Argonaute proteins are key determinants of RNAi efficacy, toxicity, and persistence in the adult mouse liver (original) (raw)
Here, our aim was to further unravel the cellular determinants of in vivo RNAi efficacy, toxicity, and persistence, sparked by previous findings by us and others that high-level shRNA expression can cause cytotoxicities and fatalities in animals. As a whole, our new data verify and expand our prior conclusion that adverse in vivo shRNA effects are highly complex and at least partly due to saturation of cellular rate-limiting components (3). Still, we do not rule out additional previously indicated explanations that may also be consistent with our phenotypical observations and that require further consideration in the design of RNAi therapies, including innate immune responses and off-targeting (2, 38). Also, since our model was derived in the liver, future experiments must be performed to comprehensively determine the concentrations of key RNAi components, including the dsRNA trigger and target mRNAs, in further clinically relevant cell and tissue types.
Importantly, we now believe that in vivo shRNA overexpression potentially saturates not only Xpo-5, but also all 4 human Ago proteins (Figure 4). Among these, Ago-2 sequestration is probably most toxic due to its key role in numerous cellular pathways, from miRNA biogenesis, nuclear RNAi, embryogenesis, and oogenesis to hematopoiesis (33–35, 39–43). Strong additional support for this conclusion is provided by prior findings that expression of most miRNAs was reduced by more than 80% in Ago-2–knockout or –knockdown cells, causing substantial global dysregulation of thousands of genes (22, 32, 41). Likewise, in mouse oocytes, Ago-2 loss resulted in decreases in endogenous siRNAs (esiRNAs) and elevated expression of retrotransposons and other esiRNA-regulated transcripts (44). Thus, unsurprisingly, homozygous Ago-2–knockout mice display severe developmental abnormalities and are embryonic lethal (22, 35, 45). Besides, Ago-2 saturation may promote shRNA entry into non-Slicer RISCs and thus enforce off-targeting, in turn reducing the specificity of therapeutic RNAi and further increasing risks of adverse side effects (28, 30).
Comparison of the 3 major classes of therapeutic RNAi triggers (vector-encoded shRNA or miRNA, or delivered siRNA) and their potential risk for saturation (indicated by different colors; see key) of the cellular RNAi machinery (gray box). Specific steps for which we have provided experimental evidence in cells and/or mice in the present study are indicated by the respective symbols. Blue boxes denote that saturation is theoretically possible at this step, but experimental proof is lacking.
Co-sequestration of the other 3 Ago proteins may even further potentiate shRNA cytotoxicity, a hypothesis supported by findings that in vitro knockout of all 4 Ago proteins triggers apoptosis (29). This would also explain the eventual RNAi loss in our mice despite Xpo-5/Ago-2 codelivery (Figure 3, A and C), as complete rescue from toxicity may require simultaneous overexpression of all 4 Ago proteins. Mechanistically, it is likely that Ago-1 to -4 cosaturation can further perturb miRNA function and/or cause global changes in the transcriptome. The latter was indeed observed in cells depleted of individual Ago proteins (32), and altered miRNA expression and activity were also among our most striking phenotypes in shRNA-expressing cells and mice (3, 4).
Interestingly, one earlier study found that selective Ago-1 or -3 depletion impaired only up to 50% of mRNAs compared with Ago-2, and the effect of Ago-4 depletion was even smaller (32). This adds to the ongoing controversy regarding the biological role of the 4 human Ago proteins, including the question as to whether at least Ago-1, -3, and -4 are redundant. Arguing for this might be our notion that wild-type human Ago-3 and -4 are expressed suboptimally. Together with their location on the same chromosome, this may suggest that they are Ago-1 pseudogenes. On the other hand, all 4 human Ago variants are expressed in a highly tissue- and developmental-specific manner (35, 46–48). Moreover, some groups reported association of Ago-2/3 with specific miRNAs (49), or of Ago-1/2 with unique mRNAs or proteins (31, 50, 51). We and others also found distinct roles of Ago proteins in human viral infection (52, 53). Furthermore, Ago-1/2 may preferably load perfect siRNAs/shRNAs (28, 29), while Ago-3/4 may more potently engage miRNAs (30). Finally, Ago-1 and -2 are involved in nuclear RNAi, while the role of Ago-3/4 is unclear (33, 42, 43). These data may indicate that all 4 human Ago proteins are indeed crucial and thus further support our belief that their cosaturation and the associated perturbation of cellular gene silencing pathways are detrimental and must be avoided in clinical RNAi therapies.
