Screening of efficient siRNA carriers in a library of surface-engineered dendrimers

Polymers are widely used as non-viral carriers for siRNA delivery, but concern has also arisen in their limited efficacy and inherent toxicity. Whilst many of previous efforts have been documented towards improving the performance of polymers via chemical modifications, the structure-activity relationships (SAR) of these ligand-modified polymers are not well understood. To address this issue, we systemically prepared a library of surface-engineered dendrimers (>300) as the screening pool to discover efficient siRNA carriers. The modified ligands include alkyls and fluoroalkyls, amino acids, benzene derivatives and heterocyclic compounds. Gene silencing results showed that the lead material shows excellent efficacy even in hard-to-transfect cells such as mesenchymal stem cells. The SAR studies revealed that ligands containing appropriate hydrophobicity, or ligands with both hydrophobic and functional atoms/groups are essential for polymers to achive efficient knockdown efficacy. A second-generation library designed based on the above principles further confirms the proposed design criteria. The results enable the future rational design of potent siRNA carriers.

Scientific RepoRts | 6:25069 | DOI: 10.1038/srep25069 to achieve efficient gene silencing. The use of helper lipids make it less feasible to prepare siRNA formulations with reproducible gene silencing efficacy. In this study, we prepared a library of ligand-modified polymers, which is used as a screening pool to discover siRNA carriers which can achieve efficient gene silencing in the absence of helper lipids. Besides alkyl chains, the ligands in this study include fluoroalkyl chains, benzene derivatives, heterocyclic compounds, amino acids and etc. Dendrimer, a class of synthetic polymers with a well-defined molecular structure [41][42][43] was selected as the model polymer due to its globular shape, good solubility and monodispersity 44,45 . The use of dendrimer as the polymer scaffold can exclude the effects of polymer structure, conformation and polydispersity on the siRNA delivery efficacy 40 . In addition, cationic dendrimer with primary amine as the surface functionality can be functionalized with various ligands in a multivalent fashion via facile and general chemical reactions 28 . These unique properties make dendrimer an ideal candidate in the study of SAR of ligand-modified polymers.

Results and Discussion
Synthesis of surface-engineered dendrimers. Generation 5 (G5) polyamidoamine (PAMAM) dendrimers with a theoretical number of 128 surface primary amine groups were conjugated with 129 types of ligands listed in Fig. 1. The conjugated ligands include several sub-groups, like alkyl and fluoroalkyl chains, amino acids, heterocyclic compounds, benzene derivatives and other compounds. A total number of 301 surface-engineered dendrimers were obtained by simply reacting the primary amine groups of dendrimer with carboxyl, anhydride, isocyanate, isothiocyanate, halogen or epoxy groups (see Supplementary Table S1).
Gene silencing efficacies of the surface-engineered dendrimers in the library. Gene silencing efficacy of the resulting materials was first tested on HeLa-luc cells (HeLa cells stably expressing a firefly luciferase gene) using a siRNA specifically targeting luciferase (siLuc). In the screening experiments, the siLuc concentration is fixed at 50 nM and the polymer concentration varies in the polymer/siRNA weight ratios (w/w) of 1 to 20. As shown in Fig. 2, a total number of 22 materials were screened as efficient materials (shown as red dots). About 7.31% of the materials in the library induced more than 50% gene knockdown efficacy on the HeLa-luc cells. We further compared the five most efficient polymers in the library with a commercial transfection reagent Lipofectamine 2000 (Lipo 2000). The screened top 5 performing materials form small-sized polyplexes with favorable biophysical properties, e.g. they are able to form positively charged nanoparticles below 300 nm with siRNA (Fig. 3). Parallel experiments were carried out using scrambled siRNA (siNC) that is inactive in silencing luciferase as negative controls for each material. As shown in Fig. 4a, without surface modification, G5 PAMAM dendrimer only shows extremely low gene silencing efficacy (<10%), while the screened top 5 performing materials show comparable efficacies to Lipo 2000 on the HeLa-luc cells. Additionally, the polymers with siNC show negative gene silencing