Artificial miRNAs mitigate shRNA-mediated toxicityin the brain: Implications for the therapeuticdevelopment of RNAiJodi L. McBride*, Ryan L. Boudreau*, Scott Q. Harper*†, Patrick D. Staber*, Alex Mas Monteys*, Ines Martins*,Brian L. Gilmore*, Haim Burstein‡, Richard W. Peluso‡, Barry Polisky§, Barrie J. Carter‡, and Beverly L. Davidson*¶�**

Departments of *Internal Medicine, ¶Molecular Physiology and Biophysics, and �Neurology, University of Iowa, Iowa City, IA 52242; ‡Targeted Genetics, 1100Olive Way, Suite 100, Seattle, WA 98101; and §Sirna Therapeutics, 1700 Owens Street, San Francisco, CA 94158

Communicated by David E. Housman, Massachusetts Institute of Technology, Cambridge, MA, February 27, 2008 (received for review February 10, 2008)

Huntington’s disease (HD) is a fatal, dominant neurodegenerativedisease caused by a polyglutamine repeat expansion in exon 1 ofthe HD gene, which encodes the huntingtin protein. We and othershave shown that RNAi is a candidate therapy for HD becauseexpression of inhibitory RNAs targeting mutant human HD trans-genes improved neuropathology and behavioral deficits in HDmouse models. Here, we developed shRNAs targeting conservedsequences in human HD and mouse HD homolog (HDh) mRNAs toinitiate preclinical testing in a knockin mouse model of HD. Wescreened 35 shRNAs in vitro and subsequently narrowed our focusto three candidates for in vivo testing. Unexpectedly, two activeshRNAs induced significant neurotoxicity in mouse striatum, al-though HDh mRNA expression was reduced to similar levels by allthree. Additionally, a control shRNA containing mismatches alsoinduced toxicity, although it did not reduce HDh mRNA expression.Interestingly, the toxic shRNAs generated higher antisense RNAlevels, compared with the nontoxic shRNA. These results demon-strate that the robust levels of antisense RNAs emerging fromshRNA expression systems can be problematic in the mouse brain.Importantly, when sequences that were toxic in the context ofshRNAs were placed into artificial microRNA (miRNA) expressionsystems, molecular and neuropathological readouts of neurotox-icity were significantly attenuated without compromising mouseHDh silencing efficacy. Thus, miRNA-based approaches may pro-vide more appropriate biological tools for expressing inhibitoryRNAs in the brain, the implications of which are crucial to thedevelopment of RNAi for both basic biological and therapeuticapplications.

gene therapy � Huntington’s disease � RNAi � AAV

The ability of siRNAs to silence target genes was first demon-strated in 1998 by Andrew Fire et al. (1) and has since emerged

as a revolutionary strategy to reduce target gene expression. RNAioccurs naturally in cells as a posttranscriptional regulatory mech-anism mediated by endogenous miRNAs (2–5). RNAi is hypoth-esized to have evolved as a cellular coping mechanism providing thecell a means to decrease the expression of various deleteriousviruses and transposons (6, 7). In recent years, scientists havecoopted this biological process to reduce the expression of targetmRNAs by using exogenously applied siRNAs, shRNAs, or artifi-cial miRNAs. Aside from the widespread basic biological applica-tions of RNAi, the ability to reduce gene expression marks a majoradvance toward the development of disease therapies, particularlyfor dominantly inherited disorders.

Among the dominant diseases that may benefit from RNAi-based therapies is Huntington’s disease (HD). Our laboratory (8)and others (9) have previously demonstrated that partial reductionof mutant huntingtin expression by viral delivery of shRNAs isefficacious in preventing the development of motor deficits andneuropathology in transgenic mouse models of HD. In theseproof-of-principal studies, the therapeutic effect on disease pheno-type was studied by knocking down a mutant human HD transgene

in the setting of two normal mouse HD homolog (HDh) alleles.Although allele-specific targeting of disease transcripts for HDtherapy would be ideal, to date no prevalent SNP residing on themutant transcript has been identified. Therefore, we undertookstudies to identify inhibitory RNAs that would target both mouseHDh and human HD transcripts, with the intention of testing theefficacy of reducing the expression of both alleles in a knockinmodel of HD (10). Here we describe the surprising finding ofneurotoxicity in mouse brain caused by some, but not all, shRNAexpression vectors screened in vivo and the notable reduction intoxicity after moving those toxic inhibitory RNAs into miRNA-based delivery systems.

ResultsshRNAs Cause Striatal Toxicity in Mice. We first designed andscreened shRNAs (driven by the mouse U6 promoter) that targetconserved sequences spanning human HD and mouse HDhmRNAs [Fig. 1A and supporting information (SI) Table S1],taking into consideration the most recent siRNA design rules(11–13). Silencing of HD mRNA measured by quantitativereal-time PCR (QPCR) and dot blot analysis revealed a decreasein huntingtin protein expression after transfection of shRNAexpression plasmids into mouse C2C12 and human-derived HEK293 cell lines (data not shown). Of the 35 shRNAs tested, threewere chosen for further study based on silencing efficacy. TheshRNAs target sequences in exons 2, 8, and 30 of HD mRNAsand are henceforth referred to as sh2.4, sh8.2, and sh30.1,respectively (Fig. 1B). Western blot analysis demonstrated thatthese shRNAs, but not mismatch (mis) control shRNAs, reduceendogenous huntingtin protein expression in mouse C2C12 cells(Fig. 1C). Similar results were seen in human-derived HEK 293cells.