Particularly notable in this context are two recent studies that reported global loss of miRNA expression and function in livers of mice, caused by conditional Dicer knockouts (54, 55). Curiously, the resulting phenotypes were remarkably mild in the early postnatal phase, implying that loss or dysregulation of hepatic miRNA expression can be tolerated at least for a limited period; still, affected mice also eventually exhibited progressive hepatocyte death and severe liver damage. Moreover, expression of more than 1,600 miRNA-regulated genes was significantly altered, including those controlling lipid and glucose metabolism, and particularly fetal stage–specific and cell cycle–promoting genes (likely explaining why mice that escaped toxicity later developed hepatocellular carcinoma). This implies that hepatocytes lacking mature miRNAs function normally for a short period of time, before the loss of miRNA-mediated gene regulation causes cell death and tissue failure, strongly supporting our own previous (3) and present conclusions.
Of note, the fact that toxicity manifested later in the Dicer-knockout mice as compared with our shRNA overexpression studies (several weeks versus a few days, respectively) is fully consistent with an expected much more gradual relief of miRNA-dependent gene regulation after conditional Dicer ablation, which will mainly affect miRNA processing, a consequence that may initially go unnoticed due to the long half-life of many miRNAs. In contrast, shRNA-mediated overloading of the Ago/RISC complex will instantly block miRNA function and in the case of Ago-2, moreover, impact miRNA biogenesis and maturation (35, 39). Thus, although the phenotypes of Dicer or Ago loss are typically very similar (also in other cells and tissues, e.g., oocytes; ref. 41), the underlying molecular mechanisms may differ. This may also include specific classes of small regulatory RNAs that require Ago proteins yet are generated Dicer-independently, such as those derived from tRNAs that we discovered recently (28). This could help to explain why only up to one-third of genes were similarly dysregulated in Dicer- or Ago-2–knockdown cells (32) or in knockout oocytes (41). Another particularly striking example is spermatogenesis, which was retarded in Dicer- yet not in Ago-2–knockout mice (56). Last but not least, our own findings that Dicer, unlike Xpo and the 4 Ago proteins, does not appear to rate-limit shRNA efficacy in cells or in mice (Figure 1, D and E) add further evidence for marked principal differences between the various key RNAi proteins. Notably, our data on Dicer are fully congruent with previous reports that it restricts the potency of neither transfected pri-miRNA constructs in cultured cells (21) nor of siRNAs (13). Still, Dicer is of course essentially required for principal processing and activity of longer RNAi duplexes (shRNAs and miRNAs), as confirmed in Dicer-knockout studies (13, 57, 58).
Importantly, our saturation model also suggests that caution must be exercised in using siRNAs and vectors based on miRNAs (Figure 4), since we found that both RNAi triggers effectively compromise Ago-2 when delivered or stably expressed at high doses, as may be required for certain clinical applications. Evidence (Figures 2 and 5, and Supplemental Figure 7) was that (a) transient or stable Ago-2 overexpression increased siRNA (and shRNA) activity in cells and relieved si-siRNA or si-shRNA competition; (b) miRNAs were also enhanced by Ago-2 (at least when directed against a perfect target); and (c) siRNAs and shRNAs were inactive in Ago-2–knockout cells and rescued only by Ago-2, but not any other Ago protein. Together, these data not only confirm, extend, and help to explain prior notions of, for example, si-si/shRNA competition in cells (12–14), but also validate that Ago-2 is vital for and saturable by all classes of RNAi triggers. Hence, as with shRNAs, care must likewise be taken especially with the latest proficient siRNA formulations (59, 60) to avoid in vivo Ago-2 saturation and associated adverse effects in clinical settings.
