Assessment of penetrance and expressivity of RNAi-mediated silencing of the Arabidopsis phytoene desaturase gene


  • This article is dedicated to the memory of Rakesh Pancholy, who suffered a fatal aneurism prior to its completion.


  • • RNA interference (RNAi) is of great value in plant functional genomics. However, the absence of RNAi phenotypes and the lack of uniform level of RNAi silencing has complicated gene identification. Here, the penetrance and expressivity of RNAi-mediated silencing of the phytoene desaturase (PDS) gene in Arabidopsis thaliana were examined quantitatively to provide a reference for the likely severity and distribution of silencing effects.
  • •  Arabidopsis plants were transformed with an RNAi construct targeting PDS. Transgenic plants were examined for frequency of RNAi-mediated silencing and various silencing phenotypes. mRNA depletion level and RNAi expressivity were assayed by relative reverse transcription polymerase chain reaction (RT-PCR).
  • • High penetrance and variable expressivity of RNAi were demonstrated. An inverse correlation between PDS mRNA level and RNAi phenotype was seen. No direct relationship between copy number for the RNAi-generating transgene and phenotype was evident. Decreased RNAi penetrance in T2 plants was observed.
  • • It is suggested that variability in RNAi expressivity and postmeiotic decrease in RNAi penetrance constitute barriers for high throughput plant gene characterization.


An exciting challenge of modern biology is how to decipher the vast amount of raw information from genome sequencing so that individual genes can be identified and their biological function can be revealed. Among the various gain- or loss-of-function approaches available for interpreting gene function, RNA interference (RNAi) is especially powerful and well-suited for functional analysis of Arabidopsis and rice, plants for which physical sequencing of the genome is essentially complete.

This double-stranded RNA (dsRNA)-induced gene-silencing phenomenon, conserved among many organisms, including animals and plants, has several advantages over other approaches. In contrast to virus-induced (Baulcombe, 1999) and agroinfiltration-mediated (Schob et al., 1997) plants, RNAi silencing is stable, allowing its effects to be studied in progeny (Carthew, 2001). Unlike other mutagenesis methods such as T-DNA insertion (Sallaud et al., 2003), transposon tagging (Brutnell, 2002) and tilling (McCallum et al., 2000), RNAi silencing can be made inducible and reversible (Guo et al., 2003; Gupta et al., 2004), attributes that are especially useful in studying genes crucial to early development. A single RNAi construct can silence duplicated genes or genes sharing coding regions of sequence identity, making it possible to characterize genes with redundant copies in the genome (Waterhouse & Helliwell, 2003). Given that some 60% of known genes in the small Arabidopsis genome are duplicated (The Arabidopsis Genome Initiative, 2000), this is an important consideration. In 2002, the AGRIKOLA project (Arabidopsis Genomic RNAi Knock-out Line Analysis) was initiated to study the function of 25 000 Arabidopsis genes using RNAi (Hilson et al., 2004).

Despite its many attributes, the value of RNAi for gene discovery and characterization is diminished where debilitation of gene function fails to produce a visible phenotype. Another caveat to its use is that the effects may vary for individual transformants. For example, Wesley et al. (2001) found that transformation of Arabidopsis and rice plants with constructs that generated hairpin-RNA (hpRNA) yielded a series of independent lines with various phenotypes and degrees of target mRNA reduction. Indeed, levels of the targeted mRNA have been reported to range from wild type to undetectable (Kerschen et al., 2004). We encountered a similar situation in the use of RNAi-induced silencing to determine the function of a series of SET domain-containing genes in Arabidopsis as an altered morphological phenotype was detected for only three of some 20 different constructs. In such cases, quantitative determination of transcript reduction in the silenced population is essential to confirm functionality of the RNAi construct. For the present studies, reverse transcription polymerase chain reaction (RT-PCR) was chosen for estimation of transcript abundance as it is more sensitive than are typical genomic RNA blots and requires much less tissue, permitting large-scale analysis of many plants. This technique has been successfully used in several studies to measure transcript abundance, usually expressed as a percentage of transcript depletion, defined in relation to controls for the specific investigation. The PDS gene that encodes phytoene desaturase (PDS) was chosen as the target because its silencing results in photobleached leaves (Goodwin, 1988), a readily visible phenotype. Although silencing of PDS has been used as a qualitative reporter of RNAi vector-based silencing in various plants, quantitative analysis of RNAi silencing in these studies is limited to the objective of establishment of efficiency of the proposed RNAi construct or reporter system under constitutive (Miki & Shimamoto, 2004; Tao & Zhou, 2004) or inducible systems (Guo et al., 2003). Because PDS silencing has been used by several investigators, our results can be compared with their work to provide a general guide to assessment of expectations for RNAi-mediated silencing.