To examine the long-term effects of brain-delivered shRNAs inthe CAG140 knockin mouse model of HD (10), U6-shRNAexpression cassettes were cloned into adeno-associated viral vectors(AAV serotype 2/1) (Fig. 2A). AAVs also contained a humanizedRenilla GFP (hrGFP) expression cassette to identify the distribu-tion and types of cells transduced. Five-week-old CAG140 knockinmice were injected bilaterally into the striatum with AAVsh2.4-GFP, AAVsh8.2-GFP, AAVsh30.1-GFP, or AAV-GFP (viral con-trol) and killed 15 weeks later. Robust expression of GFP was

Author contributions: J.L.M., R.L.B., and S.Q.H. contributed equally to this work; J.L.M.,R.L.B., S.Q.H., B.P., B.J.C., and B.L.D. designed research; J.L.M., R.L.B., S.Q.H., P.D.S., A.M.M.,I.M., B.L.G., H.B., and R.W.P. performed research; J.L.M., R.L.B., S.Q.H., and B.L.D. analyzeddata; and J.L.M., R.L.B., S.Q.H., and B.L.D. wrote the paper.

Conflict of interest statement: B.L.D. was a consultant for Sirna Therapeutics, Inc..

†Present address: Center for Gene Therapy, Department of Pediatrics, Ohio State University,Columbus, OH 43205.

**To whom correspondence should be addressed. E-mail: beverly-davidson@uiowa.edu.

This article contains supporting information online at www.pnas.org/cgi/content/full/0801775105/DCSupplemental.

© 2008 by The National Academy of Sciences of the USA

5868–5873 � PNAS � April 15, 2008 � vol. 105 � no. 15 www.pnas.org�cgi�doi�10.1073�pnas.0801775105

Dow

nloa

ded

by g

uest

on

May

18,

202

0

observed in cells throughout the rostral/caudal extent of the stria-tum and within fibers of the globus pallidus (Fig. 2B). Immuno-fluorescence analyses indicated that GFP-positive cells colocalizedwith a neuronal marker (NeuN), but not with markers for astrocytes(GFAP) or oligodendrocytes (RIP1) (Fig. S1). QPCR performedon RNA isolated from GFP-positive striatal tissue showed asignificant and statistically similar reduction of HDh mRNA ex-pression (�60%) among the different active shRNA-expressingvectors, compared with mice injected with AAV-GFP [F(3, 11) �32.3, P � 0.001 for post hoc analyses comparing each AAV-shRNAgroup to the AAV-GFP control] (Fig. 2C). Moreover, Western blotanalysis demonstrated a significant reduction in huntingtin proteinlevels after AAVshRNA-GFP administration, compared with mis-match controls [t(8) � 3.9, P � 0.01] (Fig. S2).

Unexpectedly, immunohistochemical analyses for dopamine-and cAMP-regulated protein (DARPP-32), a marker of medium-sized spiny projection neurons in the striatum, revealed striataltoxicity in mice injected with AAVsh2.4-GFP and AAVsh30.1-GFP (Fig. 2D Upper). Reduction in DARPP-32 immunoreactivitywas largely confined to the transduced (GFP-positive) regions ofthe striatum. Interestingly, this toxicity was not seen in mice injectedwith AAVsh8.2-GFP (Fig. 2D Upper). Striata from these mice weresimilar to AAV-GFP-injected control mice.

To assess whether the observed loss of DARPP-32 staining wasassociated with microglial activation, tissue sections were stainedwith an anti-Iba1 antibody to identify both resting and reactivemicroglia throughout the brain. AAVsh2.4-GFP- and AAVsh30.1-

GFP-injected striata demonstrated high Iba1 expression, whereasAAVsh8.2-GFP-injected striata were similar to control mice (Fig.2D Lower). Moreover, AAVsh2.4-GFP- and AAVsh30.1-GFP-injected mice demonstrated dramatic reactive astrogliosis, com-pared with AAVsh8.2-GFP- and control-injected mice, as evi-denced by robust GFAP staining in areas of the striatumcorresponding to high GFP positivity (data not shown). Notably, amismatch control for the HD2.4 sequence, AAVsh2.4mis-GFP,induced toxicity similar to sh2.4 and sh30.1 without reducing HDhmRNA expression. This, in addition to the sh8.2 data, indicates thatthree (two active, one inactive) of four shRNAs were toxic and thattoxicity is not caused by silencing huntingtin.

Although all U6-shRNA expression cassettes were cloned intothe same viral vector, we tested for the possibility that toxicitycorrelated with steady-state levels of the expressed products.

sh30.1

sh8.2

A1 2 8 301

B

C

sh8.

2 m

issh

2.4

mis

sh30

.1 m

issh

8.2

sh2.

