Mucosal Immunity

Freely disseminating standalone viral particles are considered the optimal agents
for spreading infection. Santiana et al. discover that enteric viruses are also shed
in feces as viral clusters cloaked in vesicles.

Summary Plant expression systems have proven to be exceptional in producing high‐value complex polymeric proteins such as secretory IgAs (SIgAs). However, polymeric protein production requires the expression of multiple genes, which can be transformed as single or multiple T‐DNA units to generate stable transgenic plant lines. Here, we evaluated four strategies to stably transform multiple genes and to obtain high expression of all components. Using the in‐seed expression of a simplified secretory IgA (sSIgA) as a reference molecule, we conclude that it is better to spread the genes over two T‐DNAs than to contain them in a single T‐DNA, because of the presence of homologous recombination events and gene silencing. These T‐DNAs can be cotransformed to obtain transgenic plants in one transformation step. However, if time permits, more transformants with high production levels of the polymeric protein can be obtained either by sequential transformation or by in‐parallel transformation followed by crossing of transformants independently selected for excellent expression of the genes in each T‐DNA. Introduction Secretory immunoglobulin A (SIgA) is a heterodecameric Ig of 400 kDa that is predominantly involved in the protection of the mucosal surfaces against pathogenic infections. Application of pathogen‐specific SIgAs at the mucosal surfaces can provide immediate protection (Corthésy, 2010) but requires bulk amounts of the SIgA protein in the range of gram amounts per day and per patient. However, recombinant manufacturing of SIgA has been challenging because of its complex assembly. SIgA is composed of two monomeric IgAs (mIgAs) each made up of two heavy and two light chains, a joining (J‐)chain keeping the two mIgAs together in a dimeric IgA (dIgA), and the secretory component (SC), wrapped around the dIgA. Of the expression systems evaluated, plants have been shown to be the most successful platform for SIgAs (Ariaans et al., 2014; Ma et al., 1995). The bacterial and yeast systems currently do not efficiently secrete assembled Igs, and the levels of correctly post‐transcriptionally modified SIgAs are too low (Fischer et al., 2004; Vasilev et al., 2016). Also the few milligram production levels of SIgAs attained in mammalian or insect cell fermenters are not sufficient or cost effective to be translated into the clinic. In contrast, several different SIgAs have been expressed successfully in plants, and one of these SIgAs, that is, CaroRx, has been approved for use as medical device for dental care (Paul and Ma, 2011). Plant‐made SIgAs can be produced either via transient expression or after stable genomic integration. The transient expression system has allowed successful extraction of SIgAs against rotavirus from leaves of Nicotiana benthamiana already five days post‐infiltration (Juarez et al., 2013) and its production capacity has been thoroughly validated, for instance, with the production of millions of vaccine doses in a month against H1N1 VLP influenza by the company Medicago (D'Aoust et al., 2010). However, for passive immunization applications, we believe that stable transformation is the most suited system for the cheap production of SIgA in plants because of the ease of scalability through the already established infrastructure of commercial varieties, the inheritable expression and the possibility to produce the recombinant protein in specific compartments and tissues, such as seeds, with advantageous properties (Stoger et al., 2014; Virdi et al., 2015). To obtain stably transformed SIgA‐producing plants, several strategies have been described. The first is the Nicotiana tabacum SIgA‐producing line obtained by sequentially crossing individual transformants with, the heavy chain, light chain, J‐chain and SC encoding genes (Ma et al., 1995). High SIgA levels were obtained, but three sequential crossings were needed to generate the stably transformed line containing the four different loci in homozygous condition. Using a single T‐DNA system in which all the four components required for SIgA production are clustered, significantly reduces the time to develop SIgA‐producing stable transformants as shown for Lemna (duckweed; Ariaans et al., 2014) and Arabidopsis thaliana (Arabidopsis; Nakanishi et al., 2017). However, the cloning process of this single T‐DNA can be challenging because of the large size of the construct. Previously, we evaluated a third strategy, which is based on the simultaneous cotransformation of all T‐DNAs with the required constructs for the in‐seed production of light‐chain devoid simplified SIgAs (sSIgAs) in Arabidopsis (Virdi et al., 2013). Four different variable domains (named V1, V2, V3 and V4) of llama heavy chain‐only antibodies (VHHs) against enterotoxigenic Escherichia coli (ETEC) were genetically fused to the Fc part of a porcine IgA. Upon triple cotransformation with Agrobacterium cultures bearing T‐DNAs with one of the four VHH‐IgA fusions, the J‐chain and the SC, each driven by the seed‐specific β‐phaseolin promoter, transformants producing sSIgA in seeds were identified. This strategy was much faster than the sequential crossing method, but the frequency with which triple cotransformants were obtained was about 1% of that for obtaining the single T‐DNA transformants. Later, X‐ray crystallography revealed that V1 and V2 bind to the same epitope, whereas V3 binds to another conformational epitope (Moonens et al., 2015). Moreover, V3 and V4 most likely bind the same epitope because they differ by a single amino acid (Virdi et al., 2013). In comparison with a monomeric protein, the production of a recombinant hetero‐multimeric protein requires the appropriate contribution of several transgenes and their simultaneous transformation into the expression system (organism). Which transformation strategy to follow for the production of complex hetero‐multimeric proteins such as sSIgAs depends on several parameters such as time, desired protein expression level, tissue specificity and/or number of available selection markers. At first sight, cloning all the different sSIgA components in a single T‐DNA seems to be the most straightforward strategy. However, as the cloning process can be challenging because of the large size of the construct (about 20 kb), cotransformation seems to be a good alternative. On the other hand, sequential crossing allows a step‐by‐step screening for transformants with the best expression for each T‐DNA, but can be labour intensive because of the need to screen for double and triple homozygotes. Thus, when the aim is to produce complex hetero‐multimeric proteins in long‐life‐cycle plants, the proper choice of the transformation strategy plays a pivotal role in determining the feasibility of the project. Here, we compared these three above‐mentioned strategies with a sequential transformation strategy (Figure 1), and evaluated the straightforwardness to obtain plants with high sSIgA accumulation levels. We used the V2 and V3 VHH‐IgA variants against ETEC developed by us previously (Virdi et al. 2013) and Arabidopsis as the perfect model to test the stable production of complex hetero‐multimeric proteins in plants because of its short life cycle. As such, we wanted to pinpoint the advantages and drawbacks for each of the strategies, and which transformation method to choose for the production of a SIgA and alike hetero‐multimeric protein in plants. Results Cloning of the sSIgA‐encoding components To evaluate these alternative transformation strategies for sSIgA production in stably transformed Arabidopsis plants, different T‐DNA constructs were assembled, containing the three transgenes (i.e. VHH‐IgA, J‐chain and SC) required for sSIgA production in one T‐DNA or in two different T‐DNAs (Figure S1). Previously, we have designed four different VHH‐IgA antibodies against ETEC and expressed them in A. thaliana seeds to enable oral passive immunization of weaned piglets by adding them to the feed (Virdi et al., 2013). Here, we included two of them, V2 and V3, in the cloning and subsequent plant transformations to evaluate the reproducibility of our results. The J‐chain, SC and VHH‐IgA encoding sequences were each fused with the N‐terminal 2S2 signal peptide and the C‐terminal KDEL motif sequences for ER targeting and retention, respectively. The coding sequences were transcriptionally