In general terms, rhizobia are defined as soil bacteria that fix nitrogen in symbiotic association with legumes and Parasponia. The Proteobacteria is an important phylum that contains diazotrophic organisms and phylogenetic studies using 16S ribosomal RNA sequences indicate that the best-known rhizobial genera are from the α-proteobacteria group (Rogel et al., 2011; Weir, 2016), including the genera: Rhizobium, Mesorhizobium, Sinorhizobium (renamed Ensifer, Martens et al., 2007; Judicial Commission of the International Committee on Systematics of Prokaryotes, 2008), Bradyrhizobium, Azorhizobium, and Allorhizobium. Some other α-proteobacteria genera also contain one or more rhizobial species, such as Aminobacter, Methylobacterium, Devosia, Ochrobactrum, Phyllobacterium, Microvirga, and Shinella (Rogel et al., 2011; Ormeño-Orrillo et al., 2015; Weir, 2016; ICSP Subcommittee on the taxonomy of Rhizobium and Agrobacterium http://edzna.ccg.unam.mx/rhizobial-taxonomy/). Recently, Neorhizobium and Pararhizobium have been proposed as new genera (Mousavi et al., 2014, 2015). Several rhizobial species belong to the β-proteobacteria genera, including Burkholderia, Cupriavidus, and Herbaspirillum (Moulin et al., 2001; Chen et al., 2003; Lloret and Martínez-Romero, 2005; Masson-Boivin et al., 2009; Rogel et al., 2011; Weir, 2016). Indeed, the taxonomy of rhizobia has recently been revised (Peix et al., 2015; Shamseldin et al., 2017).

In a first instance, a comparison of glutamine synthetase (GS) genes I and II allowed the time of divergence among the α-proteobacteria genera of rhizobia to be estimated (Turner and Young, 2000). The data from GSII sequences suggest that Rhizobium and Ensifer are the most recent genera, and Bradyrhizobium and Mesorhizobium the most ancient. Based on GSI, Rhizobium, Ensifer, and Mesorhizobium genera appear to have separated at the same time, and Bradyrhizobium is the most ancient genus. Based on the analysis of GS genes and the amino acid substitution rates in their orthologs, the Bradyrhizobium genus probably diverged from the last common ancestor of all rhizobia some 500 Mya, before the appearance of land plants (about 400 Mya). Similarly, the most recent genus Ensifer diverged about 200 Mya (Turner and Young, 2000; Morton, 2002; Lloret and Martínez-Romero, 2005), before the appearance of Angiosperms (dated more than 150 Mya; Martin et al., 1989) and legumes (about 70 Mya; Lavin et al., 2005). When phylogenetic analyses of the 16S rRNA gene and the intergenic spacer region was combined, slightly but not significantly more recent divergence times were found for rhizobia: about 385 Mya for Bradyrhizobium, 344 Mya for Mesorhizobium, 201 Mya for Ensifer, 145 Mya for Rhizobium/Agrobacterium, and 54 Mya for Neorhizobium (Chriki-Adeeb and Chriki, 2016).

Evolutionary studies of α-proteobacteria indicate that while the evolution of a reductive genome has been observed in intracellular animal-associated bacteria, genome expansion is observed in plant symbionts (as well as in several animal and plant pathogens, such as Rickettsia, Brucella, or Bartonella). Rhizobia are among the α-proteobacteria with the largest genomes (MacLean et al., 2007) and genes involved in nitrogen fixation and nodulation (or pathogenicity) have become integrated for symbiosis, often arranged on auxiliary replicons in genomic islands (mobile elements). The genome size and the diversity among rhizobia are due to the presence of these highly dynamic auxiliary replicons and to a high degree of paralogy (Batut et al., 2004).

Genome plasticity and instability in rhizobia is due to largescale recombination events (the presence of repeated DNA sequences, insertion elements and multiple replicons), and in fact, lateral gene transfer is the primary source of genetic diversity in rhizobia (Flores et al., 2000; Guo et al., 2003; MacLean et al., 2007; Provorov et al., 2008). It has been proposed that the genomes of rhizobia have evolved by expansion as a means to adjust to the challenges imposed by their multiphase lifestyle, principally through horizontal gene transfer and gene duplication (Batut et al., 2004; MacLean et al., 2007; Provorov and Andronov, 2016). In some rhizobia-legume symbioses, up to 15-20% of the rhizobial genome is activated in symbiosis (Udvardi et al., 2004; Tikhonovich and Provorov, 2009). Different models of co-evolution in the rhizobia-legume symbiosis have been proposed or are under study (but they are still controversial); especially in relation to the selection of rhizobial symbiotic traits by the host legume (Provorov et al., 2008; Martínez-Romero, 2009).

An evolutionary step from free-living diazotrophs related to Rhodopseudomonas to the symbiotic diazotroph Bradyrhizobium through the acquisition of fix genes was proposed as the first stage of rhizobial evolution (Provorov, 2015; Provorov and Andronov, 2016). It is noteworthy that when compared to other well-known rhizobia, Bradyrhizobium displays several particular genomic and physiological characteristics related to diazotrophy and symbiosis. For example:

Bradyrhizobium japonicum strains have some of the largest bacterial chromosomes sequenced to date (9.1–9.6 Mb), and the largest of all rhizobia (Kündig et al., 1993; Kaneko et al., 2002, 2011; Batut et al., 2004; Siqueira et al., 2014).

All angiosperms that perform symbiotic nitrogen-fixing symbioses (except Gunnera) are included in the Rosid I clade (Soltis et al., 2000). This clade includes actinorhizal plants and plants that are nodulated by rhizobial bacteria. Recent phylogenetic and molecular data suggest that these nitrogen-fixing plants are derived from a common ancestor of the Rosid I clade with a genetic predisposition for nodulation (Soltis et al., 1995; Pawlowski and Sprent, 2008; Hocher et al., 2011). It was proposed that rhizobial symbioses has evolved four times independently within the Rosid I clade, three times for legumes and once for Parasponia (Doyle, 1998; Pawlowski and Sprent, 2008; Sprent, 2008). More recently, it was postulated that there might have been six to seven separate origins of nodulation in legumes (Doyle, 2011).

