A two unit antisense RNA cassette test system for silencing of target genes
A two unit antisense RNA cassette test system for silencing of target genesHilde M. Engdahl*, Tord Å. H. Hjalt1 and E. Gerhart H. Wagner
Department of Microbiology, Swedish University of Agricultural Sciences, Genetic Center, Box 7025, Genetikvägen 1, S-75007 Uppsala, Sweden and 1Department of Microbiology, Uppsala University, Biomedical Center, Box 581, S-75123 Uppsala, SwedenReceived May 19, 1997;Revised and Accepted June 26, 1997
ABSTRACT
This communication describes a two unit antisense RNA cassette system for use in gene silencing. Cassettes consist of a recognition unit and an inhibitory unit which are transcribed into a single RNA that carries sequences of non-contiguous complementarity to the chosen target RNA. The recognition unit is designed as a stem-loop for rapid formation of long- lived binding intermediates with target sequences and resembles the major stem-loop of a naturally occurring antisense RNA, CopA. The inhibitory unit consists of either a sequence complementary to a ribosome binding site or of a hairpin ribozyme targeted at a site within the chosen mRNA. The contributions of the individual units to inhibition was assessed using the lacI gene as a target. All possible combinations of recognition and inhibitory units were tested in either orientation. In general, inhibition of lacI expression was relatively low. Fifty per cent inhibition was obtained with the most effective of the constructs, carrying the recognition stem-loop in the antisense orientation and the inhibitory unit with an anti-RBS sequence. Several experiments were performed to assess activities of the RNAs in vitro and in vivo: antisense RNA binding assays, cleavage assays, secondary structure analysis as well as Northern blotting and primer extension analysis of antisense and target RNAs.The problems associated with this antisense RNA approach as well as its potential are discussed with respect to possible optimization strategies.
INTRODUCTION
Antisense strategies for gene silencing have attracted much attention in recent years (1 ,2 ). The underlying concept is simple yet (in principle) effective: antisense nucleic acids (NA) base pair with a target RNA resulting in inactivation. Target RNA recognition by antisense RNA or DNA can be considered a hybridization reaction, although this view is sometimes clearly misleading (see below). Since the target is bound through sequence complementarity, this implies that an appropriate choice of antisense NA should ensure high specificity. Inactivation of the targeted RNA can occur via different pathways, dependent on the nature of the antisense NA (either modified or unmodified DNA or RNA) and on the properties of the biological system in which inhibition is to occur, i.e. the sum of the metabolic activities that mediate inhibition.
Major differences between exogenous [antisense oligo(deoxy)- ribonucleotides; AONs] and endogenous antisense RNA strategies lie in: (i) delivery of the inhibitor; (ii) presumed mode of binding to the target RNA; (iii) details of the inhibitory reaction. AONs are generally applied extracellularly and taken up, whereas antisense RNAs are most often transcribed intracellularly, after transient or stable introduction of an appropriate antisense gene (1 ,3 ). AONs are designed to contain a low degree of secondary structure in order to permit efficient hybridization, whereas antisense RNAs are often larger and structurally more complex. The latter implies that binding reactions, unlike the AON case, occur between folded RNAs and therefore may follow different rules. Antisense DNA approaches often rely on endogenous RNase H activity to inactivate the target RNA. Numerous modifications of AONs have been introduced, resulting in metabolically more stable compounds and those with altered stereochemistry (see for example 4 ). In contrast, antisense RNA-mediated silencing can occur by duplex-dependent blockage of a ribosome binding site (RBS) within an mRNA, duplex-dependent facilitated mRNA decay, antisense RNA-induced premature termination of transcription and cleavage of the mRNA by an antisense RNA with ribozyme activity. Inhibition of gene expression by introduced antisense RNA genes has been achieved in, for example, bacteria (5 ), plants (6 ) and mammalian cells (7 ).
Numerous naturally occurring antisense RNA systems have been found in bacterial accessory DNA elements, such as phages, plasmids and transposons (for a review see 8 ). These systems are useful sources of information regarding binding rate, specificity and mechanisms of target RNA inactivation. The copy number control circuit of bacterial plasmid R1 with its key elements CopA (antisense RNA) and CopT (target) has been studied extensively. Its efficiency, both in terms of binding rate and specificity (9 -12 ), prompted us to design antisense RNA cassettes that incorporate favourable characteristics of this system. In particular, it is known that stem-loop II (SL II) of CopA suffices for `kissing complex' formation with the target RNA and that formation of this transient, but long-lived, intermediate is rate limiting for pairing (11 ).
