Peter M. Waterhouse & Christopher A. Helliwell   about the authors

CSIRO Plant Industry, Canberra, Australian Capital Territory 2601, Australia.

correspondence to: Peter M. Waterhouse peter.waterhouse@csiro.au

The nucleotide sequences of several animal, plant and bacterial genomes are now known, but the functions of many of the proteins that they are predicted to encode remain unclear. RNA interference is a gene-silencing technology that is being used successfully to investigate gene function in several organisms — for example, Caenorhabditis elegans. We discuss here that RNA-induced gene silencing approaches are also likely to be effective for investigating plant gene function in a high-throughput, genome-wide manner.

The genomes of several animal species and two plant species, Arabidopsis thaliana¹ and Oryza sativa (rice)^{2, 3}, have been sequenced. Analyses of sequence data from these two plant species indicate that there might be 25,000 protein-encoding genes in Arabidopsis and up to 55,000 in rice. A key challenge is now to identify how each of these genes functions in the growth and development of these plants.

Reverse genetics, in which a gene is disrupted so that the effect (if any) of its loss on an organism can be observed, is a simple way to investigate gene function. A reverse-genetic approach called insertional mutagenesis has been a key in studying Arabidopsis. Two insertional mutagenesis strategies — one based on transferred DNA (T-DNA)⁴, the other on transposon tagging^{5, 6} — have been developed in this plant and have given rise to several large, publicly available collections of Arabidopsis insertion mutants (such as the Arabidopsis Stock Centre and the Salk Collection at The Arabidopsis Information Resource (TAIR)). Although these collections have provided the plant research community with a superb resource of null mutants, the approach of using insertion mutants has several limitations. For example, it cannot be used to investigate the functions of duplicated genes, and many mutant phenotypes in these lines are caused by disruptions to genes other than those into which the DNA tag is inserted. Antisense and CO-SUPPRESSION⁷ technologies have also been useful tools for investigating gene function in plants, but they are labour intensive and somewhat unpredictable⁸.

One approach that can circumvent these limitations is RNA-induced gene silencing. It is variously termed post-transcriptional gene silencing (PTGS) in plants, quelling in fungi and RNA interference (RNAi) in animals^9-18. The essence of RNA-induced gene-silencing is the delivery of double-stranded RNA (dsRNA) into an organism, or cell, to induce a sequence-specific RNA degradation mechanism that effectively silences a targeted gene. How RNAi operates and its natural role for virus defence and endogenous gene regulation in plants have been reviewed elsewhere^{8, 19-23}, and so are not the focus of this review. Instead, we discuss and compare the various ways in which RNAi can be carried out in plants and assess how effective each technology is for high-throughput, genome-wide studies of plant gene function. Although we use the term RNAi, in place of PTGS, to describe this process in plants, the use of RNAi to describe RNA-induced gene silencing in plants remains a matter of debate.

Inducing RNA interference in plants

When naturally occurring viral RNA (that produces dsRNA during its replication), self-complementary, single-stranded 'hairpin' RNA (hpRNA) or dsRNA, is introduced into a plant, it is degraded into 21-nucleotide dsRNA fragments, known as small interfering RNAs (siRNAs). These siRNA fragments are then incorporated into a nuclease-containing complex called RISC (RNAi silencing complex), which degrades mRNAs that are complementary to the single-stranded siRNA that is associated with the complex²³ (Fig. 1).

		Figure 1 |  The current model of RNA-mediated gene silencing in plants.  This model is based on the results of in vitro studies of RNA-induced gene silencing, or RNA interference (RNAi), in animal extracts (reviewed in Ref. 23). RNAi is believed to operate in a similar manner in plants because small interfering RNAs (siRNAs) are found in silenced plants, and plants have homologues of the animal gene Dicer. Double-stranded RNA (dsRNA) from replicating viral RNA, viral-vector-derived (VIGS, or virus-induced gene silencing) RNA or hairpin RNA (hpRNA) transcribed from a transgene, is processed by a Dicer-containing complex to generate siRNAs. An endonuclease-containing complex (called the RNAi silencing complex, RISC), is guided by the antisense strand of the siRNA to cleave specific mRNAs, so promoting their degradation.

