Cells compartmentalize their biochemical functions in a variety of ways, notably by creating physical barriers that separate a compartment via membranes or proteins. Eukaryotes have a wide diversity of membrane-based compartments, many that are lineage- or tissue-specific. In recent years, it has become increasingly evident that membrane-based compartmentalization of the cytosolic space is observed in multiple prokaryotic lineages, giving rise to several types of distinct prokaryotic organelles. Endosymbionts, previously believed to be a hallmark of eukaryotes, have been described in several bacteria. Protein-based compartments, frequent in bacteria, are also found in eukaryotes. In the present review, we focus on selected intracellular compartments from each of these three categories, membrane-based, endosymbiotic and protein-based, in both prokaryotes and eukaryotes. We review their diversity and the current theories and controversies regarding the evolutionary origins. Furthermore, we discuss the evolutionary processes acting on the genetic basis of intracellular compartments and how those differ across the domains of life. We conclude that the distinction between eukaryotes and prokaryotes no longer lies in the existence of a compartmentalized cell plan, but rather in its complexity.
“Nothing epitomizes the mystery of life more than the spatial organization and dynamics of the cytoplasm.”Tim Mitchison 
Elaborate and dynamic spatial organization is a ubiquitous property of living matter and the key to life's distinct condition . Spatial structure is most apparent in the form of separate compartments or compartmentalization, a phenomenon studied in its abstraction as ‘modularity’ in diverse fields ranging from psychology to engineering, from arts to biology .
In biology, modular or compartmentalized systems are found across all scales. The separation of functions inherent to compartmentalization results in robust  yet evolvable  systems, which may explain the prevalence and evolutionary success of this organizational paradigm: ecological networks are compartmentalized, individuals are composed of multiple parts such as organs, organs are composed of multiple cell types, cells themselves are subdivided into compartments, proteins are composed of domains, and so on.
Among these entities, the cell plays a special role in biology and has for long time been understood as the basic unit from which all living beings are built (the ‘Cell Theory’; see  for a historical perspective). The cell compartmentalizes processes in multiple ways, e.g. by using gradients , via lateral diffusion barriers such as the septins in membranes , by bringing multiple reactions together via protein–protein interactions as for example in metabolic channelling , or by using membranes and proteins to physically separate distinct areas within the cell. Hereinafter, we loosely refer to physically delimited compartments as organelles, the focus of the present review.
Organelles are important cellular structures that perform many essential functions. They allow multiple biochemical environments to coexist in the cell and to be adapted independently . For example, proteins can be translated and degraded at the same time without undesired cross-talk . Moreover, spatial confinement of toxic metabolites limits the propagation of potential damage.
In the present paper, we consider the evolution of organelles and physical compartmentalization. Our main aim is to discuss how new organelles are formed and new organellar functions evolve. To do so, we illustrate the broad diversity of organelles and contrast the various mechanisms of compartmentalization in prokaryotes and eukaryotes (Figure 1). We review current theories about the origins and evolution of different types of organelles. Finally, we briefly discuss the genetic basis of different compartments, i.e. the pathways for assembly, maintenance, inheritance, transport from/to and across the physical barriers, and the evolutionary mechanisms shaping those genetic systems.
Diversity of physically delimited compartments
The tremendous diversity of organelles precludes an exhaustive enumeration, as well as an in-depth discussion of each one in particular. Instead, we chose to focus on selected compartments which illustrate distinct aspects, and point the reader to recent reviews for deeper exploration of this fascinating topic.
EVOLUTION OF ORGANELLES: PICTURES, REACTIONS AND GENES
How many types of organelles are there? How do they arise? These are old and surprisingly difficult questions, tightly linked to the emergence of cell biology itself.
As early as the 19th Century, light microscopy had shown the existence of the nucleus, ER (endoplasmic reticulum) or ‘ergatoplast’, Golgi, mitochondria and chloroplasts . Yet, the full extent of eukaryotic cellular complexity was only revealed with the advent of electron microscopy in mid-20th Century . In parallel, biochemical fractionation led to the discovery of other compartments such as the peroxisome and the lysosome [13,14]. Most importantly, biochemical approaches allowed the assignment of specific biochemical activities to distinct organelles, establishing the notion of biochemical or functional compartmentalization of the cell.
Microscopy showed very early on that similar compartments were present in many organisms. For example, the early Cell Theory postulated that all plants, and later also all animals, were composed of nucleated cells (reviewed in [6,12]), and fuelled the first debates about the origins and inheritance of an organelle [15,16].
Besides morphological and functional similarity, comparative genomics has recently opened a new window that allows for the comparison of the genetic basis of organelles. The steady growth of fully sequenced genomes and the accumulation of data that map organellar functions to gene products allows us to ask whether two organelles are homologous, because they are built from homologous gene products, or whether the apparent similarity is based on entirely different machineries, suggesting a functional and/or morphological analogy instead.
Not surprisingly, comparative genomics has its own limitations. The starting point for a comparative genomic study is always a set of sequenced and annotated genomes. As sequencing is biased towards organisms of biomedical or industrial interest, many evolutionarily interesting taxa are not represented by a single sequenced genome. The second step in comparative genomics is the selection of genes that are informative about a given organelle. As we will see below, a significant fraction of organellar diversity is not associated to any molecular data that can be mapped to a genome and, as a result, comparative genomics is simply impossible. Having genomes to compare, and representative genes selected, the homology between genes need to be established. For closely related organisms, this is a rather simple task. However, after hundreds of millions of years, the evolutionary relationships may be less obvious. After establishing homology, the potential gene content of ancestral hypothetical species lying on the interior branching points of the species tree can be reconstructed. This is mostly done using parsimony, a straightforward application of Occam's razor. In brief, the ancestral state is chosen which requires the fewest evolutionary events that alter its gene content (i.e. gene gains and losses) to explain its descendants. However, there is no intrinsic reason why evolution would have to always proceed parsimoniously, i.e. in a minimal number of steps, and statistically more sophisticated model-based approaches exist. Lastly, the interpretation of the results requires judging the cell-biological consistency of the comparative genomics predictions . There is no general recipe to do so, and any reasoning depends on what is known about the cell biology of the compartment. Reconstructing the evolution of organelles is thus a perilous adventure. However, as we discuss below, a series of general principles regarding the origins and evolution of intracellular compartments are emerging. For practical reasons, we divide the discussion into three types of organelles: IMSs (internal membrane systems), endosymbiotic and protein-based organelles.
