smith et al. 2011b) – as well as the complex plastids of secondary endosymbiotic origin (Curtis et al., 2012). Through EGT, the host gains ever more control while the endosymbiont loses its autonomy.

The presence of a single organelle, whether it is a plastid or mitochondrion, significantly slows down the rate of subsequent EGT, trapping the genomes of the organelles evolutionarily (Barbrook et al., 2006; Curtis et al., 2012). For Paulinella chromatophora, it was suggested that horizontal gene transfer (HGT) complemented EGT early during chromatophore origin (Nowack et al., 2016). The frequency of HGTs among eukaryotes is debated (Huang, 2013; Ku et al., 2015), but if true for P. chromatophora, HGT ceased with the switch from phagocytosis-based heterotrophy to photoautotrophy (Nowack et al., 2016). HGTs therefore have little, if any, impact on subsequent steps of organelle integration, for which examples are the synchronization of organelle and host division and the transfer of the remaining genetic material of the endosymbiont to the nuclear genome. In summary, the presence of multiple cyanobionts during the early stages of plastid origin appears likely and may be a prerequisite for successful plastid inheritance. Yet, the majority of extant algae appear to have a host cell (nucleus) to plastid number ratio of one and that ratio is rarely altered. As outlined below, we propose that a monoplastidic bottleneck was part of the evolutionary origin of the plastid (Fig. 2).

A plant or algal cell needs to ensure inheritance of its plastids and mitochondria as the loss of these organelles during cytokinesis is lethal. There are two possible solutions for the control of the inheritance of at least one organelle (of each type) by the offspring. The first is the presence of numerous organelles in the cytosol, which results in a rather passive inheritance based on their stochastic distribution. While this might be feasible for the mitochondria of mammals (Mishra and Chan, 2014), it carries the risk of random failure (Mishra and Chan, 2014). In multicellular organisms, this can be tolerated, but is selected against in single-celled eukaryotes. The second solution is a synchronization of organelle and nuclear division and a controlled distribution of compartments during cytokinesis. If multiple endosymbionts foster the transition of endosymbiont to organelle, because it promotes EGT events (Smith 2011a; Smith et al., 2011b; Barbrook et al., 2006; Curtis et al., 2012), then the first solution appears to be correct. However, this is only the case for the earliest stages of plastid evolution, when plastid-to-nucleus communication and coordination of their division had not yet been established. Basal branching (i.e. an early divergence from the last common ancestor) algae of the red and green lineage are all monoplastidic (Fig. 1A,B), and a few non-vascular land plants are monoplastidic at least during some stages of their life cycle (Brown and Lemmon, 1990; Vaughn et al., 1992). One might even consider some polyplastidic-appearing glaucophytes to be monoplastidic too, but it is difficult to form a definitive conclusion (Box 2) owing to the lack of structural and molecular data available for glaucophytes. Species that harbour two plastids, such as the glaucophyte Cyanophora, are worth a second look (Box 2) and may be viewed as a deferred case of monoplastidy.

Monoplastidy is thus the common and ancestral character state of plastid-bearing eukaryotes. It suggests that during the endosymbiont-to-plastid transition, plastid numbers per cell were reduced to one and this monoplastidic bottleneck occurred before the split into the three main archaeplastidal lineages (Fig. 1A). The archaeplastidal ancestor was likely to be a heterotrophic single-celled eukaryote, a protist. Such a heterotrophic protist will have possessed mechanisms for ensuring mitochondrial, but not yet plastid, inheritance. It is conceivable that evolution selected for monoplastidy to secure proper vertical inheritance of the photosynthetic organelle through the synchronization of the division of plastid and host. The same cannot be said about mitochondria, because their division is not strictly coupled to that of the nucleus; rather, it is coordinated by the endoplasmic reticulum in both animals (Lewis et al., 2016) and plants (Mueller and Reski, 2015). The situation for mitochondria differs because their origin is irrevocably linked to the very origin of eukaryotes and their hallmarks of meiosis, the cell cycle and sex (López-García and Moreira, 2015; Gould et al., 2016; Garg and Martin, 2016).

