Helper-dependent Adenovirus-mediated Short Hairpin RNA Expression in the Liver Activates the Interferon Response*

The use of RNA interference has proven to be an effective means to study the function of genes. Constitutive synthesis of small interfering RNA molecules can be accomplished with the use of viral vectors expressing short hairpin RNA (shRNA). Binding of shRNA to the target mRNA promotes transcript degradation. So far, little is known about the effects that shRNA induce in vivo. To determine the feasibility of using helper-dependent adenoviral vectors for expression of shRNA in liver, we have designed an shRNA construct to mouse fabp5 (fatty acid-binding protein 5). Intravenous administration of this vector resulted in ∼75% silencing of fabp5. Increasing the dose of vector did not result in higher levels of silencing, indicating that there is a threshold for the level of knockdown that can be achieved. Synthesis of high levels of shRNA molecules did not alter the levels of cellular micro-RNA, such as miR-122 and let-7a, suggesting that the exportin-5 pathway was not affected. However, high level shRNA expression resulted in activation of the interferon response. Thus, an important consideration when using shRNA-based vectors in vivo is to closely monitor signs of interferon-stimulated gene expression, since a narrow window exists between gene silencing efficacy and nonspecific effects.

RNA interference (RNAi) 2 is a highly effective mechanism to regulate gene expression by post-transcriptional silencing. RNAi is mediated by short interfering RNA (siRNA) that binds to mRNA targets and inhibits translation or induces RNA degradation. siRNAs have emerged as a new tool to study gene function and identify gene targets. Conventional technologies to validate the role of genes in disease involve the development of animal models with null alleles for the target gene. In some instances, gene compensatory effects or embryonic death that results from generating a null animal impairs studying the role of the gene of interest. Consequently, siRNAs have become a powerful alternative approach to study the effects of gene silencing, because it allows reducing gene expression in adult animals, thereby bypassing problems originating from gene expression abrogation at the embryo stage.
The liver is a critical tissue controlling glucose and lipid homeostasis and is a target for functional studies directed at understanding the molecular basis of prevalent human conditions, such as the metabolic syndrome, diabetes, and cardiovascular disease. Compared with other viral and nonviral systems, adenoviruses are the most efficient vectors for delivery of genes to the liver because of their high tropism for hepatocytes (1). Down-regulation of gene expression in the liver of mice has been achieved by adenovirus-mediated delivery of short hairpin RNA (shRNA) expression cassettes (2)(3)(4)(5)(6). These reports demonstrate that double-stranded RNA (dsRNA) molecules can be produced in vivo by gene transfer approaches. However, in these studies, constructs were delivered by an E1-deleted or first generation adenoviral vector. One problem inherent to this type of vector is that even in the absence of E1 functions, viral proteins can be expressed, resulting in the development of hepatotoxic effects at high doses (7)(8)(9)(10)(11). Additionally, E1-deleted adenovirus vectors express VA1 noncoding RNA, shown to inhibit micro-RNA (miRNA) processing by interfering with both nuclear export and Dicer functionality (12). These problems have been overcome by the use of helper-dependent adenoviral vectors, which are devoid of viral coding sequences. The lack of viral genes eliminates the possibility of leaky production of viral proteins in the host, reducing the chances for toxic effects (11,(13)(14)(15)(16)(17)(18)(19)(20). Although this type of vector elicits innate immune responses to an extent similar to that observed with E1-deleted adenovirus vectors, the response is transient and lasts less than 7 days (21). Thus, helper-dependent adenoviral vectors possess the desired characteristics to deliver shRNA-expressing constructs to the majority of hepatic cells, resulting in liver-specific gene silencing. This approach has the potential to circumvent generating time-consuming conditional knock-out mice in gene function studies.