An important aspect of using RNAi in plant genomics research is the delivery of the silence-inducing dsRNA or hpRNA. This RNA can be delivered by stably transforming plants with transgenes that encode hpRNAs or viral RNAs. It can also be transiently delivered by bombarding plants with nucleic-acid-coated beads, by infiltrating plant cells with transgene-carrying AGROBACTERIUM TUMEFACIENS or by infecting plants with a virus, either on its own or together with a satellite virus. Each delivery method has its own advantages and disadvantages (see below).

Microprojectile bombardment and agroinfiltration. The bombardment of plant tissues with gold or tungsten particles that have been coated with DNA is an approach that has been used routinely for studying gene expression in plants and for transformation of several MONOCOTYLEDON (monocot) plant species, such as rice²⁴. The DNA or RNA that coats each particle is released and expressed in the cells where the particles come to rest. This procedure has also recently been used to induce RNAi in plants by delivering dsRNA or DNA constructs that encode hpRNA into leaf epidermal cells of the cereal species, maize, wheat and barley²⁵. The bombardment of monocot leaves with in vitro transcribed dsRNA of A1 (a gene from the anthocyanin biosynthetic pathway), MLO (a negative regulator of the resistance to barley powdery mildew) or -GLUCURONIDASE (GUS) sequences²⁵, resulted in the reduced activity of each of these target genes. RNAi has also been induced against a GFP (green fluorescent protein) transgene in Nicotiana benthamiana. In this study, N. benthamiana plants were bombarded with particles coated with one of several types of silence-inducing molecules: dsRNA; siRNA; a plasmid DNA that encoded single-stranded GFP RNA²⁶; and dsDNA that targeted the transgene²⁷. In each case, the visually scorable phenotype of suppressed GFP fluorescence was observed. In another study, a repressor of GUS in a transgenic tobacco plant (Nicotiana tabacum) was silenced by bombarding the plant with dsRNA and siRNA, and with sense or antisense RNAs. Each approach induced the production of siRNAs that silenced the repressor, resulting in GUS expression²⁶.

When Agrobacterium tumefaciens infects a plant, it transfers part of its T-DNA plasmid into the genome of the infected plant cells. The genes on the T-DNA seem to be expressed in the plant cell both during the transfer process and after the T-DNA has integrated into the plant genome. Infiltrating leaves²⁸ (using a needle-less syringe to pass liquid through the STOMATA into the intercellular spaces) with a culture of Agrobacterium, in which the T-DNA plasmid contains a transgene that encodes an endogenous plant gene sequence, can trigger RNAi against the target endogenous gene, although it is not clear how this single, sense transcript generates dsRNA. Nevertheless, RNAi by this method has been shown in transgenic N. benthamiana plants that express GFP^{29, 30}. In the first days after infiltration, GFP is overexpressed at the site of infiltration, but after about three days this expression subsides to undetectable levels, concomitant with a reduction of endogenous GFP expression. This localized silencing subsequently spreads throughout the plant. Similar, but more potent, silencing has been found to occur when T-DNAs that contain hpRNA-encoding sequences are used for agroinfiltration³⁰.

Virus-induced gene silencing. Most plant viruses have single-stranded RNA genomes, which are released from the protein coat of their virus particles as they enter a cell. Their genomic RNA is then replicated by the virus-encoded RNA-dependent RNA polymerase to produce sense and antisense RNA. These viral RNAs (which have the potential to hybridize to form dsRNA) trigger an RNAi response against their sequences^31-33 (Fig. 1).

Viruses have several features that make them particularly useful to plant researchers. The naked RNAs — that is, viral RNAs without the protection of a virus particle — of several plant viruses can be applied to plants to cause infection. These RNAs can be generated by in vitro transcription of a cDNA clone that encodes a complete virus sequence (Fig. 2a). Similarly, a full-length virus cDNA clone that has been placed into a T-DNA plasmid, expressed under the control of a cauliflower mosaic virus (CAMV) 35S PROMOTER (Fig. 2b,c) and delivered to plants by agroinfiltration, can produce viral infection when it is transcribed by the plant³⁴.