However, none of these studies have focused on the issue that, using the same RNAi vector system, a uniform population of plants exhibiting an equal level of silencing is rarely obtained. This situation is not likely to be unique to PDS and therefore has implications for all RNAi-mediated silencing studies in plants. Thus, in this study, we have used the terms ‘penetration’ and ‘expressivity’ in an attempt to address the issue of variable silencing effects and their quantitative assessment in a more global connotation. These terms are conventionally used in population genetics (Zlotogora, 2003). Classically, penetrance is defined as the percentage of individuals with a given genotype that exhibit the phenotype associated with that genotype, whereas expressivity measures the extent to which a given genotype is expressed in an individual at the phenotypic level (Griffiths, 1996). Thus, for PDS silencing, the percentage of transgenic plants displaying an identifiable photobleached phenotype represents penetrance and the percentage depletion of endogenous PDS mRNA defines expressivity.

Materials and Methods

Plant material

Arabidopsis thaliana (ecotype Columbia) seeds were germinated in soil (Redi Earth, Scotts, Marysville, OH, USA) and, following vernalization at 4°C for 48 h in the dark, grown at 22°C under a 16-h/8-h light/dark cycle. Transformants were selected on Murashige and Skoog (MS) medium (Gibco™ Invitrogen Corporation, Carlsbad, CA, USA) containing 100 µg ml−1 Timentin (ticarcillin disodium and clavulanate potassium, SmithKline Beecham Pharmaceuticals, UK), 50 µg ml−1 kanamycin (Sigma, St Louis, MO, USA), 3% (w/v) sucrose and 0.2% (w/v) phytagel (Sigma). After two weeks, resistant plants were transferred to soil and grown in a growth chamber. Pictures of plants were taken 21 d after transfer to soil using an Olympus C-3040ZOOM digital camera (Olympus, Melville, NY, USA).

Plasmid construction and transformation

Two primer sets with different restriction enzyme recognition site overhangs were used to amplify a 179-bp coding region (spanning exons 8 and 9) of PDS. One set, with XhoI and KpnI overhangs, was inserted in a sense orientation into pHannibal (Wesley et al., 2001); the other, with BamHI and ClaI overhangs, was inserted in an antisense orientation. In the second step, a fragment containing 35S:PDS-s:intron:PDS-as:ocs was released by NotI and inserted into the binary vector pArt27 (Gleave, 1992) to form the RNAi construct K-1. Primer sequences were as follows.





Thermocycling conditions were 94°C for 5 min, followed by 35 cycles of 94°C for 1 min, 53°C for 1 min and 72°C for 2 min, with a final polymerization step at 72°C for 10 min. The K-1 construct was transformed into Agrobacterium (GV3101) using electroporation with a Gene Pulser (Bio-Rad, Hercules, CA, USA). Arabidopsis plants were transformed using vacuum infiltration (Bechtold & Pelletier, 1998).

Genomic DNA blot analysis

Genomic DNA was extracted from four to five leaves automatically with an AutoGenprep 850alpha (Autogen, Holliston, MA, USA). Genomic DNA (500 ng) was digested with 20 units of BamHI for 17 h. After electrophoretic separation in a 0.7% agarose gel, the DNA fragments were transferred to Hybond-N+ membrane (Amersham, Little Chalfont, UK). DNA probes were labeled using a DECAprime II kit (Ambion, Austin, TX, USA). Hybridizations were performed using ULTRAhyb solution (Ambion) according to the manufacturer's recommendations. Hybridization signals were detected by exposure to a PhosphorImager (Fuji, Stamford, CT, USA) and quantitated using the public domain NIH ImageJ program (developed at the U.S. National Institutes of Health and available at The densitometry ratios for the intact transgene to the endogenous PDS gene were calculated and the copy number of the transgene was expressed as an integer relative to the plant having the lowest ratio. The copy number of rearranged transgenes was estimated by counting the aberrant transgenic bands.


