ANTISENSE & NUCLEIC ACID DRUG DEVELOPMENT 8:249-254 (1998)Mary Ann Liebert, Inc.

State-of-the-Art Review

Antisense Drug Discovery: Can Cell-Free ScreensSpeed the Process?

ANDREA D. BRANCH

ABSTRACT

Many conditions must be satisfied for an antisense drug to function. It must colocalize with its target RNA at a

sufficient concentration for a bimolecular reaction to occur, and it must have a structure that favors associa-tion with its target. In addition, if the antisense compound is to form Watson-Crick bonds with the target RNA,it must be complementary to sites that are amenable to binding. Unfortunately, the peculiarities that cause cer-

tain sites to be especially vulnerable to antisense compounds are undefined, as discussed previously (Branch,1998). Because vulnerable target sites have no common properties allowing them to be identified by sequenceanalysis, most target sites and their antisense counterparts are found through a trial and error process inwhich oligomers—each complementary to a different site in the target RNA—are tested individually to find theone with the greatest specificity and lowest inhibitory concentration (IC50). However, testing antisense mole-cules one at a time can be a taxing process, and there is great interest in developing cell-free screening methodsthat can reduce the number of compounds that must be tested in cells and in whole animals. These cell-freescreens are designed to generate short lists of target sites that include the ideal site—the site most vulnerable toantisense ablation in vivo. They are based on the unproven assumption that ideal sites have distinctive proper-ties, such as susceptibility to RNase H-mediated cleavage, that allow them to be detected in cell-free assays.This is a review of data emerging from studies using RNase H-based screens and a summary of the challengesconfronting these and any similar methods that use naked RNAs as surrogates for intracellular RNAs. It is notyet clear if cell-free screening methods will be effective.

IIT PAYS TO SEND THE VERY BEST more effective (D. Marijourides, personal communication,

1998). In a phase III clinical trial, ISIS 2922, a phosphoroth-T now costs more than $500 million to bring a new drug to ioate oligodeoxynucleotide (S-ODN) designed to treat cy-the market, according to Drews and Ryser (1997), and costs tomegalovirus (CMV)-associated retinitis, was found to cause

are rising. High costs of research and development place a pre- reversible "mild to moderate intraocular inflammation" (Muc-mium on finding the best antisense molecule, so that a corpo- cioli et al., 1998). ISIS 2922 also caused transient inflammatoryrate competitor is unlikely to find a superior drug complemen- responses in pig and rabbit models, prompting Flores-Aguilar et

tary to the same target. Market share is only one reason to seek al. (1997) to conclude that efforts to improve its "low intraocu-the most specific and potent antisense drug. Equally important lar therapeutic index" are indicated. Because of the need tois the need to minimize toxic, dose-dependent, nonantisense ef- minimize toxicity, pharmaceutical developers require methodsfects. The true clinical significance of nonantisense effects will that identify the most potent and specific compounds. To bebecome apparent only after several human trials have been useful to industry, RNase H-based screening assays must sat-

completed. However, enough adverse effects have already been isfy this requirement. (However, laboratory research groupsreported to cause serious concern. For example, GEM 91, a may benefit from screening methods that succeed at the lessdrug for the treatment of AIDS, reached phase II clinical trials challenging task of increasing the average performance of an-

and then was discontinued, to be replaced with compounds that tisense compounds complementary to the intended targetthe manufacturer, Hybridon, hopes will be better tolerated and RNA.)

Department of Medicine, The Mount Sinai School of Medicine, New York, New York 10029.