The basic features of the cassette are as follows: a promoter is located upstream of a sequence which is either complementary to a chosen RBS or encodes a ribozyme targeted against a sequence within the mRNA; downstream of this region a stem-loop is encoded, designed to serve as a CopA-like recognition structure and a transcription terminator. The (extended) loop sequence is complementary to a short segment within the target RNA. Thus, the RNA contains two non-contiguous regions of complementarity; one (the stem-loop) represents the recognition unit, which promotes rapid and transiently stable binding (but not complete duplex formation), whereas the second is the inhibitory unit (anti-RBS or hairpin ribozyme). The overall rate of interaction with and inactivation of the target RNA is expected to be determined by the stem-loop (rapid formation of a `kissing complex'; see above). After this concentration-dependent initial step, subsequent base pairing of the inhibitor unit to its target segment will occur in principle intramolecularly, i.e. independent of concentration and, hence, should not be rate limiting.
We describe here a set of antisense RNA and (as controls) sense RNA constructs aimed at a model target in Escherichia coli, the lacI mRNA (13 ). Inhibition was measured as inhibition of LacI-LacZ fusion protein synthesis in vivo and the structural and inhibitory properties of the various RNAs were assessed in vitro.
MATERIALS AND METHODS
Bacterial strains and plasmids
The Escherichia coli strain DH5[alpha] (F-endA1, hsdR17, supE44, thi-1, recA1, gyrA96, relA1, [Delta][argF-lacZYA]U169, [Phi]80[Delta]lacZ[Delta]M15; 14 ) was used for plasmid construction. Strain XAc (F' [Delta]14[pro lac] ara, gyrA, rpoB, argE (UAG); 15 ) was used for measurements of LacI-LacZ fusion protein synthesis.
Plasmids used are listed in Table 1 . Plasmid pGW47 is a ColE1 vector that contains the complete copA gene of plasmid R1, inserted into the BamHI site of pMc5-8 (16 ; BamHI site destroyed). To create plasmid pGW47-1 two polymerase chain reactions (PCRs) were performed using primers HE3/HE5 and HE4/HE6 respectively on pGW47 DNA. The two PCR fragments were cleaved with EcoRI and BamHI (pair 3/5) and BamHI and XbaI (pair 4/6) respectively and ligated to plasmid DNA of pGW47, cleaved with EcoRI and XbaI. The resulting plasmid, pGW47-1, lacks the SmaI site of pGW47, has acquired a BamHI-HindIII sequence between the copA promoter and the 5'-end of SL II and carries an XbaI site just downstream of the SL II terminator. Subsequently we replaced the wild-type sequence of SL II by a double-stranded DNA fragment encoding a copA SL II sequence containing two SmaI/XmaI sequences (Fig. 2 ), resulting in pGW48. The DNA segment between the two SmaI sites was replaced by a double-stranded DNA fragment (HE12/HE13) containing a lacI sequence, to yield pGW48-a and pGW48-s respectively (Figs 1 and 2 ). Plasmids pGW48-a and pGW48-s were the parents of all subsequent constructs: double-stranded DNA fragments (see below) were inserted into the unique HindIII site to place anti-RBS or ribozyme sequences, in either orientation, between the promoter and the stem-loop.
Cell growth and media
Cells were grown in LB medium (17 ) supplemented with 0.2% glucose. LA solid medium was LB medium with 1.5% (w/v) agar. When appropriate, chloramphenicol (30 [mu]g/ml) was included. M9 medium (18 ) was used as minimal medium. Solid medium contained 10 g Sicomol agar/l and was supplemented with 0.2% glucose, chloramphenicol (30 [mu]g/ml), arginine (70 [mu]g/ml), proline (46 [mu]g/ml) and thiamine (1 [mu]g/ml). X-Gal (5-bromo-4-chloro-3- indolyl-[beta]-galactopyranoside) was used at a final concentration of 40 [mu]g/ml.