In mammalian cells, introduction of long double-stranded RNA (dsRNA) molecules induces activation of antiviral mechanisms that eventually lead to a general shutdown of protein synthesis. Initial studies using small interfering RNA intermediates of 21-23 nucleotides were found to bypass these limitations (22) and have been used by multiple investigators to study the effects of inhibiting gene expression (2,(22)(23)(24)(25)(26)(27)(28)(29)(30). As a result of the growing number of research applications using RNAi, it is becoming increasingly clear that siRNA molecules have the capacity to induce nonspecific and off-target effects in addition to inducing gene-specific silencing. Recent studies have shown that expression of shRNA in cultured cells can activate antiviral mechanisms that recognize the presence of dsRNA and result in type I interferon (i.e. IFN-␣ and IFN-␤) production (31,32). IFNs are part of the innate immune response against viral infections and up-regulate expression of multiple genes that contain interferon-stimulated response elements in their promoters, resulting in early blocking of viral transcription/translation and activation of apoptotic pathways (33).
To determine the impact of shRNA expression in liver using helper-dependent adenoviral vectors and the feasibility to use this approach for short term gene function studies, we have developed a vector expressing an shRNA from the U6 promoter to target fabp5 (fatty acid-binding protein 5). Fatty acid-binding proteins are 14 -16-kDa cytosolic proteins that bind long chain fatty acids with high affinity and play a crucial role in shuttling them to cellular compartments. Here we show 75% gene silencing using a helper-dependent adenoviral vector expressing shRNA, thereby providing an excellent tool for silencing gene expression in vivo and for carrying out gene function studies.
Helper-dependent Adenoviral Vector Production-Helperdependent adenoviral vectors were generated using an improved Cre-loxP system generated by Merck and Microbix (Toronto, Canada) (35)(36)(37). The system consists of plasmid pC4HSU, in which expression cassettes are cloned; Cre recombinase-expressing 293Cre4 cells; and an E1-deleted adenovirus containing loxP sites flanking the packaging signal (H14) that is used as helper virus (37). Plasmid pRES was created by cloning a 3.2-kb HindIII fragment from pC4HSU into pBluescript II-SK (Stratagene, La Jolla, CA). The shRNA expression cassette from pFABP5-242 and pSCR was amplified by PCR using primers containing AscI restriction sites (5Ј-GAACCTGGCGCGCCG-ACTGGATCCGGTACC-3Ј, 5Ј-GAACCTGGCGCGCCTAC-AAGAAAGCTGGGTC-3Ј) and cloned into the AscI site in pRES. The resulting plasmids pRES/FABP5-242 and pRES/ SCR were digested with HindIII. Simultaneously, plasmid pC4HSU (37) was digested with AscI. This plasmid and each of the HindIII fragments were co-transformed into Escherichia coli BJ5183 cells, as described previously (38). Helper-dependent adenovirus plasmids pC4HSU/FABP5-242 and pC4HSU/ SCR, containing shRNA expression cassettes, were generated through recombination between homologous regions in the HindIII fragments and pC4HSU.
To monitor the helper-dependent vector amplification process, a pC4HSU construct containing a cytomegalovirus-green fluorescent protein expression cassette (pC4HSU-GFP) was used. Approximately 5 g of DNA were transfected into 293Cre4 cells, and 24 h later, helper H14 was added at a multiplicity of infection of 3. Cells were harvested upon the presence of Ͼ90% cytopathic effect. To increase the titer of the helper-dependent adenoviral vectors, four passages were carried out in 6-cm dishes (5 ϫ 10 6 cells/dish), as described (36), followed by two passages in 15-cm dishes. Large scale preparations were obtained from cells collected from 16 triple flasks (500 cm 2 , 5-7 ϫ 10 7 cells/flask). The virus was purified by one CsCl step gradient centrifugation followed by one CsCl isopycnic separation. The helper-dependent adenovirus band was collected and dialyzed in TMN buffer (10 mM Tris-HCl (pH 7.5), 1 mM MgCl 2 , 150 mM NaCl, 10% glycerol). The level of contamination with helper was determined by a plaque assay and shown to be lower than 0.03%. Preparations were tested for endotoxin and sterility before injection into animals.
