Diazotrophic bacteria that form symbioses with legumes are referred to as rhizobia, and their presence in soils is widespread across the planet. Establishing the symbiotic relationship requires the plant root and bacteria to find each other in the soil, involving a molecular dialogue between host plant roots and rhizobia, which, in turn, induces host and bacterial processes associated with bacterial infection and nodule organogenesis. We suggest four principal steps in this process that usefully define the critical areas to consider when striving to transfer nitrogen fixation: signal perception, bacterial infection, nodule organogenesis, and the establishment of the environment for nitrogen fixation. How nodulation has evolved remains an ongoing discussion, but the latest work proposes a single origin in the evolution of nodulation, followed by multiple losses [15,16]. Associated with the multiple losses has been a consistent loss of only a very few genes: the Nod factor (NF) receptor NFP associated with rhizobial recognition; the master regulator of the root nodule symbiosis NIN and RPG, which is required for bacterial infection. This highlights a theme that runs throughout this review: Evolution of nodulation is not associated with massive increases in new genetic components; rather, a few key regulators allow novel networking of preexisting processes. These key regulators, which may be defined by those genes lost with loss of nodulation, match the processes we outline below.

Legumes appear to have evolved very stringent recognition of rhizobial-produced LCOs, which likely emerged following a whole genome duplication and further expansion of the LysM receptor-like kinase class [41]. Recent work has demonstrated the precise residues in the legume receptors that not only define an LCO-receptor, but also allow stringent recognition of specifically decorated LCOs [42], providing the framework for engineering such receptors in cereals. Activation of symbiosis signalling in legumes is sufficient to induce nodulation [27,28,43,44], something that does not appear to happen in cereals. Mycorrhizal activation of symbiosis signalling induces the expression of a mycorrhizal-specific transcription factor, REDUCED ARBUSCULAR MYCORRHIZA1 (RAM1), through the action of a transcriptional complex made up of CCaMK-CYCLOPS-DELLA proteins [45]. Similarly, the expression of the rhizobial-specific transcription factor NODULE INCEPTION (NIN) [30] is also activated through a similar CCaMK-CYCLOPS-DELLA complex [45]. How do legumes utilise the same signalling to promote quite different transcriptional outputs, and how is this specificity linked to the precise recognition of nitrogen-fixing bacteria? This question sits at the heart of utilising the preexisting signalling capabilities present in cereals, to allow the engineering of rhizobial recognition and the regulation of the relevant cellular and developmental processes to allow accommodation of bacteria in engineered cereals.

The perception of NFs plays an essential role in selecting the suitable symbiont for the legume host. However, some rhizobia can bypass this process and induce nodulation in a discrete number of legume species, in an NF-independent manner [46–51]. This does not negate symbiosis signalling; rather, an alternative mechanism to activate this signalling pathway appears to exist, and at least in one situation, this is the function of a bacterial effector delivered through the type III secretion system [51,52]. Understanding this NF-independent mechanism for activation of symbiosis signalling could provide an alternative approach to engineering the receptors in cereals to drive appropriate induction of symbiosis signalling, upon rhizobial contact.

The perception of NFs induces root hair tip growth reinitiation and curling, which encloses rhizobia and forms an infection chamber, also known as an infection pocket. The cell wall surrounding the infection chamber is degraded, which allows rhizobia to enter the root hair cell via plasma membrane invaginations. The continuous fusion of membrane vesicles at the infection site extends the membrane invagination and forms an intracellular tube, named the infection thread. The infection thread continuously elongates and branches through the root cortex, growing towards the nodule primordium. Once the infection thread reaches the nodule, “infection droplets” containing rhizobia are released into the nodule cells, resulting in membrane-bound symbiosomes, the organelle-like structures where nitrogen fixation occurs (Fig 2A).