controlled by the 5′ β phaseolin (1475‐bp) promoter and the 3′ arceline (600‐bp) terminator to obtain seed‐specific expression (Morandini et al., 2011) like in our previous study, so we could use the best sSIgA producer for the V2 and V3 variant obtained after triple cotransformation by Virdi et al. (2013) as reference seed stocks. For the single T‐DNA transformation strategy (Figure 1a), these three sSIgA expression cassettes were cloned in tandem (same transcriptional orientation) in a single T‐DNA construct (MX‐JSA, Figure S1a). By restriction characterization of individual E. coli clones, we found that some of them integrated a double copy of the VHH‐IgA cassette, and because this might result in a positive dosage effect, we included this “single T‐DNA double VHH‐IgA” (MX‐JSAA, Figure S1) construct in the comparative transformation efficiency analysis. Restriction analysis of the clones also demonstrated the frequent occurrence of a particular variant with missing fragments in the expected restriction pattern (Figure S2), and upon inspection, this corresponded to the loss of the J‐chain expression cassette in the pMX‐JSA and pMX‐JSAA T‐DNA vector plasmids (Figure S1a,b). These frequently observed deletions occurred not only upon transformation and establishment of the clones in E. coli, but also upon propagation of the E. coli and Agrobacterium clones. For the other three transformation approaches (Figure 1b–d), the constant elements, being the J‐chain and SC, were clustered in one T‐DNA (MX‐JS; Figure S1c), and the variable VHH‐IgA, either V2 or V3, was cloned in the other T‐DNA (X‐A; Figure S1d). Once the different T‐DNA constructs were obtained and transferred into Arabidopsis tumefaciens, all transformations were performed by floral dip in wild‐type Col‐0 plants as schematically depicted in Figure 1. T1 transformants were selected on phosphinothricin and/or kanamycin and grown to obtain T2 seed stocks, which were then screened for their sSIgA production level. For the crossing strategy, F2 hybrid seeds were analysed. For the heritability study, homozygous T3 seeds were analysed. Efficiencies of different transformation strategies to obtain sSIgA‐producing plants If speed is an issue, the only two methods that need a single transformation step are gene stacking of the components in a single JSA or JSAA T‐DNA and cotransformation of the JS and VHH‐IgA T‐DNAs, referred to as JS+A (Figure 1a,b). The efficiency by which all three transgenes were integrated in a single transformant, independently of the VHH variant, was recorded for the two strategies. We obtained four times more transformants with the single JSA T‐DNA [2.38% (±1.47) of the T1 seeds] as compared with double selected JS+A cotransformants [0.52% (±0.36) of the T1 seeds]. In the supertransformation strategy (Figure 1c), two subsequent transformations are required to obtain the sSIgA‐producing plants: first single‐locus homozygous plants producing either VHH‐IgA or the constant components (J‐chain together with the SC) need to be selected, after which they are supertransformed with the other T‐DNA. Here, we supertransformed the already available lines with high levels of either the V2 or V3 VHH‐IgA variant (Virdi et al., 2013) with a JS T‐DNA containing the J‐chain and SC genes. In this way, 21 supertransformants were obtained and referred to as JSinA. The efficiency by which the new T‐DNA got integrated in the recombinant plants was 4.38% (±1.87). For the crossing strategy (Figure 1d), 24 transformants with the JS T‐DNA, obtained with a frequency of 1.58% (±0.29) in the T1 generation, were selected and expression of the J‐chain and SC was screened by quantification in western blots (Figure S3). Highly variable levels of J‐chain and SC expression were observed between the JS transformants, and counter‐intuitively, also the relative levels of J‐chain and SC within each transformant were variable (Figure S3a). However, the J‐chain western blot assay revealed the predicted single band at about 20 kDa, the SC western blot showed a main band of 75 kDa corresponding to the glycosylated version (as shown later in the characterization of sSIgA) and a band at the expected molecular weight of 65 kDa, corresponding to the non‐glycosylated SC. Six transformants showing predominantly glycosylated SC in their western blot profile and relatively high expression of both SC and J‐chain were propagated to the T3 generation, and of those, four homozygous, single‐locus JS transformants were retained (Figure S3b). These were then crossed with the same VHH‐IgA‐producing lines as used in the supertransformation strategy to obtain sSIgA‐producing hybrid seeds referred to as JSxA. All the F1 hybrids obtained were resistant to both kanamycin and phosphinothricin, indicating the successful crossing of the lines. To summarize (Table 1), 15–23 transformants were obtained by the single T‐DNA (JSA or JSAA) or the cotransformation (JS+A) strategies for both the V2 and V3 variants. In the sequential transformation strategy, 21 JS supertransformants of the VHH‐IgA expressing lines V2 and V3 were selected (JSinA), and for the crossing strategy, four hybrid F2 progeny seed stocks (JSxA) were obtained by crossing the V2 or V3 VHH‐IgA‐expressing line with the four best selected events from 24 JS transformants. Transformation group V2 V3 No. % No. % JSA 21 14% 21 29% JSAA 23 43% 21 29% JS+A 15 0% 21 0% JSinA 21 0% 21 0% JSxA 4 0% 4 0% sSIgA levels in seeds obtained by different transformation strategies The relative amounts of sSIgA recombinant protein production were determined by an sSIgA‐specific ELISA (Figure S4a), in which only the antigen‐binding, fully assembled sSIgAs are measured by detecting the complex with SC‐specific antibodies, while the antigen‐binding VHH‐IgA levels present as monomeric, dimeric and secretory IgAs were determined by a VHH‐IgA‐specific ELISA (Figure S4b). First, the levels of antigen‐binding sSIgA in seed extracts from transformants belonging to the different transformation groups were compared with those in reference seed stocks for the V2 and V3 variant, producing both VHH‐IgA antibodies to levels of 0.2% of seed weight and selected as the best sSIgA producers after triple cotransformation by Virdi et al. (2013). Because sSIgAs could not be purified from a mixture of monomeric, dimeric and secretory IgAs, it prevented us from performing absolute quantifications of the sSIgA content. Conversely, the relative quantification of the sSIgA content allowed us to determine the proportion of transformants in each transformation series that surpassed the sSIgA content of the reference seed stock. The proportion of V2 sSIgA transformants accumulating more sSIgAs than the reference was 33% for the JSA single T‐DNA, 30% for the JSAA single T‐DNA, 47% for the JS+A cotransformation, 76% for the JSinA supertransformation and 100% for the hybrids obtained by crossing (Figure 2a). The same trend was seen for the V3 sSIgA transformants: 14%, 29%, 29%, 57% and 100% for JSA, JSAA, JS+A, JSinA, and JSxA, respectively (Figure 2b). Thus, the supertransformation and crossing strategies provided clearly the largest proportion of highly producing sSIgA transformants. Among these, only the crossing strategy of selected events resulted in a low coefficient of variation (CV) for sSIgA production. Second, the accumulation of antigen binding VHH‐IgA was determined in all V2 and V3 sSIgA transformants, independent of whether it was present in a monomeric, dimeric or assembled secretory format. For both the single T‐DNA and cotransformation approaches (JSA, JSAA and JS+A), a large variability in antigen‐binding VHH‐IgA accumulation levels could be observed (Figure 2c,d). The results show that in some groups of transformants, a positive correlation with the sSIgA and VHH‐IgA levels could be seen, like such as in the group of JSA(V2) transformants, but in other groups of transformants, there was no correlation with the levels of sSIgA and VHH‐IgA, such as in the JSAA(V3) group (Figure S5). This indicates that the levels of either of the three components can be limiting to obtain high sSIgA levels. Furthermore, and remarkably, a rather high percentage of up to 43% of the single T‐DNA transformants (JSA and JSAA) did not contain detectable amounts of assembled sSIgA (Table 1), whereas they did almost all contain antigen‐binding VHH‐IgA (Figure 2c,d). In contrast, all of the analysed seed extracts from JS+A or JSinA transformants expressing detectable levels of VHH‐IgA, produced also the assembled sSIgA in variable amounts. In comparison with the