Animals-Male 8-week-old C57BL/6J mice were obtained from the Jackson Laboratory (Bar Harbor, ME). Animal care guidelines set forth by the Indiana University School of Medicine were followed. Mice were kept in a BL2 facility and had free access to standard chow and water at all times. Mice were allowed to acclimate to the facility for 1 week prior to receiving helper-dependent adenovirus vectors. Injections were given in a volume of 400 l. All animals were sacrificed 1 week postinjection. Tissues were collected and snap-frozen in liquid nitrogen or fixed in 10% buffered formalin for histology analysis. Serum was obtained and frozen at Ϫ20°C.
Southern Blotting-A probe was made by digesting pC4HSU with SfoI and gel-purifying the resulting 1.8-kb fragment. As DNA loading control, a 0.75-kb probe to the homeobox-containing transcription factor meis4 gene was used. Probes were labeled using the Gene Images Alk-Phos direct labeling kit (GE Healthcare), following the manufacturer's instructions. Genomic DNA was isolated from ϳ100 mg of mouse liver using genomic DNA 500 tips (Qiagen, Valencia, CA). Fifteen g of DNA was digested for 16 h with BamHI and run on a 0.8% agarose gel. DNA was transferred to Hybond N ϩ nylon membranes (GE Healthcare) by capillary transfer and fixed with UV light (Stratagene, La Jolla, CA). Membranes were incubated with 10 ng/ml of probe at 55°C overnight. Membranes were washed and then developed using the Gene Images ECF detection kit (GE Healthcare). Images were obtained using a Storm 830 imager (GMI, Ramsey, MN).
Histology-Four-m-thick sections were cut from routinely processed paraffin-embedded tissue and stained with hematoxylin and eosin for histological analysis. Processing and staining were performed at the Immunohistochemistry Core at Indiana University Medical Center (Indianapolis, IN).
Serum Alanine Aminotransferase-Serum alanine aminotransferase was measured in 25-l samples using a microtiter plate assay (Pointe Scientific, Canton, MI).
Statistics-Numerical data is expressed as means Ϯ S.D. Significant differences are defined as p Ͻ 0.05 by two-tailed, unpaired t test.

RESULTS
In Vitro Analysis of Constructs Expressing shRNA to Silence fabp5-To silence fabp5 in the liver of mice, four 21-bp shRNA constructs under the control of the U6 promoter were designed shRNA Expression in Liver using the Invitrogen Block-it shRNA design algorithm. Optimally, a mouse hepatoma cell line should be used to test the efficacy of each construct at gene silencing, given that our goal is to deliver shRNA-expressing helper-dependent adenoviral vectors to the liver. However, the transfection efficacy of hepatoma cell lines is low (ϳ25% in the mouse cell line Hepa1c1c, based on green fluorescence protein expression). 3 This impairs detection of gene silencing effects, which can only be accurately determined when a majority of cells express the shRNA. Thus, we used a nonhepatic cell line to determine the efficiency of the shRNA constructs at silencing fabp5. Each shRNA construct was tested by co-transfecting HEK293 cells with a plasmid expressing FABP5. As a control, a construct with a scrambled sequence, pSCR, was used. One construct, pFABP5-242, resulted in ϳ90% fabp5 gene silencing, whereas two others reduced expression to ϳ50% (Fig. 1A). The pFABP5-242 and pSCR constructs were subsequently cloned into the helper-dependent adenovirus shuttle plasmid, and viruses were produced as described under "Experimental Procedures." The adenovirus construct is shown in Fig. 1B. Viral DNA from the final preparation was isolated and digested with BamHI to confirm the correct structure (Fig. 1C).