Support for the theory that monoplastidy was selected for to secure proper plastid inheritance comes from two of the three main archaeplastidal lineages. In both the glaucophyte C. paradoxa and the red alga Cyanidioschyzon merolae (Fig. 1A), plastid and nucleus division is coordinated in a manner such that mitosis only commences after the successful division of the plastid (Sumiya et al., 2016). The same is observed in eukaryotes that carry plastids of secondary origin (Hashimoto, 2005), which allows us to speculate that the same regulatory machinery (including a means of retrograde communication) which evolved in the primary host to coordinate plastid and nuclear division is now at work in the secondary host, too. Thus, within the archaeplastidal ancestor, we can speculate that a common set of regulatory factors were implemented that orchestrate plastid division and synchronize it with the host cell cycle. Furthermore, along the evolutionary trajectory of this synchrony, the number of plastids (or cyanobionts) would have gone from a non-controlled population of many to a single plastid. Next, we inspect the associated components and deduce their role with regards to the evolution of the plastid division machinery.

Plastids inherited the backbone of their division (i.e. fission) machinery from their cyanobacterial ancestor (Miyagishima and Kabeya, 2010; Dagan et al., 2013; Ponce-Toledo et al., 2017) (Fig. 1A,B). Although cyanobacterial division differs in some components from that of other bacteria, it nevertheless relies on physical constriction of the cell by a contractile ring (Miyagishima et al., 2005), which is formed by the self-assembling GTPase filamentous temperature-sensitive protein Z (FtsZ) (Bi and Lutkenhaus, 1991; de Boer et al., 1992; Osawa et al., 2008; de Boer, 2010). FtsZ is the primary component of the ‘Z’ ring that forms in the bacterial cytosol and an array of accessory factors either control formation of the Z ring or are recruited to it after the Z ring has formed (Adams and Errington, 2009). Land plant plastids have inherited many of these components, including FtsZ, ARC6 (ACCUMULATION AND REPLICATION OF CHLOROPLASTS 6) and minD/E (also known as ARC11/12) (Strepp et al., 1998; Miyagishima et al., 2014a; Osteryoung and Pyke, 2014; Grosche and Rensing, 2017). They now act in concert with the host cell cycle (Sumiya et al., 2016), not least because most of them are today encoded in the nucleus as a result of EGT (Miyagishima et al., 2012).

During plastid division in Chloroplastida and Rhodophyta (the green and red lineage, respectively), genes that are associated with plastid division share the same expression pattern (Miyagishima et al., 2012). This includes ftsZ, DRP5B (dynamin-related protein 5B), which facilitates organelle scission from the cytosolic side of the plastid and the minicell gene minD, which regulates the positioning of the FtsZ-based ring on the stromal side of the inner plastid membrane (Fujiwara et al., 2008; Osteryoung and Pyke, 2014). In the case of the chloroplastidian minD, the localisation of the gene (i.e. whether in the plastid or nuclear genome) has no impact on its expression pattern, although in the latter case, minD expression seemed to be regulated by light rather than the cell cycle (Miyagishima et al., 2012). The plastid division genes (ftsZ, ftsW and the septum development gene sepF) of the glaucophyte C. paradoxa, however, experience little fluctuation in expression (Miyagishima et al., 2012). Moreover, both C. paradoxa and C. merolae only allow mitosis to commence once plastid division is completed (Sumiya et al., 2016). Still, the differences in the expression pattern of division factors might indicate that the plastid division in Cyanophora is somewhat distinct from that of the other Archaeplastida. How then could this have occurred?