Leaves exhibiting a similar PDS silencing phenotype were used for RNA extraction. If more than one leaf on a plant exhibited the phenotype, the leaves were pooled. However, leaves from different plants were never pooled. RNA was extracted using TRIzol (Invitrogen, San Diego, CA, USA) according to the manufacturer's instructions. RNA (1 µg) was treated with DNaseI (1 unit per µl, Invitrogen) and RT-PCR was carried using the QIAGEN OneStep RT-PCR Kit (Qiagen, Valencia, CA, USA) to amplify a PDS coding region upstream of the RNAi target region. Reactions contained 250 ng RNA and were made to 0.6 µm for each PDS primer and 0.08 µm for each EF1″ primer in a final volume of 25 µl. Primer sequences for PDS were: 5′-GTATGAGACTGGTTTACATATTTTCT-3′ and 5′-CCGCAAAATAGCCCAAATACC-3′. Primer sequences for the internal control EF1″ were: 5′-TGCTGTCCTTATCATTGACTCCACCAC-3′ and 5′-TTGGAGTACTTGGGGGTAGTGGCATC-3′. Thermocycling conditions were reverse transcription at 50°C for 30 min, 95°C for 15 min, followed by 25 cycles of 94°C for 1 min, 55°C for 1 min and 72°C for 1 min, with a final polymerization step at 72°C for 10 min. The products of the RT-PCR amplification were subjected to electrophoresis through a 2.0% agarose gel, followed by staining with ethidium bromide (100 ng ml−1). The gel was then digitally imaged and was analyzed using ImageJ.


PDS is encoded by a single-copy gene in Arabidopsis

That the Arabidopsis genome contains a single copy of PDS was validated by a BLAST search (Altschul et al., 1997) using Accession NM117498.2, the original full length cDNA sequence (Scolnik & Bartley, 1993), against the Arabidopsis genome. Although two additional cDNAs were identified, they differed from the original PDS cDNA by only one or two nucleotides. Further examination revealed that PDS3 (Accession Dl3154c) is a C to G correction at position 42 of the 5′-UTR and the other (Accession NM202816.1) is a splice variant of the original cDNA that results in a difference of two amino acid residues (GV to AI, encoded by exons 7 and 8). Nevertheless, all three cDNAs (4344 bp) originated from the same 4837-bp gene locus (At4g14210) on chromosome 4. Thus, only one copy of PDS is present in the Arabidopsis genome.

High penetrance of RNAi-mediated silencing

To inactivate the Arabidopsis PDS gene, an RNAi construct (K-1, Fig. 1) containing sense and antisense orientations of a fragment spanning exons 8 and 9 of the PDS coding region flanking the pHannibal intron, was transformed into Arabidopsis thaliana (ecotype Columbia). A series of three replicate transformations generated 485 kanamycin-resistant (Kanr) T1 plants. Only 5% of the Kanr plants did not display any visible phenotype. Genomic DNA blot analysis revealed that 6 of 7 randomly selected P0 plants contained at least one copy of transgene. To simplify the calculation, all P0 Kanr were counted as bona fide transgenic plants. Thus the penetrance of the K-1 RNAi construct was 95%. This demonstrates the sensitivity of the PDS system for evaluating the efficacy of silencing induced by the RNAi construct.

Figure 1.

Organization of the T-DNA region of RNAi vector K-1 used to target PDS. RB and LB, T-DNA right and left border; 35S, cauliflower mosaic virus 35S promoter; PDS-s and PDS-as: sense and antisense orientation, respectively, of the targeted PDS fragment. NPTII, neomycin phosphotransferase II. The shaded boxes marked OCS and NOS denote terminators of Agrobacterium octopine synthase and nopaline synthase genes, respectively; the arrow labeled NOS denotes a nopaline synthase promoter. The KpnI–XhoI region (179 bp, thick bar) corresponds to the PDS coding sequence used to generate a probe for genomic DNA blot analysis. The presence of a 3103-bp BamHI-BamHI fragment in genomic blots was used to confirm the presence of the intact RNAi construct. Diagram not drawn to scale.

Relationship between expressivity and phenotype

The wide range of photobleached phenotypes present in T1 progeny as a result of PDS silencing (Fig. 2) indicated that the expressivity of the K-1 construct can be dramatically different in individual plants and often for different parts of a plant. To characterize the efficacy of RNAi-mediated PDS silencing, the plants were grouped into six classes (P0–P5), based on their phenotypes (Fig. 2 and Table 1; P5 plants were not shown in Fig. 2 because they were already dead before the photos were taken).