		Figure 2 |  DNA constructs for RNA-mediated gene silencing.  a | A DNA plasmid that can be propagated in Escherichia coli from which infectious potato virus X (PVX) RNA can be transcribed in vitro, using T7 polymerase. The PVX cassette contains sequence derived from the gene to be targeted. b | A transferred (T)-DNA plasmid that is propagated in Agrobacterium. When this plasmid-carrying Agrobacterium is inoculated onto a plant, it transfers the DNA between its left (LB) and right (RB) borders into the plant's cells. The region between the borders contains the viral sequences shown in part a, but in this vector, the T7 promoter has been replaced with the cauliflower mosaic virus promoter. This enables the transferred DNA to be transcribed by the plant's endogenous transcription machinery to generate infectious PVX (plus insert sequence) RNA. In amplicon transgene vectors, a selectable maker gene is also present between the left and right borders of this plasmid, enabling plants to be stably transformed with the transferred DNA. c | The tobacco rattle virus (TRV) virus-induced gene-silencing (VIGS) system. Two T-DNA plasmids that encode the TRV genome (one encoding TRV RNA1 and the other encoding TRV RNA2, which carries the inserted target sequence) are propagated separately in Agrobacterium and used to co-infect plant tissue. d | A typical T-DNA plasmid for the expression of hairpin RNAs (hpRNAs). This plasmid can be transiently introduced into plants by bombardment or stably introduced by agroinfiltration. A generic silencing precursor construct (pHANNIBAL) that enables hpRNA vectors to be easily constructed has different multiple cloning sites either side of the intron to enable the rapid insertion of target sequences in forward and reverse orientations. 35S, CaMV 35S promoter; CP, coat protein; M1,2,3, movement proteins 1, 2, 3; RdRP, RNA-dependent RNA polymerase; T7, T7 promoter; Term, transcription termination sequence.

It is also possible to introduce exogenous sequences into specific locations in the genome of a virus, and retain the infectivity of the RNA transcript. When these transcripts are used to infect plants, the foreign sequences also induce, and become the target of, the RNAi response of the host plant. This ability of viruses to carry and induce RNAi against foreign sequences has been harnessed into a technology referred to as virus-induced gene silencing (VIGS)³³.

The first demonstration of VIGS by Monto Kumagai and colleagues³⁵ came well before the concept of RNAi had been developed. A fragment of an endogenous plant gene, phytoene desaturase (PDS), which encodes an enzyme of the carotenoid biosynthesis pathway, was inserted into an infectious clone of tobacco mosaic virus (TMV), so that the virus would produce either a sense or an antisense PDS-derived RNA. The tobacco plants infected with either modified virus showed marked photobleaching. This result showed that the endogenous PDS gene of the plant was silenced because tissues that lack carotenoids photobleach under strong light conditions. David Baulcombe's laboratory has developed robust and refined VIGS systems that use potato virus X (PVX)³² and tobacco rattle virus (TRV)³⁴. There are now infectious clones of several plant viruses that have been used as VIGS vectors^32-46, most of which have RNA genomes. However, some DNA viruses (mainly the GEMINIVIRUSES), which possibly produce dsRNA by transcriptional readthrough beyond their terminators, have also been found to induce RNAi^39-42.

Usually, 300–800-nucleotide fragments of target gene sequences are used in VIGS systems (Table 1), but sequences as short as 23–60 nucleotides can be effective^{37, 47}. Many of the different VIGS vectors have been tested against the reporter genes GUS or GFP, and the endogenous gene PDS. TRV–VIGS has also been used against a range of endogenous genes, especially in N. benthamiana, and has been particularly useful for identifying genes involved in disease resistance and ethylene signalling (Table 1). PVX, TMV and many other viruses do not infect plant MERISTEMS, but TRV seems to infect almost all tissues, including meristems and floral organs. The widespread distribution of TRV throughout the plant probably makes TRV–VIGS^{34, 43-46}the better system.

Table 1 | Examples of endogenous plant genes silenced by VIGS and their phenotypes

The TRV genome comprises two RNAs: RNA1, which encodes several genes, including the RNA-dependent RNA polymerase; and RNA2, which encodes the coat protein. The target gene sequence is inserted into RNA2, downstream of the gene that encodes the coat protein, and this modified RNA is co-inoculated with unmodified RNA1 onto a plant to generate an infection and induce VIGS. These viral RNAs can be made by transcribing in vitro the appropriate DNA constructs, but an easier way is to use the TRV–AgroVIGS constructs (Fig. 2c). In this system⁴⁴, a T-DNA plasmid that contains the CaMV 35S promoter and encodes RNA1 is propagated in one culture of Agrobacterium, and a similar construct that encodes RNA2 (and the target sequence) is propagated in a different Agrobacterium culture. An aliquot of each culture is mixed together and infiltrated into plants to initiate infection and RNAi.