in the cells, suggesting high specificity in gene silencing mediated by the screened materials. When decreasing the siLuc dose, the gene knockdown efficacies of the screened materials are slightly reduced (see Supplementary Fig. S1). Even at a siRNA dose of 5 nM, the screened polymers still show gene silencing efficacies above 70%. Moreover, the screened materials exhibit high gene silencing efficacies in MDA-MB231-luc cells (MB231 cells stably expressing a firefly luciferase gene) (Fig. 4b). Besides the luciferase gene, the materials can also efficiently and specifically knockdown other genes such as Bcl-2 in HeLa cells (Fig. 4c) and PHD-2 in NIH3T3 cells ( Fig. 4d and Supplementary Fig. S2). We also test gene knockdown efficacy of screened materials in hard-to-transfect cells such as primary mouse mesenchymal stem cells (Fig. 4e). The lead material (E9-2, triphenylphosphonium-modified G5 dendrimer) shows high gene silencing efficacy even  in hard-to-transfect cells. For example, E9-2 efficiently knockdown the Smurf1 gene in primary mouse mesenchymal stem cells (Fig. 4f), followed by the down-regulation of Smurf1 protein (see Supplementary Fig. S3), but Lipo 2000 shows poor efficacy (<10%) in the delivery of siRNA into primary mouse mesenchymal stem cells. Anderson et al. recently also reported that alkyl substitution of regularly branched dendrimers such as PAMAM and polypropylenimine is a facile strategy to prepare efficient siRNA carriers 39,40 . However, these amphiphiles should be co-formulated with helper lipids such as DSPC, DSPC-PEG, and cholesterol, which make it difficult to prepare formulations with reproducible siRNA delivery efficacy. In comparison, the discovered materials in our library (A18-5, E5-2, E9-2, A16-8 and D45-3) can efficiently knockdown the target genes in various cell lines without any helper lipids.
SAR of the surface-engineered dendrimers in siRNA delivery. We further investigated SAR of the surface-engineered dendrimers in gene silencing. It is known that hydrophobic modification on polymers could improve serum stability, membrane fusion, cellular uptake and intracellular siRNA disassociation abilities of polymer/siRNA complexes [38][39][40]46 . According to the chemical structures of efficient materials in the library, the modification of hydrophobic ligands on dendrimer is beneficial for efficient siRNA delivery. As shown in Fig. 5a, conjugation of alkyl with proper chain length to G5 dendrimer significantly improves its gene silencing efficacy. In particular, G5 dendrimer modified with lauric acid (A7-3) shows the highest efficacy among the alkyl chain-functionalized dendrimers. The improved gene silencing efficacy can be explained by increased cellular uptake efficacy (Fig. 6). These amphiphilic materials can escape from endolysosomes through a combination of membrane fusion mechanism of the lipids and proton sponge effect of the dendrimers 39,47,48 . Conjugation of aliphatic acids with longer chains such as myristic acid (A9-3), palmitic acid (A10-3) and fatty acid (A11-3) fails to knockdown the luciferase gene probably due to the decreased charge density on the dendrimer surface (shielding effect of longer alkyl chains, Supplementary Fig. S4) 49 . Similarly, conjugation of cycloaliphatic compounds with higher hydrophobicity shows a beneficial effect in improving the efficacy of G5 dendrimer (see Supplementary  Fig. 5b). G5 dendrimer modified with 33 cyclododecyl chains shows the highest efficacy among the polymers in Fig. 5b. The compound that contains an aromatic ring as well as an alkyl group with appropriate hydrophobicity also shows much enhanced gene knockdown efficacy when conjugated onto dendrimer surface (Fig. 5c). Besides, dendrimers modified with fluorous chains show a similar phenomenon in gene silencing (Fig. 5d). G5 dendrimer conjugated with fluoroalkyl chains containing seven fluorine atoms shows the highest efficacy. Considering the more hydrophobic effect of fluoroalkyls compared with alkyls, much shorter fluoroalkyl chains are needed to achieve efficient gene silencing (C4 for fluoroalkyls versus C12 for alkyls). The improved gene knockdown efficacy after fluorination can be explained by enhanced cellular uptake and endosomal escape of the complexes due to the fluorine effect as reported in a recent study 50 . All the highest efficient materials in Fig. 5b-d exhibit the highest cellular uptake among the control materials no matter at optimal transfection conditions or equal molar concentrations (see Supplementary Figs S5-S7).