INTERNAL MEMBRANE SYSTEMS
The characteristic cellular architecture of eukaryotes is defined by extensive membrane-delimited compartments. In animals, approximately half of the volume of a eukaryotic cell is membrane-enclosed (see Table 12.1 in ), and in plants the vacuole can make up to 95% of the cellular volume. Prokaryotes also have unique and diverse membranous organelles compartmentalizing the cytoplasmic space  (Figure 2).
The eukaryotic ES (endomembrane system)
The eukaryotic ES is composed of numerous organelles, such as the ER, the Golgi apparatus, lysosomes or vacuoles, and a trafficking machinery that links these compartments via vesicular intermediates. It is functionally organized mainly along two major pathways: the secretory/exocytic pathway that synthesizes, folds, modifies and targets proteins and lipids to the plasma membrane or the outside of the cell , and the endocytic pathway which operates in the opposite direction, taking proteins and material up from the surface in order to be either recycled or degraded .
The ES probably evolved from inward budding of the plasma membrane in some ancestor of eukaryotes . The main support for this hypothesis comes from the location of the protein translocation machinery, a system to integrate or translocate proteins into or across lipid bilayers and which is a fundamental requirement for all autonomous lifeforms. In prokaryotes, it is found in the plasma membrane, whereas eukaryotes translocate proteins through the membrane of the ER.
Depending on the cellular complexity ascribed to the eukaryotic ancestor, eukaryotes are an example either for an increase in compartmentalization or even for the de novo evolution of an ES in an ancestral cell previously lacking membranous compartments. This directly prompts an essential question: what are the evolutionary forces which can lead to autogenous evolution of compartments?
Evolving a compartmentalized cell plan: the nucleus
In this section, we discuss possible evolutionary forces driving organellogenesis. We illustrate the debate using the example of the eukaryotic nucleus, which is the defining morphological feature shared by all eukaryotic cells and the organelle that has spurred the most debate. The nucleus is a specialized compartment spatially confining the chromosomal DNA and consists of the double bilayered NE (nuclear envelope), a specialized part of the ER. Nuclear pores fenestrate the NE and are the sole sites of material exchange, which are filled and stabilized by one of the largest macromolecular assemblies of the eukaryotic cell, the NPC (nuclear pore complex).
The question of how the nucleus evolved has attracted much attention, as it has to be explained by any scenario for eukaryogenesis. Multiple models based on endosymbiosis, either of viral  or prokaryotic organisms , as well as autogenous origins in the context of an evolving ES have been proposed and extensively debated [17,25,26]. Although this debate surely is not over yet, recent evidence strongly supports an autogenous origin (see  for a recent review).
In any case, the origin of the nucleus does not explain why it evolved in the first place. In the following, we highlight some suggestions for an adaptive value early in the origin of the nucleus. Lopez-Garcia and Moreira  propose a scenario in which the formation of the nucleus is initially driven by the need to separate the anabolic and catabolic pathways. Hence this hypothesis states that the enclosing of the genome is a by-product of metabolic compartmentalization. In contrast, other theories place the adaptive value on the informational function of the nucleus. Cavalier-Smith  hypothesized that as chromosomes became longer, the nucleus emerged to provide physical protection against chromosome shearing. Jékely  proposed that the nucleus allowed the cell to avoid the formation of less efficient chimeric ribosomes resulting from mixing ribosomal subunits in the chimaeric proto-eukaryote. Another scenario argues for the separation between DNA transcription in the nucleus and RNA translation in the cytoplasm as a central driver for compartmentalizing the genetic material. Assuming an early origin of mitochondria, such a separation would provide a mechanism allowing for (slow) splicing to occur before translation, important to cope with an invasion of introns in the wake of the acquisition of mitochondria . This latter view is not inconsistent with the other scenarios, and has also been suggested as a later drive for nuclear evolution in . Thus these three models argue that fidelity and efficiency of information processing provided the basic selective drive to form an NE.
An alternative to the above is to view the nucleus as a ‘frozen accident’, using the words of Bill Martin . The smaller the population size, the weaker is the force of selection in relation to drift. This opens the possibility that characters which are not advantageous in any way, or even slightly deleterious, can be fixed. It has been argued that this basic principle is able to account for much of the genomic architecture observed in eukaryotes . As it is not implausible to think that eukaryogenesis involved major population bottlenecks, non-selective processes may also explain the evolution of some of the eukaryotic cellular structures. As a consequence, although the emergence of complex features inspires us to think of selective drives, many of the features of the eukaryotic ES, and of the nucleus in particular, may have started as non-adaptive events. These would have been fixed as a consequence of small population sizes, and only later gained the essential functions that ensured their conservation in eukaryotes.
Ultrastructure of selected prokaryotic compartments
The autogenous versus exogenous debate: peroxisomes
Compartments can evolve in different ways: autogenously, as we discuss above for the nucleus, or as a result of an endosymbiotic event. For most organelles, both scenarios have been proposed, but in many cases the debate has settled: the community mostly agrees on an endosymbiotic origin of mitochondria and an autogenous, i.e. non-endosymbiotic, origin of the ES.