All plants nodulated by rhizobia are included in the family Leguminosae, except Parasponia. Leguminosae comprises more than 700 genera and about 20,000 species (Doyle, 2011), divided into three subfamilies: Caesalpinioideae, Mimosoideae, and Papilionoideae, although the legume taxonomy is currently under revision (Sprent et al., 2017). A key evolutionary study of the Leguminosae family has been performed taking into account molecular and fossil data (Lavin et al., 2005), concluding that legumes evolved about 60 Mya. It was postulated that nodulation could have developed due to an important climatic change at that time, involving an important increase in CO₂ levels that made nitrogen limiting for plant growth (Sprent, 2007). A crucial first step in rhizobia-legume symbiosis is the capacity for mutual recognition and it is thought that this capacity derived from ancient arbuscular mycorrhizal symbiosis (Szczyglowski and Amyot, 2003). In fact, arbuscular mycorrhizal fungi secrete soluble LCO signals (Gough and Cullimore, 2011; Maillet et al., 2011) that are essential for arbuscular mycrorrhiza development in legumes, indicating there is a common signaling pathway for both rhizobia-legume and arbuscular mycorrhizal symbioses (Capoen et al., 2009; Markmann and Parniske, 2009; Genre and Russo, 2016).

The macrosymbiont determines the mode of root infection by rhizobia, and the structure and morphology of the nodule. The way of infection has been related to the evolution of legume nodulation, while the structure and morphology of nodules are different among legume clades and may be markers of legume phylogeny (Sprent, 2007, 2009; Sprent et al., 2013). It was considered that the infection processes and nodule structure are more important taxonomic characteristics of legumes than their ability or inability for nodulation (Sprent et al., 2017). An evolutionary scheme of the different rhizobia infection types and the nodule structure of extant legumes has been proposed (for details of this scheme and for examples of the legume nodules on which the model is based see: Sprent, 2007, 2008, 2009; Sprent and James, 2007; and Ibáñez et al., 2017). In this scheme the origin of rhizobia infection could either be through direct epidermal infection or crack infection, which would produce two distinct branches of nodule evolution (Figure 4). The more complex evolutionary line involves the formation of transcellular ITs and their entry into some daughter cells of the meristem. In a further evolutionary step, bacteria could be retained in a modified IT (no bacteria released into the host cell and consequently, no symbiosome is formed), as observed in Caesalpinioideae and Papilionoidae legumes. Alternatively, bacteria could be released into the host cell to form the symbiosome. The infection of root hair would be a later, key event in the evolution of the determinate and indeterminate nodules found in Mimosoidae, and in some Papilionoidae and Loteae legumes. All nodules originated in this evolutionary line contain infected and uninfected cells in their nitrogen-fixing zone. About 75% of nodulated legumes, including almost all mimosoids and Caesalpiniodeae, and more than 50% of papilionoids, would have followed this strategy.

In the other branch of nodule evolution a few cells are infected by rhizobia and they divide repeatedly (Figure 4). The bacteria enter the host cytoplasm in symbiosomes but not via an IT because no such structure is formed. The most distinctive structural feature of these nodules is that the infected zone is composed of only infected cells. Nodules evolved in this way are only found in Papilionoidae legumes and they include the determinate dalbergoid nodules (crack infection, aeschynomenoid nodules), and those of many Genistae and some Crotalarieae legumes (epidermal infection and some infected cells with meristematic activity: indeterminate nodules and lupinoid nodules).

The Papilionoid crown node arose about 58 Mya, while the genistoid and dalbergioid nodes date to about 56 and 55 Mya, respectively. In comparison, galegoid legumes (a clade that includes Medicago, Vicia, and Pisum) began their spread about 39 Mya and thus, it is the genistoid and dalbergioid that have the oldest origin within the papilionoids (Sprent, 2007; Hane et al., 2017). All legumes that originated later than 40 Mya form their nodules by root hair infection (Sprent, 2009).

In the framework of this review, it is interesting to note some features of the genistoid legume lupin. The Lupinus genus includes about 300 species that can be found all over the world. Although they predominantly exist on the American continent and in the Mediterranean area, some Mediterranean species have been introduced into Australia and South Africa. Lupin species colonize different environments and they have particular agronomic potential as they are more tolerant to certain abiotic stresses than other legumes (Fernández-Pascual et al., 2007). These legumes can grow in nitrogen and phosphate depleted soils, and their capability to exploit poor, degraded, contaminated or stress-affected soils, and produce safe, protein-rich seeds make Lupinus a legume of great interest (Lucas et al., 2015). The Lupinus genus has the fastest evolution rate in plants and species from the Andes evolved less than 2 Mya (Hughes and Eastwood, 2006). Moreover, it is the only legume genus known to be unable to establish mycorrhizal symbiosis. A draft genome sequence of L. angustifolius was recently obtained (Hane et al., 2017), showing that all mycorrhiza-symbiotic specific genes have been lost, although this species has retained genes commonly required for mycorrhization and for nodulation. The lupin nodule has unique peculiarities (lupinoid) in which a lateral meristem allows the nodule to grow and surround the root (Figure 1). Beyond Lupinus spp., this type of nodule has only been found in some species of Listia to our knowledge (Yates et al., 2007; Ardley et al., 2013; Sprent et al., 2017). Using L. albus and Bradyrhizobium as a model, we described the mode of rhizobia infection of lupin roots and other early steps of nodule development in detail (González-Sama et al., 2004). Bacteria infect the root intercellularly, at the junction between the root hair base and an adjacent epidermal cell, and they invade a sub-epidermal outer cortical cell through structurally altered cell wall regions. This infected cell divides repeatedly and together with uninfected dividing cells, the nodule primordium is formed. Thus, the infected zone of the nodule originates through the division of a single infected cortical cell and therefore, the central zone of the lupin nodules has no uninfected cells.