Robust Silencing of fabp5 in the Livers of C57BL/6J Mice Using
Helper-dependent Adenovirus-Hepatic fabp5 silencing in C57BL/6J mice was accomplished by tail vein injection of 2 ϫ 10 11 , 0.6 ϫ 10 11 , or 0.2 ϫ 10 11 viral particles (vp) of gAd.sh242 (expressing an shRNA to silence fabp5) or gAd.shSCR (expressing a scrambled sequence). A control group received an equal injection volume of vehicle. One-week postinjection animals were sacrificed, and the level of down-regulation was evaluated by Western blot analysis of liver protein extracts. High level gene silencing (ϳ75%) was observed in mice that received 2 ϫ 10 11 vp of gAd.sh242 (Fig. 2). Expression levels of the liver proteins microsomal triglyceride transfer protein, cyclophilin, and FABP1 (another member of the fatty acid-binding protein family), were not affected (Fig. 2), indicating that the knockdown effect was specific to the target gene. The control virus gAd. shSCR did not affect expression of any of the examined proteins (Fig. 2).
High Doses of gAd.sh242 Lead to Inflammatory Infiltration and Toxicity in the Liver-In all animals that received the 2 ϫ 10 11 vp dose of gAd.sh242, histological examination of liver sections showed the presence of marked nonzonal inflammatory infiltrate within hepatic lobules consisting predominantly of lymphocytes; neutrophils were present in rare foci. The inflammatory infiltrate surrounded necrotic hepatocytes. In addition, apoptotic hepatocytes without associated inflammation were also seen. In some animals, there was a mild lymphocytic inflammatory infiltrate within portal tracts accompanied by endothelitis of portal and central venules (Fig. 3A). Lower doses of gAd.sh242 (0.6 ϫ 10 11 and 0.2 ϫ 10 11 vp) and all doses  shRNA Expression in Liver of gAd.shSCR resulted in scattered foci of inflammatory cells surrounding necrotic hepatocytes (Fig. 3A). In agreement with the histology observations, serum levels of alanine aminotransferase, a marker of liver function, was significantly increased in the animals that received the 2 ϫ 10 11 vp dose of gAd.sh242 (153 Ϯ 27 IU/liter compared with 53 Ϯ 3 IU/liter in vehicletreated mice and 75 Ϯ 13.5 IU/liter in the 2 ϫ 10 11 vp gAd. shSCR group; n ϭ 5/group). No alterations in alanine aminotransferase levels were seen in mice that received lower doses of gAd.sh242 or gAd.shSCR. Other organs examined histologically, including heart, lung, kidney, and duodenum, did not appear abnormal at any dose (data not shown). Altogether, our data suggest that high level shRNA synthesis may induce toxic effects in the liver.
Optimization of Helper-dependent Adenoviral Vector Dose Is Necessary to Achieve Maximum Level of fabp5 Silencing While Limiting Nonspecific Effects-Given that the second highest dose of the helper-dependent adenoviral vector (0.6 ϫ 10 11 vp) resulted in normal liver morphology but decreased FABP5 expression to only 60% of normal levels (Fig. 2), we hypothesized that an optimal level of shRNA expression exists that more effectively silences fabp5. Therefore, we performed a sec-ond experiment in which an intermediate dose, 1 ϫ 10 11 vp, was included to further refine the amount of gAd.sh242 vector that would result in maximal fabp5 silencing in the absence of toxicity. Mice received 2 ϫ 10 11 , 1 ϫ 10 11 , or 0.6 ϫ 10 11 vp of either gAd.sh242 or gAd.shSCR and were sacrificed after 1 week. Interestingly, the 1 ϫ 10 11 vp dose induced levels of fabp5 silencing similar to the 2 ϫ 10 11 vp dose of gAd.sh242 (Fig. 4). Furthermore, we did not observe the inflammation shown in liver sections of mice that received 2 ϫ 10 11 vp (Fig. 3B). Serum alanine aminotransferase levels in the 1 ϫ 10 11 vp gAd.sh242 mice were indistinguishable from control, vehicle-treated mice (39 Ϯ 3 IU/liter compared with 41 Ϯ 11 IU/liter, respectively; n ϭ 3/group). This suggests that 1 ϫ 10 11 vp was sufficient to maximally reduce expression. Increasing the dose of adenoviral vector did not lead to increased fabp5 silencing and only resulted in toxic effects. Additionally, this result indicates that it is unlikely that silencing of fabp5 per se causes liver inflammation or toxicity.