4sh

30.1

shLa

cZ

Unt

reat

ed

huntingtin

beta-catenin

GACCGUGUGAAUCAUUGUCUA

CUGGCACACUUAGUAACAGAU

U UC C

C U

A GC U

5‘

3‘-UU

CAGCUUGUCCAGGUUUAUGAA

GUCGAACAGGUCCAAAUACUU

U UC C

C U

A GC U

5‘

3‘-UU

sh2.4

GGAUACCUGAAAUCCUGCUUU

CCUAUGGACUUUAGGACGAAA

U UC C

C U

A GC U

5‘

3‘-UU

Fig. 1. In vitro screening of shRNAs targeting human HD and mouse HDhtranscripts. (A) Thirty-five shRNAs (bars above cartoon) targeting conservedsequences (Table S1) spanning human HD and mouse HDh mRNAs weregenerated with consideration for sequences that promote proper loading ofthe antisense strands into the RISC. Plasmids expressing U6-driven shRNAswere transfected into HEK 293 cells, and HD gene silencing was evaluated byQPCR and protein dot blot analyses 48 h after transfection. (B) Three candidateshRNAs targeting sequences in exons 2 (sh2.4), 8 (sh8.2), and 30 (sh30.1) werechosen for further study (red bars above cartoon in A). (C) shRNA expressionplasmids were transfected into mouse C2C12 cells, and endogenous hunting-tin protein levels were evaluated by Western blot analyses 48 h after trans-fection. Mismatch (mis) controls contain 4-bp changes that render the shRNAsineffective. �-Catenin serves as the loading control.

A

C

B

shRNA TTTT hrGFPU6 CMV polyA

ITR ITR

DA

RP

P-3

2 I

ba1

AAVsh8.2-GFP

0

0.2

0.4

0.6

0.8

1.0

sh2.4 sh8.2 sh30.1

Rel

ativ

e H

Dh

mR

NA

sh8.2 sh30.1sh2.4

Anterior

striatum

globus pallidus

D

striatum striatum striatum

striatum globus pallidus globus pallidus

AAVshRNA-GFP

GFP

GFP

* * *

Fig. 2. HD shRNAs cause sequence-specific striatal toxicity in mice. (A)Diagram of the recombinant AAV2/1 viral vectors containing shRNA andhrGFP expression cassettes. (B) Photomicrographs represent the rostral-to-caudal distribution of hrGFP-positive cells in mouse brain after direct injectionof virus into the striatum. (Scale bar: 500 �m.) (C) QPCR analysis measuringHDh mRNA levels in shRNA-treated mouse striata demonstrates similar silenc-ing efficacies among sh2.4, sh8.2, and sh30.1. Mice were injected into thestriatum with AAVsh2.4-GFP, AAVsh8.2-GFP, AAVsh30.1-GFP, or AAV-GFP,and RNA was harvested 4 months later from GFP-positive striata. All valueswere normalized to �-actin and are shown relative to AAV-GFP-treated brains.(D) Immunohistochemistry reveals that sh2.4 and sh30.1 induce striatal toxicityin mice. Mice were injected with the indicated AAVshRNA-GFP or AAV-GFPinto the striatum, and histological analyses were performed on brains har-vested at 4 months after treatment. Representative photomicrographs forimmunohistochemical staining of DARPP-32-positive neurons (Upper) andIbaI-positive microglia (Lower) are shown for each treatment group. (Scalebar: 500 �m for Upper; 100 �m for Lower.)

McBride et al. PNAS � April 15, 2008 � vol. 105 � no. 15 � 5869

MED

ICA

LSC

IEN

CES

Dow

nloa

ded

by g

uest

on

May

18,

202

0

RNA samples harvested from shRNA-treated striata were an-alyzed by small transcript Northern blot probing for the matureantisense (AS) and sense (S) RNAs generated by the respectiveshRNAs. Results demonstrate that sh2.4 AS RNA and sh30.1 ASRNA are expressed more robustly than sh8.2 AS RNA (Fig. 3),thus correlating toxicity with increased expression levels of theshRNAs in vivo. The disparity in expression levels is interesting,particularly given the fact that each shRNA was designed usingthe same rules, injected at the same viral dose, driven by the samePol-III promoter, and silenced HDh mRNA to a similar degree.Notably, the processed sense strands and unprocessed shRNAtranscripts were not detectable in brain lysates. This findingsuggests that the toxicity is due, in part, to high levels of mature

inhibitory RNAs, rather than inappropriate sense strand loadinginto the RNA-induced silencing complex (RISC) or saturation ofendogenous RNAi export machinery.

Antisense Sequence Levels Are Reduced by Using an Artificial miRNA.Because the toxic shRNAs were expressed at higher levels than thenontoxic, active hairpin, an obvious approach to reduce toxicitywould be to lower the viral titer injected. In the brain, decreasingthe titers of AAVsh2.4-GFP by a half log (1e12) or a full log (5e11)achieved silencing of HDh mRNA (47% and 51%, respectively),but did not alleviate striatal toxicity (Fig. S3). Decreasing the titerseven further (1e11 or 5e10) reduced the silencing efficacy to 15%of controls, an activity level possibly below therapeutic efficacy (Fig.S3). Thus, we tested whether levels of inhibitory RNAs could beminimized without compromising silencing efficacy by using anartificial miRNA as an siRNA shuttle (vs. an shRNA).