249

250 BRANCH

ANTISENSE SCREENS MUST OUTPERFORMSHOTGUN APPROACHES

To establish a standard for comparison, it is useful to con-

sider the potency of antisense oligomers obtained without thebenefit of prescreening methods. In an unusually complete anddetailed study, Monia (1996) measured C-raf kinase mRNAlevels in A549 lung carcinoma cells treated with a series of S-ODNs complementary to different portions of the target RNA.For RNA studies, cells were treated with 20-mers (200 nM) inthe presence of cationic lipids and extracted 24 hours later.Northern hybridization revealed that ISIS 5132 was an order ofmagnitude more effective at lowering target mRNA levels thanthe other S-ODNs. The distribution of potencies of the 34 S-ODNs is shown in Figure 1A.

ISIS 5132 had a median inhibitory concentration (IC50) fortarget mRNA reduction of approximately 50 nM. The IC50 forthe reduction of C-raf kinase protein was twofold higher, 100nM, and was measured under different conditions from the IC50of the target mRNA. To reduce protein levels, cells were giventwo doses of the S-ODN separated by 24 hours and were incu-bated for a minimum of 40 hours before extraction. As the au-

thors pointed out, it is likely that the second treatment was

needed simply because the protein has a long half-life andmRNA levels began to rise after 24 hours. However, discrepan-cies between RNA and protein decay curves could be a sign thatsuch processes as translational regulation are occurring. Dispar-ities between RNA and protein susceptibilities need to be inves-tigated, particularly if mRNA reduction is to be used as a

marker of antisense activity in the future.In the meantime, the experiments carried out by Monia et al.

(1996) highlight the ability of the shotgun method to yield po-tent and selective antisense compounds. They set a high stan-dard that cell-free screening methods must equal or surpass.

EVIDENCE THAT RNA STRUCTUREINFLUENCES ANTISENSE

SUSCEPTIBILITY—THE UNDERPINNINGSOF CELL-FREE SCREENS

Inside cells, antisense molecules do not encounter nakedRNAs. They confront ribonucleoprotein complexes that containproteins and the target RNA folded into a three-dimensionalstructure. Under these circumstances, most antisense com-

pounds are relatively ineffective at lowering levels of their tar-

get RNAs (Fig. 1A), leading to the conclusion that most targetsites are inaccessible. To date, few experiments have distin-guished between antisense resistance conferred by RNA struc-ture and resistance due to protein cloaking. When studies are

carried out in vivo, the factors responsible for variation in targetsite vulnerability can be treated as a black box. However, be-cause cell-free screening methods use naked RNA as startingmaterial, their validity requires that RNA structure per se be a

key determinant of target site vulnerability. Evidence indicatesthat it is (although this evidence does not necessarily imply thatRNAs form the same structures in cells and in test tubes).

Direct evidence that intrinsic RNA structure influences anti-sense susceptibility inside cells comes from a study carried out

by Wagner (1996). The antisense molecules used in this studycontained C-5 propyne pyrimidines and phosphorothioate inter-nucleotide linkages. The oligomers (1000 nM) were microin-jected into nuclei along with various expression plasmids. Aninitial set of experiments revealed that the SV40 T antigen(TAg) mRNA was sensitive to ablation by a particular 7-mer,ON8, whereas cyclin E mRNA was resistant, even though bothmessages contain the target sequence, AAGGAGG. Subse-quently, the investigators produced plasmids in which the targetsequence and either the flanking sequences of TAg mRNA or

those of cyclin E mRNA were inserted into the 5'-untranslatedregion of a luciferase gene. Following antisense treatment, lu-ciferase levels were reduced when the TAg RNA flanking se-

quences were present but not when those of cyclin E mRNAwere present. As the experimental design controlled for othervariables, this result powerfully demonstrates that RNA struc-ture can modulate antisense susceptibility within cells.