Enzymes and chemicals
Restriction enzymes for cloning procedures were bought from Amersham, New England Biolabs or Pharmacia. Radiochemicals were purchased from Amersham or DuPont NEN.
copA gene of plasmid R1 (Sau3A fragment) cloned in ColE1-vector
pMc5-8 + pKN505
This work
pGW47-1
Removed unique SmaI site in vector, introduced unique BamHI and HindIII sites
pGW47
This work
pGW48
SL II sequence with flanking SmaI sites
pGW47-1
This work
pGW48-a
SL II with antisense lacI sequence inserted
pGW48
This work
pGW48-s
SL II with sense lacI sequence inserted
pGW48
This work
pGW48-a-rbs
Contains antisense lacI RBS sequence
pGW48-a
This work
pGW48-a-sbr
Contains sense lacI RBS sequence
pGW48-a
This work
pGW48-a-rib
Contains antisense lacI ribozyme sequence
pGW48-a
This work
pGW48-a-bir
Contains inverted ribozyme sequence
pGW48-a
This work
pGW48-s-rbs
Contains antisense lacI RBS sequence
pGW48-s
This work
pGW48-s-sbr
Contains sense lacI RBS sequence
pGW48-s
This work
pGW48-s-rib
Contains antisense lacI ribozyme sequence
pGW48-s
This work
pGW48-s-bir
Contains inverted ribozyme sequence
pGW48-s
This work
aConstructions are described in Materials and Methods.
DNA manipulations
Purification of plasmid DNA, restriction enzyme cleavages and other DNA techniques were essentially according to Sambrook et al. (19 ).
Oligodeoxyribonucleotides
Oligodeoxyribonucleotideswere synthesized on an Applied Biosystems 394 DNA/RNA Synthesizer or bought from Pharmacia (Sweden). Sequencing primers for all plasmid inserts were HE7 (5'-GTG ATC TTC CGT CAC A) or HE8 (5'-AGG AGC CTT TAA TTG TAT). For construction of pGW47, two pairs of PCR primers were used, HE3 (5'-ATA TGA ATT CCT CGG GAT CAG TCA C) and HE5 (5'-ATA TGG ATC CTG CGG GGA GTA TAG TTA TAT) and HE6 (5'-ATA TGG ATC CAA GCT TGG GCC CCG GTA ATC TTT T) and HE4 (5'-TAT ATC TAG AGG CAA GGA ACT GGT TCT GAT). For insertion of the copA SL II sequence containing two SmaI sites we used HE12 (5'-AGC TTG GGC CCG GGT AAT CTT TTC GTA CTC GCC AAA GTT GAA GAA GAT TAC CCG GGT TTT TGC TTT T) and HE 13 (5'-CTA GAA AAG CAA AAA CCC GGG TAA TCT TCT TCA ACT TTG GCG AGT ACG AAA AGA TTA CCC GGG CCC A). To introduce lacI sequences into the SL II loop of pGW48 we used the pair HE18 (5'-CCG GGT AAT CTT TTC TGA CTC TCT TCA GTT GAA GAA GAT TAC) and HE19 (5'-CCG GGT AAT CTT CTT CAA CTG AAG AGA GTC AGA AAA GAT TAC). For insertion of lacI rbs/sbr sequences into pGW48-a/pGW48-s the pair HE16 (5'-AGC TTA GGG TGG TGA ATG TGA AAC CAG TAA CGT TAT ACG ATG TA) and HE17 (5'-AGC TTA CAT CGT ATA ACG TTA CTG GTT TCA CAT TCA CCA CCC TA) were used. HE14 (5'-AGC TTG AAA CGA GAA GAT AAC CAG AGA AAC ACA CGT TGT GGT ATA TTA CCT GGT A) and HE15 (5'-AGC TTA CCA GGT AAT ATA CCA CAA CGT GTG TTT CTC TGG TTA TCT TCT CGT TTC A) served to introduce the ribozyme/inverted ribozyme sequence. As probes for Northern analyses, we used HE20 (5'-AAG ATT ACC CGG GCC CAA GCT T) and the 5S rRNA-specific oligo GW-5S (5'-TAC GGC GTT TCA CTT CTG AGT TTG GG). DNA templates for in vitro synthesis of antisense RNAs containing rbs/sbr and a- and s-SL II sequences by E.coli RNA polymerase were generated by PCR from appropriate plasmid DNA templates with primers HE27 (5'-GTG ATC TTC CGT CAC AGG TAT) and HE25 (5'-GGT GAA TTT CGA CCT CTA GA). Primers for synthesis of lacI mRNA transcription templates were HE21 (5'-GAA ATT AAT ACG ACT CAC TAT AGG AAG AGA GTC AAT TCA G) and HE22 (5'-CCG CTT CCA CTT TTT C). PCR primers for generating transcription templates containing rib and a-/s-SL II were HE23 (5'-GAA ATT AAT ACG ACT CAC TAT AGG ATC CAA GCT TGA AAC) and HE25. For the bir templates, containing the inverted sequences, primers HE24 (5'- GAA ATT AAT ACG ACT CAC TAT AGG ATC CAA GCT TAC CAG) and HE25 were used. HE21, HE23 and HE24 contain T7 RNA polymerase promoter sequences. Primer extension on ribozyme-cleaved lacI mRNA was done with HE22, as was in vivo 5'-end mapping of lacI mRNA.