Some geminivirus–VIGS systems also seem to be promising tools for genomic research. The tomato golden mosaic geminivirus-based VIGS system is an effective initiator of targeted RNAi³⁹ in N. benthamiana, and has been shown to be capable of silencing the expression of the endogenous proliferating cell nuclear antigen (PCNA) gene in meristem tissues⁴¹. The cabbage leaf curl geminivirus (CbLCV) has the potential to be even more useful, because a VIGS system based on this virus has recently been shown to infect and induce RNAi in Arabidopsis⁴².

A new viral silencing system, using SATELLITE RNAS, has also been recently developed. In this satellite-virus-induced silencing system (SVISS)⁴⁸, the target sequence is inserted into the satellite RNA, which is then co-inoculated with the associated virus. SVISS has been shown to work with more than ten genes, including PDS, plastid transketolase and CESA (cellulose synthase A), using a satellite of the U2 strain of TMV in tobacco⁴⁸ (Fig. 3).

		Figure 3 |  Tobacco plant phenotypes after infection with a satellite-virus-induced silencing system.  Results show plant phenotypes four weeks after infection. Phenotypes caused by the silencing of the genes that encode a | cellulose synthase, b | transketolase and c | phytoene desaturase are shown. Images courtesy of M. Metzlaff, Ghent, Belgium. Reproduced with permission from Ref. 48 © (2002) Blackwell Publishing.

Amplicon and hairpin RNA transgenes. Amplicon and hpRNA transgenes are stably transformed in plants, and they ensure that the RNAi they induce is inherited by subsequent generations. An amplicon transgene encodes a virus-derived transcript that contains a target gene sequence, but not necessarily that of all the genes of the native virus, and can direct its own replication. The transgene is usually under the control of the CaMV 35S promoter (Fig. 2), so that the amplicon RNA is expressed in almost every cell of the plant, although the transfected plants do not show symptoms of being infected with virus⁴⁹. Amplicon constructs have not been as widely used as VIGS. They have been based mainly on PVX and have been used successfully to silence the reporter transgenes, GUS⁴⁹ and GFP, and the endogenous genes DWARF, RUBISCO and PDS⁵⁰ in tobacco, N. benthamiana and tomato. An amplicon construct based on tobacco yellow dwarf geminivirus has also been used to silence chalcone synthase in petunia⁵¹. Interestingly, a PVX–GFP amplicon transgene induced GFP-specific RNAi in Arabidopsis⁵², even though Arabidopsis is not a host of PVX. This indicates that amplicons might overcome the host specificities that restrict the use of VIGS to the host range of the virus.

An alternative method of inducing RNAi in plants is to use transgenes that express a self-complementary hpRNA. A hpRNA forms when an RNA hybridizes with itself to form a hairpin structure that comprises a single-stranded loop region and a base-paired stem, which mimics dsRNA and provides the specificity of the RNAi. (The loop plays no part in directing RNAi.) The frequency with which RNAi can be effectively induced in transformed plants can be increased by using hpRNA transgenes in which the loop size is minimized by using an intron^{52, 53} (Fig. 2d). The induction of RNAi by a transgene that encodes an hpRNA was first shown by using the GUS reporter gene in rice¹². In this experiment, the construct was under the control of the strong and constitutive maize ubiquitin promoter. So far, most hpRNA constructs have been used in DICOTYLEDONS and have been expressed under the control of the strong, constitutive CaMV 35S promoter^53-56. Indeed, a generic vector, pHANNIBAL⁵⁴, incorporating the CaMV 35S promoter, has been created, which enables hpRNA constructs to be easily generated (Fig. 2d). However, seed-specific promoters, such as the napin and lectin promoters^{53, 54, 57} have also been effective at silencing seed-expressed genes. A wide variety of genes, ranging from transcription factors to metabolic biosynthesis enzymes, as well as viral sequences, have been effectively silenced using hpRNA transgenes (Table 2), and many of the phenotypes obtained have been similar to those of counterpart insertion mutants. As with VIGS, most of the gene fragments that have been encoded by hpRNA vectors have been between 300 and 800 nucleotides, although a fragment as small as 98 bases has been effective⁵⁴. Results with hpRNAs indicate that 5' or 3' untranslated regions or the coding region of an mRNA might all be good silencing targets⁵⁴.