Glaucophytes such as Cyanophora are the most strongly diverging clade of the Archaeplastida (Burki, 2014) (Fig. 1A). Although only 15 glaucophyte species have been described (Guiry, 2012), it is likely that more species exist (Jackson et al., 2015; Takahashi et al., 2016). The most basal-branching glaucophyte genus is Cyanophora (Chong et al., 2014). Cyanophora cells tend to harbour either one or two plastids (in glaucophytes these are referred to as muroplasts; see Glossary in Box 1); if two are present, then they are semi-connected as if frozen in the act of division (Jackson et al., 2015; Box 2). Other glaucophyte genera, such as Cyanoptyche, Gloeochaete and Glaucocystis are polyplastidic (Jackson et al., 2015), but the muroplast morphology in Glaucocystis warrants attention as it has two stellate muroplast clusters within the cell (Schnepf et al., 1966). These clusters consist of individual muroplast lobes that at one end are held together through unknown mechanisms (Schnepf et al., 1966). Therefore, together with the basal position of Cyanophora in a clade of polyplastidic algae, these findings further support the idea that harbouring two (physically connected) plastids per cytosol might be an intermediate stage between monoplastidy and polyplastidy. The bundling of plastids in Glaucocystis might well be a relic of this transition (Boxes 2 and 3). Glaucophytes are generally considered to have retained ancestral features of the earliest Archaeplastida (Fathinejad et al., 2008; Steiner and Löffelhardt, 2011; Facchinelli et al., 2013), including a thick peptidoglycan (PG) layer between the two membranes of the plastid organelle (Steiner et al., 2001). There is little molecular research being carried out on glaucophytes; however, the retention of some ancient traits and the diversity of their plastid morphology for such a small group could prove useful for future research in light of the recent identification of a PG layer in moss (Hirano et al., 2016).

The cyanobacterial progenitor of the plastid had a layer of PG, which separated its inner and outer membranes. This is evident by the retention of the PG layer – sometimes referred to as a murein layer – by the muroplasts of glaucophytes (Steiner et al., 2001) and the recent identification of a thin PG layer containing D-amino acids that surrounds the plastids of the model moss species Physcomitrella (Hirano et al., 2016). Previously, there was a consensus that the murein layer had been lost early during plastid evolution after the green and red lineage had diverged from the glaucophytes. One reason for the loss of the PG layer was the idea that it interfered with protein import of nuclear-encoded plastid proteins (Steiner and Löffelhardt, 2002). It has always been intriguing that photosynthetic eukaryotes contain genes such as murE and mraY, which encode enzymes that synthesize PG, and the resulting PG layer is very likely to be associated with plastid division (Machida et al., 2006; Garcia et al., 2008; Takano and Takechi, 2010). The function of MurE diverged in angiosperms and gymnosperms – where it is one of the plastid RNA polymerase-associated proteins (PAPs; Pfalz and Pfannschmidt, 2013) – whereas it has retained its ancestral function in the moss Physcomitrella (Garcia et al., 2008; Lin et al., 2017). This raises the question of whether the retention of even only a thin murein layer could be associated with the regulation of plastid division.

Independent lines of evidence connect the PG layer with the regulation of plastid division. For instance, PG-inhibiting antibiotics, such as ampicillin or fosfomycin, affect the division and morphology of plastids in streptophyte algae (Matsumoto et al., 2012), lycophytes (Izumi et al., 2003) and mosses (Katayama et al., 2003). Additionally, plastid division is altered in Physcomitrella knockout lines for enzymes synthesizing PG (Homi et al., 2009). Such effects are much less pronounced in euphyllophytes (Takano and Takechi, 2010): Izumi and colleagues (Izumi et al., 2008) showed that three different fern species had a reduced number of plastids upon treatment with fosomycin, but not ampicillin, whereas their lycophyte control (Selaginella nipponica) responded to both. In addition, fosomycin and ampicillin have no effects on the plastid number of angiosperms (Kasten and Reski, 1997; Izumi et al., 2008). These differences might thus reflect and add to the specific evolutionary changes experienced by the embryoplast (de Vries et al., 2016). A link between the complete loss of the PG layer and the switch from mono- to polyplastidy in embryophytes is feasible, but cannot be its sole reason because of the absence of a PG layer in rhodophytes (Grosche and Rensing, 2017), which are predominantly monoplastidic (Fig. 1B). In bacteria, there is a correlation between the loss of the PG layer and loss of FtsZ (Miyagishima et al., 2014a), and the division and PG synthesis in bacteria is driven by GTP-dependent treadmilling of FtsZ filaments (Bisson-Filho et al., 2017; Yang et al., 2017). This is in line with the observation that the guidance of synthesis of the PG layer is one of the main functions of FtsZ (de Pedro et al., 1997; Aaron et al., 2007; Typas et al., 2012).