Ago-2 is critical for basal and competitive activity of perfect RNAi duplexes. (A) Scheme depicting shRNA (luc29) or siRNA (siluc) against Firefly luciferase, plus competitors (H25 shRNA and sigfp siRNA). (B) New Huh-7 cell lines stably overexpressing human Ago-2 (8 representative examples; all studies were performed in clone 1 [arrow] and validated in clone 2). Arrow indicates the best-expressing cell line, in which all studies were performed. (C) Huh-7 cells were cotransfected with: (a) Firefly (target) and Renilla (normalization) luciferase, (b) shRNA expression plasmid (encoding H25 or luc29 shRNA, or only U6 promoter), (c) plasmids encoding Ago-1 or -2 (or CMV promoter as control), (d) anti-gfp competitor siRNA (sigfp), and (e) anti-luciferase siRNA (siluc). Ago-2 overexpression enhanced basal luc siRNA activity and diminished competition with gfp siRNA or H25 shRNA (green arrows). Ago-2 likewise enhanced luciferase knockdown when cotransfected with luc29 shRNA (plus or minus luc siRNA; orange arrows). Ago-1 overexpression had no significant effect on the siRNA, but also reduced si/shRNA competition (red arrow), perhaps by sequestering the shRNA and thus freeing Ago-2/RISC for the siRNA. P < 0.05, all Ago-2 bars labeled with green/orange arrows versus controls. (D) Confirmation of the Ago-2 effects in a different cell line. (E) Validation in stable Ago-2 Huh-7 cells (see B), transfected with Firefly/Renilla, various RNAi triggers (siRNA, esiRNA, or shRNA, as indicated) against Firefly, and Ago expression plasmids (or a control). Basal siRNA, esiRNA (enzymatically created siRNA), and shRNA activities were all elevated (red) compared with parental Huh-7 (blue), confirming that Ago-2 restricts all classes of perfect RNAi duplexes. (F) Repeat of the transfections from C in stable Ago-2 cells. Competition between 2 siRNAs, or between luc siRNA and H25 shRNA, was greatly diminished compared with that in the parental cells (C). (G) Transfection of all perfect RNAi duplexes into Ago-2–deficient MEFs confirmed their strict Ago-2 dependence.
Concerning miRNAs as vector templates, intriguing data from the Rossi and Davidson laboratories imply that their in vitro and in vivo safety may in fact be relatively high as compared with that of shRNAs (5, 6, 12, 20). While counterintuitive in view of their potential for Ago-2 sequestration noted here, the expressed steady-state levels of active RNA strands are reportedly far below those achieved with conventional shRNAs (6, 20). In addition, as miRNA vectors are being shunted through the cellular miRNA pathway, it was speculated that this may slow their loading into RISC and thereby also help to reduce their competition with other small RNAs or RNAi factors (12). Finally, as mentioned, it is possible that Ago-2 competition occurs on multiple levels and that miRNA vectors specifically overload the final step (binding to, and cleavage of, a perfect target). All explanations (which are not exclusive) thus further support our saturation model and underscore the necessity of avoiding obstruction of RISC and other RNAi factors at any level in therapeutic settings.
In this regard, our results and models allow us to envision multiple new avenues to concurrently improve the efficacy, safety, and persistence of future RNAi gene therapies in humans, especially those involving shRNA vectors (Figure 6). For their routine applications, stably coincreasing all limiting factors is impractical, as it would require staggered high-dose multiple vector delivery to patients and may also yield complex results, involving the opposing roles of Ago proteins for shRNAs and potentially for miRNAs. Still, deliberate upregulation of Xpo-5 and Ago-2 (e.g., using our AAV vectors) may benefit short-term RNAi applications, such as tumor therapies, where transient rapid and potent knockdown are desirable and where limiting toxicity is typically not the highest priority. Of note, we have found no evidence thus far for general adverse effects from Ago/Xpo expression in mice, and our stable Ago-2 cell lines have not displayed abnormalities after more than 1 year in culture (data not shown).