Table 2 | Examples of endogenous plant genes silenced by hpRNAs and their phenotypes

Transcriptional gene silencing

A recently discovered feature of RNAi in plants is that dsRNA induces and directs not only sequence-specific RNA degradation, but also sequence-specific DNA methylation^58-60. The VIGS or hpRNA constructs described so far in this review use sequences contained in the mRNA of target genes to bring about silencing. This often results in the concomitant methylation of the coding region of the target gene, which usually has no direct effect on its transcription. However, in three RNAi studies in plants, a gene promoter has been targeted, resulting in promoter methylation. This promoter methylation, or changes in chromatin conformation as a result of methylation at the promoter, presumably affect the binding of transcription factors, leading to reduced transcription (Fig. 4). In these experiments, hpRNAs were targeted against transgene promoters^{58, 61} or against an endogenous promoter that controls the expression of dihydrofolate reductase⁶¹. In another study, a viral vector was used to target the CaMV 35S promoter of a transgene⁶². This process of hpRNA-directed transcriptional gene silencing might become a useful weapon in the armoury of plant genome research.

		Figure 4 |  The current model of hairpin-RNA-directed transcriptional gene silencing in plants.  Hairpin RNA (hpRNA) directed against promoter sequences gives rise to the methylation of cytosine residues (shown as red circles) at the targeted DNA sequence. This methylation directly, or indirectly, causes changes in the conformation of local chromatin, resulting in gene silencing by loss of transcription. dsRNA, double-stranded RNA; Met, de novo methyltransferase; siRNA, small interfering RNA.

Comparison of silencing technologies

Some of the advantages and disadvantages of transient and stably integrated gene-silencing systems in plants are compared in Table 3. Microprojectile bombardment and non-viral agroinfiltration induction of RNAi have the potential to be valuable tools for rapidly identifying gene functions where cellular, rather than whole-plant, assays can be used. Their main disadvantages are that the silencing they induce usually only persists for a few days and, as yet, non-viral agroinfiltration-mediated silencing has only been shown in N. benthamiana and tobacco.

Table 3 | Advantages/disadvantages of transient- and stable-integration gene-silencing systems

So far, there are only a handful of examples in which amplicon-mediated RNAi and directed transcriptional gene silencing (TGS) have been used to silence genes, making it difficult to assess their potential impact on plant genomic research. The rules for TGS are far less resolved than those for RNAi. For example, answers are needed about which, and what size, promoter regions to target, and whether the methylation of specific cytosines or the alteration of the local chromatin structure by the methylation of some cytosines in the region, is sufficient to inhibit transcription. Amplicons share with hpRNA-encoding transgenes the advantages and disadvantages of delivery from stably inherited transgenes, and they share with VIGS the advantages and disadvantages of being induced by a modified replicating virus (see below).

At present, the two most widely used systems are VIGS in N. benthamiana and hpRNA transgenes in Arabidopsis. Clearly, both systems effectively and efficiently produce phenotypes (Tables 1,2) in plants, so providing insights into the functions of the target genes. Transformation of plants with hpRNA constructs gives stable silencing that is inherited from generation to generation, thereby enabling the continued study of a phenotype. For example, the silencing of the Arabidopsis fatty acid desaturase 2 (FAD2) gene by a hpRNA has been maintained for five generations and has been found to be consistent in its effectiveness⁵⁷. VIGS is particularly useful in plant species that are difficult or impossible to transform (if there is an VIGS system available for the species) or when analysing genes that are essential, either for housekeeping functions or for embryonic development (which cause embryonic lethality when knocked out), as the infectious transcripts can be applied to mature plants. VIGS is also well suited to the high-throughput analysis of genes, as VIGS vectors can be delivered by simply rubbing an infectious transcript onto a plant, although the cost of transcribing each construct in vitro might be prohibitively expensive on a large scale. The development of Agro-VIGS circumvents this problem because, in this approach, Agrobacterium cultures of the VIGS constructs can be inexpensively grown and injected^43-46.

Most viruses used for VIGS have a limited number of hosts, and the virus–host combination seems to be a crucial factor in determining the efficacy of silencing. Some of the viruses used in VIGS can cause symptoms that might mask the phenotype caused by the silencing of the target gene. Moreover, many viruses do not infect the growing points or floral parts of plants, especially the seed, precluding gene silencing in these tissues. TRV–VIGS in N. benthamiana overcomes many of these problems. TRV infects almost all tissues of the plant and produces mild symptoms. However, N. benthamiana seems to be a special plant as it somehow enables several viruses to maintain their foreign inserts and to disperse and cause silencing throughout the plant. By contrast — for example, in inoculated tobacco leaves — TMV⁴⁵ and PVX vectors that contain GFP do not spread from these leaves unless they recombine out the GFP sequence. However, the same chimeric viruses spread rapidly and express GFP throughout N. benthamiana plants.