Conversely, recent studies by Milner et al. (1997) provideconclusive evidence that RNA structure influences antisensebinding in solution. They measured the binding of 1938 ODNsto a 122-nucleotide-long RNA representing the 5'-end of rabbitß-globin mRNA. These ODNs ranged in length frommonomers to 17-mers and contained all possible sequencescomplementary to the target RNA. When analyzed in a cell-freetranslation assay, the 17-mer that bound the target RNA mostefficiently was five times more potent than any other ODN. Thebinding site of the most effective ODN, BG1, overlapped withthose of other, much less potent ODNs, such as BG2 (Fig. 2).Thus, the ideal target site had sharp boundaries, although a pri-ori, this site had no obvious features that marked it as a particu-larly good binding site. Together, the sharp boundaries and theidiosyncratic nature of the binding site suggest that a very largenumber of compounds may need to be tested to find the bestone. Such implications account for the desire to develop facilecell-free screening methods. However, although it is clear thatRNA structure can influence antisense vulnerability in vivo andthat RNA structure can influence antisense binding in solution,it remains to be determined whether the sites that are maximallyvulnerable in vivo also bind antisense compounds most effi-ciently in cell-free assays. This question is currently being ex-

plored in studies of RNase H-based screens.

CELL-FREE RNASE H-BASED SCREENS MAYPROVE USEFUL BUT HAVE NOT

BEEN VALIDATED

Evidence that antisense effects are often mediated by RNaseH in vivo has led to several cell-free screening procedures thatmap RNase H-sensitive sites. These screens supplant previousefforts to select vulnerable target sites based on computer-gen-erated models of RNA structure or on thermodynamic estimatesof hybrid stabilities.

Ho et al. (1998) recently carried out an extensive RNase H-based screening assay. They first used over 4 million chimeric11-mers to identify sites in angiotensin type 1 receptor mRNAthat were highly susceptible to RNase H-mediated cleavage.They then obtained antisense S-ODN 20-mers whose 5'-endswere in proximity to either accessible sites (25 sites) or inacces-sible sites (8 sites) and tested the ability of these oligomers to

PROBING FOR ANTISENSE TARGET SITES 251

iOQtr>Qtr>otf>o

uojsssjdxgjo |9Aan U9AI0 B BupaiA SNQO-S %

CO

s S S S 2uoissejdxB

¿o leAen U8A|9 e ßupeiA SNQO-S %

* 2 S° ». Ê3= ET"a 7t &> g »

u E 2

SES•-OU

rt o "^

s<

!S otí ^

S3 !»S___ 3

O

2 •£

SS -

fr E03 O w

c -S s

¡ S Sc c uC 3 uO — 3«» a!» * ¡2

S -I S

<! s« s

z g xo * s"ï -r ^c .ss <'S Z§ 0¿

i » .

- á 73u c X> E o— i. X^ o _u -s P

>».£>

* -~ ta

OsiSi cc oa u

.5 o

252 BRANCH

20—c J^g il g u u[\caca >V Je g ú ÚI ¡aü>—cu c\ hgu y aVy«

« g g g

10

FIG. 2. Location of binding sites for a highly potent ODN, BG1, and for two less potent ODNs, BG2 and BG3, on a secondary structure map ofrabbit ß-globin mRNA. (From Milner et al., 1997).

reduce levels of angiotensin type 1 receptor in Chinese hamsterovary (CHO) cells stably transfected with an AT, receptor ex-

pression plasmid. Cells were treated with ODN-liposome com-

plexes for 4 hours, and AT, receptor levels were determined thefollowing day using a radioligand binding assay. When admin-istered at 1000 nM, none of the oligomers reduced receptor lev-els to <20% of the control, as shown in Figure IB. In contrast,using no prescreening methods, Monia et al. (1996) found an S-ODN that reduced C-ra/kinase mRNA levels to <10% of con-

trol levels (Fig. 1A), and Cioffi et al. (1997) found one that re-

duced rat C-raf kinase protein levels to 12% of controls.Although differences between the experimental systems pre-clude a direct comparison, it is not evident that the prescreeningstep used by Ho et al. identified the ideal target site nor that itwas faster than the old try-a-little-of-everything method. Fur-thermore, because most of the S-ODNs complementary toRNase H-sensitive sites were administered at doses four timesthe level of those complementary to RNase H-resistant sites(Ho et al., 1998), the results were difficult to interpret, under-scoring the need to use a standard set of experimental condi-tions.