Plastid division involves FtsZ in all Archaeplastida species that have been analysed thus far (Yang et al., 2008; Miyagishima et al., 2012; Chen et al., 2017). Here, two slightly different FtsZ proteins (FtsZ1 and FtsZ2) heteropolymerize to form the inner division ring (Yoshida et al., 2016). Streptophyta and Glaucophyta possess another FtsZ protein, FtsZ3 (Grosche and Rensing, 2017). Localization studies in P. patens have shown that FtsZ3 assembles into ring-like structures in the cytosol and plastid (Kiessling et al., 2004). Recently, the loss of the PG-synthesising enzymes (and sensitivity to PG-inhibiting antibiotics) was correlated with the loss of FtsZ3 (Grosche and Rensing, 2017), in line with the aforementioned pattern in bacteria (Miyagishima et al., 2014a). Surprisingly, although FtsZ is clearly involved in plastid division, it is not essential (Schmitz et al., 2009; Miyagishima et al., 2014a). The plastids in Arabidopsis ftsZ-knockout lines still divide, but intriguingly, their cells harbour a reduced number of plastids per cell (Schmitz et al., 2009). This suggests that the additional copy of FtsZ3 is strictly connected to a PG-associated function and that whereas the remaining copies of FtsZ are crucial for the fine-tuning of plastid division, they are not essential for the process per se.

Plastid division commences within the organelle (Miyagishima et al., 2014a) because of its cyanobacterial origin. The prokaryotic division machineries act first on the cytosolic face of the plasma membrane (Errington et al., 2003). The cyanobacterial ancestor rests in the cytosol of the host cell, and with the concomitant reduction or even loss of the PG layer as seen in higher land plants, plastids became accessible and amenable to external forces that act within the cytoplasm of the host. The most substantial cytosolic forces are the outer division rings: one is formed by a protein of the dynamin family (Gao et al., 2003; Miyagishima et al., 2003; Yang et al., 2008; Miyagishima and Kabeya, 2010), whose origin and function are connected to the origin and fission of mitochondria (Purkanti and Thattai, 2015; Gould et al., 2016); the other is the plastid division ring (Kuroiwa et al., 1998) that consists of polyglycan filaments in the red alga C. merolae (Yoshida et al., 2010).

In eukaryotes, dynamins are conserved GTPases that exert mechanical forces, with some dynamins mediating the contraction and scission process during organelle division (McFadden and Ralph, 2003; Purkanti and Thattai, 2015; Leger et al., 2015). Dynamin-mediated plastid division arose in the common ancestor of Rhodophyta and Chloroplastida, as it is present in all studied members of the red and green lineage, but not glaucophytes (Miyagishima et al., 2014b). Interestingly, mutations in genes coding for the inner division machinery components, such as ARC6 (a J-domain protein homolog to the cyanobacterial protein Ftn2; Vitha et al., 2003), have a more pronounced influence on plastid number (Fig. 3) than mutations in components acting on the outside (Robertson et al., 1996; Gao et al., 2003; Sakaguchi et al., 2011). Moreover, the division of proplastids – plastids that can differentiate into different types and are in this form only present in land plants – is not affected in dynamin-knockout lines (Robertson et al., 1996). Regardless of whether outer or inner division components are manipulated, proplastid division still occurs (Miyagishima et al., 2014a). Knockout of components such as ftsZ in land plants tends to reduce the number of plastids (Fig. 3) to a degree that it allows only the proplastids to divide (Schmitz et al., 2009; Miyagishima et al., 2014a). The latter might be facilitated by alternative plastid division processes, such as budding (Miyagishima et al., 2014a), which was proposed to allow plastid division in mutants with a macrochloroplast (Pyke, 2010). Plastid budding has been described for non-chlorophyllous plastids of the Arabidopsis arc6, crumpled leaf (crl; which exhibits a severe phenotype with some cells maybe even lacking plastids altogether; Asano et al., 2004; Chen et al., 2009) and Solanum lycopersicum (tomato) suffulta plastid division mutants (Forth and Pyke, 2006). Thus, plastid division is evidently quite robust.