Where it has been measured, TRV–VIGS reduces the target mRNA level by 80–90% (Refs 44–46). By contrast, transforming plants with hpRNA constructs typically generates a series of independent lines that have different phenotypes and degrees of target mRNA reduction⁶³. In many hpRNA experiments, levels of the targeted mRNA have ranged from wild type to undetectable^{57, 63}. In experiments in which genes have been targeted with intron-containing hpRNA constructs, silencing usually causes easily visible or measurable effects in 70–100% of the resulting plants^{54, 63}. For example, each of 31 independent plants transformed with an intron-containing hpRNA construct against flowering locus C (FLC) flowered earlier than wild-type plants⁵⁴. FLC represses flowering and provides an easily measurable phenotype; the shorter the period between germination and flowering, the more profound the silencing. These lines showed a range of flowering times, indicating that each line has a different degree of silencing. A similar range of silencing has also been seen in hpRNA constructs targeted against the Arabidopsis genes: hpRNA silencing of PDS (Fig. 5) and of FAD2 causes photobleaching and the production of a range of seed oil profiles, respectively⁵⁷. In such cases, 10–20% of the independent transformants had phenotypes that were indistinguishable from those of the corresponding null mutation. This heritable range of degrees of silencing in hpRNA plants is a useful feature that can give a range of phenotypic effects that might provide new insights into gene function.

		Figure 5 |  Degrees of silencing produced by hairpin-RNA-encoding transgenes.  The stable transformation of Arabidopsis plants with the same hairpin RNA (hpRNA) construct that is targeted against phytoene desaturase gives rise to lines that show a heritable photobleaching phenotype in: a | all tissues; b | sectors of tissue; or c | the cotyledons, but not the rest of the plant. Images courtesy of C.A.H. and P.M.W., CSIRO, Australia. Reproduced with permission from Ref. 63 © (2002) CSIRO Publishing.

Future prospects

Researchers using Caenorhabditis elegans as a model system have led the way in using RNAi as a tool to test gene function on a genome-wide scale^64-66. This has been feasible largely because of the ease with which RNAi can be delivered to worms, by either soaking them in dsRNA⁶⁵ or feeding them with bacteria that express dsRNA⁶⁶.

In plants, most of the published work on RNAi has been devoted to describing studies that investigated its mechanism, the development of RNAi delivery systems or its use to validate the already known or inferred functions of genes. However, genome-wide gene-function and gene-discovery studies using VIGS^{43, 44} and hpRNA transgenes are under way (Box 1). The pHELLSGATE^{54, 63} and the pTRV–attP/R⁴⁴ vector series (Box 1), which use the rapid recombination-based Gateway^TM technology (see link to Gateway Technology), promise to facilitate the high-throughput cloning and manipulation of gene libraries that is required for creating RNAi constructs. Compared with conventional restriction enzyme cloning, the transfer of gene inserts into Gateway vectors is reliable and efficient, enabling numerous inserts to be moved to the vector of choice (Box 1). Gateway-cloned Arabidopsis cDNA libraries (see link to Construction of an Arabidopsis open reading frame library) and Arabidopsis gene sequence tags (Box 1) are obvious sources of the gene fragments that are required for RNAi-based studies of gene function. Coordinated genomic investigations built around these libraries can be envisaged. Experiments using microarrays are beginning to identify many of the genes that show expression changes in response to external stimuli⁶⁷, that are expressed in a particular tissue⁶⁸ or that are altered in a mutant⁶⁹. So far, RNAi in nematodes has revealed phenotypes and possible functions for 13–27% of the 7,000 genes examined, including gene products with previously unknown biochemical properties^64-66. Hopefully, similar insights will be made into the functions of plant genes.

Although Gateway cloning and the pHellsgate vectors can facilitate the rapid and automatable generation of RNAi constructs (Box 1), the stable delivery of these transgene libraries into plants is only practically feasible in Arabidopsis, for which an easy, non-tissue-culture, floral-dip method of transformation is available. Fortunately, Arabidopsis is now the foremost plant for genomic research, given the wealth of genomic information and resources that are available for it (see the link to TAIR). Before hpRNA transgenes can be used for genome-wide research in any other plant species, similarly easy and efficient transformation systems will need to be developed.