In corresponding work, we have found that artificial miRNAseffectively silence target gene expression relative to shRNAs with-out generating excessive levels of inhibitory RNAs (R.L.B. andB.L.D., unpublished data). Consequently, we cloned two of thetoxic sequences (HD2.4 and HD2.4mis) into an artificial miRNAscaffold based on human miR-30 (14), thus creating mi2.4 andmi2.4mis (Fig. 4A). We first compared the expression levels of mi2.4and sh2.4 by small transcript Northern blot analysis at 48 h aftertransfection of RNAi-expressing plasmids into HEK 293 cells.Probing for the HD2.4 antisense strand revealed that mi2.4 pro-duces substantially lower levels of inhibitory RNAs relative to sh2.4.Notably, sh2.4 generates an abundance of precursor and processedRNAs even at a 10-fold lower dose (Fig. 4B). Despite the dramaticdifference in expression levels, mi2.4 reduced endogenous HDtranscripts almost as effectively as sh2.4 (50% and 60% silencing,respectively) in HEK 293 cells (Fig. 4C).

Artificial miRNAs Mitigate Striatal Toxicity in Mice. We next gener-ated AAV2/1-expressing mi2.4 or the mi2.4 mismatch control (Figs.2B and 4A) to test whether the development of striatal toxicitycould be prevented relative to AAVsh2.4-GFP. Because shRNA-induced toxicity was not dependent on the disease model, subse-quent studies were performed in wild-type mice. Mice were injectedinto the right striatum with AAVsh2.4-GFP, AAVmi2.4-GFP, orAAVmi2.4mis-GFP and killed 4 months after injection. The timecourse, volume, and titer were identical to those used in our earliershRNA studies (Fig. 2). QPCR performed on RNA isolated frommouse striata showed a statistically significant reduction of HDhmRNA (�70%) after treatment with either sh2.4- or mi2.4-

sh2.4

sh8.2

sh30.1

shRNA

Sense

Probe for Antisense Strand

A

sh2.4 AS

sh8.2 AS

sh30.1 AS

0

50

100

150

Rel

ativ

e A

S e

xpre

ssio

nB

Antisense

Pos. Control Standard

Fig. 3. The nontoxic sh8.2 generates lower levels of processed antisenseRNA. (A) Small transcript Northern blot was performed to assess AS RNA levelspresent in mouse striata treated with the indicated AAVshRNA-GFP. (Left)Two separately treated striatal tissue samples. (Center and Right) Positivecontrols loaded as standards [10-fold dilutions for both S (Center) or AS (Right)strands]. (B) Densitometry analysis was used to quantify the relative levels ofHD AS RNAs. Signals were quantified by using Image J software, and expres-sion is shown as femtomoles per microgram of total RNA.

GACCGUGUGAAUCAUUGUCUA

CUGGCACACUUAGUAACAGAU

U UC C

C U

A GC U

5‘ +1

3‘-UU

5‘-G...NNN A CAGUG GCG ACCGUGUGAAUCAUUGUCUAACU

3‘-...NNN C AUCAU CGC UGGCACACUUAGUAACAGAUUGG

UAAAG G

C C

G AUAGAC

sh2.4

mi2.4

0

Rel

ativ

e H

D m

RN

A

sh2.4mi2.4

0.4

0.8

0.2

0.6

1.0

1.2

2.4 AS

A B sh2.4mi2.40.2 0.22 2DNA (ug)

U6

Pre-

C

EtBr

+1

Fig. 4. An artificial miRNA approach naturally reduces precursor and mature inhibitory RNAs. (A) Sequences and comparison of sh2.4 and mi2.4 containingthe core HD2.4 sequence (shaded boxes). Each transcript starts with the �1-G nucleotide natural to the U6 promoter. The major Drosha and Dicer cleavage sitesare shown by hash marks. (B) HEK 293 cells were transfected with HD2.4 RNAi expression plasmids at the indicated amounts, and small-transcript Northern blotwas performed 48 h later. Results demonstrate that sh2.4 generates abundant levels of unprocessed precursor (Pre-) and processed antisense RNAs (2.4AS) evenat a 10-fold-lower dose relative to mi2.4. Ethidium bromide (EtdBr) staining is shown as the loading control. (C) HD2.4 RNAi expression plasmids were transfectedinto HEK 293 cells, and QPCR analysis was performed 48 h later to measure endogenous HD mRNA levels. Results demonstrate that mi2.4 silences HD transcriptsefficiently, relative to sh2.4, despite being expressed at considerably lower levels.

5870 � www.pnas.org�cgi�doi�10.1073�pnas.0801775105 McBride et al.