Despite the advantages associated with the colonization of nitrogen poor environments, the ability of many legumes to nodulate may have evolutionary benefits in terms of alleviating abiotic stress. However, this issue has been little explored. Accordingly, a range of nodulated legumes are found in desert ecosystems and in high altitude areas, suggesting that nitrogen-fixing symbiosis confers an advantage in these ecosystems (Sprent and Gehlot, 2010). Nitrogen-fixing legumes make more efficient used of the available water and their fitness is enhanced in arid and semi-arid climates relative to non-fixing plants (Adams et al., 2016). Some putative adaptations of symbiosis to the environment have been reported and for example, some Mimosa species prefer to nodulate with certain rhizobia species rather than others, a preference that may be influenced by soil fertility and pH (Elliot et al., 2009; Garau et al., 2009). The semiaquatic legume Sesbania rostrata displays phenotypic plasticity for legume nodulation driven by environmental conditions. Thus, Sesbania can develop nodules of the indeterminate or determinate type depending on the environmental conditions (Fernández-López et al., 1998). Similarly, rhizobia infection is via an IT in non-flooding conditions whereas flooding switches the infection mechanism to crack entry, favoring nodulation in conditions of water stress in this legume (Goormachtig et al., 2004). On the other hand, the mode of infection may also be determined by the rhizobia in certain legumes. For example the intercellular via was used by a S. fredii strain in Lotus burttii (Acosta-Jurado et al., 2016) as well as by a strain of R. leguminosarum (Gossmann et al., 2012), whereas a M. loti strain enters by IT (Gossmann et al., 2012).

Several biochemical, genetic, and proteomic studies have set out to characterize the composition of the symbiosome (or peribacteroid) membrane and the peribacteroid space (Whitehead and Day, 1997; Panter et al., 2000; Hinde and Trautman, 2002; Saalbach et al., 2002; Wienkoop and Saalbach, 2003; Catalano et al., 2004; Limpens et al., 2009; Clarke et al., 2014, 2015; Emerich and Krishnan, 2014).

Some membrane microdomain-associated proteins can be found in the SM and they seem to play a key role in the regulation of the nodulation process. Flotillins are markers for membrane microdomains called “lipid rafts.” Flotillin genes are induced during early nodulation events in M. truncatula (Haney and Long, 2010). Some of these proteins are involved in infection thread invagination and elongation and they could be involved in endocytosis and trafficking of bacteria and nodule organogenesis (Haney and Long, 2010). Flotillin-like genes are induced in soybean nodules (Winzer et al., 1999) and flotillin-like peptides have been identified and isolated from SM of soybean and pea nodules (Panter et al., 2000; Saalbach et al., 2002). A remorin gene encoding another membrane microdomain-associated protein (MtSYMREM1) is specifically and strongly induced during the rhizobial infection and nodule organogenesis of M. truncatula (Lefebvre et al., 2010). This protein was located in plasma membrane of ITs and in the SM and may be a scaffolding protein required for infection and bacterial release into the host cytoplasm (Lefebvre et al., 2010). FWL1 is another interesting membrane microdomain-associated protein identified in soybean symbiosomes (Clarke et al., 2015). FWL1 interacts with remorins, flotillins and other proteins associated with membrane microdomains, regulating legume nodulation (Qiao et al., 2017).

Even at early stages of formation the SM has particular characteristics (Whitehead and Day, 1997), and both the composition and the function of the SM change as it develops (Hinde and Trautman, 2002). In principle, the SM is derived from the plant cell membrane and several plasma membrane markers can be found in the peribacteroid membrane, such as a plasma membrane H⁺-ATPase (Wienkoop and Saalbach, 2003) and the SNARE (N-ethylmaleimide-sensitive factor attachment protein receptor) protein SYP132 (Catalano et al., 2007; Limpens et al., 2009). It is noteworthy that the activation of H⁺-ATPases was also detected in the arbuscular membrane at the AM symbiosis interface (Harrier, 2001).

Symbiosome formation and division induces the activation of the endomembrane system of the host cell (Roth and Stacey, 1989), and it has been proposed that the endoplasmic reticulum (ER) and Golgi vesicles fuse with the SM (Whitehead and Day, 1997; Ivanov et al., 2010; Gavrin et al., 2017). Several proteins from the endomembrane system can be detected in the SM (e.g., cytochrome P450 and a luminal binding protein), and calreticulin, a disulphide-isomerase protein, and some chaperonin-like proteins of the ER have also been identified in symbiosomal fractions and they are probably located in the symbiosome lumen (Saalbach et al., 2002; Wienkoop and Saalbach, 2003; Catalano et al., 2004; Verhaert et al., 2005). Other endomembrane-related proteins in the symbiosome are annexin and syntaxin, which are involved in vesicle transport and secretion, as well as small GTPases involved in the regulation of membrane fusion (Wienkoop and Saalbach, 2003; Catalano et al., 2004; Limpens et al., 2009; Ivanov et al., 2012; Gavrin et al., 2017). It is interesting to note that many of these ER and Golgi proteins, as well as small Rab7 GTPases, have also been found in phagosomes, an organelle compartment of macrophages (Garin et al., 2001; Verhaert et al., 2005), suggesting that symbiosome and phagosome membranes may form in a similar way. Carbohydrate epitopes associated with Golgi-derived glycoproteins and glycolipids have been identified in the inner face of the SM (Perotto et al., 1991). These glycoconjugated molecules, collectively known as the glycocalyx, are involved in physical interactions with the bacterial surface inside the symbiosome and they are important in symbiosome development (Bolaños et al., 2004).