Figure 2.

Wide range of RNAi-induced PDS phenotypes in T1 plants. (a) A tray containing a representative variety of silencing phenotypes (21 d post transfer). Representative plants exhibiting a mild (P1) silencing phenotype and a severe (P4) phenotype are shown in (b) and (c), respectively. (d) A wild-type leaf. (e–i) Representative leaves showing P0–P4 phenotypes, respectively. A classification of phenotypes is given in Table 1.

Table 1. High penetrance and various expressivities of RNAi-mediated PDS silencing in T1 plants
Class T 1 plantsPhenotype descriptionExpressivity (%)
  1. The penetrance of RNAi-mediated silencing was calculated as the percentage of kanamycin-resistant T1 plants displaying an identifiable PDS silencing phenotype P1–P5. The PDS transcript depletion level was used as a measure of expressivity for the K-1 construct. dpt, days post transfer; nd, not determined.

P0 25 5no visible phenotype 4
P1 7415mild symptoms, ∼1 mm white patches on the leaves 21
P2158333–5 mm white patches on the leaves 48
P3 27 6leaves predominately white, with green patches64
P4 22 5white leaves85
P517937white leaves, very stunted, died within 14 dpt.nd

Approximately 20% of the transformants displayed readily discernable mixed phenotypes; these were classified according to their severest phenotypes. An example is shown in Fig. 3(a), in which six relatively old rosette leaves of a P2 plant had only very small (< 1 mm) white patches, a typical P1 phenotype, whereas the cauline leaf and three relatively young rosette leaves showed a P2 phenotype, with bigger white patches (3–5 mm). Similarly, the T1 plant in Fig. 3(b) displayed a P3 phenotype in old rosette leaves and a P4 phenotype in young rosette and cauline leaves. Two T1 plants (out of 485 plants) displayed rare phenotypes. In one plant (Fig. 3d), about 70% of rosette and cauline leaves variously displayed P0, P3 and P4 phenotypes and the other 30% were variegated, having one side (25–50%) of the leaf area completely white and the other completely green.

Figure 3.

Expressivity of RNAi-mediated silencing in different plant parts. Representative plants exhibiting different PDS silencing phenotypes in various tissues are shown. (a) Plant displaying a mixture of P1 and P2 phenotypes. Arrow 1 indicates a cauline leaf showing a P2 phenotype; arrows 2 and 3 indicate rosette leaves exhibiting P1 and P2 phenotypes, respectively. (b) Plant showing P3 (in older rosette leaves) and P4 (in younger rosette leaves) phenotypes. (c) A wild-type plant. Rare phenotypes (d) showing leaves variegated for P0 and P4 and (e) a plant with P0 and P2 rosette leaves and P4 shoots.

Although it is tempting to think that these effects reflect the spread of silencing induced by RNAi, current evidence does not support this possibility. For example, because Vaistij et al. (2002) have shown that Arabidopsis PDS mRNA cannot be used as a template by the RNA-dependent RNA polymerase SDE1, the production of dsRNA, which could trigger the spread of PDS silencing (Himber et al., 2003), is lacking. The plant shown in Fig. 3(d) would have been valuable for studying the possibility that systemic gene silencing of PDS occurred. Unfortunately, it was sterile and two attempts to regenerate it through young silique culture were not successful. Similarly, the plant shown in Fig. 3(e), with rosette leaves that displayed no or mild silencing while the rest of the shoot system was completely white, was also sterile.

Penetrance and expressivity of RNAi-mediated silencing in T2 progeny

Compared with agroinfiltration (Schob et al., 1997) and other transient methods for expressing transgenes, Agrobacterium-mediated stable transformation has the advantage that progeny transgenic plants can be obtained. Opportunity was taken of this advantage to inspect whether RNAi-induced silencing was heritable and, if so, to determine the penetrance of K-1 in T2 progeny. Flowers and small siliques developed and T2 seeds were obtained from six P0, 10 P1, 10 P2 and five P3 plants (P4 plants were sterile and P5 plants were dead before reaching a reproductive stage). A portion of seed (20–30 from each plant) was germinated under kanamycin selection. T2 progeny derived from a given class of T1 parental plants were grouped together and screened for PDS silencing (Table 2). The penetrance of K-1 in T2 progeny of P0 plants was 0 because none displayed detectable PDS silencing. T2 progeny of P1 plants showed both P0 and P1 phenotypes, with a penetrance of ∼46%. For P2 and P3 plants, PDS silencing was decreased in most T2 progeny but enhanced in a few (< 8%). Overall, penetrance of the K-1 construct in T2 progeny dropped sharply to 46% from 95% in T1 plants (Table 2).