However, for example, in the case of the peroxisome, deciding between the two competing models of compartmental origin is not straightforward, and owing to its hybrid characteristics, arguments in favour of both modes of origin persist. Peroxisomes are single-membrane-bound organelles characterized by the presence of hydrogen-peroxide-producing oxidases, and catalases that break down hydrogen peroxide . In animals, peroxisomes are involved in lipid metabolism and free radical detoxification , among other processes. In protists, the diversity and specializations of their functions sometimes blurred their common identity as peroxisomes, and motivated names such as glycosome, glyoxysome or Woronin body .
An endosymbiotic origin of peroxisomes is suggested by several lines of evidence. Peroxisomal membrane and matrix proteins are translated by free cytosolic ribosomes independently of the ER, and their import into the peroxisome is mediated by short peroxisome-targeting signals , which are bound by cytosolic receptors. Such ER-independent protein translation and targeting mechanisms are also found in mitochondria and plastids, with which peroxisomes share certain metabolic pathways. Peroxisomes normally arise from fission of existing peroxisomes, an inheritance strategy shared with endosymbiotic organelles.
However, this model has recently been challenged by both genomic  and cell-biological  evidence. Comparative genomic analysis of peroxisomal proteomes revealed that approximately half of the genes are of eukaryotic origin. Furthermore, mechanisms for the recruitment of entire mitochondrial pathways have been proposed [36,38], able to account for the presence of the bacterial genes in the peroxisomal proteome. Particularly relevant is that the import machinery used by peroxisomes has been found to share similarities with the ER misfolded protein machinery. More generally, peroxisomal maintenance proteins were found to be homologous with the ERAD (endoplasmic-reticulum-associated protein degradation) pathway [34,36], suggesting an evolutionary connection between peroxisomes and the ER and hence to the ES. A second major line of evidence supporting an autogenous origin of the peroxisome relates to its inheritance. Besides fission, peroxisomes can be formed de novo  from membranes and vesicles clearly originating from the ER . Hence current evidence strongly supports an autogenous origin.
One organelle, many shapes: Golgi
Organelles of the ES mostly have complex shapes, with spherical, flat or tubular regions, and the tendency of these shapes to be conserved throughout evolution suggests that they are important for organelle function . Yet, organelle structure can be very dynamic and vary across organisms and tissues. The Golgi apparatus is an example of such a dynamic and morphologically heterogeneous compartment.
The Golgi is the central hub of the membrane-trafficking system and the site where the two major pathways, secretory and endocytic, intersect. It receives secretory lipids and proteins from the ER, which are post-translationally modified, sorted and targeted to their final destination. Vesicles for constitutive secretion are formed at a specialized subcompartment of the Golgi, the TGN (trans-Golgi network).
The Golgi is nearly ubiquitous in eukaryotic cells  and performs essential functions. Yet, it shows great diversity in biogenesis, morphology and inheritance. In mammals, it consists of a series of flattened cisternal membrane structures responsible for the Golgi's large and regular structure. The cisternae form Golgi stacks, which are interconnected by lateral tubules and organize into the Golgi ribbon . At the onset of mitosis, the Golgi loses its characteristic shape and fragments into tubulovesicular structures, which are partitioned during cytokinesis and later reform the Golgi . Interestingly, Drosophila cells usually lack a Golgi ribbon, but clearly have the potential to form it, as is observed for example in spermatids . In Saccharomyces cerevisiae, the Golgi does not have the stacked morphology, but rather consists of single isolated cisternae distributed throughout the cytoplasm . However, this type of organization appears to be an exception, as most fungi possess Golgi stacks [43,44]. Yet another shape is observed in plants: their Golgi apparatus is made of hundreds of individual motile stacks which are not disassembled during cell division . Most surprising for an organelle with essential functions, the Golgi has been lost altogether on at least eight independent occasions . For example, the protozoan parasite Giardia lamblia lacks a morphologically discernible Golgi, and secretory functions are performed at specific ER sites .
One organelle, many functions: lysosomes
Lysosomes are the primary catabolic site of the animal cell  that receive and degrade macromolecules from multiple cellular pathways (secretory, endocytic, autophagic and phagocytic)  in their characteristic acidic lumen. The maintenance of lysosomal function depends on constant influx of hydrolases and lysosomal membrane proteins delivered by endosomes coming from the TGN .
The lysosome is a good example of an evolutionarily plastic compartment which evolved additional functions. In fungi and plants, the functional equivalent of the lysosome is the vacuole, which has adapted to many additional roles, e.g. in calcium homoeostasis and osmotic control, long-distance transport of nutrients through the mycelial filaments, storage of organic and inorganic nutrients, and in generating turgor pressure [50–52]. In plants, at least three distinct vacuole types have been described, the lytic, protein storage and the senescence-associated vacuole that coexist in some cell types [53,54].
In animals, the lysosome diversified into a whole family of distinct compartments, the LROs (lysosome-related organelles) . Some LROs evolved to function as secretory compartments, yet without losing the ability to perform classic lysosomal functions, as for example the lysosomes of cytotoxic T-lymphocytes. In contrast, other LROs have lost degradative capacity and coexist with lysosomes in one cell, e.g. melanosomes and diverse types of granules. They are still formed from maturing endosomes, and are therefore distinct compartments different from other organelles involved in regulated secretion such as secretory granules. LROs and lysosomes are biogenically related organelles , suggesting an ‘organelle paralogy’ that still awaits confirmation at the genetic level. LROs are highly taxon- and tissue-specific, and with growing knowledge of the diversity of eukaryotic cell biology more of these organelles may be found. A candidate LRO is for example the apicomplexan rhoptry, an apical secretory organelle involved in host cell invasion .