JSA and JS+A obtained seed stocks, the VHH‐IgA accumulation levels in supertransformed JSinA and JSxA crossed seed stocks were much more similar in the V2 variant, as pointed by their coefficient of variation (CV) (Figure 2c). A similar pattern was seen for the V3 VHH‐IgA levels, except for the JSinA transformants that experienced a moderate increase in the CV as compared with the V2 (Figure 2d). Then, the four best sSIgA‐accumulating transformants from each strategy were chosen and the relative accumulation of sSIgA was compared in an ELISA titration experiment. We found that the V2 sSIgA accumulation levels ranged, on average, from more than double to almost 32‐fold higher than the reference line (Figure 2e). More specifically, the best JSA and JSAA transformants contained two‐ to fourfold more sSIgA accumulation than the reference line. For the JS+A cotransformants, the sSIgA levels were about eightfold higher, and among the JSinA supertransformants and JSxA hybrids almost 32‐fold higher sSIgA levels were found. Outliers from the supertransformation and crossing strategies even contained 64‐fold higher sSIgA levels than the reference line. Similar results were obtained for the four best V3 sSIgA expressors. In this case however, JSA and JSAA transformants reached slightly better accumulation levels (between four‐ to eight‐fold difference than the reference) and differences were not always statistically significant between these groups and the JSinA and JSxA transformants (Figure 2f). Molecular characterization of sSIgA‐expressing plants The seed extracts of the best sSIgA transformants in each strategy were further analysed in different western blot setups. After sodium dodecyl sulfate‐polyacrylamide gel electrophoresis (SDS‐PAGE) of the seed extracts in reducing conditions, SC‐specific antibodies revealed a band at about 65 kDa corresponding to a non‐glycosylated form of SC and higher bands corresponding to glycoforms (Figure 3a) as shown by a PNGase F digest (Figure S6). Similar blots revealed the presence of the J‐chain (Figure 3b) and the single polypeptide VHH‐IgA and proteolysed version of it (Figure 3c). Quantifying the SC, J‐chain and VHH‐IgA signals after SDS‐PAGE in reducing conditions further revealed that there was no clear correlation between the sSIgA amounts as found by ELISA (Figure 2e,f) and the accumulation levels of either the SC, J‐chain or VHH‐IgA component (Figure 3a–c). Further, we attempted to demonstrate the presence of assembled sSIgA complexes by SDS‐PAGE in non‐reducing conditions (Figure 3d–f). We postulate that the faint band at ~250 kDa corresponds to covalently linked sSIgA complexes of four VHH‐Fc polypeptides, one J‐chain and one SC. For most samples, the blots showed similar relative amounts of assembled sSIgA as found in the sSIgA‐specific ELISA (Figure 2e,f). Discrepancies found in the band intensities between ELISA and western blot could be attributed to the likely presence of functional, non‐covalently‐linked sSIgA, which is detectable by ELISA but disaggregated during the western blot procedure, and to the normalisation process applied to the OD values in the ELISA. Nevertheless, the high proportion of SC, J‐chain and VHH‐IgA found in other forms than in the sSIgA complexes is remarkable. For instance, apart from complexed with sSIgA, SC could be found as single polypeptide and as a postulated dimer in different glycoforms (Figure 3d). This was different for the J‐chain: in reducing conditions, the J‐chain was present at the expected molecular weight (Figure 3b), but in non‐reducing conditions, the J‐chains were detected in high‐molecular weight complexes, covalently linked among themselves or with antibody chains or with endogenous proteins, while no free J‐chain molecules were detectable (Figure 3e). Finally, the VHH‐IgA main bands corresponded to monomeric and single VHH‐IgA polypeptides and truncated or glycosylated versions of these molecules (Figure 3f). Differences in the level of glycosylation and proteolysis of the analysed molecules were observed between transformants, however, no specific patterns could be assigned to any specific group. Heritability of sSIgA production To check the heritability of the expression, we compared the V2 and V3 sSIgA amounts in some segregating T2 seed stocks and their homozygous T3 progeny (Figure 4). Overall, the production of sSIgA in T2 and T3 seed stocks was comparable, except for several transformants in the JSAA series. In these T3 progeny seed stocks, both for the V2 and V3 series, the sSIgA amounts were decreased below 50% of the levels achieved by their parents. Thus, only when the three constructs were cloned in tandem in a single T‐DNA, they seemed more susceptible to gene silencing in subsequent generations, but we did not confirm this at the molecular level. Discussion The transformation process to produce a hetero‐multimeric protein can follow different paths, leading to certain advantages and disadvantages related to workload, protein production level or transformation efficiency, among other parameters. To test this, we used sSIgA as model. Previously, Virdi et al. (2013) cloned the three encoding genes (J‐chain, SC and VHH‐IgA) in three different T‐DNA vectors, and transformants were obtained by triple cotransformation. Here, we evaluated four alternative strategies for the stable seed‐based production of sSIgA: a JSA single T‐DNA transformation, cotransformation of the JS and VHH‐IgA T‐DNAs, supertransformation of a selected VHH‐IgA transformant with the JS T‐DNA and crossing of selected JS and VHH‐IgA transformants. Table 2 summarizes the main advantages and disadvantages encountered for each strategy. Single T‐DNA Double cotransformation Supertransformation Parallel transformation & Crossing Cloning process Demanding Easy Easy Easy Transformation efficiency High Moderate High High Plant manipulation steps One One Two (or one†) Two Time to get T2 sSIgA‐accumulating plants Moderate Short Long (or short†) Long High‐level sSIgA producers Few Medium Many Many Selection markers One Two Two Two Selection of single‐locus events One locus One or two loci Two loci Two loci Selection of stable lines Easy Easy (if single locus) Easy Demanding † Depending on whether a stable recombinant plant is available to be supertransformed. In terms of complexity, the DNA assembly and cloning of the three components in a single T‐DNA vector turned out to be very difficult, because this T‐DNA vector is close to 20 kb, resulting in very low cloning efficiencies. Combining the genes for J‐chain and SC in one JS T‐DNA and the other VHH‐IgA component in another T‐DNA, made the cloning process much easier. Another important factor to consider when introducing multiple genes is to determine if a number of the genes can be reused to produce families of heteromeric proteins. For instance, in our sSIgA model, the J‐chain and SC remain constant within the same species, while only the VHH‐IgA differs for every different epitope (Janeway et al., 2001). Therefore, the reusability of the same J‐chain and SC to produce different SIgAs allows a simplification of the system by joining both expression cassettes (J‐chain and SC) in a single T‐DNA to use in a double cotransformation process (together with a VHH‐IgA bearing plasmid) or to get a stably transformed JS line producing the J‐chain and SC, which subsequently can be supertransformed with a VHH‐IgA bearing plasmid or crossed with a VHH‐IgA‐producing plant. Thus, plants with good expression of the constant elements can be reused in future projects with different VHH‐IgAs, whereas applying the single T‐DNA strategy implies that a new 20‐kb construct needs to be generated for each new VHH. Alternatively, the use of emerging cloning systems such as Golden Braid (Vazquez‐Vilar et al., 2017) or MoClo (Weber et al., 2011), both based on Golden Gate multiple assembly technology, may ease the assembly of huge constructs. Another problem we encountered was the loss of the J‐chain expression cassette during clone establishment and propagation. These deletions are most likely the result of homologous recombination between the repetitive sequences flanking the different genes. Indeed, as a consequence of using identical transcriptional regulatory sequences, the three or four genes in the pMX‐JSA and pMX‐JSAA constructs, respectively, are flanked with stretches of more than 1400‐bp homology, and even in low‐recombination E. coli strains such as DH5α, these deletions were observed, as previously described (Bzymek and Lovett, 2001). This results in bacterial cultures