Table 2. Penetrance of RNAi-mediated silencing in T2 progeny
PhenotypeKanamycin- resistant T2 progeny No.P0P1P2P3P4RNAi penetrance in T1 plant T2 progeny (%)
P0 62 62 0 000 0
P1 72 3933 00046
P2 70 3027 93157
P3 53  914206483

Relationship between transgene copy number and severity of RNAi-mediated silencing

The number of transgene copies present can be either positively or negatively associated with the level of transgene expression. Usually, increased expression is observed in correspondence with higher transgene copy number if the copies of the transgene are all intact and the copy number is below a threshold whose value is dependent on transgene itself (Hobbs et al., 1993; Lechtenberg et al., 2003). However, transcript levels that are above this threshold or aberrant transcripts can trigger post-transcriptional gene silencing (PTGS) (Garrick et al., 1998; Lechtenberg et al., 2003). Therefore we considered the possibility that the level of hpRNA (Wesley et al., 2001), the RNAi silencing trigger generated from the K-1 construct, was related to transgene copy number.

To determine the K-1 transgene copy number, genomic DNA was isolated from more than 70 randomly selected kanamycin-resistant plants, digested with BamHI and subjected to Southern (Southern, 1975) analysis. A single 7083-bp fragment corresponding to the endogenous PDS gene was detected in all of the transformants tested. Figure 4 is a representative blot for 12 independently transformed plants, hybridized with a PDS probe (Fig. 1). One to three copies of a 3103-bp fragment, representing the intact transgene, was detected in all Kanr plants except for one P0 plant (Fig. 4, lane 4), in which only a rearranged copy of the transgene (∼5.5 kb) was found. One to four partial or rearranged transgene fragments were present in many transformants. Phenotypes P0–P3 were represented in this sample, but P4 and P5 plants were excluded because they provided insufficient plant material.

Figure 4.

Transgene copy number shows little correlation with phenotypic severity. DNA extracted from 12 T1 plants (lanes 1–12) randomly selected from a total of 485 plants transgenic for the K-1 construct, and a wild-type plant (wt, lane 13), was digested with BamHI and subjected to genomic DNA blot analysis. The PDS phenotypes and number of intact and rearranged PDS transgene fragments are indicated at the bottom of each lane. The probe used for hybridization corresponded to the PCR amplicon (XhoI-KpnI fragment, Fig. 1) employed in the construction of the K-1 vector. Intact transgene copy numbers were calculated as described in the Materials and Methods, with the signal ratio of transgene to endogenous PDS for lane 10 set at 1. The positions of the endogenous gene (7083 bp) and the transgene (3103 bp) are indicated by arrows. Lane 14 contained a 1-kb DNA ladder (New England Biolabs, Beverly, MA, USA).

Figure 4 shows the phenotype and copy number (see the Materials and Methods section) for each of the plants for which Southern analysis was conducted. From these data, it appears that a single-copy intact transgene is not necessarily associated with a severe silencing phenotype. For example, it can be seen in Fig. 4 that single-copy intact transgenic plants displayed P0 (lane 2), P1 (lanes 5, 11), P2 (lanes 6, 7 and 9) and P3 (lane 10) phenotypes. A range of phenotypes was also evident from the data shown in Fig. 4 for plants containing two or more copies of an intact transgene: P0, lane 1; P1, lanes 8 and 12; P2, lane 3. Indeed, the copy number of rearranged transgenes also showed little correlation with the severity of the phenotype as plants with 3 or 4 copies of rearranged transgene displayed P0 (Fig. 4, lanes 1 and 2), P1 (Fig. 4, lane 12), P2 (Fig. 4, lanes 3, 6, and 9) and P3 (Fig. 4, lane 10) phenotypes. Another blot of XhoI-digested DNA from a different set of 12 independent transformants showed similar results (data not shown). Taken together, the results from these plants do not provide any evidence that a correlation exists between transgene copy number and severity of RNAi-mediated silencing.

