Agrobacterium tumefaciens enables genetic transformation of many plant species via the movement of transferred-DNA (T-DNA) of Ti plasmids into the plant nucleus. Despite its use in both basic research and molecular breeding in several crops, many species and varieties are still recalcitrant to Agrobacterium-mediated transformation. To overcome this limitation requires not only optimization of cell and tissue culture conditions but also elucidation of the molecular mechanisms of all the events that occur during Agrobacterium-mediated transformation.

In the conventional Agrobacterium-mediated transformation system in rice, callus derived from the scutellum of mature seeds or immature embryos is generally used for Agrobacterium inoculation (Nishimura et al., 2006; Hiei and Komari, 2008). Since these tissues have a stereo architecture consisting of many cells, Agrobacterium can only infect cells on the surface, and, unlike protoplasts, it is quite difficult to isolate single cells from these tissues without any selection pressure. Indeed, there are far fewer transformed than non-transformed cells when Agrobacterium-inoculated primary calli derived from mature seeds are cultured without selection pressure in a conventional transformation system (Saika and Toki, 2009; Saika et al., 2011b). Therefore, transformed cells in which antibiotic- and herbicide-resistance genes located on T-DNA are present, but not expressed to a sufficient level, cannot be distinguished from non-transformed cells.

Agrobacterium-mediated transformation has many steps (Gelvin, 2010; Pitzschke and Hirt, 2010; Magori and Citovsky, 2011; Shiboleth and Tzfira, 2012). The last step, i.e., stable transgene expression that results from T-DNA integration into the host genome (and the avoidance of the gene silencing system) is crucial to the clonal propagation of transformed cells of rice as referred to above. Meanwhile, transient expression of transgenes is often observed at an earlier stage when transgenes are expressed stably and constantly from T-DNA that has not integrated into the rice genome. After import into the nucleus of the T-complex, which consists of single stranded T-DNA (ssT-DNA) and proteins such as virD2 and virE2, ssT-DNA is replicated to double-strand T-DNA (dsT-DNA). Hypothetical models of dsT-DNA formation have been proposed as shown in Figure 1 (Liang and Tzfira, 2013). Subsequently, transgenes located on the dsT-DNA are expressed. However, this expression is not continuous because naked T-DNA that has not integrated into the host genome is susceptible to degradation (Gelvin, 2010). The transient expression of a cytokinin biosynthesis gene, which leaves no selection marker or vector backbone in the host genome, has been exploited in a transformation system for Solanaceous plants (Richael et al., 2008). However, it is thought that transient transgene expression is not crucial to the production of clonally propagated transformed cells in the case of rice.

The molecular mechanisms of T-DNA integration into the plant genome in Agrobacterium-mediated transformation remain unclear. The sequences of T-DNA integration sites were determined in Arabidopsis and tobacco plants over 20 years ago (Gheysen et al., 1991; Mayerhofer et al., 1991). Based on these pioneer reports, two major models—the strand-invasion model and the double strand break (DSB) repair model—were proposed (Gelvin, 2010; Pitzschke and Hirt, 2010; Magori and Citovsky, 2011; Shiboleth and Tzfira, 2012). In the strand-invasion model, T-DNA integration is thought to be caused by microhomology-mediated repair between ssT-DNA and the plant genome. On the other hand, in the DSB repair model, dsT-DNA is thought to integrate via non-homologous end-joining (NHEJ)—one of the main DSB repair pathways—into the DSB sites that occur randomly in the plant genome (Figure 1).