with a mixed population of correct and truncated versions of the desired T‐DNA. Even though cultures were checked via restriction digestion before plant transformation, the occurrence of plants producing high concentrations of VHH‐IgA but no detectable levels of sSIgA suggests that these plants received a truncated version of the T‐DNA. These plants had to be discarded, resulting in a waste of time and resources. However, previous groups producing SIgA via the single T‐DNA method did not experience homologous recombination in their constructs (Ariaans et al., 2014; Juarez et al., 2013), and other groups, such as Ruiz‐Lopez et al. (2015) and Malik et al. (2015), managed to assemble up to six and four expression cassettes, respectively, without apparent complications. This suggests that the regulatory sequences used in this project favour the occurrence of recombination events. To lower the risk of homologous recombination, it is possible to work with a pool of different promotor and terminator sequences. DNA cloning systems such as Biobricks or GoldenBraid are specially designed for that purpose, and they allow to choose á la carte between a collection of promoters and terminators and combine them at will (Smolke, 2009; Vazquez‐Vilar et al., 2017). However, care has to be taken that these different transcriptional regulatory sequences are synchronised (switched on/off) in the same cells and at the same time. Although cloning every gene in separate T‐DNA vectors is an approach that works well for transient expression (Juarez et al., 2013), for stable transformation, we do not advice to distribute the genes over more than two T‐DNAs because of the time and screening efforts required. For instance, the transformation efficiency registered for the triple cotransformation strategy by Virdi et al. (2013) (0.028%) was approximately 100‐fold lower as compared with the efficiencies achieved by introducing a single T‐DNA at a time (supertransformation and single T‐DNA strategies) and 20‐fold lower than that obtained by double cotransformation. Logically, the efficiency to obtain cotransformants after A. tumefaciens‐mediated transformation is inversely proportional to the number of different T‐DNAs that are required to integrate and varies depending on the plant species and target plant tissues (De Buck et al., 2000, 2009). The direct consequence of obtaining a lower number of cotransformants is the higher number of transformation projects that need to be started. In this project with Arabidopsis, a fourfold reduction in obtaining the desired transformants, when comparing the double cotransformation to the single T‐DNA strategies, is not an issue, but would be a serious limitation in plants recalcitrant to transformation (Birch, 1997). It should also be noted here that every T‐DNA commonly requires a different selection marker. However, a single T‐DNA needs only one selection marker, cotransformation and supertransformation require two selection markers. Another point to consider is the time required to obtain sSIgA‐producing plants: whereas the single T‐DNA and cotransformation methods can give sSIgA‐producing plants in one plant transformation cycle, supertransformation and crossing require a first cycle to obtain stably transformed plants, and a second cycle for supertransformation or crossing. This extra cycle thus results in a substantial increase in time and efforts required to obtain the desired sSIgA‐recombinant plants. Taking into account that cloning a single T‐DNA with several constructs takes more time than cloning simple T‐DNA vectors, we conclude that cotransformation is the fastest and least work‐demanding option to obtain transformants producing sSIgA. On top of that, transforming plants such as tobacco and soybean may ease biomass upscaling of sSIgA‐transformed lines, but it comes at the cost of longer life cycles and more demanding tissue culture requirements to obtain the transformants. For instance, Ma et al. (1995) could stably produce up to 500 μg of SIgA per gram of fresh weight in the easy‐to‐scale tobacco leaves, however the process required (i.e. the sequential crossing of four transgenic tobacco plants) implied years of work until the first plant producing SIgA was obtained. On the other side, in plant species such as duckweed, in which a transformation process takes only a few weeks (Yamamoto et al., 2001), Ariaans et al. (2014) reached an SIgA production level of 16% of the total soluble leaf protein by transforming all the constructs in a single T‐DNA. By determining the assembled sSIgA levels by ELISA and Western blot analysis, we were able to determine the variability in expression within each group and more importantly the number of high‐level sSIgA producers in each group. We found that the single T‐DNA transformants (JSA and JSAA) overall had the lowest performance in sSIgA accumulation, followed by cotransformation that only performed slightly better in one of the two VHH‐IgA variants. The low expression level of the three transgenes within one T‐DNA (JSA) might be due to the triple repeat of the regulatory elements, making them more sensitive to transcriptional gene silencing and position effects (Matzke and Matzke, 1998). Indeed, adding a second VHH‐IgA expression cassette to the single T‐DNA (JSAA) did not provide higher sSIgA levels. On the contrary, the transformants selected for the highest accumulation of sSIgA suffered a drastic reduction in the assembly levels of this molecule in the next generation most likely because of gene silencing effects. It has been described how the transgene copy number, the transgene expression level and the repetitive use of DNA sequences increase the occurrence of silencing (Assaad et al., 1993; Lessard et al., 2002; Matzke and Matzke, 1998). Moreover, the high frequency of silencing in the single T‐DNA transformants might be related to the floral dip transformation method for which it is known that multiple T‐DNA copies are frequently integrated (De Buck et al., 2009). Thus, using another transformation method or using a rapid screen for single‐copy T‐DNA inserts might improve the expression performance of transformants with T‐DNAs encoding all the components of a hetero‐multimeric protein. On the other hand, the results clearly demonstrate that supertransformation and parallel transformation followed by crossing are the best suited strategies to find excellent sSIgA producers among a limited number of transformants, most likely because of the independent selection of good events for both T‐DNAs. Indeed, if an optimal event occurs in about 1 of 20 transformants, then the probability of optimal expression from both T‐DNAs will only be found in 1 of 400 transformants. It is indeed known that transgene expression levels are variable but also, especially for single‐copy inserts, to a high degree heritable (De Buck et al., 2004). Therefore, after crossing two parentals with high‐expressing transgenes, hybrids with high sSIgA levels can be obtained, whereas, upon supertransformation, the first generation transformant is selected and screened for the best expression, and subsequently the best expressor for the second T‐DNA is selected among the second generation transformants, increasing the chance to isolate a transformant in which all transgenes have high expression rates. However, a more detailed look at the expression level of each component via western blot revealed that we are far from understanding the most suitable ratios for the assembly of sSIgA. The western blot results show how, even in the selected top sSIgA‐expressing seeds, several components are not covalently assembled in a functional sSIgA. VHH‐IgA and SC are also found as single polypeptides and as dimers in their glycosylated and non‐glycosylated form, truncated versions of VHH‐IgA are present as previously described by Virdi et al. (2013), and the J‐chain is covalently linked into aggregates that might limit the amount of free J‐chains available to form sSIgA. A bibliographic research showed us how the described effects have been experienced previously by other authors, such as the proteolysis of IgA (Nakanishi et al., 2017; Paul et al., 2014; Wieland et al., 2006), the simultaneous presence of different glycoforms (Juarez et al., 2013; Wieland et al., 2006) and the proteolysis of the SC (Ma et al., 1995; Nakanishi et al., 2017), the latter not observed here. These results emphasize the need to thoroughly study molecule stability and protein interactions when working with hetero‐multimeric proteins. Another important aspect is the ease with