Synthesis of PCR fragments for use as in vitro transcription templates
We used suitable pairs of primers (see above) to synthesize all in vitro transcription templates. PCR to obtain templates for T7 RNA polymerase transcription of lacI RNA (149 nt) and ribozyme-containing RNAs (148 nt) was performed as described (20 ). The same protocol was used to obtain DNA templates of rbs-/sbr- containing RNAs (296 nt). These fragments contain the copA promoter. PCR products were eluted from gels with DEAE-cellulose membranes (KEBO) according to Dretzen et al. (21 ). DNA fragments were dissolved in 30 [mu]l TE buffer (10 mM Tris-HCl, pH 7.9, 1 mM Na2EDTA).
In vitro transcription of RNAs
For secondary structure analysis of 5'-end-labelled SL II RNAs (Fig. 3 ), supercoiled plasmid DNA of the pGW47/48 series was transcribed by E.coli RNA polymerase (Pharmacia) with [[gamma]-32P]ATP and unlabelled NTPs as in Persson et al. (9 ). The RNAs used for binding studies (Fig. 5 ) were transcribed from linear, PCR-generated templates according to the same protocol, but including [[alpha]-32P]UTP and unlabelled NTPs. Ribozyme sequence-containing RNAs were transcribed by T7 RNA polymerase from PCR-generated DNA templates as described (9 ,20 ). [3H]UTP was included to permit quantification of the amounts purified. Transcription of lacI RNA was with [[alpha]-32P]UTP (Fig. 6 A) or [3H]UTP (Figs 5 and 6 B).
Secondary structure probing of SL II RNAs with RNases A, T1 and T2
Structural mapping of 5'-end-labelled RNAs was performed as described previously (20 ,22 ,23 ).
Northern blot analysis
Isolation of total cellular RNA and Northern analysis were as described (23 ,24 ). As a probe, an end-labelled oligodeoxyribonucleotide complementary to a sequence present in all RNAs investigated (HE20) was used. Reprobing for 5S rRNA was done as described (23 ). Band intensities were quantified using a Molecular Dynamics PhosphorImager.
In vitro binding assays
Binding of uniformly 32P-labelled antisense RNAs to unlabelled lacI target RNAs was assayed by gel shift analysis and calculations of the apparent second order binding rate constants (kapp) performed as described previously (9 ,11 ). In all cases, a high (>10-fold molar) excess of target RNA (concentration 1.3 * 10-8 M) over labelled antisense RNA was used to ensure pseudo-first order kinetics.
In vitro ribozyme assay
In vitro cleavage experiments with hairpin RNA constructs were performed according to Chowrira and Burke (25 ). Reactions contained 20 pmol 3H-labelled ribozyme and 0.5 pmol 32P-labelled lacI RNA in standard reaction buffer (40 mM Tris-HCl, pH 7.5, 12 mM MgCl2, 2 mM spermidine) at 37oC. Aliquots were withdrawn at different time points and quenched by adding 4 [mu]l ice-cold formamide dye. Samples were boiled for 1 min and electrophoresed on 8% sequencing gels, followed by autoradiography. For determination of the cleavage sites (Fig. 6 B), reactions were done as above except that lacI RNA was unlabelled.
Primer extension analysis
Primer extensions (Fig. 7 ) were performed on ~10 [mu]g total cellular RNA. Samples were boiled for 1 min in the presence of 32P-end-labelled primer HE22 (~100 000 c.p.m.) in hybridization buffer (50 mM Tris-HCl, pH 7.9, 50 mM KCl). Mixtures were snap frozen in liquid nitrogen and subsequently thawed on ice. Incubation was at 45oC for 60 min in extension buffer (50 mM Tris-HCl, pH 7.9, 10 mM MgCl2, 50 mM KCl, 10 mM DTT) containing 0.2 mM dNTPs, 13 U human placental ribonuclease inhibitor and 3.5 U AMV reverse transcriptase (both Amersham). After ethanol precipitation, samples were dissolved in water. An equal volume of formamide dye was added and samples boiled prior to loading (8% sequencing gel). Sequencing ladders were obtained on lacI DNA using a Thermal Cycle Sequencing kit (New England Biolabs). Primer extension on ribozyme-cleaved lacI RNA (Fig. 6 B) was also performed with primer HE22.