Dow

nloa

ded

by g

uest

on

May

18,

202

0

expressing vectors, compared with uninjected striata or striatatreated with mi2.4mis [F(2, 8) � 77.6, P � 0.001 for post hocanalyses comparing sh2.4 and mi2.4 vs. uninjected and mi2.4mis](Fig. 5A). Importantly, the degree of HDh mRNA silencing be-tween sh2.4 and mi2.4 was similar and not significantly different(P � 0.05). Additional QPCR analyses were performed on thesesamples to measure CD11b mRNA, a readout for microglialactivation, as an initial assessment for toxicity. Striata treated withsh2.4 showed nearly a 4-fold increase of CD11b mRNA relative touninjected striata, whereas mi2.4- and mi2.4mis-treated striatashowed only minimal induction [F(2, 8) � 23.6, P � 0.001 for posthoc analyses comparing sh2.4 to all other groups] (Fig. 5B). Todetermine whether these differences in toxicity could be attributedto levels of HD2.4 inhibitory RNAs, we performed Northern blotanalysis on the same RNA samples used for the QPCR analyses.Although silencing efficacies between the sh2.4- and mi2.4-treatedgroups were comparable, Northern blot analysis, probing for the

HD2.4 antisense strand, demonstrated considerably more matureHD2.4 antisense RNAs in sh2.4-treated mice relative to mi2.4-treated mice (Fig. 5C). These results corroborate our in vitrofindings and correlate the improvement in toxicity with reducedlevels of HD2.4 antisense RNA.

We further assessed striatal toxicity by histological analyses.Immunolabeling for DARPP-32 expression revealed significantattenuation of striatal toxicity in AAVmi2.4-GFP-injected cohortsrelative to AAVsh2.4-GFP-injected mice (Fig. 5D Middle). More-over, the intense microglial activation (Iba1-positive cells) seen inAAVsh2.4-GFP-injected mice was scarcely present in AAVmi2.4-GFP-injected mice (Fig. 5D Bottom and Fig. S4). Of note, mi2.4mis-treated brains also showed no apparent toxicity by these analyses,whereas HD2.4mis was toxic when delivered as an shRNA (data notshown). Thus, sequences encoding HD2.4 and HD2.4mis were toxicin the setting of an shRNA in the brain, but not in the context ofa miRNA scaffold.

DiscussionHere, we show that some shRNAs cause toxicity in mouse striatumindependent of HDh mRNA silencing. Similar to our work, Grimmand colleagues (15) observed acute liver toxicity and mortality inmice after systemic shRNA delivery, which correlated with in-creased mature antisense RNA levels. However, there are impor-tant differences between our findings. First, Grimm et al. found thatlowering the vector dose by �10-fold significantly improved thelethal effects of some shRNAs on liver function and animalviability. In our studies, reducing the dose led to lower transductionthroughout the striatum, but did not abrogate toxicity. Second, thedata by Grimm and colleagues show significant buildup of shRNAprecursors in liver cells. They attributed the liver toxicity, in part, tothe saturation of endogenous RNAi export machinery. In our work,we detected abundant levels of unprocessed shRNAs in vitro, but,interestingly, low to undetectable levels in vivo. This finding suggeststhat export was likely not limiting in our studies. Alternatively, thestriatal toxicity may be caused by the buildup of antisense RNAsand subsequent off-target silencing of unintended mRNAs. Ourdata on sh8.2 also are consistent with this; sh8.2 was not toxic whendelivered at the same dose as sh2.4 and sh30.1. Although silencingactivity was similar among the three shRNAs, levels of matureproduct for sh8.2 were significantly lower.

We found that moving the HD2.4 and HD2.4mis sequences, bothof which caused toxicity in the context of a shRNA, into a miRNAscaffold significantly reduced neurotoxicity within the striatum withno sacrifice in gene-silencing efficacy. We correlated this positiveeffect to lower steady-state levels of mature antisense RNAsprocessed from the artificial mi2.4 relative to sh2.4. Whether thisdisparity in expression levels results from the differences in tran-scription or the stability between shRNAs and artificial miRNAsremains unknown. However, the latter provides a more likelyexplanation because sh2.4 and mi2.4 are expressed from the samemouse U6 promoter and only differ in size by �100 nucleotides.

In addition to improved safety profiles, artificial miRNAs areamenable to Pol-II-mediated transcription. Conversely, shRNAshave limited spacing flexibility for expressing shRNAs fromPol-II-based promoters (16). This advantage of miRNA-basedsystems allows for regulated and cell-specific expression ofinhibitory RNAs. These versatile expression strategies advancethe application of artificial miRNAs as biological tools and mayfurther limit potential toxicity in therapeutic applications.