Besides the hydrophobic effect, we also found some functional atoms or ligands in improving the siRNA delivery efficacy. In Fig. 5e, the dendrimers conjugated with different numbers of benzoic acid show extremely low gene silencing efficacy (<10%), while significantly improved gene silencing efficacy was observed if the same structures were incorporated with bromine atoms or boronic acid groups. Similarly, butyric acid-and lysine-modified G5 dendrimers show poor gene knockdown efficacies, but performances of the materials can be improved by replacing the hydrogen atoms on butyric acid with fluorine atoms or replacing the amine groups on the side chain of lysine with guanidine groups. These results suggest the beneficial effects of fluorine/bromine atoms, boronic acid and guanidine groups in siRNA delivery. The presence of these atoms or ligands significantly improves the cellular uptake efficacy of the complexes (Fig. 7). The hydrophobic effect of the ligand on dendrimer surface is also essential for the efficient siRNA delivery. For example, G5 dendrimer modified with butylboronic acid with decreased hydrophobicity compared to phenylboronic acid show limited gene silencing efficacy (see Supplementary Fig. S8). A combination of hydrophobic ligands (such as aromatic rings or alkyl chains) and functional ligands (such as boronic acid or guanidine) generates efficient materials for siRNA delivery.
A second-generation library to confirm the material design principle. To further confirm the material design criteria described above, we synthesized a small second-generation library of surface-engineered  Table S2). The ligand in the library either has appropriate hydrophobicity (fluorous chains or aliphatic rings) or combines both hydrophobic ligands (phenyl group) and functional ligands (fluorine and bromine atoms/ligands). In such a way, we could expect a large fraction of this second-generation library to be efficacious in comparison with the first-generation one. When tested the luciferase gene silencing efficacy of the materials on HeLa-luc cells at a siRNA dose of 50 nM, it was found that more than half of the designed materials (52.9%) in the second-generation library show gene silencing efficacy above 50% (Fig. 8b). These results suggest that the proposed material design criteria are able to significantly increase the probability of success in preparing efficient surface-engineered dendrimers for siRNA delivery. For cationic polymers, there is usually a transfection efficacy-cytotoxicity correlation associated with the carriers. We further evaluated the cytotoxicity of the efficient materials from the first-and second-generation libraries with or without siRNA by an MTT assay. All the materials show minimal toxicity on the transfected cells (> 90% viability) at their optimal transfection conditions (Fig. 8c), suggesting that the materials feature high efficacy and low cytotoxicity when

Conclusions
In summary, we have synthesized a library of surface-engineered dendrimers via facile and simple chemical reactions and screened their performances in siRNA delivery. The screened efficient materials can specifically knockdown several target genes in different cell lines. Efficacy of the lead material E9-2 is much superior to commercial lipid carriers such as Lipo 2000 in hard-to-transfect stem cells. SAR studies demonstrated that hydrophobic ligands such as aliphatic chains and aromatic rings, as well as functional ligands such as fluorine and bromine atoms, boronic acid and guanidine groups are essential for the high gene silencing efficacy of screened materials. Above all, the screened materials in the library enable efficient delivery of various siRNA into different cell lines with minimal cytotoxicity. The SAR results in this study successfully paved the way towards the design of a second-generation library of surface-engineered dendrimers with high gene silencing efficacy.

Materials and Methods
Materials. Ethylenediamine-cored G5 PAMAM dendrimer with primary amine groups was purchased from Dendritech, Inc. (Midland, MI). The received dendrimer was characterized by 13 C NMR and polyacrylamide gel electrophoresis to confirm its quality. Lipo 2000 was purchased from Invitrogen (Carlsbad, CA).