Tonoplast proteins have also been identified in the SM, including a vacuolar H⁺-pyrophosphatase, a vacuolar type H⁺-ATPase (V-ATPase) and an intrinsic tonoplast protein of the Nod26 group (Saalbach et al., 2002; Wienkoop and Saalbach, 2003; Catalano et al., 2004). The presence of active H⁺-ATPases in the SM drives proton accumulation and the establishment of a membrane potential (Whitehead and Day, 1997; Fedorova et al., 1999; Hinde and Trautman, 2002; Clarke et al., 2014). A vacuolar cysteine protease that could be involved in protein turnover and/or the adaptation to changes in cell turgor was also identified in the symbiosome lumen (Vincent and Brewin, 2000; Vincent et al., 2000). This cysteine protease is also involved in nodule organogenesis and function (Sheokand et al., 2005). The vacuolar SNAREs SYP22 and VT111 were also found in senescent symbiosomes (Limpens et al., 2009; Emerich and Krishnan, 2014; Gavrin et al., 2014).

Several proteins originating from mitochondria and chloroplasts are also associated with the SM. Among the chloroplast proteins identified are the peripheral membrane protein F1 ATPase α- and β- subunits, the chloroplast outer envelope protein 34 and a chloroplast nucleoid DNA-binding protein. Mitochondrial membrane proteins have also been found, such as a membrane anion channel (porin) and a nucleotide translocator (malate dehydrogenase), as well as mitochondrial processing peptidases, probably located in the symbiosome lumen (Panter et al., 2000; Saalbach et al., 2002; Wienkoop and Saalbach, 2003; Catalano et al., 2004). Bacterial proteins can also be detected in SMs and the peribacteroid lumen, including several nitrogenase components, chaperones, the α-subunit of bacteroid ATP synthase and others (Whitehead and Day, 1997; Saalbach et al., 2002; Catalano et al., 2004; Emerich and Krishnan, 2014).

The SM is a regulated interface with a key role in nutrient exchange between both symbiotic partners, and different types of proteins and transporters are specifically located at this membrane (White et al., 2007; Clarke et al., 2014; Emerich and Krishnan, 2014). The SM has specific integral membrane proteins, such as nodulin 24 (a glycine-rich protein; Sandal et al., 1992), nodulin 26 (an aquaporin; Dean et al., 1999), and others (Clarke et al., 2015). The sulfate transporter gene (Sst1) that is expressed in a nodule-specific manner in Lotus japonicus, is essential for nodule symbiosis (Krusell et al., 2005). This transporter seems to reside in the SM (Wienkoop and Saalbach, 2003) and it is thought to transport sulfate from the plant cell cytoplasm to the bacteroids (Krusell et al., 2005). Similarly, a proteomic analysis of the SM from nodules of L. japonicus revealed the presence of a putative sucrose transporter of the SUC family (Wienkoop and Saalbach, 2003). More recently, another sucrose transporter (MtSWEET11) was proposed to be located at the symbiosome membrane in M. truncatula nodules (Kryvoruchko et al., 2016), suggesting the possible transport of sucrose toward the rhizobia. However, specific transporters for some crucial molecules for nitrogen fixation seem not to be located in the SM. For example molybdenum is a key element for the bacteroidal nitrogenase but the molybdate transporter has not been identified in the SM (Tejada-Jiménez et al., 2017). The sulfate transporter Sst1 (Krusell et al., 2005) could be involved in molybdenum delivery to the symbiosome, as some sulfate transporters can transfer molybdate across membranes (González-Guerrero et al., 2016). Similarly, specific ammonium transporters have not yet been identified in the SM. Although a symbiotic ammonium transporter1 (SAT1) was seen to localize to the SM (Kaiser et al., 1998), it was recently shown that this protein to actually be a membrane-localized basic helix–loop–helix DNA-binding transcription factor involved in ammonium transport (Chiasson et al., 2014). However, ammonium may enter the symbiosome via the aquaporin-like nodulin 26 channel, or through a cation channel that transports K and Na, as well as by diffusion (Tyerman et al., 1995; Hwang et al., 2010; Courty et al., 2015).

The roles and functions of several proteins located at the symbiosome membrane and the peribacteroid space remain unknown (Kereszt et al., 2011; Emerich and Krishnan, 2014). However, the information available provides some markers of the symbiosome membrane identity. Evidence suggests that secretory pathways play an important role in the formation of the symbiosome and perimicrobial compartments, i.e., an exocytosis-related pathway already present in arbuscular mycorrhizal symbiosis. In fact, an exocytotic pathway for endosymbiosis was defined (Ivanov et al., 2012), providing the first evidence that symbiosomes are generated through exocytosis and that they could therefore be considered apoplastic compartments rather than endocytotic compartments. Rhizobia are confined to plasma membrane protrusions, compartments that rapidly increase in surface area and volume due to microsymbiont expansion. Because the plasma membrane is not elastic and it is unable to stretch more than 3%, exocytosis of new membrane material is crucial to increase the membrane's surface area (Grefen et al., 2011). Membrane fusion is achieved through the action of SNARE proteins in the targeted compartment (t-SNAREs) and the vesicle-associated membrane protein (VAMP or v-SNAREs) that form a SNARE complex, small GTPases of the Rab family that control the transport and docking of vesicles to their target membrane, and Ca²⁺-sensors from the synaptotagmin group involved in membrane repair (Catalano et al., 2007; Limpens et al., 2009; Ivanov et al., 2010, 2012; Wang et al., 2010; Gavrin et al., 2016, 2017; Harrison and Ivanov, 2017). Briefly, a plasma membrane t-SNARE (SYP123) is present in the SM throughout the life of the symbiosome (from when the rhizobia is released from the IT to symbiosome senescence) and only when the symbiosome has stopped dividing does the SM acquire a late endosomal/vacuolar marker (Rab7), which persists until senescence. At the onset of senescence, the SM acquires a lytic vacuolar identity due to the appearance of the two vacuolar t-SNAREs (SYP22S, VTI11). These SNAREs allow the symbiosome to fuse and form lytic compartments in which the rhizobia are eventually killed. On the other hand, transporters may have a third, new identity for SM (Emerich and Krishnan, 2014) and it could be speculated that a sulfate transporter like-Sst1 should be considered at this point.