In some diseases, it is possible to specifically target disease-linked SNPs that exist on the mutant transcript (17, 18). For HD,however, no prevalent SNP has been reported. Because earlierwork showed that a minimum of 50% huntingtin expression isrequired to offset the embryonic lethality noted in huntingtin-null mice (19), knowing the consequences of reducing huntingtinexpression in adult brain is important to moving non-allele-specific RNAi forward as a HD therapy. Our data with sh8.2 and

mi2.4 mis mi2.4 sh2.4

hrG

FP

DA

RP

P-3

2Ib

a1

0Rel

ativ

e H

Dh

mR

NA

sh2.4mi2.4 mi2.4

0.4

0.8

0.2

0.6

1.0

1.2A

NS

mis

2.4 AS

0Rel

ativ

e C

D11

b m

RN

A

2

4

1

3

5B

sh2.4mi2.4 mi2.4mis

C sh2.4mi2.4

mi2.4mis

D

EtBr

Fig. 5. Artificial miRNAs mitigate striatal toxicity in mice. (A and B) QPCRanalyses were performed to measure mouse HDh (A) and CD11b (B) mRNAlevels in AAV-RNAi-injected striata harvested 4 months after treatment (NS,not significant). Samples were normalized to �-actin. Results, shown relativeto uninjected striata, demonstrate that mi2.4 silences HD transcripts as effec-tively as sh2.4, but avoids induction of CD11b, a marker for microglial activa-tion. (C) Small-transcript Northern blot analysis for mature HD2.4 AS RNAspresent in AAV-RNAi-treated striatal lysates reveals a robust disparity be-tween the levels generated from sh2.4 and mi2.4 vectors. EtdBr staining isshown as the loading control. (D) Histological analyses demonstrate theimproved safety profile of mi2.4. Mice were injected with the indicatedAAV-RNAi-GFP viruses into the striatum, and histological analyses were per-formed on brains harvested at 4 months after treatment. Photomicrographsrepresenting hrGFP (Top), immunohistochemical staining of DARPP-32-positive neurons (Middle), and IbaI-positive microglia (Bottom) are shown foreach treatment group. (Scale bar: 500 �m.)

McBride et al. PNAS � April 15, 2008 � vol. 105 � no. 15 � 5871

MED

ICA

LSC

IEN

CES

Dow

nloa

ded

by g

uest

on

May

18,

202

0

mi2.4 are encouraging and suggest that the mammalian brain cantolerate �50% reduction in HD mRNA for 4 months, the lasttime point studied. The long-term safety and efficacy of sh8.2 iscurrently being tested in a study including histochemical, bio-chemical, and behavioral readouts in CAG140 HD mice.

In summary, we show that reducing HDh mRNA levels in adultmammalian brain is tolerated. We also make the important obser-vation that the toxicity of shRNAs after their expression in braincould be alleviated by moving the inhibitory RNA sequences intoan artificial miRNA scaffold. Thus, miRNA-based approaches aremore suitable for achieving RNAi in the brain to address basicresearch questions or develop disease therapies.

Materials and MethodsExpression Vectors and AAV. shRNA expression cassettes were generated by PCRas described (8) and cloned into pCR-Blunt-II TOPO vectors (Invitrogen). Eachcandidate shRNA expression cassette consisted of a mouse U6 promoter, anshRNA[targetinghuntingtin sequences,mismatchcontrol sequences (containingfour base pair changes relative to the respective huntingtin shRNAs), or E. coli�-gal; shLacZ], and an RNA polymerase III termination sequence (six thymidinenucleotides). For artificial miRNAs, siRNA sequences based on HD2.4 or HD2.4miswere embedded into an artificial miRNA scaffold comparable to human miR-30togeneratemi2.4andmi2.4mis (general structureshowninFig.4A). TheartificialmiRNA stem loops were cloned into a mouse U6 expression vector so that �30 nt(5� and 3�) flank the stem loop in the transcribed product.

AAV shuttle plasmids pAAVsh2.4-GFP, pAAVsh2.4mis-GFP, pAAVsh8.2-GFP, pAAVsh30.1-GFP, pAAVmi2.4-GFP, and pAAVmi2.4mis-GFP contain therespective RNAi expression cassettes driven by the mouse U6 promoter. TheAAV shuttles also contained a hrGFP gene under the control of the humancytomegalovirus immediate-early gene enhancer/promoter region, a chimerichuman �-globin eGFP expression cassette followed by the splice donor/humanIg splice acceptor site, and a bovine growth hormone poly (A) signal. Thesetranscriptional units are flanked at each end by AAV serotype 2 145-bpinverted terminal repeat sequences. The transpackaging plasmids,pBSHSPR2C1, were constructed as follows: genomic DNA was extracted fromAAV1 (American Type Culture Collection), and the cap coding sequence wasamplified by PCR using Pfx polymerase (Invitrogen). The AAV2 cap gene wasexcised from the AAV2 helper plasmid pBSHSPRC2.3 and replaced with theamplified AAV1 cap sequence by using a Swa I restriction site in the rep/capintergenic junction and a BsrG I site engineered just upstream of the AAV2poly(A) signal. The resulting transpackaging construct, pBSHSPR2C1, containsthe AAV2 rep gene under the control of a minimal eukaryotic promoter andthe AAV1 cap ORF positioned between the AAV2 rep/cap intergenic junctionand the AAV2 poly(A) signal. The plasmid pAd Helper 4.1 expresses the E2a,E4-orf6, and VA genes of adenovirus type 5 (Ad5) for AAV amplification.

Recombinant AAV vectors were produced by a standard calcium phosphatetransfection method in HEK 293 cells by using the Ad helper, transpackaging, andAAV shuttle plasmids as described (20). Vector titers were determined by real-time PCR and were between 5 and 20 � 1012 DNase-resistant particles per ml.Vector infectivity was assessed in a TCID50 assay by using the HeLa-based B50 cellline (21).