Homology versus analogy: granules and bacterial intracytoplasmic membranes
Functional and/or morphological similarity is most readily explained by homology, i.e. common ancestry. Thus the common interpretation of similarities observed among organelles is that of a single evolutionary origin. However, convergent evolution also plays a role in subcellular evolution, giving rise to analogous organelles defying this interpretation. Below, we discuss two examples: DCGs (dense-core granules), which probably arose multiple times in eukaryotes, and the relationship between the IMSs of Plantomycetes and those of eukaryotes.
DCGs are specialized secretory vesicles with a characteristic condensed protein core involved in regulated secretion [58,59]. Membranous compartments resembling mammalian DCG are found in diverse eukaryotic lineages where they serve a wide range of functions, e.g. the trichocysts in Paramecium tetraurelia, which function in predator defence, or encystation secretory vesicles in G. lamblia, which provide protection from the environment . They share important hallmarks with mammalian granules: biosynthesis from the TGN, subsequent maturation and regulated secretion upon specific stimuli . The comparison of the molecules essential for the structural scaffold of DCG in ciliates and mammals showed that the genes in ciliates are lineage-specific innovations. Moreover, key proteins important for maturation, e.g. endoproteases, and organellar identity, e.g. SNAREs (soluble N-ethylmaleimide-sensitive fusion protein-attachment protein receptors) and Rabs, are not orthologous. Hence, comparative genomics hints at an independent origin of DCG at least in ciliates and mammals, implying that an entire organelle can arise repeatedly by convergent evolution.
Another important debate regarding homology versus analogy is the relationship between the intracytoplasmic membranes of Planctomyetes (see, for example, 13 in Figure 2) and the eukaryotic ES, in particular the nucleus. Planctomycetes are a major bacterial phylum that compartmentalize cytoplasmic space by internal membranes . In its simplest form, e.g. in the genus Pirellula, a lipid bilayer sequesters chromosomes and ribosomes from other cellular components and forms an organelle coined the pirellulosome . In some Planktomycetes, the pirellulosome is itself compartmentalized and contains a double-bilayer membrane fenestrated by small pore-like openings surrounding the compacted chromosome . These membranes hence form a structure with obvious morphological parallels to the eukaryotic nucleus, although the functional implications of those IMSs are currently unclear. For example, a clear separation of transcription and translation cannot be observed, as ribosomes are present both inside and outside the nuclear body . The discovery of membrane coat-like proteins in these bacteria , the description of endocytosis-like processes  and several other similarities to eukaryotes have fuelled speculations that Plantomycetes may represent a frozen intermediate stage between prokaryotes and eukaryotes . This interpretation has, however, been strongly opposed . This is a situation that exemplifies well the limitations of comparative genomics: the claims of homology of some components rely on very sensitive methods to identify distant relationships, necessarily susceptible to false positive errors. Carl Sagan popularized the idea that “extraordinary claims require extraordinary evidence”, considering the time that separates us from these events, any claim of homology based on remote similarities is to be treated with caution.
Thylakoids, bacterial intracellular trafficking?
Planctomycetes provide a good illustration of the fact that bacteria are able to evolve IMSs fully detached from the plasma membrane. Yet, a major difference from the eukaryotic ES is the lack of obvious vesicular trafficking, conceptually separating the endomembranes observed across the two domains of life. Bacteria do not share the same need for active trafficking as eukaryotes, as they are in general small enough to rely on diffusion . However, some use motor-driven intracellular transport for cellular movement  and indications exist that some cyanobacteria may have evolved dynamic vesicular trafficking pathways in the context of thylakoid membranes.
The thylakoids are one of at least three major types of photosynthetic membranes  (examples of thylakoids and the other two types, chlorosomes and chromatophores, are shown in Figure 2: 4 and 5 show thylakoids, 3 and 6 show chlorosomes, and 10 shows chromatophores). They form a proper compartment with characteristic lipid composition [71,72] harbouring the photosynthetic systems in Cyanobacteria and chloroplasts. In Cyanobacteria, they mainly consist of long lamellae appearing as several flattened and stacked layers. Unlike bacteria performing anoxygenic photosynthesis, thylakoid membranes are not connected to the plasma membrane .
The discontinuity with the plasma membrane raises an important question: the precursors of the photosystems are pre-assembled in the plasma membrane, but how are those proteins and lipids transferred to the thylakoids? The obvious possibilities are via transient membrane connections or by vesicle transport [71,73], but the definitive mechanism has not yet been determined. Interestingly, cytoplasmic vesicles have been reported in fixed samples of some cyanobacterial species, where they were found to be located near and partially fused to thylakoids . This means that bacteria with thylakoids may be capable of long-range vesicular trafficking, but independent confirmation is still lacking. In support of this hypothesis, thylakoid membranes showed numerous perforations through which substantial flow of cellular material such as ribosomes and storage granules could be observed. These features could be adaptations to allow for efficient intracellular transport, possibly also of membrane-bound vesicles . Hence this could mean that when confronted with the corresponding cell-biological challenge, prokaryotes may have the potential to evolve an ES organized according to the same principles as eukaryotes.
Lateral organelle transfer: bacterial magnetosomes
Magnetosomes are membranous compartments of a diverse paraphyletic group of MB (magnetotactic bacteria) (see, for example, 12 in Figure 2). They are usually organized in one or more chains of 15–20 organelles. They allow the sensing of geomagnetic field lines, which bacteria use to find preferred redox conditions . Magnetosomes are formed by invaginations of the plasma membrane  which house a ~50 nm magnetic crystal. They contain a characteristic set of proteins in various quantities, which suggests the existence of a dedicated sorting pathway . The arrangement in chains is achieved by filaments formed by homologues of eukaryotic actin. Hence magnetotactic bacteria provide an example for prokaryotes, which, like eukaryotes, use the cytoskeleton to position their organelles within the cell and actively shape and maintain their intracellular organization .
Sequencing of several MB revealed that most of the genes believed to be involved in biogenesis and functioning of magnetosomes are comprised within a large genomic island [76a], which may be sufficient to form a magnetosome . It could be shown by comparative genomics that the island has been laterally transferred between diverse bacterial species [76b] accounting for the patchy phylogenetic distribution of MB.