gAd.sh242 and gAd.shSCR Vectors Have Similar Infectious Titers-It was interesting that the hepatotoxicity observed with gAd.sh242 was not induced by gAd.shSCR, which also expresses an shRNA. To eliminate the possibility of a discrepancy in infectious rates between the two vectors, we examined the levels of viral DNA in the liver. As shown in Fig. 5A, similar levels of vector DNA were present in the livers of mice treated with gAd.sh242 or gAd.shSCR, which indicates that the infectious titers of the two vectors were not significantly different. Approximately 13, 10, and 7 copies of vector DNA/diploid genome were present in the livers of the animals that received 2 ϫ 10 11 , 1 ϫ 10 11 , or 0.6 ϫ 10 11 vp, respectively (data not shown). To determine whether both vectors expressed equal  shRNA Expression in Liver levels of shRNA molecules, we carried out Northern analysis using an antisense primer as a standard (Fig. 5B). Based on this assay, ϳ0.05-0.2 pmol are expressed from the gAd.sh242 vector, whereas the gAd.shSCR vector resulted in 0.05-0.1 pmol (Fig. 5B). Thus, although similar copy numbers of vector were present in the liver, it is possible that the lack of toxicity with the gAd.shSCR is the result of lower levels of shRNA expression. Both vectors are identical, except for the orientation of the expression cassette in the vector genome and sequence of the shRNA expressed. The reason for the different levels of expression between the two vectors is unknown at this point. Previous studies using helper-dependent adenoviral vectors have shown no effect of the orientation of the expression cassette on the level of transgene expression (39).
Expression of shRNA Using Gutless Adenoviral Vectors Does Not Influence Levels of Cellular Micro-RNAs-Cellular toxicity in mouse liver with shRNA-expressing adeno-associated viruses has been recently reported (40). It was found that saturation of the exportin-5 pathway, which shuttles cellular miRNA from the nucleus to the cytoplasm (41,42), was the limiting factor (40). To rule out the possibility that saturation of exportin-5 induced alterations to levels of cellular miRNA, we examined the levels of miR-122, a highly expressed liver miRNA (43), in the livers of mice receiving 2 ϫ 10 11 , 1 ϫ 10 11 , or 0.6 ϫ 10 11 vp of gAd.sh242 or gAd.shSCR. Although there was some variation of miR-122 levels between samples, there was no reduction in the group that received the 2 ϫ 10 11 vp dose of gAd.sh242 compared with vehicle-treated mice (Fig. 6A). To confirm these data, miR-122 levels were also measured in livers of mice from our first in vivo experiment (supplemental Fig. S1). No decrease of miR-122 levels was detected in the high dose gAd.sh242 group. Similar data were observed for another less abundant micro-RNA, let-7a (Fig. 6, A and B). This result indicates that the miRNA processing pathway is functional and is not disturbed by expression of shRNA using high doses of helper-dependent adenoviral vector. Furthermore, we estimated that the approximate level of miR-122 expression (a highly abundant micro-RNA in liver) is 0.2-0.4 pmol in mice that received the 2 ϫ 10 11 -vp dose of gAd.sh242 (Fig. 6C), which is above the level of shRNA expression in these animals (0.05-0.2 pmol) (Fig. 6A). Thus, at the highest dose of gAd.sh242 vector, the level of shRNA expressed does not seem to have reached levels above those of endogenous micro-RNA. To further confirm that cellular micro-RNA activity is not influenced by the shRNA expressed from the adenoviral vectors, we measured expression of genes controlled by miR-122. The mRNA levels of aldoA and collagen p4ha1 are regulated by miR-122 (44, 45). We did not observe any correlation between expression of these genes and vector dose with neither gAd.sh242 nor gAd.shSCR (Fig. 6D), although there was a tendency for all gAd.sh242treated animals to express p4ha1 at levels higher than vehicletreated mice.