The activity of the vacuolar H⁺-ATPase in the symbiosome membrane leads to the accumulation of protons, which should generate an acidic pH in the symbiosome (Whitehead and Day, 1997; Hinde and Trautman, 2002). Several symbiosome enzymes have an acidic optimum pH, including the proteases, acid trehalase, protein protease inhibitor, and alpha-mannosidase isoenzyme II that are typically found in vacuoles (Mellor, 1989; Panter et al., 2000). In fact, certain mutant and senescent bacteroids are degraded by these proteases and glycosidases, suggesting that the survival of these bacteroids is dependent on them avoiding acid digestion in the symbiosome compartment (Mellor, 1989; Parniske, 2000). As mentioned above, a functional cysteine protease with proteolytic activity has been characterized in the symbiosome lumen (Vincent and Brewin, 2000; Vincent et al., 2000). In 1989, it was proposed that since symbiosomes (which can be considered to be “temporary but independent organelles”) are morphologically different from the plant central vacuole, they may represent organ-specific modifications of lysosomes, analogous to the protein bodies of seeds (Mellor, 1989). Nitrogen activity counteracts the tendency of the ATPase to acidify the lumen of the symbiosome and thus, if the bacteroids stop fixing nitrogen the pH will drop to a level that favors the lysis of the symbiosome. Again, this phenomenon would support the notion of the symbiosome as a modified lysosomal compartment (Brewin, 1991; Hinde and Trautman, 2002).

Symbiosomes do not fuse with lytic vacuoles but they remain as individual units within the cytosol. In fact, it was suggested that vacuolar formation is altered in nodule infected cells in order to allow the expansion of the bacteria in the cytoplasm (Gavrin et al., 2014). Indeed, the vacuoles in infected cells are non-functional and have a neutral pH, or they are degraded (Gavrin et al., 2014, 2016). This facilitates the maintenance of symbiosomes as individual nitrogen-fixing organelles (Limpens et al., 2009; Emerich and Krishnan, 2014).

Rab7 GTPase is thought to be required for the formation of lytic compartments in different organisms (Bucci et al., 2000). In nodules, the plant late endosomal marker Rab7 has been localized in symbiosomes after division stops and it persists until the symbiosome reaches the senescence stage. Therefore, it seems to be involved in symbiosome maintenance (Cheon et al., 1993; Son et al., 2003; Limpens et al., 2009; Clarke et al., 2015). Symbiosome senescence occurs when symbiosomes fuse and form lytic compartments (Hernández-Jiménez et al., 2002; Van de Velde et al., 2006). During senescence, symbiosomes acquire a lytic vacuolar identity, evident through the presence of vacuolar SNAREs and the vacuolar proteins of the HOPS complex at the symbiosome membrane, making it competent for trafficking similar to that of a lytic vacuole (Gavrin et al., 2014).

It has been postulated that metabolic innovations may be important for organelle-producing endosymbiosis (O'Malley, 2015). Rhizobia-legume symbiosis depends on the highly regulated exchange of carbon and nitrogen sources, and nutrients, across the bacteroid and SMs. Specific transporters in these membranes that are critical for symbiosis have been identified through transcriptome and proteome analyses (Udvardi et al., 1988; Vincill et al., 2005; White et al., 2007; Clarke et al., 2014). Most rhizobial species only exhibit highly efficient nitrogen fixation when they are endosymbiotic in the host nodule cells. This suggests that the host plant controls rhizobial nitrogen fixation. It was reported that the host plant has overcome the lack of a bacterial gene necessary for symbiotic nitrogen fixation, a homocitrate synthase gene, a key genetic adaptation needed to establish efficient nitrogen-fixing symbiosis in legumes and rhizobia. In L. japonicus, a legume host nodule-specific homocitrate-synthase is exclusively expressed in infected cells and it supplies homocitrate to the symbiosome. This tricarboxylic acid is an essential component of the iron-molybdenum co-factor of nitrogenase, although it is not itself required for plant metabolism and it is absent from almost all rhizobia species. This homocitrate makes the nitrogen-fixing activity of the endosymbiont possible and it represents an example of the co-evolution of metabolic pathways in the two symbiotic partners (Hakoyama et al., 2009; Terpolilli et al., 2012). It is interesting to note that photosynthetic bradyrhizobia interacting with Aeschynomene legumes can synthesize bacterial homocitrate for free-living and symbiotic nitrogen fixation, and that the plant enzyme is not usually induced. A. caulinodans, which form nodules with S. rostrata, also has this enzyme. These data suggest that different rhizobia-legume symbioses could have co-evolved differently.

A complex amino acid cycle has been observed in pea nodules, whereby the plant cell supplies amino acids to the symbiosome, which can shut down nitrogen fixation, and in return the latter acts like a plant organelle supplying amino acids back to the plant cell for asparagine synthesis. It has been postulated that this exchange induces mutual dependence, preventing the symbiotic relationship from being dominated by the plant and generating selective pressure for the evolution of mutualism (Lodwig et al., 2003). Further studies into amino acid metabolism suggest that symbiosomes in the indeterminate nodules of pea (carrying Rhizobium leguminosarum bv. viciae as a microsymbiont) and alfalfa (E. meliloti), and in the determinate nodules of soybean (Bradyrhizobium japonicum), display metabolic dependence on the host for branched-chain amino acids (Prell et al., 2009, 2010; Dunn, 2014). Thus, symbiosomes become symbiotic auxotrophs and they behave like facultative plant organelles. It was suggested that this enabled the plant to control the degree of bacterial infection (Prell et al., 2009, 2010; Terpolilli et al., 2012; Haag et al., 2013).