In vivolacI-lacZ expression assays
LacI-LacZ fusion protein synthesis was measured in cell extracts from exponentially growing cultures. The protocol was essentially as in Berzal-Herránz et al. (26 ), except that cultures were grown in M9 minimal medium supplied with 0.2% glucose.
RESULTS
Construction of plasmids encoding two unit antisense RNA cassette genes
Secondary structures of the recognition unit segment of the antisense RNAs
Since the basic concept of the recognition unit was derived from previous studies of structural features that contribute to the binding rate of CopA (9 -11 ,20 ,31 ), we compared the secondary structures of SL II (stem-loop II of CopA) to stem-loops containing lacI sequences in the antisense or sense orientation respectively (a-SL II and s-SL II; Fig. 2 ). Structure-specific RNases were used to map single-stranded regions of in vitro transcribed, 5'-end-labelled RNAs. Figure 3 (upper panel) shows autoradiograms of partial cleavage patterns obtained with RNase A (cleavage 3' of single-stranded pyrimidines), T1 (single-stranded G residues) and T2 (single-stranded nucleotides with some preference for adenosines) and a schematic representation of the results (lower panel). All three stem-loops show indications of similar folding of the upper stem-loop, however, this region within s-SL II RNA is more exposed to single-strand-specific reagents, suggesting that the top three base pairs are not formed.
In vivo analysis of the plasmid-encoded antisense RNAs
The relative inhibitory efficiency of an antisense RNA depends on both its activity (e.g. binding/cleavage rate) and its intracellular concentration. Since it was conceivable that the stability of the various antisense RNAs could differ greatly, we performed an analysis of RNA extracted from strains that carried plasmids of the pGW48 series (as in Fig. 1 ). The use of Northern analysis permits quantification of relative RNA levels and permits an asessment of the intactness of the RNAs, i.e. provides information on specific decay intermediates and on aberrant termination. The probe used was directed against an RNA segment present in all RNAs and reprobing of the membrane for 5S rRNA permitted correction for loading.
The autoradiogram in Figure 4 (upper) and the summary of the relative signal intensities (lower) indicate that all antisense RNAs accumulate to similar intracellular concentrations. Stem-loop II of CopA (pGW47.1; Fig. 4 ) appears to be somewhat more stable than SLII containing the SmaI site sequences in the lower stem (pGW48, pGW48-a and pGW48-s; Fig. 4 ). All RNAs with insertions in front of SL II were ~2-fold more stable than the parental stem-loop RNA alone (encoded by pGW47-1).
The band pattern also shows that read-through of the SL II terminator occurs to an appreciable degree, ~7-9% in pGW48, pGW48-a and pGW48-s, as indicated by the appearence of bands consistent with termination at the plasmid-encoded fd terminator ~70 bp downstream of SL II (Fig. 4 , open squares). This phenomenon is more pronounced when additional sequences are present in front of SL II (rib or rbs in either orientation; cf. Fig. 1 B); signals of the `read-through' bands are up to 2.5 times more intense than the bands of the correctly terminated ones (Fig. 4 , open circles). Thus, either most transcripts fail to terminate at SL II, but terminate at the fd terminator, or the longer transcripts are more stable than the shorter ones. Since all RNAs are initiated at the same promoter, the latter possibility is more likely. In addition to the major bands, whose sizes are consistent with their expected lengths, we see a series of minor bands that most likely represent degradation intermediates. In conclusion, the Northern analysis shows that RNA sizes are as expected (taking into account a termination defect) and intracellular concentrations do not vary greatly in dependence on any of the sequences being present or absent.
Antisense RNA-mediated inhibition of LacI-LacZ fusion protein synthesis in vivo
In pilot experiments we transformed all cassette plasmids separately into an E.coli strain carrying a complete lac operon. Transformation and spreading on X-Gal indicator plates was expected to permit assessment of antisense activity by the appearance of blue colonies, since inactivation of lacI mRNA should derepress lacZ expression. The colonies obtained showed variations in blue colour intensity, but it proved difficult to use this test for quantification of effectiveness (data not shown).