Facilitating intracellular bacterial infection carries risks for the host plant because multiple different rhizobia present in the soil might also enter the plant tissues, including poor nitrogen fixers or potential pathogens. Therefore, mechanisms of selecting efficient symbiotic nitrogen-fixing bacteria strains are required for host plants to overcome this dilemma. During root hair infection, generally, single bacteria are entrapped, creating a clonal population of bacteria within the infection thread [55]. Subsequently, host plants can also apply conditional sanctions to inhibit the growth of underperforming nodules that are colonised by ineffective nitrogen fixers [56]. Notably, all these processes require the host plant to have the ability to recognise and distinguish specific rhizobia. Therefore, additional levels of stringency are attached to the recognition of bacteria before the infection is activated. Multiple studies demonstrate the importance of NF perception for the initiation of infection threads, and in some species, this stage requires a much higher degree of specificity for NF recognition [41,57]. Alternatively/in addition, other bacterial signals appear to play a role, with plant perception of exopolysaccharides on the surface of the rhizobia providing an additional level of stringency in bacterial recognition at this infection stage [58]. Two scaffold proteins, FLOTILLIN 4 (FLOT4) and the remorin protein SYMREM1, facilitate the formation of membrane nanodomains, reducing the mobility of the NF receptors, and this appears vital for appropriate rhizobial infection [59]. While the symbiosis signalling pathway is important for infection, there is little evidence that this transduction pathway is the primary link between receptor activation and the initiation of infection threads. Rather, a single calcium transient across the plasma membrane [60] that correlates with a burst of reactive oxygen species [61–63], appears associated with infection. Receptor-associated ROPs [64,65] activate Respiratory Burst Oxidase Homologs (RBOHs), and this is a likely mechanism for provoking cytoskeletal remodelling [66], initiating the infection thread [67] and regulating the dynamics and assembly of actin needed to redirect root hair polar growth [66,68]. ROPs also modulate actin dynamics via the SCAR/WAVE (Suppressor of cAMP receptor defect/WASP family verpolin homologous protein) complex [69–71], which allows actin nucleation, required for infection thread initiation [71–75] (details in Fig 2). Translating an activated microdomain of receptors at the site of bacterial attachment into an invagination of the plasma membrane requires physical changes to the membrane, and the first indicators of how this might happen has come with the description of a SYMBIOTIC FORMIN 1 (SYFO1) that provides a cell wall-plasma membrane-cytoskeleton continuum [76] (Fig 2) that can perhaps provide the initial scaffold to invaginate the membrane.

A mature cell wall is a barrier to the redirection of polar cell growth, and to achieve cell wall remodelling, the plasticity of the cell wall needs to be altered. Precisely, localised cell wall degradation at the site of the infection pocket is facilitated by directed secretion of NODULE PECTATE LYASE (NPL), which is transcriptionally induced by NFs and accumulates at the site of the infection pocket [53,77]. NPL degrades pectin, reinstating plasticity to the cell wall, which, in part, allows the formation of new sites for polar growth [78,79]. Analogous to an infection thread is the invasion of plant cells by mycorrhizal fungal hyphae. It has been demonstrated that during the formation of an arbuscule (Box 1), all vesicle transport is redirected to the sites of fungal penetration. As such, any secreted protein that is expressed during this time is targeted to the peri-arbuscular space [80]. Although not proven, an analogous situation could easily explain the unique proteins that are delivered to the growing bacterial infection thread: Any protein expressed at a time of infection thread growth, such as NPL, will, by default, be targeted to the tip of the growing infection thread [79].