In a previous investigation, Ho et al. (1996) also used a cell-free RNase H-based assay to seek potent antisense oligomerscomplementary to human multidrug resistance 1 mRNA, whichencodes P-glycoprotein. This study is not considered in detailhere because an indirect, functional assay—inhibition of rho-damine export—was used to assess antisense potency. Steinand Krieg (1994) have stated that it "is virtually imperative thatthe experimenter demonstrate a decrease in the target protein ifan antisense mechanism is proposed."

The relationship between target site vulnerability inside cellsand RNase H susceptibility in solution was directly addressedin studies by Wagner et al. (1996). They compared the ability ofvarious oligomers to reduce SV40 T antigen expression insidecells with their ability to induce RNase H-mediated cleavage ofTAg mRNA. They found examples of oligomers that were lesseffective than others at inhibiting the production of TAg in thein vivo assay but more effective at promoting RNase H-medi-

ated cleavage in the cell-free assay. Thus, the rate of RNase Hcleavage did not predict activity in vivo. The authors concludedthat vulnerable target sites in cells did not produce hot spots forRNase H cleavage in their cell-free extract. Many more studiescomparing in vivo vulnerability with cell-free RNase H suscep-tibility are needed to determine whether the lack of correlationreported by Wagner et al. is a widespread phenomenon. How-ever, these results, which were obtained using HeLa cell nu-

clear extracts as the source of RNase H, raise doubts about thevalue of cell-free RNase H-based screening methods.

CHALLENGES CONFRONTING CELL-FREESCREENING METHODS: HOW THEY MIGHT

FALL SHORT

¡deal sites may be relatively common

It is possible that cell-free screening cannot confer a benefitbecause ideal binding sites are common enough that they can beidentified by the brute force method—by testing 30-50oligomers in cells. The density of ideal sites is not known.However, Monia et al. (1996) found one highly vulnerable sitewhile probing 680 nucleotides (34 times 20), Milner et al.(1997) found one high-affinity site while probing 122 nu-

cleotides, and Ho et al. (1998) found four vulnerable sites in a

1253 base-long RNA. According to these studies, an ideal sitemay occur about once every 400 nucleotides. If this is the ap-proximate density of ideal sites, cell-free screening methodsmay fail because there is simply no way to recover the time theyconsume.

The predominant conformation of a target RNA insolution may differ from its functional conformation

The ability of RNA molecules to fold into multiple confor-mational isomers is well established. In fact, this structural vari-

PROBING FOR ANTISENSE TARGET SITES 253

ability is the bane of RNA biochemistry: the presence of multi-ple structures makes long RNA molecules nearly impossible to

crystallize and often causes ribozyme populations to be mix-tures of active and dead enzymes. The presence of multipleconformational isomers impedes efforts to determine the actualstructure of functional RNA. To the extent that misfolded struc-tures predominate in solution, all structural probing methods,including RNase H sensitivity tests, will yield inaccurate infor-mation about the structure that exists in vivo. As a further com-

plication, RNA structure in vivo may vary depending on thefunctional state of the RNA. For example, the eukaryotic pro-tein synthesis initiation factor elF4A is thought to unwindmRNA structure upstream of the initiation codon (Rozen et al.,1990). Thus, proteins may alter RNA structure. In addition,proteins may bind to target RNAs and conceal portions of them.Ribosomes can definitely protect target RNAs from ribozyme-mediated cleavage in Escherichia coli (Chen et al., 1997). It islikely that ribosomes and additional ribonucleoprotein com-

plexes, such as spliceosomes, protect portions of target RNAsin human cells.