Many proteins involved in the NHEJ pathway, such as Ku70, Ku80, Ligase 4 (Lig4), and Xrcc4, have been characterized in model plants such as Arabidopsis and rice (West et al., 2000; Riha et al., 2002; Kimura and Sakaguchi, 2006; Hong et al., 2010; Nishizawa-Yokoi et al., 2012). To summarize previous reports in Arabidopsis, on the whole, the frequency of stable transformation tends to decrease in NHEJ mutants except for xrcc4, although this point remains controversial. It was reported initially that the frequency of stable transformation increased twofold in in planta floral dip transformation experiments in Arabidopsis ku80 mutants (Gallego et al., 2003), while another report demonstrated a two to threefold decrease (Friesner and Britt, 2003). However, a root tumorigenesis assay in Arabidopsis ku80 mutants confirmed that the stable transformation frequency was decreased severely, although the frequency of transient transformation remained comparable to controls (Li et al., 2005b). In addition, overexpression of Ku80 enhanced stable but not transient transformation frequency (Li et al., 2005b). Moreover, recent reports have shown that stable transformation frequency was decreased two to fourfold in Arabidopsis ku80 mutants in both root tumorigenesis assay and in in planta floral dip assay (Jia et al., 2012; Mestiri et al., 2014), and stable transformation frequency was also decreased fourfold in Arabidopsis ku70 mutants in the in planta floral dip assay (Jia et al., 2012). Thus, Arabidopsis Ku70/80 plays a positive role in stable transformation. It was proposed that Ku80 proteins interact with dsT-DNA molecules and direct them to DSB sites (Li et al., 2005b). On the other hand, there are contrary reports in Arabidopsis lig4 mutants: some authors found that stable transformation frequency was decreased to 40–70% in in planta floral dip assay (Friesner and Britt, 2003), but others found comparable rates to wild-type in both root tumorigenesis assay and in planta floral dip assay (van Attikum et al., 2003). These inconsistencies are probably due to differences in cell type (somatic cells vs germ cells) and/or experimental procedures (Gelvin, 2010; Nishizawa-Yokoi et al., 2012; Lacroix and Citovsky, 2013). Interestingly, the possibility that Xrcc4 plays a negative role in stable transformation via interaction with virE2 has been raised (Vaghchhipawala et al., 2012). In this latter theory, active Xrcc4 proteins are inhibited by virE2 proteins and T-DNA is integrated into DSB sites that are either not repaired or that occur de novo by suppression of the NHEJ pathway at a higher frequency. Agrobacterium might bring stable transformation to a successful conclusion by the good use of host proteins involved in the NHEJ pathway, among others Ku80 and Xrcc4. Just recently, Mestiri et al. (2014) reported that DSB repair pathways including NHEJ are redundantly involved in stable transformation. Stable transformation frequencies were decreased in the single mutants, ku80, xrcc1, xrcc2, and xpf (Mestiri et al., 2014), which are deficient in NHEJ, Ku-independent NHEJ (known as alternative NHEJ to distinguish it from classical Ku-dependent NHEJ), microhomology-mediated end joining/single strand annealing and homologous recombination, respectively. Interestingly, stable transformation was markedly but not completely suppressed in a quadruple mutant (Mestiri et al., 2014), possibly because the quadruple mutant did not lose DSB repair activity completely (Charbonnel et al., 2011).

Unlike Arabidopsis, rice is not a natural host plant of Agrobacterium; however, the combination of a sophisticated tissue culture system and activation of Agrobacterium using the phenolic compound acetosyringon, allows transgenic rice plants to be produced via this method (Hiei et al., 1994). Human cells and yeast—also not natural hosts of Agrobacterium—are also susceptible to Agrobacterium-mediated transformation (Bundock et al., 1995; Kunik et al., 2001). Moreover, the NHEJ pathway is involved in T-DNA integration in yeast (van Attikum et al., 2001; van Attikum and Hooykaas, 2003). These findings suggest that Agrobacterium uses a fundamental DNA repair system that is common to both host and non-host organisms, including rice, in the step leading to stable transformation. Since no conventional root transformation system or in planta floral dip transformation system has been established in rice, we are currently restricted to rice callus for analyzing stable transformation frequencies. Tumorigenesis assays are not applicable to rice callus because this tissue has already dedifferentiated. In Arabidopsis and Nicotiana, stable transformation frequencies are estimated by visualization of transformed cells (Mysore et al., 1998; Anand et al., 2007). In rice, transient and stable transgene expression can also be observed using green fluorescent protein—a non-destructive and visible selection marker (Toki et al., 2006; Saika et al., 2012). Visualization of transformed cells can quantify stable transformation events rapidly and easily compared to clonal propagation of transformed cells using antibiotics. We have established a sequential monitoring system for stable transformation events in rice callus that uses the enhancer trap of the click beetle luciferase gene, which is 30 times more sensitive than firefly luciferase (Saika et al., 2012). Using this system, we showed that the stable transformation frequency is decreased in knock-down lines of OsKu70, OsKu80, and OsLig4 genes and the knock-out line of the OsKu70 gene (Nishizawa-Yokoi et al., 2012; Saika et al., 2012), suggesting that the NHEJ pathway is involved in the stable transformation process also in rice.