Animals. Allanimalprotocolswereapprovedbythe InstitutionalAnimalCareandUse Committee at the University of Iowa. CAG140 heterozygous knockin mice(10) and wild-type littermates were bred and maintained in the animal vivariumat the University of Iowa. Mice were genotyped and repeat length identified byseparate PCRs using primers flanking the CAG repeat. Mice were housed ingroups of either two or three per cage and in a controlled temperature environ-ment on a 12-h light/dark cycle. Food and water were provided ad libitum.

AAV Injections. CAG140 knockin or wild-type mice were injected with AAVsh-RNAs or AAV-miRNAs (at the indicated titer) at 5 weeks of age and killed at 4months after injection. Procedures were performed as reported previously (8)with the following exceptions. In the initial study, 5-�l injections of eitherAAVsh2.4GFP, AAV30.1sh-GFP, AAVsh8.2-GFP, or AAV-GFP were made bilat-erally into striata (coordinates: 0.86 mm rostral to bregma, �1.8 mm lateral tomidline, 3.5 mm ventral to the skull surface). For the miRNA/shRNA compar-ison study, 5-�l injections of vector were injected unilaterally. Injection rates

for all studies were 0.2 �l/min. Mice used in histological analyses were anes-thetized with a ketamine/xylazine mix and transcardially perfused with 20 mlof 0.9% cold saline, followed by 20 ml of 4% paraformaldehyde in 0.1 M PO4

buffer. Brains were removed and postfixed overnight, and 40-�m thick sec-tions were collected. Mice used for molecular analyses were perfused with 20ml of 0.9% cold saline, and brain was removed and blocked into 1-mm-thickcoronal slices. Tissue punches were taken by using a tissue corer (1.4 mm indiameter). All tissue punches were flash frozen in liquid nitrogen and storedat 80°C until used.

Molecular Studies. For in vitro shRNA screening, shRNA expression plasmids weretransfected (Lipofectamine 2000; Invitrogen) into human HEK 293 cells or mouseC2C12 cells, which naturally express full-length human or mouse huntingtin,respectively. Huntingtin levels were assessed by protein dot blot (anti-huntingtinprimary antibody MAB2166, 1:5,000; Chemicon) or Western blot (protein loadingcontrol, anti-�-catenin, 1:4,000; AbCam). Knockdown also was assessed by QPCRusing a human huntingtin-specific TaqMan primer/probe set with normalizationto a human GAPDH primer/probe set. This QPCR strategy also was used toevaluate HD knockdown mediated by sh2.4 and mi2.4 in Fig. 4B.

For in vivo QPCR analyses, tissue was dissected from GFP-positive striatum, andrelative gene expression was assessed by using TaqMan primer/probe sets formouse HDh, CD11b, and �-actin. All values were quantified by using the CT

method (normalizing to �-actin) and calibrated to either AAV-GFP-injected stri-ata (screening study) or uninjected striata (miRNA-shRNA comparison study).

For Northern blot analyses, tissue was dissected from GFP-positive striatum.RNA was harvested by TRIzol reagent and RNA (1–5 �g and 15 �g for in vivo andin vitro studies, respectively) was resolved on 15% polyacrylamide/urea gels, andRNA was visualized by ethidium bromide staining and UV exposure to assessloading and RNA quality. Samples were then transferred to Hybond-N�/XLmembranes (Amersham Pharmacia) and UV cross-linked. Blots were probed with32P-labeled oligonucleotides at 30–36°C overnight, washed in 2� SSC at 30–36°C,and exposed to film.

For in vivo Western blot analysis, tissue was dissected from GFP-positivestriatum and lysed in 150 �l of lysis buffer, and protein level was quantified withthe DC protein assay (Bio-Rad). Then 10 �g of total protein was separated on 8%SDS polyacrylamide gel before transferring to a 0.45-�m PVDF membrane. Themembrane was blocked with 2% milk in PBS-Tween 20 (0.05%) and incubatedwith either an anti-huntingtin antibody (1:5,000; Chemicon) or an anti-�-actinantibody (1:10,000; Sigma), followedbyaconjugatedgoatanti-mousesecondaryantibody (1:10,000; Jackson ImmunoResearch) and an ECL-Plus substrate (Amer-sham Biosciences), and then exposed to film.

Immunohistochemical Analyses. Briefly, 40-�m-thick, free-floating coronal brainsections were processed for immunohistochemical visualization of striatal neu-rons (DARPP-32, 1:100; Cell Signaling Technology) and microglia (Iba1, 1:1,000;WAKO) by using the biotin-labeled antibody procedure. Primary antibody incu-bations were carried out for 24 h at room temperature. Sections were incubatedin goat anti-rabbit biotinylated IgG secondary antibodies (1:200; Vector Labora-tories) for 1 h at room temperature. In all staining procedures, deletion of theprimary antibody served as a control. Sections were mounted onto SuperfrostPlus slides and coverslipped with Gelmount (Biomeda). Images were captured byusing an Olympus BX60 light microscope and DP70 digital camera, along with anOlympus DP Controller software.