High Level shRNA Expression Stimulates the Cellular Antiviral Defense Pathway-A potential mediator of cellular toxicity by shRNA is stimulation of the interferon signaling pathway upon production of dsRNA. To test this possibility, we examined the expression of three known interferon-stimulated genes, oas1b (31,46), isg56 (47), and isg20 (32), by real time reverse transcription-PCR. Mouse livers receiving the 2 ϫ 10 11 -vp dose of gAd.sh242 had severalfold increases of each of the genes tested, compared with vehicle-treated animals (Fig.  7). All other doses and the control viruses displayed negligible differences compared with vehicle-treated mice. This suggests that the inflammation and toxicity observed with the 2 ϫ 10 11 dose of gAd.sh242 is mediated by induction of the interferon pathway.

DISCUSSION
Expression of shRNA to knock down a gene has become a valuable alternative approach to study gene function. It has the advantage of being a much faster and cost-effective method than generating conditional knock-out mice. Furthermore, multiple mouse strains can be studied, thereby eliminating the need for tedious and expensive back-crossing experiments. The liver is a main target for gene function studies directed at deciphering the molecular mechanisms of common human conditions, such as metabolic syndrome and cardiovascular disease. Use of shRNA-expressing viral vectors has the potential to be used as a therapy for treatment of human diseases (48). Currently, there is little information on the impact of expressing shRNA in vivo. Here we have shown that it is possible to induce 75% gene silencing in the liver using helper-dependent adenoviral vectors. A dose of 1 ϫ 10 11 vp was found to be sufficient to maximally reduce expression of our target protein, FABP5 (Fig.  4), without triggering nonspecific effects, as shown by histology analyses and measurement of the liver function marker alanine aminotransferase. In addition, this dose induced a level of gene silencing similar to that observed in animals that received 2 ϫ 10 11 vp, which suggests there is a limit to the level of silencing that can be achieved in liver. To our knowledge, this is the first study to establish such a threshold effect in vivo.
A recent publication reported the use of adeno-associated virus type 8 to deliver shRNA constructs to the liver. Stem  Fig. 3B. DNA was isolated from liver, and 15 g were digested with BamHI and transferred to a nylon membrane. A 1.8-kb fragment from pC4HSU was used as probe (see "Experimental Procedures"). As internal control, a probe detecting a 3.5-kb fragment of the meis4 gene was used. Veh, vehicle-treated animals. B, small RNA were isolated from livers of mice that received 2 ϫ 10 11 vp of gAd.sh242 or gAd.shSCR vector. A primer with the antisense sequence of the shRNA was used to generate a standard of 0.05-0.2 pmol, and the intensity of bands in the samples was compared with determined approximate levels of shRNA expressed from each vector. Each lane represents an individual mouse sample.
lengths of 21 bp resulted in severe toxicity for multiple shRNA constructs (40). It should be noted that a 21-bp stem length was also used in our study. The authors hypothesized that high levels of vector-produced shRNAs were capable of overloading the endogenous miRNA processing pathway, probably at the level of exportin-5 (40). This protein mediates the transport of unprocessed miRNA and shRNA out of the nucleus (41,42). Since we did not observe a reduction in miR-122 and let-7a levels (Fig. 6A), overloading of exportin-5 does not appear to be a factor in the toxicity observed with the helper-dependent adenoviral vector system. We did not observe alterations in the mRNA levels of two genes known to be regulated by miR-122 (Fig. 6C), supporting the observations made on miR-122 levels. An important difference between the helper-dependent adenovirus vector used in our study and the adeno-associated virus vector used by Grimm and colleagues (40) is the fact that the DNA of the latter virus forms high molecular weight concatemers within a single cell (49), potentially resulting in extremely high levels of shRNA expression. Consequently, the amounts of shRNA produced by an adeno-associated virus vector may have overloaded the exportin-5 pathway.