Once initiated, progression of the infection thread appears to function in a manner analogous to the invasion of the plant cell by mycorrhizal fungal hyphae. Similar cellular structures have been reported during colonisation by both organisms [81,82], with the nucleus lining up to the site of infection and cytoplasmic bridges becoming apparent between the infection site and the nucleus, predicting the path of the infection structures. Beyond these commonalities in cellular structures are also commonalities in the genetic components or homologous genetic components required for the extension of both mycorrhizal fungal and rhizobial infection structures. These genetic components make up a protein complex, the infectosome, which forms at the tip of infection threads regulating exocytosis and is composed of VAPYRIN (VPY), the E3 ligase LUMPY INFECTION (LIN)/CERBERUS, and EXOCYST subunit H4 (EXO70 H4) [83–85]. The infectosome governs polar growth of infection threads by regulating the fusion of vesicles (Fig 2). In addition, the coiled-coil protein RHIZOBIUM-DIRECTED POLAR GROWTH (RPG) [15,86], which is essential for rhizobial infection, is located at both the perinuclear cytoplasm and the tip of the growing infection thread [79], providing a tantalising link to the nucleus that appears to guide the growth of the infection thread. Note that RPG is one of the few genes shown to be lost with the loss of nodulation, highlighting the central role it appears to play in allowing the existence of the RNS [15,16].

Intercellular rhizobial infection is observed in approximately 25% of legume species [87,88]. Here, the rhizobia enter the root tissue by degrading the middle lamella and primary cell wall and progressing through the intercellular space between epidermal and cortical cells (Fig 2B). In some interactions, rhizobia proliferate in the intercellular spaces and only become intracellular at the point of colonising the cells of the nodule [89,90]. In other cases, rhizobia induce cell death to form intercellular infection pockets in dead or collapsed cells, from which intracellular infection pegs or threads are initiated [89,90].

Intercellular infection appears simpler than intracellular infection through infection threads and thus is an attractive target for engineering, when considering transferring RNS to other crops. Since intercellular infection has been observed in the nitrogen-fixing lineages across the nitrogen-fixing clade: Fabales; Fagales; Cucurbitales; and Rosales, intercellular infection is considered a more ancient pathway for flowering plants that established a nitrogen-fixing symbiosis with bacteria [90], which likely evolved from intercellular diazotrophic endophytes [91,92]. Interestingly, recent studies have shown that Fabales and Fagales are sister clades [93,94], and the root hair–based intracellular infection likely evolved before their diversification because several reports indicate that the basal clades in Fabaceae [95,96] and several nodulating actinorhizal lineages in Fagales utilise intracellular infection mechanisms [97,98]. Therefore, the most recent common ancestor of Fabales and Fagales that accepted rhizobium as a microsymbiont might already have had the genetic framework to allow root hair–based intracellular infection [99]. Consequently, intercellular infection observed within legume lineages is currently considered a derived trait in Fabales. Most of these host plants in Cucurbitales and Rosales that show such alternative infection strategies are challenging genetic systems, making understanding this intercellular infection mode difficult. However, recent advances in studying species in Fabales and Fagales are changing this situation. The discovery of situations whereby the model legume Lotus japonicus undergoes intercellular rather than intracellular infection [100] has allowed a genetic dissection of this infection route [101], and the development of Aeschynomene evenia as a genetic model is also allowing the dissection of alternative means of rhizobial infection [48,49]. To date, these studies have only shown the importance of genetic components already known to be important for intracellular infection, but these platforms should soon allow novel genetic components to be discovered with specific functions in intercellular infection. Compared with intracellular infection, our understanding of intercellular infection is still relatively limited. It remains unclear whether intercellular infection will be easier to engineer, although it logically appears that way. Further studies and knowledge of these intercellular colonisation systems could provide a potential alternative path for crop engineering. This information might be precious, especially when our engineering target crop species are monocots, like maize, rice, and wheat, which are distantly related to the nitrogen-fixing clade of plants.

Nodule development must be temporally and spatially coordinated with bacterial infection, and underpinning this coordination is cell-to-cell communication linking activation of the symbiosis signalling pathway in the root epidermis to the promotion of cell divisions in the root cortex and pericycle [22,108]. NFs themselves seem unable to diffuse into the deeper cortical cell layers [109], and evidence supports the notion that a product of NIN in the epidermis may act as the mobile signal [22]. This could be cytokinin, and, indeed, genes involved in cytokinin biosynthesis and accumulation are induced in epidermal cells upon NF perception [110,111] (Fig 2C). However, recent work has demonstrated that infection by mycorrhizal fungi and treatments with elicitors from mycorrhizal fungi promote cell divisions in the cortex and the emergence of lateral roots in both legumes and cereals [81,112]. This suggests that the cell-to-cell communication linking symbiosis signalling at the root surface to the induction of root organogenesis may be another mechanism in nodulation derived from the more ancient mycorrhizal symbiosis. If correct, this places the novelty for nodulation being defined by the ability to amplify this signal and drive a novel aspect of development. Whether cytokinin or a component derived from the mycorrhizal symbiosis, the nature of this diffusible signal linking root tissues during nodule development, is both interesting and strategically important.