which stable lines can be generated from the selected transformants by reaching homozygosity of the inserted T‐DNA(s). This is straightforward if there is only a single locus involved: then one in four of the segregating T2 seeds will be homozygous. A single‐locus insert is also frequently found after cotransformation, because cotransformed T‐DNAs frequently cointegrate at the same genetic locus (De Neve et al., 1997). On the other hand, supertransformation and stacking two events by crossing automatically involves two genetic loci. However, whereas screening for a double homozygous plant among the progeny of a double hybrid requires intensive screening, getting double homozygous plants in the supertransformation strategy is simple if a homozygous plant is used for supertransformation. In conclusion, single T‐DNA and cotransformation approaches may represent the fastest options to obtain several transformants in a screening project. However, cotransformation gives less problems during DNA cloning, single T‐DNA cloning requires only one selection marker and single‐locus homozygous plants are easy to obtain. On the other hand, supertransformation and crossing are more time‐demanding but show an increased chance to obtain good sSIgA producers. When the aim is to generate a production platform for a library of sSIgAs, the best approach seems to be to first select a good stable, homozygous line producing high levels of the constant J‐chain and SC, and then to use this line to supertransform with several VHH‐IgAs to finally obtain a homozygous lines producing high levels of sSIgAs. In all, our work here gives a clear overview of the advantages and disadvantages of transformation strategies that can be used for complex protein production in plants, and this can facilitate and guide important decisions about how to obtain high expression of hetero‐multimeric proteins in different plant production systems. Experimental procedures Construction of T‐DNAs for in‐seed sSIgA expression All cloning experiments were performed in E. coli DH5α. The different transgenes required for in‐seed sSIgA production are described in Virdi et al. (2013). Briefly, the fusion VHH‐IgA and the porcine SC expression cassettes were introduced in the pPhasGW destination vector (Morandini et al., 2011), which has the nptII gene for kanamycin resistance, resulting in the expression vector pX‐A and pX‐S, while the porcine J‐chain expression cassette was contained in the pBm43GW,0 plasmid, which has a bar gene that confers resistance to phosphinothricin (PPT; Karimi et al., 2005), to obtain pX‐J. The combination of restriction enzymes XbaI, for pX‐S, and AvrII, for pX‐J, served to transfer the SC expression cassette from the original plasmid to pX‐J, to obtain pMX‐JS with both transgenes in tandem. Because XbaI and AvrII are isocaudomers, the resultant sequence lost the reassembled recognition sites. Then, the same combination of restriction enzymes was used to transfer the VHH‐IgA expression cassette for the V2 and V3 VHH‐IgA variants from the pX‐A to the pMX‐JS plasmid, to obtain the pMX‐JSA plasmid with all three components in tandem in a single T‐DNA. Unexpectedly, a plasmid with a double copy of the VHH‐IgA expression cassette (pMX‐JSAA) was also found. Agrobacterium‐mediated transformation via floral dip in A. thaliana The T‐DNA plasmid vectors obtained were transformed into the Agrobacterium strain C58C1RifR with the helper plasmid pMP90 (Koncz and Schell, 1986) via electroporation (Shen and Forde, 1989). The Agrobacterium strains bearing the correct plasmids were used for floral dip of Col‐0 Arabidopsis plants, as described in Clough and Bent (1998). Transformants were selected on Murashige and Skoog agar medium supplemented with nystatin (50 mg/L), vancomycin (750 mg/L) and PPT (10 mg/L) and/or kanamycin (50 mg/L), depending on the selection marker of the plasmids used. Finally, single‐locus T2 transformants were identified and homozygous T3 seed stocks were obtained by segregation analysis as described previously (De Jaeger et al., 2002) for the heritability study. Characterization of J‐chain and SC transformants The seed protein extraction was performed as described by De Jaeger et al. (2002). The levels of J‐chain and SC in the soluble extract were determined by SDS‐PAGE using 4%–15% Mini‐PROTEAN® TGX™ Precast Protein Gels (Biorad, Hercules, CA, USA) under reducing conditions. Then the proteins were blotted onto a polyvinylidene difluoride membrane using the Trans‐Blot® TurboTM Transfer System (BioRad). The SC was detected with mouse anti‐porcine SC (Thermo Fisher MA1‐80544, Waltham, MA, USA) followed by anti‐mouse IgG conjugated to horseradish peroxidase (HRP; GE Healthcare GENXA931). The porcine J‐chain was detected with mouse anti‐human J‐chain (Thermo Fisher MA1‐80527) followed by anti‐mouse IgG conjugated to HRP (GE Healthcare GENXA931, Chicago, IL, USA). Recombinant proteins were visualized and quantified using the Bio‐Rad ChemiDoc™ Imaging System (Bio‐Rad) and the Image Lab™ software (Biorad). A PNGase F (New England Biolabs, Ipswich, MA, USA) glycosidase treatment was performed in some samples to study the glycosylation state of SC, following the manufacturer's instructions. Crossing to obtain sSIgA‐producing hybrids Four stable, high‐accumulating JS transformants were selected as mother plants to be crossed with the pollen of stably transformed lines producing the highest level of V2 and V3 VHH‐IgA as isolated by Dr. Virdi (Virdi et al., 2013). After the crossing process, the F1 hybrids were selected on Murashige and Skoog agar medium supplemented with PPT (10 mg/L) and kanamycin (50 mg/L), and were then propagated to harvest the segregating F2 seeds for subsequent protein analysis. Characterization of T2 and F2 sSIgA‐producing seeds The levels of antigen‐binding antibodies in seed protein extracts from the T2 and F2 seed stocks were determined by ELISA with immobilized FaeGac (F4‐fimbriae most abundant adhesin). The wells were incubated with serial dilutions of 1 mg/mL seed protein extract (commonly 1/10, 1/40, 1/160 and 1/640) made in 2% skimmed milk in PBS. In the sSIgA‐specific ELISA setup, sSIgA complexes binding to FaeG were detected by using a mouse anti‐porcine SC (Thermo Fisher MA1‐80544) followed by an anti‐mouse IgG antibody conjugated to HRP (GE Healthcare GENXA931). In the VHH‐IgA‐specific ELISA setup, all different forms of IgA that bind to FaeG were detected by applying an anti‐pig IgA conjugated to HRP (Biorad AAI40P). TMB (3, 3′, 5, 5′ tetramethylbenzidine, Sigma) was used as substrate for the HRP, followed by adding 0.18 m H2SO4 45 min post‐incubation to stop the reaction. Finally, both ELISAs were measured at 450 nm in a VERSAmax ELISA microplate reader (Molecular Devices). The results were normalized against a reference provided by Dr. Virdi (Virdi et al., 2013). After the analysis, the best sSIgA‐accumulating transformants were used in a titration experiment to calculate precisely the relative accumulation of sSIgA. In the assay, the sSIgA‐specific ELISA setup was applied to a broader serial dilution of the samples (1/80, 1/160, 1/320, 1/640, 1/1280, 1/2560, 1/5120 and 1/10240). End point titres were determined within the threshold of two times the average of a blank sample and then transformed into relative units compared to the dilution value recorded for the reference. The protein seed extracts were also evaluated via western blots after SDS‐PAGE, similar to the evaluation of the JS transformants. The VHH‐IgA and the VHH‐IgA‐derived complexes were detected with an anti‐pig IgA antibody conjugated to HRP (Biorad AAI40P), while the same antibodies as used for the sSIgA‐specific ELISA were used for SC detection. Isolation of stable lines producing sSIgA The best sSIgA expressors in T2 seed stocks among the JSA, JSAA, JS+A, and JSinA transformants for both V2‐ and V3‐containing VHH‐IgAs were propagated to the next (T3) generation to obtain stable lines. To simplify the procedure, a segregation analysis was applied and only single‐locus transformants (i.e. showing a 3:1 ratio of resistant to susceptible transformants) were kept. Analysis of the T3 offspring of these allowed to identify the homozygous lines. On the other hand, the F1 hybrids resulting from the crossing were propagated to F2 seed stocks, but stable lines based on screening for homozygosity of both loci (1/16) were not isolated. Acknowledgements This work was supported by grants from Research Foundation Flanders (FWO project G0C9714N, predoctoral fellowship to J.P.) and the Agency for Innovation by Science and Technology (IWT‐innovation mandate 140851, postdoctoral fellowship to V.V.). Author contributions A.D. conceived the idea. J.P., V.V. and A.D. designed the research; J.P. performed research; J.P., V.V. and A.D. analysed the data; and all authors were involved in writing the manuscript. Conflict of interest The authors declare no conflict of interest. Supporting Information References