Options to improve human RNAi gene therapies and other RNAi applications. Red box: Preferred setting (due to highest specificity and potency) for therapeutic RNAi, i.e., perfect RNAi duplexes against perfect targets. As shown, this is also the trigger-target combination that is most prone to saturating Ago-2 and thus causing cytotoxicity. As further implied by our data, this risk may be lowered by reducing the homology either between the two dsRNA strands or to the target. Hence, imperfect triggers against imperfect targets were least dependent on Ago-2. Yet they may also be least useful for human RNAi therapies. We thus favor two alternative strategies that both aim at avoiding saturation and thus alleviating shRNA toxicity and for which we have provided proof-of-concept in this article. One (a) is to transiently overexpress (e.g., using AAV vectors) limiting cellular key RNAi components, especially Xpo-5 and Ago-2. This should for instance benefit cancer-directed RNAi gene therapies, where instant and maximally potent target knockdown might be critical. It should likewise improve functional in vitro RNAi screens, where short-term and robust siRNA or shRNA activity is typically desired. An alternative (but not exclusive) second strategy (b) is to utilize minimal yet effective trigger doses in humans, by employing weak and/or tissue-specific promoters for shRNA expression. Based on our mouse data, we predict that this can also mitigate saturation-induced toxicity and prevent vector loss and thus enhance both efficacy and persistence of RNAi. Accordingly, this specific strategy should be particularly useful for long-term RNAi gene therapies against persisting exogenous targets, such as HIV or hepatitis viruses.
In contrast, persisting human viral pathogens as another clinically relevant target may instead require long-term strategies aimed at stabilizing RNAi. Avoiding global saturation is then particularly critical to avert toxicity and to preserve combinatorial RNAi therapies designed to thwart viral escape (12, 61). One strategy outlined here is the use of weak pol III promoters for shRNA expression, which yielded more than 1 year of potent HBV silencing in adult mice in this study. Beyond viral infections, we consider such promoters highly useful for any RNAi therapies requiring stable and safe in vivo gene silencing. Further assets are their small sizes (~95/239 for H1/7SK versus ~500 bp for U6), making them ideal for combined gene silencing/addition strategies employing viral gene therapy vectors (61). An alternative strategy already indicated above is expression within the context of a miRNA vector, which will likewise help to avoid RNAi saturation and thus attenuate in vivo toxicities, as recently demonstrated in mouse brain (5). Yet in contrast to straightforward shRNA expression from pol III promoters, many aspects of artificial miRNA vectors, including designing robust vectors and elucidating their intracellular processing, remain uncertain, and substantial further optimization is warranted to improve this particular strategy. Also, the limited silencing efficacy that characterizes the current generation of these vectors inherently precludes their use in clinical RNAi applications requiring maximum potency. Moreover, an additional looming concern is potential in vivo competition for Drosha or other nuclear factors that are specific to miRNA processing.
Finally, we and others recently documented that in vivo RNAi safety can be further enhanced by judicious shRNA selection, use of minimal effective vector doses, or shRNA expression from tissue-specific pol II promoters (3, 8, 62). It will be exciting to juxtapose these avenues with other promising advances, such as molecularly evolved viral vectors with inherent assets for safe and long-term therapeutic RNAi in humans (63). Such combinations of experimental strategies will then provide an additional wealth of options that researchers and clinicians can empirically test for their given RNAi application before selecting the most suitable system based on careful consideration of the balance of efficacy, specificity, delivery, persistence, and, last but not least, toxicity.