Thus magnetosomes are an example for the lateral transfer of an entire compartment. This process is unique to the prokaryotic domain, where it may be rather common: comparative genomics suggests that, for example, the proteinaceous compartments discussed below have been laterally transferred among as much as one-fifth of sequenced bacterial species .
ORGANELLES OF ENDOSYMBIOTIC ORIGIN
The universal mitochondria
One of the most distinctive features of eukaryotic cells, and potentially the critical step in the establishment of the eukaryotic cell plan , is the presence of mitochondria. They are the result of an ancient endosymbiosis predating the radiation of eukaryotes, as all eukaryotic cells have mitochondria or MROs (mitochondria-related organelles) such as hydrogenosomes or mitosomes [79–81]. This second group is sometimes genome-less, but biochemical similarity and molecular composition have established them as derived mitochondria (, and reviewed in [80,81]). In most cases, mitochondria and MRO seem to be involved in the generation of ATP, either by aerobic or anaerobic respiration . Sometimes, this function may be shared with other prokaryote-derived organelles, such as in ciliates living under anaerobic conditions, which have acquired multiple archaeal endosymbionts that consume produced hydrogen and use it to reduce carbon dioxide . These observations have fuelled the hypothesis that mitochondria provided the energy necessary for the establishment of the complex eukaryotic cell plan . However, the only mitochondrial function that is never lost, even in very derived MRO, is the synthesis of iron–sulfur clusters [80,84]. This suggests that the dependence of the cell on the energy produced by mitochondria may be reduced, for example in the context of parasitic relationships.
Same organelle, repeatedly: plastids
Plastids are found in a variety of phylogenetic lineages. However, their evolutionary relationships are not straightforward, as successive endosymbiotic events resulted in cells consisting of several nested organisms.
A single ancient endosymbiosis gave rise to the chloroplasts found in the Plantae supergroup (Glaucophyte, and red and green algae) (reviewed in [85–87]). This stable relationship was established between a putative phagotropic mitochondria-containing host cell and a cyanobacterium. Subsequent independent symbiotic events of chloroplast-containing algae with several protozoa gave rise to the plastids found in Chromalveolata, Rhizaria and Excavata (see [85–88] for reviews). These secondary endosymbioses are well illustrated by the existence of typically four plastid membranes, as opposed to the two membranes observed in organelles derived from a primary endosymbiosis (mitochondria and chloroplasts). Furthermore, in Dinoflagellates, even tertiary endosymbiotic events have occurred, leading to organisms with an intriguingly complex compartmentalized cell plan . Interestingly, the amoeba Paulinella chromatophora acquired a cyanobacterial endosymbiont which has been argued to represent a second independent origin of a primary plastid. It is, however, still a matter of debate whether the level of genetic integration  and the protein-targeting system  qualify the endosymbiont as a bona fide organelle .
Primary plastids (chloroplasts) are photosynthetic organelles that can differentiate in a reversible manner into storage functions. Some secondary plastids retained their photosynthetic function, as, for example, is the case for members of the genus Euglena, while others lost it (see  for a review). The function of these latter plastids is unclear, even though their existence in human pathogens as, for example, the causative agents of malaria, toxoplasmosis and cryptosporidiosis has made them subject of intense interest (see [92,93] for reviews). Thus plastids represent yet another example of one organelle with multiple taxon-specific, cell-type-specific and condition-specific functions, illustrating how the principle discussed for lysosomes applies also to organelles of endosymbiotic origin.
Parasites versus endosymbionts
Mitochondria and chloroplasts represent one extreme of full integration between endosymbiotic organelle and host. At the opposite end of the spectrum are compartments formed by intracellular parasites that invade and parasitize eukaryotic cells for several resources, occupying different intracellular niches (phagosomes, lysosomes, cytosol, etc.) and having a deleterious effect on the host [94–96]. Examples are the agents of Lyme disease, typhus or leprosy. In between these two extremes, we find multiple independent cases of endosymbiosis of bacteria from diverse taxonomic groups with various degrees of integration with organisms from the majority of eukaryotic taxa.
It is unclear where to draw the line between parasite and endosymbiont. In the case of the associations of Buchnera sp. with several insects, the endosymbiont supplies the host cell with essential amino acids that it cannot synthesize or obtain from the environment, and its presence has no obvious fitness cost, suggesting (endo)symbiosis . On the other hand, the association of Wolbachia with several insects is less clear: it provides protection against viral and bacterial infections, while inducing a series of reproductive alterations on the host organisms via feminization, male-killing, parthenogenesis and/or cytoplasmic incompatibility . Hence fitness trade-offs exist and the distinction between parasite and symbiont is less obvious. However, unlike Buchnera, Wolbachia is not essential, suggesting that the parasite has not fully integrated with the host. It is tempting to speculate that endosymbiotic compartments may start off as a parasitic relationship that evolved to mutual benefit, to the point where the endosymbiont becomes an essential compartment of the host cell. Endosymbiotic associations have been reviewed extensively elsewhere [99–104].
Moving genes and proteins around: EGT (endosymbiotic gene transfer)
Endosymbiotic and obligate endoparasitic relationships are accompanied by extensive gene loss, as these bacteria ‘outsource’ many of their normal metabolic requirements to the host . Endosymbionts such as Carsonella rudii exemplify how far this genome miniaturization can go, retaining only 182 open reading frames and a 160 kb genome . Genes from the bacterial endosymbiont are frequently transferred to the nucleus, and infrequently retained. For example, Wolbachia-derived transfers of various sizes have been detected in a large range insects and nematodes, ranging from individual genes to almost complete genomes .