Abstract Mucosal‐associated invariant T (MAIT) cells develop in the thymus and migrate into the periphery to become the largest antigen‐specific αβ T‐cell population in the human immune system. However, the frequency of MAIT cells varies widely between human individuals, and the basis for this is unclear. While MAIT cells are highly conserved through evolution and are phenotypically similar between humans and mice, they represent a much smaller proportion of total T cells in mice. In this review, we discuss how MAIT cells transition through a three‐stage development pathway in both mouse and human thymus, and continue to mature and expand after they leave the thymus. Moreover, we will explore and speculate on how specific factors regulate different stages of this process. Introduction Mucosal‐associated invariant T (MAIT) cells are a highly conserved population of T lymphocytes that exhibit innate‐like functions similar to those described for natural killer T (NKT) cells.1-4 MAIT cells are defined by their invariant T‐cell receptor (TCR)‐α chain, and their restriction to the nonpolymorphic major histocompatibility complex class I‐like molecule, MHC‐related protein 1 (MR1).2 In mice, MAIT cells express a TCR comprising of Vα19 joined to Jα33, which pairs with Vβ6 or Vβ8.5-7 In humans, MAIT cells predominantly express Vα7.2 joined to Jα33, Jα20 or Jα12, and these pair with Vβ2 or Vβ13.3, 5, 7 Smaller subsets of MR1‐restricted Vα7.2− T cells have been identified in humans,8-10 although this review will focus on the major population of Vα7.2+ MAIT cells, herein simply described as MAIT cells. In this issue, Lantz and Legoux11 provide a historical perspective of the MAIT cell field, including the initial identification and the description of MAIT cells as “mucosal‐associated” due to their presence within the gut lamina propria.2 It is now appreciated that MAIT cells can be found in all sites where conventional T cells are present,6, 12, 13 although functionally distinct subsets of MAIT cell exist at mucosal sites, suggesting an important role for these cells in barrier tissue immunity.12, 14 MAIT cells are abundant in humans, making up to 10% of T cells in peripheral blood and up to 50% in the liver,14, 15 yet their frequency can significantly vary between individuals.13, 15-17 In mice, the frequency of MAIT cells is much lower, and it is currently unclear what factors regulate MAIT cell numbers and cause this disparity between species.6, 7, 16 Our understanding of human MAIT cells has progressed predominantly through the study of CD3+ Vα7.2+ CD161hi T cells, which incorporates most MAIT cells in peripheral blood and organs.13, 18, 19 Conversely, because MAIT cell numbers in laboratory mice are very low, and monoclonal antibodies against the invariant mouse MAIT TCRα chain do not exist, Vα19Jα33 transgenic mice were established to investigate the biology of mouse MAIT cells.16, 20 The discovery of riboflavin derivatives as antigens (Ags) for MAIT cells21, 22 (described in this issue by Kjer‐Nielsen, Corbett and colleagues23), coupled with the production of MR1 tetramers that incorporate these Ags,5 has fueled new studies into the development and function of MAIT cells. These tetramer reagents now provide a basis for distinguishing MAIT cells from other T‐cell populations that share similar phenotypic markers.8, 9, 24, 25 MR1 is expressed at low levels in a wide range of tissues26, 27 and shows strong evolutionary conservation between species.28 This, together with the broad microbial reactivity of MAIT cells suggests that they play a key role in host immunity.29-32 The role that MAIT cells play in microbial immunity is extensively reviewed in this issue by Ussher, Willberg and Klenerman (MAIT cells and Viruses)33 and Meermeier, Harriff, Karamooz and Lewinsohn (MAIT cells and Microbial Immunity).34 Given the wide variation in MAIT cell frequency between individuals13, 16, 19, 35 and evidence that reduced MAIT cell numbers are associated with aberrant immune responses in humans and mice,29-31, 36-39 it is important to understand how MAIT cells develop and determine what factors regulate this process. In this review, we describe how MAIT cells undergo a three‐stage development pathway within the thymus and continue to mature after they enter the periphery, in both mice and humans. In addition, we examine similarities and differences with the development of NKT cells, which may compete with MAIT cells for a similar environmental niche. Thymic Selection Studies in mice revealed that MAIT cells arise intrathymically following interaction of the MAIT TCR with MR1.2, 7, 16, 40 Furthermore, analysis of Vα19‐Jα33 Cα−/− and Vβ6 transgenic mice revealed that MAIT cells are selected by MR1‐expressing DP cortical thymocytes, compared to thymic epithelial cells that select MHC/peptide reactive T cells.40 It is currently unclear whether specific Ags must be presented in association with MR1 for intrathymic selection, and if so, the nature of the selecting Ags. Selection may require agonist ligands like the MAIT activating riboflavin derivative Ags such as 5‐(2‐oxopropylideneamino)‐6‐d‐ribitylaminouracil (5‐OP‐RU) and 5‐(2‐oxoethylideneamino)‐6‐d‐ribitylaminouracil (5‐OE‐RU).22 Alternatively, MAIT cells might be selected by a weak or nonagonist Ag, such as 6‐formylpterin (6‐FP), a vitamin B9 derivative that binds to MR1 but does not activate the majority of MAIT cells.21, 41 Although these vitamin‐derivative Ags have microbial or dietary origins, it is also possible that MAIT cells are selected in the context of an endogenous self‐Ag produced in the thymus. Early evidence supported a role for microbial derived Ags for the selection of MAIT cells as these cells could not be detected in germ‐free (GF) mice using RT‐PCR.2 We recently re‐examined GF mice using MR1 tetramers and detected the presence of normal numbers of immature MAIT cells in the thymus, but these cells were unable to fully mature and expand in the periphery.24 These data suggest that microbial colonization is not required for the initial selection of MAIT cells, but is essential for their subsequent expansion.16, 24 Another important question surrounding the selection of MAIT cells is whether their TCR repertoire is shaped by negative selection. Like MAIT cells, NKT cells are selected by DP cortical thymocytes,42 and in vivo injection of the NKT cell agonist Ag α‐galactosylceramide or the addition of this lipid Ag to fetal thymic organ cultures ablated the development of mouse NKT cells, suggesting these cells had undergone negative selection.43, 44 It will be important to establish if similar selection criteria also exist for MAIT cells. For instance, would the overexpression of MR1 or the presence of a high affinity Ag lead to the deletion of MAIT cells? Accordingly, further studies are required to examine the types of Ags (if any) that govern the intrathymic selection of MAIT cells. A Three‐Stage Pathway for MAIT Cell Development in Mice and Humans Analysis of MAIT cells from the periphery of WT mice using MR1‐5‐OP‐RU tetramers revealed that they expressed high levels of CD44 and had a memory phenotype, whereas most MAIT cells from Vα19‐Jα33 Cα−/− transgenic mice lacked CD44 expression and were described as naïve.6, 16 Moreover, and in contrast to previous findings,16 MAIT cells from WT mice expressed the transcription factor, promyelocytic leukemia zinc finger (PLZF).6, 45 PLZF was previously reported to be required for the development of other innate‐like T cells such as NKT cells,46, 47 innate lymphoid cells48, 49 and some γδ T cells.50, 51 These data highlight important differences in the phenotype of MAIT cells from WT and Vα19 transgenic mice and suggest that the overexpression of the mouse MAIT TCR α‐chain likely alters the development of MAIT cells. Our studies of mouse thymus revealed three populations of MAIT cells based on their expression of CD24 and CD44, including CD24+CD44−, CD24−CD44− and CD24−CD44+ MAIT cells.24 Through a combination of phenotypic analysis, ontogeny experiments and in vitro development studies, we determined that the CD24+CD44− population were least mature, defined as stage 1 MAIT cells. These give