We propose that this robustness, including unconventional plastid division mechanisms such as budding, is due to the framework that evolved at the monoplastidic bottleneck. Other layers of additional control, such as the ARC6-interacting PLASTID DIVISION proteins (PDVs) that control the rate of plastid division, (Miyagishima et al., 2006; Glynn et al., 2008), were added to the existing plastid division machinery later in evolution. The use of systematic knockouts to strip these additional layers of control that were implemented to better command plastid division should reveal the ancient molecular chassis that traces back to the time when plastid and nuclear division were synchronized.

Components of the plastid division machinery were lost or modified several times independently during evolution (Miyagishima et al., 2014a). Above, we discussed a handful of genes and properties that might have been crucial for the orchestration of plastid division. Studies have shown that the knockout of single genes such as ARC6 can transform a polyplastidic embryophyte cell to a cell carrying a single embryoplast (Pyke et al., 1994; Vitha et al., 2003). This raises the question of whether there may be just a few key factors that determine organelle numbers. The mutants mentioned above often harbour a single plastid that is massively increased in size, which is reminiscent of large algal plastids such as those of Chlamydomonas, whose plastid takes up about half of the cell volume (Gaffal et al., 1995). In fact, large single plastids are quite frequently observed when plastid division is impaired, for example in ftsZ knockouts in both A. thaliana (Schmitz et al., 2009) and P. patens (Martin et al., 2009), as well as in P. patens knockouts for murA and mraY, which are involved in PG synthesis (Homi et al., 2009) (Fig. 3). In summary, single gene mutations are capable of reverting polyplastidy back to (macroplastidic) monoplastidy and the list of involved proteins is probably not yet complete.

Plastid division requires the interplay of two genetic compartments: the plastid and the nucleus. Whereas in land plants all plastid division proteins are nucleus encoded, the situation is quite different in algae: their plastid genomes can include ftsI, ftsW, sepF, minD and minE genes (Miyagishima et al., 2012). Recently, we asked whether the absence of minD and minE from the plastid genome – now encoded by the nucleus or entirely lost – could be a prerequisite for the evolution of polyplastidy (de Vries et al., 2016). We speculated that EGT of minD to the nucleus could be essential to gain more control over plastid function and division. In the context of the monoplastidic bottleneck (Fig. 2), this means that a desynchronization (‘de-sync’) of plastid and nuclear division was only feasible through additional control mechanisms exercised by the nucleus. This stands in stark contrast to the morphologically similar, but autonomy-based primordial presynchronization (‘pre-sync’) state of the early endosymbiont inside the host.

We screened all 999 species of Archaeplastida and secondary plastid-bearing lineages for which plastid genomes were available for: (1) the presence of minD and minE genes in their plastid genomes and (after excluding most of the numerous chlorophyceaen and land plants species with sequenced plastid genomes) the morphological characters of the remaining 131 species for (2) mono-, bi- or poly-plastidy, and (3) unicellularity, colonial morphology or multicellularity (de Vries et al., 2016) (Fig. 1B). Algae that contain minD and/or minE in their plastid genome are exclusively monoplastidic and limited to primary green Chlorophyta and secondary red Hacrobia (see Glossary in Box 1). In contrast, secondary green chloroplast genomes of Euglenozoa lack minD and minE, and the cells are predominantly polyplastidic (Fig. 1A,B). The case of secondary red Hacrobia is peculiar. Only the plastid genomes of Hacrobia (Okamoto et al., 2009; Burki et al., 2012) and P. chromatophora encode minD and minE. In fact, among all Archaeplastida-derived plastids, the plastid genomes of cryptophytes are the only ones to still encode minE. The deep branching of cryptophytes among photosynthetic eukaryotes (Stiller et al., 2014; Burki, 2014) and certain traits, such as the coding of plastid protein import machinery components in the nucleomorph and their plastid membrane topology (Gould et al., 2015), suggest they represent an ancestral state to which the presence of minD and minE in the plastid genome corresponds.