In the case of the mitochondria and chloroplasts, this genome reduction has been even more extreme and was accompanied by a substantial transfer of the organelle's genes to the nuclear genome and by the loss of the corresponding genes from the organelle genome. This led to a reliance on the host to synthesize the gene products and target them to the symbiont, requiring the evolution of targeting and transport systems. This process is termed EGT (reviewed in ), and was argued to represent the defining step in the transition from an endosymbiont to an organelle (for a discussion, see ).
Mitochondrial proteins are synthesized in the cytosol and maintained in an unfolded translocation-ready conformation by the action of chaperones. Their import into mitochondria is mediated by the TIM (transporter of the inner mitochondrial membrane)–TOM (transporter of the outer mitochondrial membrane) complexes via sophisticated pathways (reviewed extensively in ). Proteins destined to the plastid are mostly synthesized in the cytosol and targeted to the TIC (transporter of the inner chloroplast membrane)–TOC (transporter of the outer chloroplast membrane) complexes. These complexes are molecularly unrelated to each other, suggesting independent evolutionary origins. Bacteria contain many proteins with similarity to the mitochondrial translocation and targeting machineries, suggesting that the latter were assembled from the former [110,111]. Similarly, the TIM–TOM system appears to have evolved from pre-existing prokaryotic and eukaryotic components .
Interestingly, a subset of proteins that are destined to the chloroplast are also trafficked through the ES, by an as yet unknown mechanism [113–115]. This has been hypothesized to represent an ancient transport pathway, which appeared before the establishment of the TIC–TOC pathway . Supporting this argument is a recent endosymbiosis of a Cyanobacteria by the amoeba P. chromatophora  (reviewed in [86–88]), where several genes suffered EGT , and some have been shown to be translated in the amoeba's cytosol and then transported to the endosymbiont in a Golgi-mediated pathway . There is growing evidence that mitochondria can sort material into and from vesicles that are then transported to peroxisomes (reviewed in ). Recent work revealed further that oxidized proteins can be trafficked from mitochondria to the lysosome via vesicular intermediates . Thus, although bona fide organelles have evolved specific targeting and transport systems, these coexist with vesicle-mediated transport. It is interesting to note that the ES is frequently subverted by intracellular pathogens [119,120], supporting the notion that endosymbiosis may, in some cases, derive from a parasitic association.
Prokaryotes inside prokaryotes
Less appreciated than endosymbiosis between pro- and eu-karyotes is the fact that bacteria have also found replication niches inside other prokaryotes, which lends support to the idea that the endosymbiotic event that resulted in mitochondria did not necessarily require a pre-existing ES. One type of bacteria that contain prokaryotic endosymbionts are Gammaproteobacterial Beggiatoa spp. (8 in Figure 2), sulfide-oxidizing organisms, found in freshwater and marine habitats. They harbour a variety of prokaryotic endosymbionts, but the metabolic nature of the relationship is still elusive . A better studied case is a secondary endosymbiosis in the endosymbiont of the mealy bug (7 in Figure 2). This nested constellation consists of a Gammaproteobacteria living as an endosymbiont of a Betaproteobacteria, which in turn is an endosymbiont of an insect, the mealybug . Genomic analysis revealed extreme metabolic integration between the two endosymbionts, with the Gammaproteobacteria complementing several amino acid biosynthetic pathways of the Betaproteobacteria . Even mitochondria have been targeted by other prokaryotes, such as ‘Candidatus Midichloria mitochondrii’, an endosymbiont of the tick Ixodes ricinus. Although it resides in the cytosol, it can invade and replicate inside mitochondria and consume them [123,124].
Our understanding of the associations between prokaryotes is still in its infancy. We do not know how frequent they are, the type of metabolic trade-offs they may involve or how inheritance of the endosymbiont is achieved after division of the host cell. Furthermore, the entry mechanisms involved in these associations remain to be determined. They may be similar to those used by the Deltaproteobacteria endoparasite Bdellovibrio bacteriovorus that invades and replicates inside other bacteria . Bacterial predators are, in fact, quite frequent and phylogenetically widespread , and could represent the origins of some or even most of the prokaryotic endosymbionts of bacteria.
Although the creation of physical separation between two environments has been most visibly achieved by lipid membranes, both prokaryotes and eukaryotes have also used proteins for the same purpose.
Protein shells everywhere
Prokaryotes possess a variety of protein-bound organelles. These compartments present a relatively recent paradigm of prokaryotic organization, and few of these organelles have been studied thoroughly. Better known are for example gas vesicles found in archaea (see, for example, 1 in Figure 2) and bacteria , and a certain class of BMCs (bacterial microcompartments) defined by homologous shell protein families which we discuss below.
BMCs are a widely distributed functionally diverse organelle family which is defined by its characteristic proteinaceous shell . The shell is assembled from a few thousand members of the conserved BMC family of shell proteins, giving rise to organelles up to ~150 nm in size. Some BMCs store special compounds , whereas others create an internal microenvironment in which metabolic enzymes and auxiliary proteins are concentrated. This enhances the catalytic efficiency by enabling substrate channelling and due to increased local metabolite concentrations which help to overcome poor enzyme affinities for their substrate, or serves to spatially confine toxic or volatile intermediates . First described in 1956 [128a], BMCs have long been studied mostly by electron microscopy, which makes it difficult to estimate their abundance and diversity throughout bacteria. Examples of BMCs include carboxysomes (see, for example, 4 and 5 in Figure 2), which can be found in all Cyanobacteria and concentrate Rubisco (ribulose-1,5-bisphosphate carboxylase/oxygenase), the Pdu compartment found in Salmonella enterica (see, for example, 9 in Figure 2) or the Eut microcompartment also found in Escherichia coli. Relatively little is known about the cell biology of this special class of organelles. A basic challenge is the sorting of enzymes to the interior of the protein shell. Most interestingly, in Salmonella, short N-terminal targeting signals have been described that are sufficient to direct proteins to the lumen of these organelles . Thus proteinaceous organelles can employ a system analogous to the targeting system known from secretion in bacteria and the two major types of lipid-bound organelles in eukaryotes.