Nitrogen fixation is uncoupled from bacterial nitrogen stress metabolism in rhizobia-legume symbiosis, such that bacteria generate “excess” ammonia and release this ammonia to the plant, a case of metabolic integration in this symbiosis (Yurgel and Kahn, 2008). The switching to ammonia synthesis by symbiosomes is accompanied by the switching off of ammonia assimilation into amino acids (Patriarca et al., 2002). Because mature bacteroids deplete nitrogen and release ammonia to the plant without assimilation, it was proposed they could be considered as ammoniaplasts (Oldroyd et al., 2011; Downie, 2014).

The appearance of an organelle-specific protein import mechanism is considered a key step in the conversion of a symbiont into a permanent organelle (Cavalier-Smith and Lee, 1985; Cavalier-Smith, 1992; Theissen and Martin, 2006; Archibald, 2015). Indeed, chloroplasts and mitochondria have developed the specific TIC/TOC and TIM/TOM protein transport systems, respectively. The presence of a signal peptide specific for protein targeting is a distinctive trait of cell organelles. Although strictly referring to targeting in order to reimport proteins back from organelle genes that were transferred to the nucleus, it is interesting to consider the specific targeting of protein products to symbiosomes as an organelle-related process. N-terminal sequence comparisons of some SM proteins, like nodulin 26B and HSP60, suggest that N-terminal signal sequences have been removed from these proteins (Panter et al., 2000). Mitochondrial processing peptidases, homologs of which have been identified in the symbiosome, catalyse the cleavage of leader peptides in precursor proteins, although their function in symbiosomes remains unknown (Catalano et al., 2004). The N-terminal processing of proteins may target them to the symbiosome (Panter et al., 2000; Catalano et al., 2004), although these proteins might be targeted to the ER or Golgi, loosing their signal peptide and later being delivered to the SM via the endomembrane system (Panter et al., 2000). A N-terminal signal peptide in nodulin MtNOD25 specifically translocates this protein to the symbiosomes (Hohnjec et al., 2009), the first clear role for a signal peptide in protein targeting to the symbiosome in nodule infected cells. Other nodulins and calcium-binding proteins from Medicago, Vicia, and Lupinus carry signal peptides (Hohnjec et al., 2009; Meckfessel et al., 2012), although no conserved N-targeting signal for SM or symbiosome space proteins has yet been identified. Moreover, these symbiosome targeting signal peptides cannot account for the majority of proteins identified in symbiosomes (Hohnjec et al., 2009). Thus, other targeting systems must be available for protein translocation to the symbiosome (Catalano et al., 2004; Clarke et al., 2014).

Vesicle trafficking to the symbiosome via the endomembrane system is not fully understood. It has been postulated that protein delivery to the symbiosome relies on the plant secretory system (Catalano et al., 2007; Limpens et al., 2009; Ivanov et al., 2010; Maunoury et al., 2010; Mergaert and Kondorosi, 2010; Wang et al., 2010) and it is interesting that proteins lacking plastid-targeting signals might also be targeted to the chloroplast via the secretory system (Bhattacharya et al., 2007 and references therein; Mergaert and Kondorosi, 2010). The syntaxin SNARE SYP132, which localizes to the SM (Catalano et al., 2004), may be involved in site-specific vesicle fusion for the delivery of cargo vesicles to the SM (Catalano et al., 2007). Indeed, some tonoplast proteins involved in symbiosome maturation appear to be retargeted to the symbiosome by a mechanism that involves membrane fusion, as observed in infected cells of Medicago truncatula nodules (Gavrin et al., 2014, 2017).

In M. truncatula, the DMI2 gene that encodes a receptor kinase plays a critical role in the Nod factor signaling cascade during the early stages of nodulation, and it is also a key regulator of symbiosome formation, allowing bacteria to be released from the infection thread into the host cell. In nodules, this kinase is found in the host cell plasma membrane and in the membrane surrounding the ITs. If DMI2 expression is compromised in plants, infected nodule cells are occupied by large intracellular ITs that do not release the bacteria rather than organelle-like symbiosomes, a phenotype that is reminiscent of the nodules of primitive legumes and Parasponia (Limpens et al., 2005; Op den Camp et al., 2011).

In galegoid legumes of the Inverted Repeat Lacking Clade (IRLC), all of which form indeterminate nodules (like Medicago and Pisum), a legume family of nodule-specific cysteine-rich (NCR) peptides are targeted to the endosymbiotic bacteria. These peptides are responsible for the bacteroid differentiation that involves the induction of endopolyploidy, cell cycle arrest, terminal differentiation, and a loss of bacterial viability. It was recently demonstrated that a nodule specific thyoredoxin (Trx s1) is targeted to the bacteroid, controlling NCR activity and bacteroid terminal differentiation (Ribeiro et al., 2017). The NCR gene family is estimated to have appeared between 51 and 25 Mya, the time at which IRLC legumes separated from the other legumes (Lavin et al., 2005; Alunni et al., 2007; Yokota and Hayashi, 2011).

All IRLC species tested induce terminal differentiation of their rhizobia endosymbionts, resulting in different morphotypes. NCR genes were also identified in all these species, although the number of NCR peptides was highly variable, ranging from over 630 in M. truncatula to only 7 in the most basal IRLC legume Glycyrrhiza uralensis (Montiel et al., 2016, 2017). The nodules of this latter legume lack cationic NCR peptides, which could indicate that the ancestral NCRs were neutral or anionic and that they originated from a single evolutionary event in IRLC legumes (Montiel et al., 2017).