rise to CD24−CD44− stage 2 cells and ultimately these differentiate into CD24−CD44+ stage 3 cells, which more closely resemble MAIT cells in the periphery (Figure 1). Importantly, MR1 expression appears to be required at each stage of development, as progression from stage 1 to stage 3 in vitro, is severely impaired by the addition of an anti‐MR1 blocking antibody.24 These distinct MAIT cell precursor populations were further characterized for expression of differentiation markers, cytokine receptors and transcription factors (Table 1). For example, PLZF is not expressed in stage 1 MAIT cells, while mature populations of stage 3 MAIT cells express PLZF.24 This contrasts with the development of NKT cells, as early NKT cell progenitors express high levels of PLZF, and expression levels decline to moderate levels as NKT cells become more mature, with the notable exception of NKT2 cells that express high levels of PLZF (discussed below).46 Analysis of PLZF‐null mice revealed that the development of stage 1 and 2 MAIT cells was unperturbed, while mature stage 3 cells failed to develop within the thymus and periphery. Moreover, residual MAIT cells from PLZF‐null mice were unable to produce cytokines and were hence functionally incompetent.24 Therefore, PLZF is a critical factor governing the transition from stage 2 to stage 3 MAIT cells, as well as the acquisition of MAIT cell functional potential. Stage 1 Stage 2 Stage 3 CD4 and CD8 coreceptor CD4+, CD8+ or CD4+CD8+ h: CD4+, CD8+ or CD4+CD8+ CD4+, CD8+ or m: CD4+, CD8+ or CD4−CD8− CD4−CD8− CD8αβ CD8αβ+ CD8αβ+ CD8αβ+ CD24 (m) Positive Negative Negative CD27 (h) Negative Positive Low to positive CD44 (m) Negative Negative Positive CD62L (m) Intermediate n.d. Negative CD69 (m) Positive n.d. Negative CD103 (m) Negative n.d. Low to positive CD161 (h) Negative Negative Positive (h) NK1.1 (m) Low to positive (m) CD278/ICOS (m) Negative n.d. Positive Cytokine receptors CD122/IL‐2Rβ (m) Negative n.d. Intermediate CD127/IL‐7R (m) Negative n.d. Positive CD218/IL‐18R Negative n.d. Positive Transcription factors PLZF Negative Low Positive T‐bet and RORγt Low Low T‐bet+ or RORγt+ (m) T‐bet+ RORγt+ (h) n.d. = not determined; m = mouse; h = human. Most studies on MAIT cell development in humans have utilized surrogate markers, such as Vα7.2 or co‐expression of Vα7.2 and CD161, to identify MAIT cells. Using this strategy, MAIT‐like cells could be detected within human thymus and were shown to express PLZF.16, 52, 53 Moreover, Leeansyah and colleagues examined a panel of human fetal tissues including the thymus and cord blood, and described a maturation pathway of microbial reactive Vα7.2+ CD161+ cells.52 Our laboratory examined MAIT cells from human thymus using MR1 tetramers and similar to our mouse studies, we identified three populations of MAIT precursors based on the differential expression of cell surface markers.24 Ontogeny and functional studies showed that stage 1 human MAIT cells could be defined as CD27−CD161−, followed by stage 2 CD27+CD161− and then stage 3 CD27lo/+CD161+ (Figure 1 and Table 1). Moreover, the expression of CD161 on MAIT cells from human thymus also coincided with the expression of IL‐18R, the latter expressed by stage 3 thymic MAIT cells from both humans and mice (Table 1).24 Notably, the vast majority of human thymic MAIT cells lack expression of CD161 and these cells would have been excluded from previous studies that relied on CD161 as a surrogate marker of MAIT cells.16, 24, 52 In addition, we also found many MAIT‐like Vα7.2+ CD161+ T cells in human thymus that do not bind MR1‐5‐OP‐RU tetramer, a result that was recently observed by another group with human cord blood and neonatal blood samples.25 The majority of peripheral blood MAIT cells have upregulated CD161 and IL‐18R, although residual CD161− IL‐18R− MAIT cells could still be found in cord blood and young peripheral blood.24 These data reveal that MAIT cells upregulate CD161 and IL‐18R as they exit the thymus or soon after they enter the periphery. Factors That Regulate MAIT Cell Development The identification of a three‐stage development pathway for MAIT cells now allows for a detailed assessment of factors that influence this process. In addition to PLZF, we also showed that microRNAs are critical for MAIT cell development, as mice deficient for a member of the RNase III superfamily, Drosha, have drastically reduced MAIT cells and accumulate stage 1 thymic MAIT cells.24 NKT cells were also diminished in these mice, supporting previous data that microRNAs are required for NKT cell development.54 Future studies should investigate which specific microRNAs are important for MAIT cell development and whether there is overlap with the microRNAs shown to be important for the development of NKT cells (e.g. miR150,55 miR‐181 family56 and miR17–92 family cluster57). Let‐7 microRNAs will be an important candidate to investigate, as they have been shown to directly target the Zbtb16 gene and downregulate PLZF expression in NKT cells.58 The loss of IL‐18 and commensal microbiota from GF mice impaired MAIT cell numbers and progression to stage 3 MAIT cells.24 A previous study also showed that B cells were not required for MAIT cell development, but were required for peripheral expansion.16 Given that thymic precursors for MAIT cells and NKT cells express similar (but not identical) phenotypic markers and both require PLZF and microRNAs for their development, future studies will likely assess whether other factors are common for their development. For instance, several members of the NFkB family of transcription factors have been shown to be important for NKT cell development.59 Furthermore, mice deficient for the transcription factors Rorγt, HeLa E‐box binding protein and c‐Myb, have fewer NKT cells as they are unable to rearrange the distal TCR gene segments due to reduced lifespan of DP thymocytes.59 Hence, factors that affect thymocyte survival may impact on the development of MAIT cells as the gene encoding the MAIT TCR Vα‐chain (Vα19) is located at the extreme end of the 5′ end of the TCRα locus.60 Inhibitors of DNA transcription factors (Id2 and Id3), which interact with E proteins, have also been implicated in the development and differentiation of NKT cells,61-64 and more recently, we showed that members of the linear ubiquitin chain assembly complex, Hoil and Hoip, were required for normal NKT cell development in the thymus.65 Interestingly, the SLAM/SAP/Fyn pathway does not appear to be necessary for the development of MAIT cells, whereas it is important for the development of NKT cells.16, 66-69 The use of MR1 tetramers in candidate gene knockout mice should lead to the discovery of key molecules that are important for the development of MAIT cells and these can be compared with those that regulate the development of NKT cells. MAIT Cell Heterogeneity At least two functionally distinct subsets of stage 3 MAIT cells exist in mice and these develop following the expression of PLZF.24 The major population of MAIT cells in mice expresses RORγt and secretes IL‐17 upon activation, while a smaller subset expresses T‐bet and produces IFN‐γ upon activation.6, 24 With the exception of PLZF which is high on both these populations, they are reminiscent of mouse PLZFlo T‐bethi NKT1 cells and PLZFint Rorγt+ NKT17 cells, which secrete IFN‐γ and IL‐17, respectively.70 Notably, mouse thymus contains a third population of NKT cells, termed NKT2 cells, which are PLZFhi T‐betlo cells and predominantly produce IL‐4.70 While low levels of IL‐4 production were detected upon stimulation of mouse thymic MAIT cells,6 a defined population of MAIT cells that are equivalent to NKT2 cells has yet to be identified. In human thymus and peripheral blood, stage 3 MAIT cells coexpress intermediate levels of T‐bet and RORγt and secrete predominantly IFN‐γ and TNF upon activation.12, 18, 24, 71-73 This differs to stage 3 MAIT cells from naïve mice, which are mutually exclusive for T‐bet and RORγt expression, although intranasal infection of mice with Salmonella Typhimurium causes MAIT cells to coexpress these transcription factors. Thus, previously activated MAIT cells in mice appear to more closely resemble their human counterparts.32 While very few MAIT cells from human blood appear to produce IL‐17, MAIT cells from other human tissues can secrete IL‐17. For instance, MAIT cells isolated from the female genital tract express more IL‐17 in response to microbial stimuli compared to MAIT cells from peripheral blood.12 Moreover, tissue resident MAIT cells isolated from human liver vascular