Currently, the common practice to obtain a potent siRNA is to design several siRNAs according to the siRNA design algorithm and test their functionality. However in some cases, siRNAs have to be designed to target particular sequences. For example, highly conserved sequences shared by different virus species have to be used for broad-spectrum antiviral activity. Such conserved regions are few and are often not the ideal sequences predicted by the currently available algorithms for siRNA design. Our method provides an alternative method wherein generally any given sequence, regardless of predictability by algorithms can be optimized for gene silencing just by introducing mismatches in the passenger strand. It has been reported previously that introducing mismatch to the end of siRNA could enhance the intended strand functionality 
. In this study, we systematically examined the effects of introducing both end and internal mismatches and found that internal mismatches, in combination with end mismatches can significantly increase the functionality of siRNA. This method could also be applied to shRNA design.
Why does introducing mismatches improve siRNA function? Since siRNA has to be loaded into RISC to be functional, mismatches may increase the loading efficiency and thereby improve functionality. One of the major discoveries related to RISC loading of siRNA is that end thermodynamic stability dominates strand selectivity of RISC loading with the strand with less stable 5′ end showing the greatest propensity for loading 
. While this rule appears to be generally true, it is not rare that siRNAs that defy this rule are also potent in gene silencing. Moreover, siRNAs with perfect asymmetric thermodynamic ends might also not function well 
. Our study showed that in addition to the end thermodynamic stability, changing the internal thermodynamic stability by introducing mismatches at certain internal positions can further increase siRNA functionality. Kawamata et al
reported that mismatches at the central position enhance siRNA loading into pre-RISC while mismatches at the seed region and mid region enhance siRNA unwinding 
. It has also been reported that potent conventional siRNAs usually have lower internal thermodynamic stability 
. A recent paper showed that lowering the thermodynamic stability in the central position (position 9–12) by either introducing mismatches or chemical modifications, could significantly improve the siRNA potency 
. Gu et al also showed that decreasing internal thermodynamic stability of shRNA could enhance RISC maturation by noncleaving Ago proteins, suggesting that unwinding of the 2 strands by Agos 1, 3 and 4 was enhanced by decreasing thermodynamic stability 
. Taken together, these studies suggest that the overall thermodynamic pattern of siRNA is what really determines the siRNA functionality, not only the end thermodynamic stability. Introducing mismatches at certain positions could optimize the overall thermodynamic pattern to make the siRNA a better substrate for RISC loading and maturation.
Several previous studies have tried to improve functionality and strand-selecting accuracy of siRNAs. Chen et al
reported that 5′ phosphate is an important determinant for strand selection and 5′-O-methylation of siRNA passenger strand could effectively reduce passenger strand functionality 
. Several other chemical modifications have also been used to enhance functionality and decrease the selecting of passenger strand 
.Other methods used to improve functionality and strand-selecting accuracy include incorporating 2 nt overhangs only in the guide strand, making the passenger stand into two pieces or by shortening the passenger strand by 3–4 nt 
. However, none of these methods can be applied to shRNA design. Our strategy provides yet another way to enhance the functionality and strand selecting accuracy of both si and shRNAs.
It is intriguing that introducing mismatches improved functionality in some siRNAs while not in other siRNAs. Our data presented in this study could not provide a clear explanation, although some of our data might provide clues to the question. As shown in , introducing a single mismatch at a certain position in two different siRNA appears to have different effects. For example, introducing a mismatch at position 4 showed no improvement in siR-21 compared to m0, but it had a significant functionality improvement in siCS2. Moreover, different mismatch combinations have very different effects on siR-21 and siCS2, suggesting that the effect of mismatches on siRNA functionality might be siRNA sequence-dependent. Different siRNA sequences need mismatches at different positions to improve the thermodynamic feature. Our results shown in , also supports this hypothesis. It is also worth mentioning that introducing mismatches did not further improve the functionality in conventional siRNAs that already worked well (such as FVEjw as shown in ). The reason might be that the thermodynamic pattern of these potent siRNAs might already be optimal and can't be further improved by introducing mismatches. Thus, whether introducing mismatches to improve functionality and in which position mismatches should be introduced might depend on siRNA sequences. A larger scale of study on the effect of mismatch-mediated thermodynamic pattern changes in different siRNA sequences might shed light on what is the optimal thermodynamic stability structure of siRNA for RISC loading.
How the guide strand of siRNA or miRNA is loaded into Argonaute proteins has been under intensive investigation. All 4 mammalian argonaute proteins are ubiquitously expressed and are involved in miRNA-mediated gene repression (reviewed in reference 
). Recently, Wang et al
reported that bacterially expressed human Ago1 and Ago2, but not Ago3 and Ago4, possess strand-dissociating activity of miRNA duplexes and passenger strand cleavage activity of siRNA duplex 
. However, Yoda et al
reported that although all 4 human Ago proteins showed remarkably similar structural preferences for miRNA-like duplexes, only Ago2 could load and unwind siRNA duplexes efficiently to generate mature RISC in vitro 
. Our results, however suggest that the different Ago proteins do not differentiate between si/miRNA-based structure to generate mature RISC. The discrepancy between these studies might be due to the different experimental systems used. Since we performed experiments in cell lines rather than Ago proteins purified in vitro
, our conclusion might reflect a more physiological context.
In summary, introducing mismatches in the passenger strand generally improves the efficacy of siRNA by changing the end as well as internal thermodynamic stability. Moreover this method can also be applied to shRNA design to improve efficacy.