Correlation between PDS transcript level and PDS silencing phenotype

Because the photobleaching phenotype in PDS plants is caused by diminution of endogenous PDS mRNA level by the K-1 RNAi construct, we conjectured that the depletion level of PDS mRNA is in positive correlation with PDS silencing phenotypes. The level of PDS transcripts in Arabidopsis has been reported to be below the detectable limit of Northern blot analysis (von Lintig et al., 1997; Wetzel & Rodermel, 1998). Therefore, a semiquantitative relative RT-PCR technique (Dean et al., 2002) was used. In this technique, the gene of interest is coamplified with an internal control gene to determine the relative abundance of endogenous PDS transcripts in each class of PDS plants. EF1″ was chosen as the internal control because it produces stable transcripts and its amplification remains in a log-linear stage at the same optimal conditions as those for PDS. The amplified products were subjected to DNA gel analysis followed by densitometry quantitation and the relative PDS : EF1″ expression ratio was calculated in wild-type and PDS plants. As expected, a close relationship between severity of phenotype and depression of PDS transcript level was displayed (Fig. 5). In plants displaying no- (P0) or mild-silencing (P1) PDS plants, PDS transcripts levels averaged 96% and 79% of wild-type plants, respectively. This number had dropped to ∼ 40% in medium-silencing PDS plants (P2 and P3) and only 15% in severe-silencing plants (P4). Thus, the depletion level of PDS transcript, a direct result form K-1 expression, was consistent with the silencing phenotype and was used as a measure of expressivity of K-1 construct (Table 1).

Figure 5.

PDS transcript levels diminish in correspondence with severity of photobleaching. (a) Relative RT-PCR analysis in wild-type and transgenic plants displaying various degrees of bleaching (P0–P4: see text and Table 1). Arrows denote the predicted position of PDS amplicon and the control (EF1″) amplicon. (b) Normalized PDS transcript levels for the various phenotypes. RDI: relative densitometric intensities (pixels mm−2), normalized relative to EF1″, was obtained using MacBAS v2.5 software (Fuji, Tokyo, Japan). The RDI for wild-type plants was set as 1.0. Error bars denote standard error of the mean.


The apparent penetrance of RNAi inactivation is influenced by phenotype

Using the intron-containing vector pHannibal (Smith et al., 2000), high RNAi penetrance (95%) was observed (Table 1), which may be attributed to the ease of identification of the silencing phenotype, permitting the detection of even a mild degree of PDS silencing. For example, if P1 and P2 plants could not be visually identified, RNAi penetrance would drop to 47%, less than half of the original level. High RNAi penetrance is unlikely to be limited to PDS and may be routinely achieved if optimal target regions are used in the construction of RNAi vectors. Conversely, RNAi penetrance could be greatly underestimated for genes whose loss-of-function mutants result in little or no visible phenotype. In regard to this, an important consideration is that expression of the targeted gene may be limited to specific tissues or certain developmental stages and/or certain environmental conditions. In such cases, phenotypic changes resulting from down-regulation may only be detectable in the relevant tissues and conditions. These caveats complicate gene discovery or characterization by RNAi-mediated gene silencing because biochemical, rather than phenotypic, analysis may be required. However, for genes whose function is predicted but not proven, RNAi remains a valuable discovery tool as it permits a guided analysis of the predicted function.

Endogenous gene expression level and RNAi efficacy

It has been suggested that RNAi phenotype or RNAi efficiency is related to the nature of the target gene. For example, in Caenorhabditis elegans, RNAi phenotypes were shown to be more scoreable for highly expressed genes than for genes expressed at low levels (Cutter et al., 2003). However, although Kerschen et al. (2004) found that transcript levels were effectively reduced by RNAi in Arabidopsis for several moderately and highly expressed genes, they found that RNAi was effective for HDA9, HDT4 and SGA1, genes normally expressed at low levels. Similarly, despite the low endogenous level of PDS mRNA expression in Arabidopsis (Wetzel & Rodermel, 1998), very high RNAi penetrance (95%; Table 1) and expressivity (Fig. 2i) were observed, suggesting that RNAi efficiency and the endogenous transcription level of the targeted gene are not necessarily related.