Instead, plasmids were transformed into a bacterial strain that carried an F plasmid encoding a LacI-LacZ fusion protein (15 ). The fusion gene carries the wild-type lacI promoter and the 5'-portion of the lacI frame, as shown in Figure 2 . LacI-LacZ fusion protein synthesis was measured in bacterial extracts and the data are presented in Table 2 . Overall, inhibition by antisense RNAs was unexpectedly inefficient. The most efficient construct, pGW48-a-rbs, conferred an ~50% reduction in lacI-lacZ expression. Thus, the most efficient antisense RNA of the set tested carried both an anti-lacI recognition and an anti-RBS unit, i.e. both in antisense orientation to lacI mRNA. Generally, all anti-lacI recognition units promoted weak, but significant, inhibition, whereas the sense units were inefficient. Contributions of ribozyme units were minor or negligible, whereas an anti-RBS unit in conjunction with a recognition unit in sense orientation inhibited to ~41%.
In vitro antisense/target RNA binding
The poor in vivo efficiency indicated in Table 2 could be a result of slow binding rates, in particular in the case of rbs units, and of ineffective cleavage in the case of ribozyme units. Complex formation between complementary (even partially complementary) RNAs can be conveniently studied using gel shift assays. An excess of unlabelled target RNA (Materials and Methods) was incubated with 32P-labelled RNA, either a-rbs, s-rbs or, as controls, a-sbr and s-sbr RNA. Aliquots were withdrawn at different time points and complex formation measured (Fig. 5 ). Second order binding rate constants were calculated for the two rbs-containing RNA species (9 ). Both rbs RNAs, containing 38 nt complementary to the lacI RBS region, showed complex formation as a function of time, although the rate constants obtained (~3-6 * 104/M/s) were almost two orders of magnitude lower than those obtained with, for example, CopA/CopT (see for example 9 ). The contribution of the recognition stem-loop to binding rate was small: a recognition unit complementary to lacI (a-rbs, Fig. 5 ) was barely 2-fold more effective than the non-complementary one (s-rbs, Fig. 5 ). As expected, the sbr RNAs, which carry the 38 nt lacI segment in the sense orientation, did not form duplexes (Fig. 5 ).
aPlasmids encoding antisense RNA gene cassettes present in strain XAc(F'[lacI- lacZ fusion]). bAssays were performed as in Materials and Methods and represent averages of five independent determinations. The values obtained are represented in Miller units (18). Standard deviations are given. cMiller units were converted to relative activities. The activity obtained with pMc5-8 was set to unity.
In vitro cleavage activity of ribozyme-containing RNAs
Primer extension analysis of target RNAs extracted from cells carrying antisense RNA-encoding plasmids
DISCUSSION
This communication describes a set of antisense RNA cassettes designed to silence chosen target genes, here exemplified by the E.colilacI gene. The antisense RNA genes were built from cassettes (a promoter, an inhibitory unit and a recognition unit), all separated by suitable restriction sites. The rationale for this design was that an RNA element modelled according to recognition structures in naturally occurring antisense RNAs should promote rapid binding to a target sequence. Subsequently, an inhibitory segment of the RNA encoded by the gene cassette should bind to additional target sequences and promote inhibition of target RNA function. This could either be accomplished by sequestering of ribosome binding sites or by cleavage of the target RNA. In either case, inclusion of the recognition element was expected to render the base pairing rate of the inhibitory unit concentration independent. The antisense RNAs are thus non-contiguous in their complementarity to the target RNA and units can in principle be optimized separately.