It was proposed that the differentiated polyploid bacteroids might have a more efficient metabolism, like polyploid eukaryotic cells (Van de Velde et al., 2010). NCR peptides are derived from antimicrobial, defensin-related peptides, and these antimicrobial peptides have different mechanisms of action and drive different states of bacteroid differentiation (Haag et al., 2013; Maróti and Kondorosi, 2014; Pan and Wang, 2017). This may be an evolved mechanism by which the host legume dominates microsymbiont proliferation (Mergaert et al., 2006; Mergaert and Kondorosi, 2010; Van de Velde et al., 2010; Maróti and Kondorosi, 2014; Yang et al., 2017). NCR peptides optimize bacteroid metabolism and the nitrogen fixation process (Van de Velde et al., 2010), and they control discrimination against incompatible microsymbionts (Yang et al., 2017). It has also been suggested that this control of bacteroid proliferation by the host plant can avoid the spreading of rhizobia to tissues other than the nodule (Mergaert et al., 2006).

Until recently, it was thought that bacteroids of non-galegoid, non-IRLC legumes, do not undergo terminal differentiation nor is their replication restricted. Indeed, they are comparable to free-living bacteria in cell size, DNA content and proliferation (Mergaert et al., 2006). It is noteworthy that in indeterminate nodules of the mimosoid legume Leucaena glauca elicited by Bradyrhizobium, no NCR peptides have been detected and the bacteroids display a moderate differentiation phenotype; it is an “intermediate” state relative to that of IRLC and non-IRLC legumes with determinate nodules (Ishihara et al., 2011). The presence of swollen (differentiated) bacteroids has been noted in five out of the six major papilionoid subclades, although each of these subclades also includes species with non-swollen or non-differentiated bacteroids (Oono et al., 2010). Moreover, there was no consistent relationship between nodule type and the host's effects on bacteroid differentiation. Accordingly, it would appear that legumes inducing bacteroid differentiation have evolved independently on five occasions from an ancestral papilionoid legume that hosts non-swollen (non-differentiated) bacteroids (Oono et al., 2010). This repeated evolution of the host's legume traits suggests a possible advantage for the plant in terms of fitness. It has been hypothesized that differentiated bacteroids fix nitrogen more efficiently than non-differentiated bacteroids (Oono et al., 2009, 2010). In fact, Oono and Denison (2010) demonstrated that swollen bacteroids confer net benefits to the host legume due to their more efficient nitrogen fixation and the higher return on the cost of nodule construction (host biomass growth per total nodule mass growth).

It was recently shown that NCR antimicrobial peptides are involved in the permeability of the SM to diverse metabolites. NCR peptides might contribute to metabolic integration between the symbiosome and plant host and in the past, similar antimicrobial peptides may have contributed to the metabolic integration and organellogenesis of mitochondrial and plastid ancestors (Mergaert et al., 2017). This hypothesis emphasizes the importance of metabolic integration in organelle development (see O'Malley, 2015). It was recently discovered that nodules of dalbergioid legume species of the Aeschynomene genus (which establish symbiosis with Bradyrhizobium spp.) carry polyploid and enlarged bacteroids, and that these plants also express NCR peptides. However, these peptides are not homologous to NCR peptides from IRLC legumes, suggesting an independent evolutionary origin (Czernic et al., 2015).

New plant and bacterial factors that induce bacteroid differentiation remain to be identified (Mergaert et al., 2006; Oono and Denison, 2010; Oono et al., 2010; Van de Velde et al., 2010; Ishihara et al., 2011). A bacterial conserved BacA (bacteroid development factor A) protein that forms an ABC transporter system is produced by rhizobia, and it is required for bacteroid development and survival in IRLC and Aeschynomene legumes. BacA may protect rhizobia and bacteroids from the antimicrobial activities of NCR peptides, antagonizing NCR peptides, or it may be involved in the uptake of these antimicrobial peptides by bacteroids (Haag et al., 2013; Guefrachi et al., 2015; Pan and Wang, 2017, and references therein).

It is now assumed that the fate of bacteroids is controlled by the host plant (Mergaert et al., 2006; Maróti and Kondorosi, 2014), although some data suggest that a particular genotype of the microsymbiont might be required, most probably related to their surface polysaccharides. Terminal bacteroid differentiation of Ensifer fredii strain HH103 does not take places in nodules of the IRLC legume Glycyrrhiza uralensis, (Crespo-Rivas et al., 2016), whereas it does occur when Mesorhizobium tianshanense forms the nodules (Montiel et al., 2016). Notably G. uralensis is the IRLC legume with the fewest NCR peptides reported to date (Montiel et al., 2017).

Interestingly, species within the genus Lupinus may host either swollen (L. angustifolius) or non-swollen (L. albus, L. diffuses, and L. bicolor) bacteroids, suggesting that the effects on bacteroid differentiation might have changed during the evolution of the Lupinus genus (Oono et al., 2010). Thus, it is possible that the host legumes have regained non-differentiated bacteroids in these latter three species, because bacteroid differentiation is no longer beneficial (for some unknown reason). Alternatively, some rhizobial strains that nodulate Lupinus may have evolved traits to overcome host-induced swelling and the loss of reproductive viability (Oono et al., 2010). To our knowledge, there is no data currently available about NCR peptides or any other similar molecules in the nodules of Lupinus (or in other legume nodules with dividing infected cells).