The unique ability of legumes to respond to cytokinin with the promotion of nodule initiation appears to be defined by the cytokinin regulation of NIN, which, in turn, recruits LATERAL ORGAN BOUNDARIES DOMAIN 16 (LBD16), a central regulator of lateral root development, to facilitate nodule organogenesis [113,114]. LBD16 coordinates the accumulation of a local auxin maximum, through the regulated expression of SHORT-INTERNODES/STYLISH (SHI/STY) transcription factors that, in turn, initiate expression of the YUCCAs, rate-limiting enzymes involved in auxin biosynthesis [113]. NIN also promotes the expression of Nuclear Factor-Y (NF-Y) subunit genes, such as NF-YA1 and NF-YB1 [115–121], which, in both plants and animals, play vital roles in the regulation of the cell cycle and cell proliferation [122,123]. These genes have also been demonstrated to further promote auxin biosynthesis during nodulation via up-regulation of SHI/STY [116,117,119] (Fig 3). The local accumulation of auxin at the site of nodule formation is further facilitated by the suppression of polar auxin transport [124,125], allowing accumulation of auxin within the local site of NIN induction (Fig 3). Auxin accumulation appears to be further promoted by auxin activation of LBD16 expression, creating a feedforward loop. These multiple avenues for NIN regulation of auxin accumulation allow the amplification of a local cytokinin signal into an auxin “hub,” creating a new developmental centre for de novo meristem formation.

While initiation of a nodule converges on multiple aspects of lateral root development [113,114], what emerges from a nodule meristem is quite different to a lateral root; in particular, nodules of legumes possess peripheral vascular strands and contain many cells able to harbour intracellular bacteria. The demonstration of a single origin for the evolution of nodules [15,16] suggests that the structure of nodules in non-legumes, so called actinorhizal species, may represent the primitive state of nodules: Such nodules possess a centralised vasculature [97,126], implying that the peripheral vasculature observed in legume nodules, probably evolved from an intermediary structure more analogous to a lateral root. This, alongside the genetic dissection of nodulation, suggests that lateral root development underpins the formation of a nodule, but this dictates the need for developmental regulators that impart a nodule identity onto the cells of the nodule primordia. NODULE ROOTs (NOOTs), orthologs of the BLADE-ON-PETIOLE1/2 (AtBOP1/2) genes encoding ankyrin repeat and BTB/POZ domain-containing cotranscriptional regulators in A. thaliana, are essential for maintaining nodule identity, after initiation: In their absence, nodules revert back to a lateral root [127,128] (Fig 3), with lateral roots emerging from nodules of noot mutants. This implies that NOOTs are required to maintain the identity of the nodule; however, the regulators that initially drive this nodule identity have yet to be defined.

Several reports have indicated that rice and Brachypodium can initiate nodule-like structures upon auxin treatment [129–131] but do not do so following treatments with cytokinins [132]. This supports the notion that NIN induction, as a function of cytokinin [106], is one of the key aspects of novelty within legumes. NIN promotes the initiation of a meristem, with many parallels to a lateral root, yet what emerges is developmentally quite different. We do not yet know the regulators that allow specific aspects of nodule development to emerge from a lateral root meristemic program. NOOTs are clearly a component of this nodule identity, but their principal function appears to be maintaining nodule identity, rather than activating it. There must be a developmental regulator, likely promoted by NIN, which acts in parallel to LBD16, to drive the novel development we see in a nodule. This novel regulator likely controls the expression of NOOTs. Understanding these regulator(s) is central to our ability to engineer this novel mode of root development into our target cereal crops.