Evolutionary processes shaping the genetic basis of an organelle, illustrated using the nucleus as an example
Recent molecular characterization of the shell proteins allowed the computational screening of bacterial genomes, which revealed homologous domains in more than 400 or one-fifth of the sequenced bacteria analysed . The phylogenetic distribution suggests frequent horizontal gene transfer of BMC shell genes, which are organized in gene clusters facilitating their modular transfer. It could be shown by the artificial expression of an empty microcompartment in E. coli that the genetic information in such a cluster is indeed sufficient for the formation of BMCs .
Although less known, eukaryotes also have protein shells forming small organelles, termed ‘vaults’. Vaults are ribonucleoprotein particles consisting of two identical cups, each with a characteristic eight-fold symmetrical shell structure. They sequester an interior volume large enough to capture hundreds of proteins and multiple copies of small RNAs , and are found mostly in the cytoplasm, although some localize to the nucleus . Homologues of the highly conserved vault constituents can be found in diverse eukaryotes , and occur in numbers potentially as high as 104–107 per cell . Since their discovery in 1986 [134a], several putative functions have been put forth, e.g. roles in multidrug resistance, signalling and innate immunity, but their exact function remains elusive. The vault shell has been shown to be dynamic, opening the possibility that the vault's interior may be able to interact functionally with the rest of cell .
Protein-based compartments in the ES: eukaryotic cilium
Above, we discussed the integration between the endosymbiotic organelles and the ES. Now, we turn to the cilium in order to illustrate the integration of different types of organelles, proteinaceous and membranous. We refer the reader to several reviews on the functions and structure of the cilium and its connection to the centrosome [136–138].
The eukaryotic cilium or flagellum (we use these terms interchangeably) is a microtubule-based protrusion from the plasma membrane that is templated by a mostly invariant nine-fold symmetrical microtubule-based cylinder termed the basal body, from which the axoneme protrudes. Active-transport mechanisms involving molecular motors move microtubule subunits and membrane receptors within, into and out of the cilium in a highly dynamic process. At the base of the cilium are protein-based physical barriers that separate an internal and external space, as well as protein-based physical barriers preventing lateral diffusion of proteins in the plasma membrane [136,139–141]. Cilia are found in all eukaryotic groups, and are thus believed to pre-date the radiation of eukaryotes (reviewed in ).
The cilium appears to be a unique organelle in having internal microtubules. However, it has many connections to the ES. The set of molecular motors which regulate the process of IFT (intraflagellar transport) [142,143] within the cilium are homologous with those functioning in endomembrane trafficking. The assembly of axonemes and delivery of receptors to the ciliary membrane depends on polarized trafficking. Cargo destined for the cilium is assembled and sorted at the Golgi and delivered at the flagellar base, in a process dependent on canonical regulators of trafficking in the ES (reviewed in [139,144]). This has led to the proposal that cilia represent an elaboration of a pre-existing polarized trafficking from the Golgi to the plasma membrane . This hypothesis is supported by the discovery that IFT components are required for polarized secretion in cells that do not form cilia . The integration of cilia with other transport systems has become even more apparent with the discovery that ciliary entry of KIF17 (kinesin family member 17), an IFT motor, is regulated by importin-β2 and RanGTP , key regulators of nucleocytoplasmic transport.
Thus the cilium represents an alternative way to create a distinct compartment within the ES. Although its biogenesis and maintenance, like that of all other organelles within the ES, requires the sorting and targeting of components using vesicular intermediates, the physical separation of the space is not achieved by enclosing it in a membrane, but rather by different protein-based macromolecular machinery.
THE MOLECULAR BASIS OF ORGANELLAR EVOLUTION
The ability to sustain a compartmentalized cell plan in general, and any one organelle in particular, requires molecular machinery for its assembly, maintenance, inheritance and for diverse trafficking routes. We discuss commonalities and major differences of how these pathways evolved at the molecular level, both between eukaryotes and prokaryotes, as well as between the different types of organelles.
In eukaryotes, organelles of the ES usually share a conserved core of genes throughout all lineages which is complemented by lineage-specific genes. This pattern arises from the interplay of two opposing evolutionary forces: lineage-specific gene losses and gains. Genes gained in specific lineages mostly derive from two processes: co-option of previously unrelated genes and paralogous expansion of protein families already involved with the organelle (see, for example, [148,149]). Co-opted genes themselves can originate from different sources: they can be endosymbiotic, horizontally transferred or the result of gene duplication. The second process contributing lineage-specific genes is paralogous expansion, which happens frequently and is the result of gene duplications followed by preservation of both duplicates mostly due to sub- or neo-functionalization . Co-option and duplication are not the only possibilities for the emergence of lineage-specific genes. Other possibilities are the de novo origin of genes and domain shuffling, i.e. the evolution of new genes by combining pre-existing protein domains in new ways [151,152]. All of these possibilities have been shown to have contributed genes forming the NE and the NPC  (Figure 3). Gene duplications may be particularly important for the functional diversification and evolution of new organelles in eukaryotes. For example, the organellar paralogy model proposes that duplication/divergence of the key gene families of an organelle could give rise to new organelles, free to diverge and acquire new functions analogous to the process at the gene level .
In endosymbiotic organelles of eukaryotes, the driving force seems to be genome reduction by extensive gene loss [100,155]. One extreme case is observed in human mitochondria, which have only 13 protein-coding genes. Although only one genome of a prokaryotic endosymbiont of prokaryotes has been well characterized, it suggests that genome reduction is also observed . Inheritance of mitochondria involved the co-option of the machinery that insures segregation of the ES to the daughter cells after division . More recent endosymbionts may have also followed the same route, as exemplified by Wolbachia's segregation to the centrosome during cell division and its microtubule-based inheritance .