Infected nodule cells are usually post-mitotic and do not divide further. However, one of the most interesting and quite unusual traits for eukaryotic cells is found in certain legume nodules whose host cells can divide after being infected by rhizobia (Figures 2, 3). The division of infected cells containing symbiosomes has been observed in nodules of Lupinus spp. and Genista tinctoria (genistoid legumes), and also in certain dalbergioid legumes (e.g., Arachis hypogea, Stylosanthes spp., Sarothamnus scoparius). All these legumes are infected by Bradyrhizobium spp. through epidermal infection or crack infection, and the infected zone of their nodules has no uninfected cells (Chandler, 1978; Chandler et al., 1982; Sprent and Thomas, 1984; Tang et al., 1993; Lotocka et al., 2000; Sajnaga et al., 2001; González-Sama et al., 2004; Kalita et al., 2006; Fedorova et al., 2007). Nodules elicited by Bradyrhizobium in the genistoid Chamaecytisus proliferus (renamed as Cytisus proliferus) also contain dividing infected cells (Vega-Hernández et al., 2001). This is the only elongated indeterminate nodule reported to date without uninfected cells in the central infected zone. Root infection of this legume occurs by a singular intercellular mechanism, and ITs are aborted and do not contribute to infection (Vega-Hernández et al., 2001). Therefore, the division of infected cells appears to be a trait restricted to nodules in which infection is independent of ITs, rather than it being influenced by the type of nodule growth (determinate/indeterminate).

Mitochondria and plastids divide in the plant cytoplasm, and cytoskeletal elements not only secure their distribution and movement but also, their correct partitioning between the daughter cells at cytokinesis (King, 2002; Sheahan et al., 2004). Symbiosomes also have the ability to divide in the host cytoplasm (Figure 1), and the accommodation of endosymbionts in host cells involves microtubule and actin microfilament rearrangements (Whitehead et al., 1998; Davidson and Newcomb, 2001a,b; Fedorova et al., 2007; Timmers, 2008; Gavrin et al., 2015; Kitaeva et al., 2016). The conformation of the cytoskeleton in dividing infected cells of legume nodules has only been studied in L. albus (Fedorova et al., 2007). We showed that in the infected cells of L. albus nodules, symbiosomes are segregated equally between the two daughter cells when the host plant cell divides, just like other cell organelles, e.g., mitochondria (González-Sama et al., 2004; Fedorova et al., 2007). The cytoskeletal dynamics of infected nodule cells during the cell cycle appear to be relatively normal. In interphase cells, thick cortical arrays of microtubules form a radial network of strands perpendicular to the cell wall to facilitate the migration of organelles and symbiosomes toward the cell periphery (Fedorova et al., 2007). During cell division, symbiosomes concentrate at opposite poles of the cell and do not interfere with the arrangement of microtubules and microfilaments, segregating evenly between the two daughter cells. These cytoskeletal rearrangements in dividing infected cells, along with the detection of an antigen of the molecular motor myosin, suggests that lupin symbiosomes are in contact with and they are driven by the cytoskeleton. Thus, the positioning of symbiosomes in lupin nodule cells seems to depend on the same mechanisms used to segregate genuine plant cell organelles during mitosis (Fedorova et al., 2007). Therefore, in this regard the symbiosome displays significant organelle-like characteristics, unlike symbiosomes from nodules in which infected cells do not divide.

It has been established that a key event in the evolution from a free-living bacteria to an organelle is the loss of bacterial genes and their transfer to the nucleus of the plant host, a fate that occurred during mitochondrial and chloroplast evolution (Douglas and Raven, 2003; Archibald, 2015). In rhizobia-legume symbiosis, the presence of duplicated prokaryotic genes in the host genome has yet to be reported, although this possibility cannot be overlooked (Raven, 1993).

In the case of nitrogen-fixing rhizobia, the absence of gene transfer to the nucleus may be due to the low oxygen concentrations required by the nitrogenase enzyme, which would generate poor ROS production and mutation rates (Allen and Raven, 1996). Thus, mutation by ROS generation is unlikely to be an evolutionary driving force in the case of symbiosomes. However, strong ROS production has been detected in nodule host cells and in the symbiosome, the electron transport chain of bacteroids generates superoxide radicals and hydrogen peroxide (Matamoros et al., 2003). Oxidation of nitrogenase and ferredoxin in bacteroids also induces ROS generation (Matamoros et al., 2003). Lipid peroxidation has been detected in the SM during senescence (Puppo et al., 1991) and it could be due to the autoxidation of leghemoglobin (a protein controlling the accurate oxygen level in nodules) that is in direct contact with the SM, as well as to a decline in the activity of antioxidants like superoxide dismutase and catalase that are also present in the bacteroid (Puppo et al., 1991; Matamoros et al., 2003). Moreover, ROS generation induces ultrastructural alterations and senescence of symbiosomes (Puppo et al., 2005; Redondo et al., 2009).

No gene loss or genome reduction has been observed in viable symbiotic rhizobia. Symbiotic rhizobia that do not undergo terminal differentiation are still capable of existing as free-living bacteria. Accordingly, they must be equipped with a number of genes to survive in different environments and to compete with other microorganisms. Moreover, in nodules containing swollen terminally differentiated bacteroids, some non-differentiated bacteria inhabit the apoplastic space and consequently, all the genes necessary for independent life are still retained (Stêpkowski and Legocki, 2001). In fact, rhizobia underwent a genome expansion during evolution (MacLean et al., 2007).

An evolutionary pathway has been proposed in symbiotic systems to shift from free-living organisms to facultative symbiosis and to ecologically obligatory symbiosis, usually involving genome expansion (Provorov et al., 2008). The following step in this evolutionary pathway would be “genetically obligatory symbiosis,” which would involve microsymbiont genome simplification or reduction, and the last stage would be a new organism (Provorov et al., 2008). The availability of more recent molecular data from microbes has driven more in-depth studies into the evolutionary transitions in bacterial symbioses, including rhizobia-legume symbiosis (Sachs et al., 2011a,b). Based on phylogenetic analyses, it was hypothesized that transitions from horizontal to obligate vertical transmission of the microsymbiont are driven by the host, the partner that most benefits from these transitions (Sachs et al., 2011b).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The reviewer OS and handling Editor declared their shared affiliation.