Intracellular accommodation of bacteria provides a stable environment for nitrogen fixation. To accommodate the rhizobia, defence responses need to be suppressed, and the regulation of the cell cycle appears critical [104,139]. NIN suppresses the defence responses by regulating several genes that appear to function in the regulation of plant defences in the nodule: Defective in Nitrogen Fixation 2 (DNF2), Nodules with Activated Defence 1 (NAD1), and Symbiotic Cysteine-rich Receptor Kinase (SymCRK) [140–142]. The genetic characterisation of these genes implies a pathway functioning in the nodule associated with the regulation of plant defences, but the detailed mechanisms of how this pathway functions are unknown. Controlling endoreduplication in nodule meristematic cells has been shown to be essential for accommodating and maintaining rhizobia, with CCS52A playing a vital role in this control of the cell cycle and the production of polyploid cells [143]. It remains to be shown whether the polyploid nature of nodule cells and the down-regulation of plant defences alone are sufficient to allow intracellular bacterial accommodation. We need to know whether additional processes are essential for this accommodating state of nodule cells. Further, we need to understand the exact regulators that function in this late-stage maturation of a nodule, if that is more than NIN alone, to coordinate these processes, if we are to transfer this stable accommodation of nitrogen-fixing bacteria.

Transporting fixed nitrogen from the bacteria to the peribacteroid space may occur via diffusion or through protein channels [137]. An H⁺-ATPase on the symbiosome membrane pumps H⁺ into the peribacteroid space creating an acidic environment, which traps ammonium by protonating ammonia, producing ammonium cations. The down-regulation of a rhizobial ammonium transporter, AMT, might be a mechanism to prevent ammonium from flowing back into the bacteroid [144]. Ammonia and ammonium cations are then exported to the plant cytoplasm via two transporters on the symbiosome membrane: a voltage-activated monovalent cation channel [145] and nodulin 26 (NOD26) [146]. The cation channel is nonselective and can transport NH₄⁺, K⁺, and Na⁺. Whether NH₄⁺ specific channels are also located on this membrane is currently unknown. NOD26 is an aquaglyceroporin and can transport H₂O and NH₃ [146] and associates with glutamine synthetase (GS), which rapidly assimilates NH₃ into glutamine (Gln), creating a sink for NH₃ transport. Exported ammonia or ammonium cations can also be assimilated into glutamate (Glu) by glutamate synthase, transferring the amide group from Gln to α-ketoglutarate. Depending on the legume species and nodule types, Gln and Glu are converted into Asparagine (Asn) or Ureides for long-distance transport through the xylem [138]. Ureide Permeases in Phaseolus vulgaris [147] and in Glycine max [148] have been demonstrated on membranes of nodule cortical cells and vascular cells, and ship ureides into the xylem (Fig 4).

The last stage of nodule organogenesis is senescence, which plays a vital role in regulating the nodule nitrogen fixation function in response to ageing or environmental signals [149–151]. Nodule senescence leads to the disintegration of bacteroids and host plant cells and ends this symbiotic association. One of the most critical aspects of this senescence process is a rise in proteolytic activity. Previous studies have shown that different cysteine proteinases (CPs) are highly expressed in nodules at the senescence stage, especially a papain peptidase (CP6) and a vacuolar processing enzyme (VPE) [152]. Both their expressions and activities are increased during nodule senescence induced by abiotic stress or age [150–152]. The early expression of CP6 and VPE promotes senescence and serves as a marker for nodule senescence (Fig 4).

Though progress has been made, detailed mechanisms and critical regulators involved in nodule senescence are not fully understood. One crucial aspect of nodule senescence is regulating CPs, which are controlled by a transcription factor NAM/ATAF/CUC (NAC)-encoding 969 (NAC969) [153] (Fig 4). Understanding the detailed mechanism of nodule senescence is important, since it allows the control of this association, ensuring it delivers benefits to the plant.