We know that Hacrobia, (e.g. the coccolithophore Emiliana), stramenopiles and some Alveolata (including the causal agent of malaria, Plasmodium; McFadden, 2011) all harbour a red complex plastid that ultimately traces back to the monophyletic, endosymbiotic incorporation of a rhodophyte by a host of unknown nature (Burki et al., 2012; Zimorski et al., 2014; Archibald, 2015). However, the chronology and number of potential tertiary and maybe even quaternary endosymbiosis events (3° ES and 4° ES, respectively; Fig. 1A) that succeeded the initial secondary ES event, which established the red complex plastid, remain disputed (Burki et al., 2012; Stiller et al., 2014; Gould et al., 2015; Archibald, 2015). It was recently suggested that haptophytes acquired their plastids from ochrophytes (which unite stramenopiles, pelagophytes and kelp) through quaternary ES (Stiller et al., 2014). However, no stramenopile plastid genome contains either a minD or minE homolog (Fig. 1B) and the nucleus-encoded minD and minE are furthermore of mitochondrial origin (Leger et al., 2015). This means that haptophytes would have acquired a stramenopile with a plastid that was unlike any of those known today. We cannot formally rule out this possibility, but it does argue against a quaternary endosymbiotic origin of the haptophyte plastid. Besides, if we accept additional layers of endosymbioses (3° and 4° ES), the important question arises: why was the same initial red complex plastid of rhodophyte origin (and similar to the one found in cryptophytes such as Guillardia theta) always passed around the individual hosts?

If none of the primary red algae have a minD or minE homolog in their plastid genome, where does the homolog in the secondary red Hacrobia come from – considering that EGT is genetically and evolutionarily a one-way process (Martin et al., 1998)? The Cyanidiales are on the deepest branch in red algal phylogenies (Yoon et al., 2006). Among the Cyanidiales, Galdieria sulphuraria is the only red alga that still harbours some remnants of a minD homolog (Leger et al., 2015) (Fig. 1). Using Galdieria theta minD as query for a tBLASTn screen versus Galdieria sulphuraria, the Galdieria minD pseudogene is retrieved with a query coverage of 83% and an average local identity of 46% over two stretches from different reading frames. When the aligned (nucleotide) sequence is used as a query for a BLASTx screen against G. theta, it returns GtMinD as the best (and only meaningful) hit. This suggests an orthologous relationship between the two (it is noteworthy that a tBLASTn screen using GtMinD as query against all bacteria in the non-redundant dataset returns Gloeomargarita lithophora minD as the top hit). The most parsimonious assumption is that this putative orthologous relationship is independent of HGT and based on plastid inheritance. Together with the deep-branching position of G. sulphuraria (Yoon et al., 2006), these data suggest that secondary red plastids probably stem from an extinct or non-sampled red algal lineage before these genes were lost or transferred to the host nucleus. In summary, minD and minE are prime candidates for future studies on (1) plastid autonomy with regard to its division, (2) its connection to the transition from mono- to poly-plastidy, and (3) the evolution of red complex plastids.

Plastids evolved in a unicellular eukaryote. Indeed, in almost any lineage, basal-branching algae are unicellular (Fig. 1B). The Glaucophyta are exclusively unicellular (Jackson et al., 2015). The deepest branching Rhodophyta, the Cyandiales (Yoon et al., 2006), are unicellular, too. In the third lineage, the Chloroplastida, the picture becomes more complicated, with the recent placement of the multicellular Palmophyllaceae (having a mass of unicells in a gel matrix) at the base of all Chlorophyta (Leliaert et al., 2016). Yet, palmophyllaceaen multicellularity probably represents a derived character state, especially because the most basal-branching streptophyte Mesostigma viride (Marin and Melkonian, 1999) is unicellular (Fig. 1B). The same pattern applies to monoplastidy: based on the characters of basal-branching streptophytes (M. viride; Marin and Melkonian, 1999) and chlorophytes (the prasinophytes; Leliaert et al., 2012), polyplastidy was not a feature of the common ancestor of all Chloroplastida (Leliaert et al., 2011, 2012).