Statistical Analyses. All statistical analyses were performed by using SigmaStatstatistical software (SYSTAT).QPCRanalyses forhuntingtinandCD11bexpressionwere performed by using a one-way ANOVA, as was Northern blot densitometryanalysis. Upon a significant effect, Bonferroni post hoc analyses were performedto assess for significant differences between individual groups. Western blotdensitometry analysis was performed by using a two-tailed Student’s t test. In allcases, P � 0.05 was considered significant.

Figure Preparation. All photographs were formatted with Adobe Photoshopsoftware, all graphs were made with Prism Graph software, and all figures wereconstructed with Adobe Illustrator software.

ACKNOWLEDGMENTS. We thank the B.L.D. and McCray laboratories for feed-back and discussion. This work was supported by National Institutes of HealthGrants NS-50210, HD-44093, DK-54759, and NS-592372; the Hereditary DiseaseFoundation; and the Roy J. Carver Trust.

1. Fire A, et al. (1998) Potent and specific genetic interference by double-stranded RNAin caenorhabditis elegans. Nature 391:806–811.

2. Reinhart BJ, et al. (2000) The 21-nucleotide let-7 RNA regulates developmental timingin Caenorhabditis elegans. Nature 403:901–906.

3. Lagos-Quintana M, Rauhut R, Lendeckel W, Tuschl T (2001) Identification of novelgenes coding for small expressed RNAs. Science 294:853–858.

4. Lau NC, Lim LP, Weinstein EG, Bartel DP (2001) An abundant class of tiny RNAs withprobable regulatory roles in Caenorhabditis elegans. Science 294:858–862.

5872 � www.pnas.org�cgi�doi�10.1073�pnas.0801775105 McBride et al.

Dow

nloa

ded

by g

uest

on

May

18,

202

0

5. Lee RC, Ambros V (2001) An extensive class of small RNAs in Caenorhabditis elegans.Science 294:862–864.

6. Ratcliff FG, MacFarlane SA, Baulcombe DC (1999) Gene silencing without DNA. rna-mediated cross-protection between viruses. Plant Cell 11:1207–1216.

7. Ketting RF, Haverkamp TH, van Luenen HG, Plasterk RH (1999) Mut-7 of C. elegans,required for transposon silencing and RNA interference, is a homolog of Wernersyndrome helicase and RNaseD. Cell 99:133–141.

8. Harper SQ, et al. (2005) RNA interference improves motor and neuropathologicalabnormalities in a Huntington’s disease mouse model. Proc Natl Acad Sci USA102:5820–5825.

9. Rodriguez-Lebron E, et al. (2005) Intrastriatal rAAV-mediated delivery of anti-huntingtin shRNAs induces partial reversal of disease progression in R6/1 Huntington’sdisease transgenic mice. Mol Ther 12:618–633.

10. Menalled LB, et al. (2003) Time course of early motor and neuropathological anomaliesin a knock-in mouse model of Huntington’s disease with 140 CAG repeats. J CompNeurol 465:11–26.

11. Khvorova A, Reynolds A, Jayasena SD (2003) Functional siRNAs and miRNAs exhibitstrand bias. Cell 115:209–216.

12. Reynolds A, et al. (2004) Rational siRNA design for RNA interference. Nat Biotechnol22:326–330.

13. Mittal V (2004) Improving the efficiency of RNA interference in mammals. Nat RevGenet 5:355–365.

14. Zeng Y, Wagner EJ, Cullen BR (2002) Both natural and designed micro RNAs can inhibitthe expression of cognate mRNAs when expressed in human cells. Mol Cell 9:1327–1333.

15. Grimm D, et al. (2006) Fatality in mice due to oversaturation of cellular microRNA/shorthairpin RNA pathways. Nature 441:537–541.

16. Xia H, Mao Q, Paulson HL, Davidson BL (2002) siRNA-mediated gene silencing in vitroand in vivo. Nat Biotechnol 20:1006–1010.

17. Miller VM, et al. (2003) Allele-specific silencing of dominant disease genes. Proc NatlAcad Sci USA 100:7195–7200.

18. Gonzalez-Alegre P, Bode N, Davidson B, Paulson HL (2005) Silencing primary dystonia:Lentiviral-mediated RNA interference therapy for DYT1 dystonia. J Neurosci 25:10502–10509.

19. White JK, et al. (1997) Huntingtin is required for neurogenesis and is not impaired bythe Huntington’s disease CAG expansion. Nat Genet 17:404–410.

20. Sandalon Z, et al. (2004) Secretion of a TNFR:Fc fusion protein following pulmonaryadministration of pseudotyped adeno-associated virus vectors. J Virol 78:12355–12365.

21. Gao GP, et al. (1998) High-titer adeno-associated viral vectors from a Rep/Cap cell lineand hybrid shuttle virus. Hum Gene Ther 9:2353–2362.

McBride et al. PNAS � April 15, 2008 � vol. 105 � no. 15 � 5873

MED

ICA

LSC

IEN

CES

Dow

nloa

ded

by g

uest

on

May

18,

202

0