Transgene copy number and variability of silencing

Several studies on RNAi-mediated gene silencing have shown a wide variability for individual plants (Chuang & Meyerowitz, 2000; Levin et al., 2000; Stoutjesdijk et al., 2002). Kerschen et al. (2004) reported little variability in target transcript reduction for single-copy RNAi lines and implicated the presence of multiple copies of the RNAi construct as being a major cause of variability. Evidence supporting this view included the observation that no multiple-copy line depleted the target transcript more than did single-copy lines. However, their data show variability in transcript depletion for both single-copy and multicopy lines targeting HDA2, HAG5 and CHR2, with the greatest variability for HAG5 being among single-copy lines. For PDS, silencing was considerably more effective for some multiple-copy transgenic plants than for single-copy transgenic plants and no correlation was found between copy number and silencing severity (Fig. 4). Another consideration is that Kerschen et al. (2004) used pooled RNA from several seedlings, thus obtaining an average transcript level that would mask any plant-to-plant variation.

The establishment of single-copy transgenic lines is usually preferred over multicopy lines because they are more readily taken to homozygosity. In general, transgene expression levels from single-copy lines are more stable than from multiple-copy lines. However, this probably reflects the organization, rather than the copy number, of the transgene (Lechtenberg et al., 2003). This follows from the finding that gene silencing typically arises as a post-transcriptional event, incited by aberrant RNA transcribed from the rearranged insert, than as a homology-dependent event (Mette et al., 2000; Matzke et al., 2002). Consequently, the presence of multiple intact copies of the RNAi-generating transgene may be beneficial, because they have the potential to provide higher RNAi transcript levels than can single-copy inserts.

High-throughput characterization of plant genes by RNAi

Whereas fabrication of RNAi constructs is rarely a limiting step in high-throughput identification of gene function using RNAi, delivery and analysis can be constraining. In C. elegans, highly efficient delivery of small interfering RNA (siRNA) occurs by ingestion, and phenotypic analysis of function is straightforward (Kamath et al., 2003). In Drosophila and mammals, the establishment of an in vitro system and the availability of numerous cell lines have simplified the delivery of siRNA or long dsRNA, and greatly facilitated screening of genes involved in particular pathways at the cellular level (Boutros et al., 2004; Foley & O’Farrell, 2004; Paddison et al., 2004).

In plants, agroinfiltration- and virus-induced gene silencing (Waterhouse & Helliwell, 2003) can provide approaches for large scale temporary and noninheritable gene silencing. However, the more important understanding of gene function at the organismal level requires Agrobacterium-mediated stable transformation. For this, the development of intron-containing hairpin RNA constructs (Smith et al., 2000) and Gateway recombination-based cloning technology (Wesley et al., 2001) have facilitated high throughput construction of RNAi vectors.

For gene discovery using RNAi, many primary transformants need to be screened. Transcriptional inactivation and an associated phenotype can be expected to range from little to complete and from no phenotype to extreme phenotype or lethality. Lines having a visible phenotype are retained for further study. When no phenotype is apparent, the functionality of the RNAi construct needs to be demonstrated by analysis of transcript depletion. The complete functional inactivation of some genes can be predicted to be lethal. Becaue the cause of lethality, rather than lethality per se, is of interest, advantage can be taken of lines that express RNAi weakly. Even more valuable is the ability to use RNAi expressed from an inducible promoter (Guo et al., 2003) because this provides flexibility for the timing and degree of gene inactivation and has the potential for reversal of silencing by withdrawal of the inducer (Gupta et al., 2004). If effective depletion is not substantiated, the use of alternative target or promoter sequences for the RNAi vector is indicated. Estimation of transgene copy number will identify single-copy transformants. If it is assumed that single-copy lines are always more effective in transcript depletion, then multicopy lines will be discarded. However, in contrast to the studies of Kerschen et al. (2004), our data show that multicopy lines can have higher expressivity than do some single-copy lines. In such situations, transcript depletion is a more meaningful selection criterion than is copy number. Even for single-copy lines, variation in expressivity can be expected and screening of these lines for those showing greatest transcript depletion is still desirable. For detailed functional investigation of the newly identified gene, well-defined stably expressing RNAi lines need to be established. Clearly, whatever protocol is followed, functional analysis will be time-consuming.


The pHannibal vector was kindly provided by Peter Waterhouse. We thank Mahesh Chandrasekharan, Prapapan Teerawanichpan, Sunee Kertbundit and other lab members for valuable discussions and suggestions. Supported by National Science Foundation grant MCB0110477 and by an REU supplement for Elena Delgado and Joshua Wollam, who provided excellent technical assistance.