Cells may differ in their interactions withantisense molecules

Although antisense failures are usually blamed on inaccessi-ble target sites, in at least some cases, the lack of efficacy maybe due to binding of the antisense compound to proteins or

other intracellular components. Such binding could inducenonantisense effects and cause ODNs to be waylaid. As thecomponents that bind particular ODNs may be present in some

cells and not others, the significance of these mishaps may varybetween tissues and between cell types. Their impact, like thatof interactions between the target RNA and cellular proteins,cannot be assessed in RNase H-based assays of naked RNAs. Itis not clear how severely this inability compromises cell-freescreening methods.

DISCUSSION

It was originally thought that antisense target site selectionwas the prerogative of the investigator. Now it is clear that idealsites exist in RNAs and that the task of the antisense developeris to find them. To determine if cell-free screening methodsspeed the identification of ideal sites, antisense molecules com-

plementary to good and not-so-good target sites need to be ad-ministered to a number of cell types under a standard set of con-

ditions. In addition, the performance of preselected antisensemolecules needs to be compared with those chosen through theshotgun approach.

Useful information about the density of ideal antisense tar-

get sites will emerge from such studies. The actual density andlength of ideal sites are of utmost importance to the antisensefield and will have a major impact on the ultimate success or

failure of antisense compounds. The characteristics of thesesites will help to determine the upper limit of antisense speci-ficity. If ideal sites in RNA occur once every 400 bases andare 10 nucleotides long, the prospects for developing highlyselective drugs are bright. Alternatively, if ideal sites occur

once every 100 nucleotide and are 7 bases long, the prospectsfor developing gene-specific antisense compounds are dim

because each antisense drug will (inevitably) bind to sites inmany different RNA molecules. Similar considerations proba-bly also apply to ribozymes. However, ribozymes are more

complex than ODNs and less is known about their ideal targetsites.

Ironically, antisense drugs (and therapeutic ribozymes) willhave the greatest potential for specificity if ideal sites turn outto be so rare that only a minority of RNAs contain them. Ifsome RNAs are terrific targets and most are poor, it will be use-

ful to find ways to identify the sensitive ones. Perhaps cellscould be treated with each of the (16,384) 7-mers, and the hu-man mRNA that is most susceptible to each one could be iden-tified by gene profiling. If RNAs containing an ideal site are

rare, this screening approach will yield a drug for only a minor-ity of genes. However, the antisense compounds for thesehighly susceptible genes may have exceptionally low IC50 val-ues compared with antisense compounds selected through cur-

rent methods. The enhanced specificity and reduced toxicity ofantisense compounds targeted against especially vulnerableRNAs may justify the arduous task of identifying the pairs ofantisense molecules and their target RNAs.

It is too early to judge whether the cell-free screening meth-ods currently in use will prove to be effective at identifying an-

tisense compounds that optimally induce antisense-mediatedgene ablation in cells. Furthermore, the ultimate tests of thesemethods will be their ability to identify optimal antisense com-

pounds for use in whole animals. The results, of these tests willnot be available for some time. However, cell-free RNase H-based screens have already advanced antisense drug develop-ment by raising important questions about the frequency andproperties of highly vulnerable target sites.

ACKNOWLEDGMENTS

I thank Drs. Nora V. Bergasa (Beth Israel Medical Center),Kerry K. Willis (Academic Press), and Mr. Decherd D. Stump(Mount Sinai School of Medicine) for insights, and Ms. TobyR. Keller for assistance. This work was supported by NIDDKgrant P01DK50759, project 2, NIDDK grant R01DK52071,and the Liver Transplantation Research Fund (Department ofSurgery, the Mount Sinai Medical Center).

REFERENCES

BRANCH, A.D. (1998). A good antisense molecule is hard to find.TIBS 23,45-50.

CHEN, H., FERBEYRE, G, and CEDERGREN, R. (1997). Efficienthammerhead ribozyme and antisense RNA targeting in a slow ribo-some Escherichia coli mutant. Nat. Biotech. 15, 432^35.