Our data suggest that induction of the interferon response is the cause of the cell death and inflammation observed at the highest dose of the gAd.sh242 vector. Interferons induce expression of genes that mediate their effects and activate apoptotic pathways (33). OAS1 induces degradation of RNA through activation of RNAse L (50), whereas ISG20 is a 3Ј-5Ј RNA exonuclease (51). Increased activity of either protein results in RNA cleavage and inhibition of protein synthesis. Despite early reports suggesting that siRNA designed to be less than 30 bp can circumvent cellular mechanisms that recognize long dsRNA (22,52), recent data clearly indicate that siRNA and shRNA as small as 21 bp can activate dsRNA-dependent protein kinase and the signaling cascade that results in up-regulation of IFN-stimulated genes (53). Although some controversy exists about whether or not siRNA/shRNA activate the exact same pathways induced by long dsRNA, it is clear that some downstream effectors/genes are shared (32). There is evi- dence that in addition to dsRNA-dependent protein kinase, Toll-like receptors and the cytoplasmic DEXD/H box RNA helicase RIG-I are the signaling molecules that recognize siRNA/shRNA and activate the interferon response (54).
One possible mediator of the hepatic interferon response is the Kupffer cell, which can be transduced by adenovirus vectors. Synthesis of shRNA in these cells may have activated the interferon response, with subsequent secretion of IFN up-regulating interferon-stimulated gene expression in hepatocytes. A second possibility is that the hepatocytes themselves synthesize type I interferon (55) when excess shRNA molecules accumulate, as may be the case of the animals that received the highest dose of gAd.sh242 vector. Future studies need to be conducted to address these possibilities.
Lentivirus-mediated expression of shRNA in human lung fibroblasts using the U6 promoter induced an interferon response in 6 of 23 constructs, and the magnitude of OAS1 induction was dependent on the multiplicity of infection; no induction was observed at a multiplicity of infection of 1, whereas at a multiplicity of infection of 3, oas1 expression was increased severalfold (31,56), suggesting that the level of shRNA expressed determines the onset of an interferon response. Our data suggest that in vivo this is also the case, since only the highest dose of gAd.sh242 vector resulted in activation of an interferon response. It is interesting that we observed such response with the gAd.sh242 but not with the gAd.shSCR vector. Our data indicate that similar levels of vector DNA were present in the liver (Fig. 5A). However, the gAd.sh242 vector expressed slightly higher levels of shRNA compared with the gAd.shSCR vector, since we observed more shRNA molecules in the livers of animals given the gAd.sh242 adenovirus (Fig.  5B). Thus, it is possible that if identical amounts of shRNA would be expressed, the gAd.shSCR vector might also induce the toxicity seen with gAd.sh242.
Another possibility is that activation of the interferon response is sequence-specific. A number of specific siRNA motifs causing cellular toxicity have been characterized. For example, activation of TLR7 (Toll-like receptor 7) by GU-rich, single-stranded RNA is known to stimulate interferon production (57). Cells expressing shRNA could feasibly produce the single-stranded RNA required for TLR7 activation. TLR7 recognition of siRNA in dendritic cells appears to involve a GUC-CUUCAA motif (58). In another example of toxic motifs, the presence of UGGC in the sense strand siRNA was found to cause cell death in HeLa cells (59). Although neither of the shRNA sequences in our study contains the toxic motifs described above, the gAd.sh242 construct may contain uncharacterized motifs that induce toxicity. Finally, it is also possible that the shRNA sequence in the genome of the gAd.sh242 vector is recognized by a cellular DNA sensor, such as DAI (DNA-dependent activator of IFNregulatory factors). DAI binds cytosolic DNA and stimulates production of type I interferon (60,61).
In summary, we have shown that silencing of fabp5 in the livers of C57BL/6J mice using a helper-dependent adenovirus vector is a feasible approach. Although shRNA-mediated toxicity was observed, lowering the amount of viral particles per animal eliminated the inflammatory reaction and the interferon response while maintaining the same level of fabp5 silencing. Thus, an important consideration when using shRNA-based vectors in vivo is to closely monitor signs of interferon-stimulated gene expression, since a narrow window exists between gene silencing efficacy and nonspecific effects. Control of shRNA expression could be adjusted via promoter strength, viral dose, or using inducible systems. Vectors that incorporate these additional elements will become better tools for gene function studies.