The choice of the major CopA stem-loop II (SL II) as the parental recognition structure was based on previous studies: (i) stem-loops like SL II of CopA (10 ,11 ), stem-loops of RNA I (ColE1; 33 ) and that of RNA-OUT (Tn10; 34 ) are implicated in the rate determining step of target RNA binding; (ii) the rate of CopA/CopT binding is essentially diffusion controlled and therefore at a maximum (11 ,12 ); (iii) the structure of CopA is crucial for rapid target RNA binding, Hjalt and Wagner (20 ,31 ) having shown that changes in loop size as well as removal of characteristic bulged out nucleotides have detrimental effects on binding rate in vitro and inhibition in vivo; (iv) a growing body of evidence indicates that antisense/target RNA complexes can be inhibitory without forming complete duplexes (see 35 ,36 and references therein). Thus, we inferred that a recognition stem-loop that promotes fast binding should resemble, as closely as possible, CopA SL II. Hence, care was taken to construct SL II elements that both carried sequence complementarity to lacI RNA in the upper stem-loop region and preserved the CopA-like structure. Secondary structure probing using structure-specific ribonucleases (Fig. 3 ) and Pb(II) acetate (data not shown) confirmed that antisense lacI SL II carried a stem-loop structure close to the parental one. Determinations of steady-state levels of the antisense RNAs in cells carrying the cassette-containing plasmids indicated ~2- to 3-fold variations in concentration, indicating that fairly great variations in sequence and (presumed) secondary structure can be tolerated without major effects on antisense RNA half-life (Fig. 4 ). This figure also indicates that the presence of ribozyme or anti-RBS RNA sequences, in either orientation, markedly affected termination at SL II.
When the series of antisense RNAs was tested for inhibiton of LacI-LacZ fusion protein synthesis, relatively small effects were observed (Table 2 ). Generally, antisense RNAs carrying SL II sequences complementary to the 5'-segment of lacI mRNA (all pGW48-a variants) showed low but significant inhibition, whether or not they were combined with inhibitory units. Similarly, the presence of sequences complementary to the lacI RBS showed inhibition, whereas the sense orientation was ineffective, as was the ribozyme sequence in either orientation. The strongest inhibition obtained, 50%, was obtained with the RNA encoded by pGW48-a-rbs, i.e. with both the recognition loop and RBS unit in antisense orientation. It is conceivable that the contribution of the anti-lacI stem-loop alone to inhibition may be underestimated, since this RNA (pGW48-a; Fig. 4 ) is ~3-fold less abundant in steady-state than either of the rbs/sbr/bir/rib-containing RNAs.
In vitro binding of rbs RNAs to target RNA was measured (Fig. 5 ) and shown to occur at a rate ~100-fold slower than that obtained with natural antisense/target RNA combinations (see for example 9 ). The rate enhancement conferred by SL II was surprisingly small (2-fold in antisense orientation; Fig. 5 ). Low rates in the 104/M/s range are indicative of RNAs whose structures are not optimized for binding. For instance, in vitro binding of two complementary RNAs derived from the coding region of the repA gene of plasmid R1 yielded similar values (Söderbom and Wagner, unpublished) and, in a study of artificial anti-HIV RNAs, Homann et al. (7 ) measured association rate constants ranging from 0.3 to 4.0 * 104/M/s. Since in vitro binding of the rbs RNAs was slow, it appears likely that the inefficiency of inhibition (Table 2 ) is mainly due to slow rates of duplex formation in vivo. Congruent with this, most of the target RNA is intact, as judged from a primer extension analysis (Fig. 7 ). Only a minor fraction of lacI mRNA appears to be cleaved within the region of complementarity (see Fig. 2 for cleavage site). This cleavage, which occurs only in cells carrying rbs-containing antisense RNAs, is most likely catalysed by RNase III, as demonstrated in a variety of systems (see for example 24 ,37 ). In line with the superior inhibitory activity of a-rbs over s-rbs (Table 2 ), the ratio of the band denoted III over the band corresponding to the 5'-end of lacI mRNA was increased.
It is intriguing that natural antisense RNAs are rapid binders with association rate constants in the 106/M/s range, whereas most artificially created antisense RNAs are ~100-fold less effective (see above). Rapid association is crucial for the function of replication inhibitors in plasmids. It is therefore not surprising that the stem-loop recognition motifs commonly found in plasmid- encoded antisense RNAs (8 ) may have arisen under selective pressure for fast interaction. Nevertheless, as our data indicate, we still have too little insight into the factors that determine this property and, hence, cannot yet tailor such structures to any chosen target sequence.
In contrast to the limited, yet significant, inhibition by rbs RNA, all units containing hairpin ribozymes were ineffective (Table 2 ). In vitro ribozyme activity assays indicated that cleavage rates were very low and that the presence of a cognate recognition stem-loop only accelerated the reaction by a factor of two. Figure 6 A indicates that, under the conditions used, >3 h are required to cleave half of the target RNA molecules. Since higher ribozyme concentrations only moderately enhanced the rate of cleavage, we infer that the chemical step rather than the association step is limiting. In agreement with this inefficient catalytic activity, we failed to observe cleavage of target RNA in vivo (Fig. 7 ) at the site mapped in vitro (Fig. 6 B).