CIOFFI, C.L., GARAY, M., JOHNSTON, J.F., McGRAW, K.,BOGGS, R.T., HRENIUK, D., and MONIA, B.P. (1997). Selectiveinhibition of A-Raf and C-Raf mRNA expression by antisenseoligodeoxynucleotides in rat vascular smooth muscle cells: Role ofA-Raf and C-Raf in serum-induced proliferation. Mol. Pharmacol.51, 383-389.

DREWS, I., and RYSER, S. (1997). The role of innovation in drug de-velopment. Nat. Biotech. 15, 1318-1319.

FLORES-AGUILAR, M., BESEN, G, VUONG, C, TATEBAYASHI,M., MUNGUIA, D., GANGAN, P., WILEY, CA., and FREEMAN,W.R. (1997). Evaluation of retinal toxicity and efficacy of anti-cy-

254

tomegalovirus and anti-herpes simplex virus antiviral phosphoro-thioate oligonucleotides ISIS 2922 and ISIS 4015. J. Infect. Dis. 175,1308-1316.

HO, S.P., BAO, Y., LESHER, T., MALHOTRA, R., MA, L.Y.,FLUHARTY, S.J., and SAKAI, R.R. (1998). Mapping of RNA ac-

cessible sites for antisense experiments with oligonucleotide li-braries. Nat. Biotech. 16, 59-63.

HO, S.P., BRITTON, D.H., STONE, B.A., BEHRENS, D.L., LEFFET,L.M., HOBBS, F.W., MILLER, J.A., and TRAINOR, G.L. (1996).Potent antisense oligonucleotides to the human multidrug resistance-1 mRNA are rationally selected by mapping RNA-accessible siteswith oligonucleotide libraries. Nucleic Acids Res. 24, 1901-1907.

MILNER, N., MIR, K.U., and SOUTHERN, E.M. (1997). Selecting ef-fective antisense reagents on combinatorial oligonucleotide arrays.Nat. Biotech. 15, 537-541.

MONIA, B.P., JOHNSTON, J.F., GEIGER, T., MULLER, M., andFABBRO, D. (1996). Antitumor activity of a phosphorothioate anti-sense oligodeoxynucleotide targeted against C-raf kinase. NatureMed. 2,668-675.

MUCCIOLI, C, GOLDSTEIN, D.A., JOHNSON, D.W., PEREZ, J.E.,MORA-DUARTE, J.F., SHEPPARD, J.D., MANSOUR, S.E.,CHAN, C.K., PALESTINE, A.G., GRILLONE, L.R., and CHAND-LER, J.W. (1998). Fomivirsen safety and efficacy in the treatment ofCMV retinitis: Phase III, controlled, multicenter study comparing

BRANCH

immediate versus delayed treatment. Abstract 770, Retrovirus andOpportunistic Infection Conference, Chicago.

ROZEN, F., EDERY, I., MEEROVITCH, K., DEVER, T.E., MER-RICK, W.C., and SONENBERG, N. (1990). Bidirectional RNA heli-case activity of eucaryotic translation initiation factors 4A and 4F.Mol. Cell. Biol. 10, 1134-1144.

STEIN, CA., and KRIEG, A.M. (1994). Problems in interpretation ofdata derived from in vitro and in vivo use of antisense oligodeoxynu-cleotides. Antisense Res. Dev. 4, 67-69.

WAGNER, R.W., MATTEUCCI, M.D., GRANT, D., HUANG, T., andFROEHLER, B.C. (1996). Potent and selective inhibition of gene ex-

pression by an antisense hetanucleotide. Nat. Biotech. 14, 840-844.

Address reprint requests to:Dr. Andrea D. Branch

Division of Liver Diseases, Box 1633Department of Medicine

The Mount Sinai School of MedicineOne Gustave L. Levy Place

New York, NY 10029

Received January 21, 1998; accepted in revised form March10, 1998.