If one draws a simple trajectory of alga and plant evolution, one might get the impression that there was a clear overall increase in morphological complexity, and that polyplastidy evolved as a by-product. This, however, is not the case (Box 3). The green and red lineages evolved multicellularity multiple times independently (Lewis and McCourt, 2004; Leliaert et al., 2012; Cock and Collén, 2015), just as several single algal classes did (e.g. the ulvophytes) based on ancestral character state inferences (Cocquyt et al., 2010). However, multicellularity and polyplastidy do not go hand in hand but rather originated multiple times independently throughout evolution (Box 3). The next section explores what these independent origins imply with regard to proper organelle inheritance.

Mitochondria and plastids face similar challenges. The genomes of the organelles are in danger of mutational meltdown as a result of Muller's ratchet (see Glossary in Box 1) (Lynch et al., 1993; Lynch, 1996; Martin and Herrmann, 1998), concomitant with an elevated risk of accumulating mutations induced by reactive oxygen species (Aro et al., 1993; Apel and Hirt, 2004; Balaban et al., 2005). One can assume that evolution favoured those organisms that evolved mechanisms that protect the integrity of the rudimentary genomes of these organelles. For mitochondria, many ideas have been formulated and some even experimentally tested (Box 4). Furthermore, the mutational baggage in mitochondrial DNA (mtDNA) accumulating over a lifetime has repeatedly been associated with aging (Harman, 1956, 1972; Balaban et al., 2005).

There are obvious limits in transferring animal-centric information to the analysis of plastids of plants and algae. First, in many plastid-housing eukaryotes, the mutation rates of organelles are quite low (Smith, 2015). This could, however, also be the result of successful genetic mechanisms that assure quality (Box 4); assuming that the mutation rate of a given mtDNA molecule is low would be a type of post hoc fallacy. Second, there is lack of an immediate (i.e. embryonic) separation of germ and soma cells. Third, most algae are unicellular (and monoplastidic) and selection acts immediately. The formulation of models on such a germline-based separation is therefore tricky, although a mechanism that ensures the inheritance of only fit organelles in protists is generally feasible. In budding yeast, especially those mitochondria with a high membrane potential (ΔΨ), which is indicative of bioenergetic vigour, are passed on to the daughter cell through their localisation to the budding site (Higuchi-Sanabria et al., 2016). Furthermore, the genetic mechanisms discussed in Box 4 also work in unicellular and monoplastidic species such as Chlamydomonas that nonetheless bear multiple plastid genome copies (VanWinkle-Swift, 1980; Birky, 2001). But what about such mechanisms in land plants?

As for land plants, the number of proplastids in a seed is rather small (∼10 versus ∼200 chloroplasts found in a mesophyll cell, Possingham, 1980) and that number might be actively reduced (Mogensen, 1996). This raises the question of whether there is a connection between the (ancestral) monoplastidic bottleneck and the genetic bottleneck during organelle inheritance, especially in polyplastidic organisms. Owing to a lack of dedicated studies on the issue this cannot be easily answered, but there are numerous mechanisms to selectively restrict plastid numbers in germ cells; furthermore, plants are generally highly efficient at obtaining homoplasmy (a stage at which all plastid genome copies are identical) after several rounds of cell division (Greiner et al., 2015). Some basal-branching embryophytes are monoplastidic in those cells that are important for spore production (Brown and Lemmon, 1990; Smith 2011a; Smith et al. 2011b) and, in the case of some lycophytes, also in their meristematic tissue (Brown and Lemmon, 1984, 1985). Furthermore, spores and gametes of the multicellular and polyplastidic brown alga Ectocarpus siliculosus are also monoplastidic, arguing for a potentially analogous need for monoplastidy as a quality control checkpoint during the reproductive cycles of Ectocarpus (Baker and Evans, 1973; Maier, 1997). Quality control mechanisms to ensure the inheritance of organelles free of mutations established at the early time point in the evolution of the monoplastidic bottleneck would benefit a genetic bottleneck during plastid inheritance. We conclude that the reduction of plastid number in gametes and spores to one is a required reversion back to monoplastidy. However, to what degree some of these functions are conserved, and whether they still act in seed plants remains to be determined – but this surely presents a worthwhile research target.