The molecular machineries underlying prokaryotic membrane-based organelles as well as protein-based compartments are still poorly understood. However, in addition to the mechanisms described for eukaryotes, a different process is important: horizontal transfer of entire genomic islands containing the genes necessary to form a compartment. This appears to be a frequent event both for membranous organelles, such as magnetosomes, and proteinaceous organelles such as BMCs, which are the most widely distributed form of compartment in prokaryotes.
The origin of eukaryotes has been identified as one of the major transitions in evolution . Whereas this is frequently equated with the evolution of compartmentalization, the present review hopefully contributes to raise awareness that compartmentalization exists in all domains of life.
The prokaryotic cell plan is emerging as increasingly more complex than previously thought. It is currently unclear how widespread compartmentalization is in prokaryotes, and how much more diversity will emerge from broader sampling of the prokaryotic world. In particular, the extent of compartmentalization of archaea is still largely unknown, as most studies have focused on bacteria (Figure 2). Especially, the discovery of archaeal compartments may have profound implications for our understanding of cellular evolution, as exemplified by recent claims that acidocalcisomes could be present in all domains of life and may thus date back as far as to the last universal common ancestor . However, these are currently speculations and other interpretations for example of eukaryotic acidocalcisomes as LROs exist .
In both prokaryotes and eukaryotes, compartments come in different flavours: they are bound either by membranes or by proteins, and can be of autogenous origin or derived from an endosymbiotic event. Examples for each of these classes exist in both domains of life (Figure 1). Yet, there are clear differences between the compartments in prokaryotes and eukaryotes. Eukaryotes are derived from a highly compartmentalized cell, frequently termed the LECA (last eukaryotic common ancestor), which already possessed an elaborate ES . All eukaryotes have more or less retained this complex compartmentalized cell plan. In contrast, prokaryotic membrane organelles are phylogenetically scattered and apparently evolved independently, although potential exceptions such as acidocalcisomes exist. Also, they have not generally reached the same level independence from the plasma membrane compared with their eukaryotic counterparts as no definitive evidence for prokaryotic vesicular trafficking exists. Endosymbiosis is apparently the exception rather than the rule in prokaryotes, whereas endosymbionts are ubiquitous in eukaryotes which all have at least one, the mitochondrion or MROs. Furthermore, prokaryotes tend to have one type of compartment and not the coexistence of multiple organelles that characterizes eukaryotes. Thus, although the presence of intracellular compartments no longer distinguishes prokaryotes from eukaryotes, the complexity of the compartmentalized cell plan is still substantially higher in eukaryotes.
Under this scenario, the debate on the origins of eukaryotes needs to be formulated differently: what is the reason for the divide in complexity between the eukaryotic and prokaryotic cell plan? It has been proposed that energy limitations in prokaryotes may constrain their genome size and therefore preclude the evolution of a more complex compartmentalized cell . Under this scenario, the acquisition of mitochondria was thus the key step en route to the more complex eukaryotic cell plan. This suggests an intrinsic limit to the complexity of extant compartmentalized prokaryotes. Yet, notably this lack of complexity has not limited the success of prokaryotes, which are clearly the dominant lifeform on our planet. In contrast, eukaryotes appear to be ‘condemned’ to a compartmentalized cell plan, which may at best be slightly simplified with the loss of a small number of organelles.
A question closely linked to the above considerations is how easy it is to evolve a compartment, and if the potential differs between the domains of life. In eukaryotes, endosymbiotic events appear to be more frequent than in prokaryotes. However, reaching the high degree of integration, as, for example, achieved by mitochondria and plastids, appears uncommon even in eukaryotes. The emergence of an ES is a rare event, having only happened once. It then served as a template to many lineage-specific organelles. This suggests that it may be simpler and sufficient to build new organelles by variations of the existing system. This is strongly supported by the examples of analogous organelles within the ES , suggesting that evolving a compartment in response to similar challenges is possible. In contrast, different IMSs exist in multiple independent prokaryotic lineages, suggesting that these rudimentary forms are somehow simpler to establish. We must then reach the conclusion that compartmentalization evolved independently more frequently in prokaryotes than in eukaryotes. Furthermore, in prokaryotes, the genetic clustering of some of the systems into genomic islands and the ease of lateral transfer result in frequent transfer of whole organelles between species. It is interesting to note that overexpression of a foreign gene in E. coli has been observed to be enough to induce the formation of massive intracellular vesicles . Yet, as argued in , it remains unclear how functionless lipid vesicles could be co-opted to form compartments without being purged by purifying selection.
At the molecular level, an issue that has attracted much attention is the evolutionary relationship between genes underlying the different types of compartments, particularly those seemingly present in both prokaryotes and eukaryotes. Yet, most prokaryotic compartments have little functional and molecular data associated to them. Thus we expect that exciting findings will arise from sequencing of more genomes of compartmentalized prokaryotes, and from the cell-biological dissection of prokaryotic compartments, potentially settling some of the debates on homology. Without a better understanding of prokaryotic compartments, i.e. their diversity and molecular basis, little progress is likely to be made in elucidating the most fascinating problems of compartmentalization, complexity and ultimately the origin of eukaryotes themselves.
endosymbiotic gene transfer
internal membrane system
nuclear pore complex
transporter of the inner chloroplast membrane
transporter of the inner mitochondrial membrane
transporter of the outer chloroplast membrane
transporter of the outer mitochondrial membrane
J.B.P.-L. thanks Damien Devos and Bill Martin for helpful discussions. We also thank José Feijó, Patrícia Brito, Artemy Kolchinsky and members of the Computational Genomics Laboratory for a critical reading of the paper.
Y.D. is funded by a Fundação para a Ciência e Technologia (FCT) fellowship [grant number SFRH/BD/33860/2009].