Agro-VIGS is free from the limitations imposed by plant transformation efficiency, and the pTRV–attP/R vectors enable the rapid generation of gene silencing libraries (Box 1). TRV–VIGS has been most widely used in N. benthamiana, for which transposons or chemically induced mutants are not widely available and about which little classical genetic or sequence information exists. However, TRV–Agro-VIGS seems to be equally effective in the tomato⁴⁴, for which there is a good genetic map, a wide range of spontaneous and induced mutants and large BAC (bacterial artificial chromosome) and EST (expressed sequence tag) libraries. For species that are difficult to transform, VIGS is the system of choice. Cereal species in particular fit this category; progress in this area depends on finding suitable virus vectors. However the demonstration of VIGS in barley using barley stripe mosaic virus (BSMV)³⁸ indicates that similar vectors should ultimately be found for all cereals.

It is likely that genome-scale RNAi studies using hpRNA constructs will initially be carried out using strong constitutive promoters to give a rapid survey of gene function. However, hpRNAs also offer the possibility of directing silencing in specific tissues depending on the promoter used to direct expression of the hpRNA. This type of approach would be particularly useful in defining the role of genes that give pleiotropic phenotypes when function is impaired at the whole plant level. Similarly, chemically induced promoter systems⁷⁰ could also be used to direct expression of hpRNAs. This type of system would be useful for defining the functions of genes that give embryonic lethal phenotypes when mutated and also for studying the roles of genes at specific stages in plant development. RNAi approaches also offer the prospect of silencing whole gene families or several, unrelated genes by either targeting conserved regions of nucleic acid sequence or including several target sequences in the same RNAi-inducing construct, respectively.

We hope that we have shown that RNA silencing has many features that make it a system of choice for plant functional genomics. The flexibility it offers is likely to be important in understanding many aspects of the function of the Arabidopsis genome. As plant research moves on to the challenges of understanding the biology of crop plants such as wheat — in which genome size, complexity and redundancy obviate the use of single-gene mutagenesis — RNA silencing, in some form, will probably have an even greater role.

Boxes

		Box 1 | Applications of RNA interference in plant genomics
		RNA interference (RNAi) is beginning to be used as a tool for plant genomic research, particularly in Arabidopsis, which is now the plant species with the most extensive array of genomic resources. Several projects are under way to produce the resources that are required for RNAi to be used as a tool for high-throughput plant genomics research. The CATMA group (Complete Arabidopsis Transcriptome MicroArray; see the link to CATMA) is generating a set of PCR products, called gene sequence tags (GSTs), that represent each Arabidopsis gene and have been designed to hybridize in a gene-specific manner on Arabidopsis cDNA microarrays. The primers that are used to generate these PCR products have 5' extensions that enable them to be reamplified and cloned into Gateway vectors by a BP clonase reaction (see figure). The AGRIKOLA consortium (Arabidopsis genomic RNAi knock-out line analysis; see link for more information on this project) is using this set of PCR products to generate, in pHellsgate vectors through an LR clonase reaction, a gene-specific RNAi construct for each Arabidopsis gene to be used in large-scale RNAi-silencing studies — a resource that will complement whole-genome microarrays. The gene fragments that have been cloned into these entry vectors could also be recombined into VIGS (viral-induced gene silencing) vectors, such as pTRV–attR (Ref. 44), by an LR clonase reaction (see figure). Identifying gene function by RNAi-induced gene silencing in plants is a key challenge for the future because many gene knockouts in plants do not produce a phenotype under normal growth conditions. Sets of growth conditions (the so-called 'Gauntlet'), in which, for example, temperature, nutrients, light quality and hormones are varied, are being developed for Arabidopsis (see link to the Arabidopsis Gauntlet Project). The aim of this type of screen is to reveal mutant phenotypes that are specific to particular growth conditions — the first step in assigning a function to a gene. The highly automated approach to identifying agriculturally useful genes that is being taken by CropDesign (see online link for more on this approach) also highlights the type of analysis that will be needed to identify such phenotypes on a large scale. attB1, attB2, attL1, attL2, Gateway recombination sites; CP, coat protein; GFP, green fluorescent protein; Term; transcription termination sequence.