The presence of dozens or hundreds of plastids per cell might be perceived as being the norm, but algae and some non-seed plants tell us otherwise. It remains to be investigated how these macrochloroplasts are integrated into signalling pathways that determine the chloroplast compartment size (space and/or volume in a cell that is devoted to chloroplasts) such as the recently described REDUCED CHLOROPLAST COVERAGE (REC) proteins (Larkin et al., 2016). In that context, it is noteworthy that mutants with macrochloroplasts appear to have a similar chloroplast compartment size (Pyke and Leech, 1992). This argues for an additional (possibly REC-dependent) layer of nuclear control over plastid function. What are the benefits of many small versus a few (or one) big plastids in the first place? A larger internal membrane system, for example, for more photosystem complexes, can easily be accommodated in a macrochloroplast. Also, the in situ control of gene regulation (in accordance with the CoRR hypothesis, see Glossary in Box 1; Allen, 2015) should be no hurdle, as nucleoids could be distributed along the intermembranes. One of the advantages of having multiple (small) plastids was previously mentioned in the context of EGT: if one organelle lyses for any reason a polyplastidic cell has a back-up. Furthermore, there is some evidence that having a large population of small chloroplasts (rather than a small population of big chloroplasts) allows for more-effective chloroplast relocation-mediated light acclimation and avoidance of photoinhibition (Jeong et al., 2002; Königer et al., 2008), which supports the idea that land plant plastid biology was shaped by stressors typical for terrestrial habitats (cf. de Vries et al., 2016).

There exists a peculiar monoplastidic bottleneck in algae and plant evolution; its reason and extent remains to be fully explored. This monoplastidic bottleneck occurred in the ancestor of Archaeplastida, at a point when the majority of the regulatory mechanisms that govern plastid division (and thereby plastid number) were implemented. Control of plastid number relies upon the modification of the fission machinery that the plastid had encoded within its genome. Subsequently, these genes were complemented by the host through proteins such as those of the dynamin family and the PVDs with the emergence of land plants. To a degree, plastid inheritance follows principles that also apply to mitochondria (rarely studied simultaneously) and both are guided by uniparental inheritance, which is, however, far more involved for organelles of cyanobacterial origin. Whether uniparental inheritance of plastids, akin to mitochondrial inheritance, is associated with the monoplastidic bottleneck is speculative, because a systematic study on the topic is currently lacking. In our opinion, the presence of a monoplastidic checkpoint, even in higher embryophytes (e.g. in the zygote), cannot entirely be ruled out. If such a checkpoint indeed exists, but is no longer present in embryophytes, how did they escape monoplastidy? Furthermore, did the same or a radically different solution evolve in those algae that escaped the monoplastidic bottleneck independently? The occurrence of polyplastidy coincides with a complex and macroscopic morphology, both in rhodophytes and streptophytes, underscoring the impact of this transition. A testable prediction from our rationale is that the molecular regulation of plastid division in those monoplastidic plant cells is homologous to that established at the monoplastidic bottleneck event. Comparative studies hold the potential to uncover the molecular basis established during the monoplastidic bottleneck, and to what degree it continues to influence and define plastid inheritance in land plants.

We thank Klaus V. Kowallik for many useful comments, as well as Pavel Škaloud (Charles University, Czech Republic), Monique Turmel (Université Laval, Canada), Eric W. Linton (Central Michigan University, USA), Matthew S. Bennett (Michigan State University, USA) and Robin Matthews (Huxley College of the Environment, USA) for kind and informative correspondence.