From the data presented, it is clear that the degree of inhibition obtained is unsatisfactory. For most gene silencing applications >>10-fold inhibition is desirable. The results reported here indicate some of the difficulties encountered and may point out what is required to improve performance. Since target RNA binding is a function of both the concentration of the antisense RNA and the binding rate constant, an increase in either one or both of these factors should improve inhibition. Daugherty et al. (5 ) obtained strong inhibition of lacZ expression with antisense RNAs directed against a large 5'-segment of the target RNA when the antisense/target RNA ratio was >100. Intracellular concentrations were not estimated, so that a direct comparison with the results presented here is not possible. From the Northern analysis in Figure 4 we estimate that antisense RNA concentrations ranged from 0.1 to 1 [mu]M, which in the case of an efficient natural antisense RNA results in >95% inhibition (12 ). If the specific inhibitory efficiency of an artificial antisense RNAs is low, a great increase in steady-state level can compensate for this deficiency. Protective 5' stem-loops could be used to protect RNAs from 5'-initiated degradation (38 ), to increase metabolic stability and intracellular concentration. Selection of rapidly binding RNAs from a pool of complementary species differing in their extent of complementarity (39 ) may be used to enhance binding substantially. Since the half-life of typical bacterial mRNAs is in the range 2-4 min (40 ), it is likely that binding rate constants in the 104/M/s range are insufficient to inactivate a substantial fraction of the target RNAs.
The experiments shown in Figure 6 and Table 2 indicate that catalytic RNAs may not be suitable as efficient inhibitors in bacterial systems. Even a considerably increased cleavage rate would only marginally improve inhibition (mRNA half-lives of 2-4 min compared with cleavage rates of more than a generation time), in particular since an increased intracellular concentration would not result in a corresponding increase in target cleavage (see above). A report of hammerhead ribozyme-dependent inhibition of proliferation of RNA coliphage SP in E.coli is pertinent to the same limitation of this approach (41 ), but a different, recent study indicates that >90% inhibition can be obtained (target lacZ; 42 ). It is likely that the differences in target RNA half-lives between prokaryotes and eukaryotes are responsible for the greater success of ribozyme strategies in the latter type of organism (43 ,44 ).
Analysis of the performance of the two unit antisense RNAs described here indicates what modifications may be beneficial for improved inhibitory potential. Since the cassettes are separated by suitable restriction sites, separate optimization protocols can be used. The choice of stronger, inducible promoters is expected to yield some increase in inhibition. Randomized sequences introduced in the region encoding the upper portion of the recognition stem-loop (bordered by SmaI/XmaI sites) can be used to select rapid binders of in vitro synthesized target RNA. The small, but significant, effect of the recognition unit alone (Table 2 ) suggests that RNAs consisting of multiple stem-loop units each with its own separate target complementarity may be effective. Inhibition units can be kinetically selected from complementary sets of anti-RBS sequences as in Rittner et al. (39 ). In addition, the introduction of RNA stability elements at the 5'- and/or 3'-ends of the antisense RNA should permit accumulation of higher concentrations of the inhibitor. Even the combination of only moderately improved units should thus yield a substantial increase in antisense RNA efficiency. Finally, it is conceivable that the choice of the target RNA may affect the degree of inhibition that can be obtained. We have therefore begun to adopt a similar strategy using plasmid-encoded bla mRNA as an inhibition model.
In conclusion, we suggest that the pitfalls of antisense RNA design require thorough study of the parameters crucial for inhibition. We believe that the approach taken here may be helpful to obtain versatile and easily optimizable antisense RNA inhibitors.
ACKNOWLEDGEMENT
This work was supported by a grant from the Swedish Board of Engineering Sciences (TFR) to E.G.H.Wagner.
18 Miller,J.H. (1972) Experiments in Molecular Genetics. Cold Spring Harbor Laboratory Press. Cold Spring Harbor, NY, pp. 352-355.
19 Sambrook,J., Fritsch,E.F. and Maniatis,T. (1989) Molecular Cloning: A Laboratory Manual. Cold Spring Harbor Laboratory Press. Cold Spring Harbor, NY.
20 Hjalt,T.H. and Wagner,E.G.H. (1992) Nucleic Acids Res., 20, 6723-6732.