According to the DSB repair model, DSBs could occur in cells successfully transformed with T-DNA. In Arabidopsis, DSBs induce cell death and endocycle, which halts cell division (Adachi et al., 2011). Cells in which these events occur cannot propagate clonally if stable transformation is successful. Thus, transformation frequency in Arabidopsis might be underestimated in mutants in which DSBs occurs at higher frequency, such as NHEJ deficient mutants. However, in rice callus, endocycle does not occur even under genotoxic conditions inducing DSBs (Endo et al., 2012). Thus, transformation frequency may be estimated more accurately in rice callus than in Arabidopsis.

Transformation frequency in Arabidopsis and rice depends on the variety (Nam et al., 1997; Saika and Toki, 2010). Furthermore, optimal conditions for cell and tissue culture differ among rice varieties (Hiei and Komari, 2008). For example, Koshihikari—an elite variety from Japan—shows lower regeneration frequency due to lower activity of nitrite reductase (Nishimura et al., 2005; Ozawa and Kawahigashi, 2006). This makes it difficult to compare transformation frequency precisely among rice varieties since different callus culture conditions can affect Agrobacterium activity and growth. Thus, in order to compare transformation frequency precisely, experiments must be performed under reliable, reproducible and identical conditions using mutants with the same genetic background. Recent technological advances have made it much easier to produce plants with knockouts of targeted genes: successful gene knockouts using artificial nucleases such as transcription activator-like effector nucleases (TALENs) and CRISPR/Cas have already been reported in rice (Li et al., 2012; Feng et al., 2013; Jiang et al., 2013; Miao et al., 2013; Shan et al., 2013; Xu et al., 2014). A series of mutants in which genes involved in the NHEJ pathway are disrupted can now be produced easily in the same variety. Moreover, targeted mutagenesis using artificial nucleases enables “null” and “truncated” mutants to be produced as necessary, unlike conventional mutagenesis approaches and T-DNA tagging lines. These new technologies could help resolve some of the conflicting results in this field. For example, the involvement in stable transformation of the Arabidopsis lig4 mutant described above remains controversial, and conflicting results have also been reported in mutants in pathways other than NHEJ. For example, the plant protein VirE2 binding protein 1 (VIP1) has long been considered important in transient and/or stable transformation (Tzfira et al., 2001, 2002; Li et al., 2005a), but results that dispute this view have been reported recently (Shi et al., 2014). A study using knockout mutants of Lig4 and Vip1 in rice produced by artificial nucleases could lead to an answer.

Gene targeting (GT) is a transformation technology that can modify a targeted gene in a predicted manner. Successful examples of gene modification via GT have been reported in several higher plants including rice and Arabidopsis. However, rice is currently the only flowering plant in which target modification via GT can be performed routinely (Endo et al., 2007; Saika et al., 2011a; Nishizawa-Yokoi et al., 2014; Osakabe et al., 2014). In particular, GT using positive-negative selection enables the introduction of desirable mutations that cause amino acid substitutions in the targeted rice gene (Wakasa et al., 2012; Dang et al., 2013; Nishizawa-Yokoi et al., 2014). Molecular analyses and structure-based protein engineering can reveal the essential amino acids involved in protein–protein interactions. For example, as mentioned above, Agrobacterium virE2 protein interacts with Xrcc4 proteins (Vaghchhipawala et al., 2012) It is easy to produce mutant rice plants in which mutated Xrcc4 proteins are expressed. The comparison of transient/stable transformation frequencies in xrcc4 mutant rice lines deficient in the interaction with virE2 protein or the enhancement of Lig4 activity will be able to provide clues as to the exact roles of Xrcc4 protein in stable transformation. Similarly, a set of rice mutants expressing mutated proteins that have lost the ability to interact with proteins derived from Agrobacterium will be useful in analyzing transformation frequency. Such studies will offer new insights into the stable transformation process. Moreover, further analysis will enable the discovery of chemicals effective in strictly activating or suppressing stable transformation, and these could be applied not only in experimental procedures in plants that are recalcitrant to Agrobacterium-mediated transformation but also in cases where T-DNA integration is not desired, such as site-directed mutagenesis using artificial nucleases and GT.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.