beds were the dominant population of IL‐17 producing T cells from this tissue14 and several studies have reported a role for IL‐17 producing MAIT cells in various autoimmune diseases (reviewed in this issue by Rouxel and Lehuen).74 Accordingly, mice and humans contain functionally distinct populations of MAIT cells, although the precise molecular mechanisms that underpin the differentiation into each distinct population remain largely unknown. Extrathymic Development of MAIT Cells MAIT cells continue to mature after they exit the thymus. While stage 3 MAIT cells from human and mouse thymus coexpress CD8α and CD8β, many peripheral MAIT cells express CD8α with low or no CD8β.5, 24, 52 These data suggest that CD8αα+ MAIT cells are likely derived from CD8αβ+ MAIT cells.18, 52 Moreover, stage 3 MAIT cells from human thymus have a limited capacity to produce cytokines compared to stage 3 MAIT cells from human blood, suggesting they undergo further maturation in the periphery. In support of this, stage 2 MAIT cells could be detected in the cord blood and the peripheral blood from young donors and stage 2 MAIT cells could be detected in the periphery of PLZF null mice, revealing that MAIT cells can exit the thymus as stage 2 cells, prior to further maturation to stage 3 cells in the periphery.24 It is currently unclear what factors drive extrathymic development of MAIT cells, whether it is direct exposure to microbial Ags or other environmental signals such as IL‐18 and/or other cytokines.24 The variation in MAIT cell frequency between humans and mice highlights important differences in the development and expansion of MAIT cells between these species. Several factors have been proposed to explain these differences. The housing of mice in specific pathogen‐free conditions likely limits their exposure to microbial Ags and, as described above, MAIT cells are drastically reduced in GF conditions.2, 24 Interestingly, attempts to reconstitute GF mice with monomicrobial flora or human microbiota only recovered MAIT cell numbers to levels akin to mice housed in specific pathogen‐free conditions, thus levels well below those found in humans.29, 45 In contrast, intranasal inoculation of mice with Francisella tularensis or Salmonella Typhimurium leads to rapid expansion of MAIT cells within the lungs of infected mice, levels more consistent with those observed in humans.31, 32 Future studies will likely examine whether mice caught from the wild or obtained from pet shops have elevated MAIT cells due to increased microbial exposure. Homeostasis and Expansion of MAIT Cells Genetic variability may also contribute to differences in MAIT cell frequency. For example, Cui et al. showed that Castaneus mice, a strain of mice originally derived from the wild but housed in specific pathogen‐free conditions, had a 20‐fold increase in MAIT cell numbers compared to common laboratory strains.45 The higher number of MAIT cells in Castaneus mice was not due to differences in thymocyte survival but rather to an increase in Vα19‐Jα33 transcripts that was linked to a genetic factor mapped to the Castaneus TCRα locus.45 Intriguingly, we recently showed that MAIT cells are over‐represented in NKT cell deficient CD1d−/− mice, suggesting that NKT cells and MAIT cells compete for a similar environmental niche (Figure 1).24 This has important implications for the field because NKT cell deficient mice are often used to assess the role of NKT cells in disease models and it is unclear how MAIT cells contribute to these diseases.75, 76 Furthermore, the frequency of MAIT cells and NKT cells in mice is inversed in humans, such that humans have high numbers of MAIT cells and low numbers of NKT cells, while the reverse is true in mice.4 Hence, it will be interesting to determine what factors these innate‐like T cells are competing for, and if other cell types are also involved in this interplay. For example, it was recently shown in mice that innate lymphoid cells compete with T cells for a limited supply of IL‐7.77 Furthermore, IL‐7 is also required for the homeostasis and survival of mouse NKT17 cells.78 In contrast to these findings in mice, we recently demonstrated a direct correlation in the frequency of MAIT cells and NKT cells in blood from the same human donor.19 A similar correlation between MAIT cells and NKT cell frequencies was shown in neonates in a recent study by Ben Youssef and colleagues.25 Moreover, they also demonstrated a parallel relationship between Vα7.2+ and Vα7.2− subsets of CD161+ T cells and while the latter population did not bind MR1 tetramer, their data suggests that both populations undergo a similar postnatal development pathway.25 These reports collectively support the concept that MAIT cells, NKT cells and related lineages share environmental maintenance factors, such that when the factors are abundant, these cell populations are increased in blood. Clearly, more work is required to fully appreciate the interactions that maintain innate‐like T cells in the immune system. The high abundance of MAIT cells in humans is likely due to ongoing exposure to microbial stimuli, which drives oligoclonal expansion of MAIT cells that have a reduced TCR repertoire.7, 45, 79 The frequency of MAIT cells increases with age such that very few MAIT cells are observed in the cord blood, presumably due to limited microbial exposure in the womb, while the numbers of MAIT cells increase in the blood from neonates and steadily rise until young adulthood (~30 years of age) before declining over subsequent decades16, 18, 24, 25, 35, 52, 80 (Figure 2). It is currently unclear how reduced numbers of MAIT cells impacts on the immune system of older individuals, although it is well described that as humans age they become increasingly immunocompromised.81, 82 Thus, one key factor that underpins the variability of MAIT cells in humans appears to be age, with optimal MAIT cell numbers coinciding with optimal immunological fitness. Taken together, these data highlight how MAIT cells continue to develop and expand after they exit the thymus and this process is likely dependent on microbial stimuli. Concluding Remarks MAIT cells, on average, represent the most abundant αβ T‐cell population with a single specificity in the human immune system. Understanding MAIT cell development and homeostasis cells may provide important clues as to why MAIT cell numbers vary between individuals (and species). The recent mapping of a three‐stage MAIT cell development pathway in mice and humans is facilitating further studies to determine the factors that govern this process. There appears to be functional overlap of MAIT cells with other innate‐like T‐cell populations and we must carefully assess the role of these different cell types in the context of microbial immunity. It will also be interesting to address how other related T‐cell subsets develop, including atypical MR1‐restricted Vα7.2− T cells and populations of Vα7.2−CD161+ T cells, that do not bind MR1 tetramer but share similar characteristics with MAIT cells. Future studies into the development and homeostasis of MAIT cells (and related subsets thereof) may lead to the identification of novel targets to modulate the numbers of these cells which would be valuable in settings where they are depleted, potentially leading to a compromised immune system, such as in cancer and HIV patients. Acknowledgments DGP is supported by a National Health and Medical Research Council (NHMRC) Career Development Fellowship (1144308). DIG is supported by an NHMRC Senior Principal Research Fellowship (1020770 and 1117766). The authors are supported by NHMRC project grants 1145373, 1122890 and 1140126 and the Australian Research Council (ARC; CE140100011). Conflict of Interest The authors declare no conflict of interest. References Citing Literature Number of times cited according to CrossRef: 2 Hui‐Fern Koay and Dale I Godfrey, MicroRNA‐managing the development of MAIT cells, Immunology & Cell Biology, 97, 2, (121-123), (2019). Wiley Online Library Erik Seth, Christine L. Zimmer, Marcus Reuterwall‐Hansson, Ammar Barakat, Urban Arnelo, Annika Bergquist, Martin A. Ivarsson and Niklas K. Björkström, Primary sclerosing cholangitis leads to dysfunction and loss of MAIT cells, European Journal of Immunology, 48, 12, (1997-2004), (2018). Wiley Online Library

Journal of Leukocyte Biology considers manuscripts of original investigations focusing on the origins, developmental biology, biochemistry and functions of granulocytes, lymphocytes, mononuclear phagocytes, and other cells involved in host defense.