Synthesis and characterization of surface-engineered dendrimers. G5 PAMAM dendrimer was reacted with the compounds via facile reactions (interacting of amine groups of G5 PAMAM dendrimers with carboxyl, anhydride, isocyanate, isothiocyanate, halogen or epoxy groups at different feeding ratios). The detailed feeding ratios and reaction conditions for the synthesized materials are listed in Supplementary Table S1. The synthesized materials were purified by intensive dialysis against distilled water or dimethyl sulfoxide according to their solubility, and characterized by 1 H NMR, ninhydrin assay or fluorine element analysis to calculate the number of conjugated ligands. Detailed synthesis and characterizations on the surface-engineered dendrimers can be found in Supplementary Table S1 and Table S2. The 1 H NMR spectra of synthesized materials are listed in Supplementary Appendix A.
Characterizations of the complexes. The synthesized materials were mixed with siRNA in diethylpyrocarbonate-treated water at different weight ratios (w/w) for 30 min. The size and zeta potential of the complexes were measured by Zetasizer Nano ZS (Malvern Instrument) and transmission electron microscope (HT7700, HITACHI, Japan). The binding of siRNA with the alkyl chain-modified dendrimers was determined by agarose gel electrophoresis assay. The weight ratio of dendrimer to siRNA ranges from 2 to 6 (10 μ L of the sample containing 1-3 μ g polymer and 0.5 μ g siRNA). The polyplexes were diluted with DNA loading buffer and the samples were run on a 1% (w/v) agarose gel at 90 V for 20 min. 150 μ L serum-free media. The cells were then incubated with the polyplex solutions for 6 h and replenished with 500 μ L serum-containing media (10% FBS). The silencing experiments of luciferase gene were further continued for 18 h. Luciferase expressions in the cells were analyzed according to the manufacturer's protocols (Promega). Protein concentration in each well was determined using a BCA Protein Assay Kit (TIANGEN, China). The data was normalized to relative luciferase light unit per mg protein (RLU/mg protein), and relative to that of untreated cells (see Supplementary Appendix B). The silencing experiments of PHD-2 gene in NIH3T3 cells and Bcl-2 gene in HeLa cells were conducted for 48 h. The silencing experiments of Smurf1 gene in primary mouse mesenchymal stem cells were conducted for 24 h. The expressions of Bcl-2 mRNA in HeLa cells, PHD-2 mRNA in NIH3T3 cells or Smurf1 mRNA in primary mouse mesenchymal stem cells was analyzed by RT-PCR using specific primers described as follows.
Unmodified G5 PAMAM dendrimer and commercial transfection reagent Lipo 2000 were tested as controls. Three repeats were conducted for each material. Students't-test was used to analyze statistically significant differences.
Total RNA was isolated from the transfected cells and reverse-transcribed into cDNA. The cDNA was subjected to RT-PCR analysis targeting Bcl-2, PHD-2 or Smurf1 using SYBR Green PCR Mix (TOYOBO, Osaka, Japan). The data were normalized to an endogenous reference, and relative to that of untreated cells.
Western-blot. Western-blot was performed to evaluate the levels of PHD-2 protein expression in NIH3T3 cells and the Smurf1 protein expression in primary mouse mesenchymal stem cells. The silencing experiments of PHD-2 gene in NIH3T3 cells and Smurf1 gene in primary mouse mesenchymal stem cells were conducted for 72 h and 48 h, respectively. Transfected cells were washed three times with cold PBS and lysed in a lysis buffer. The cell lysates were separated on 10% SDS-PAGE gels and transferred onto a PVDF membrane. The membrane was incubated with rabbit monoclonal antibodies against PHD-2 and HIF-1α or Smurf1 and MEKK2 overnight, and further incubated with HRP-linked IgG antibody (goat anti-rabbit) for 1 h. The protein bands were visualized using the Odyssey CLx Infrared Imaging System (LI-COR, USA). β -Actin was measured as the endogenous reference. Cellular uptake. HeLa-luc cells were plated in 24-well plates and grown until 50% confluent. The cells were then incubated with the polymer/siLuc-FAM complexes at concentrations equal to those used in transfection experiments. After incubation for 2 h at 37 °C, the cells were washed with cold PBS buffer for three times and trypsinized. The mean fluorescence intensity of incubated cells was quantitatively analyzed using flow cytometry (BD FACSCalibur, San Jose). Three repeats were conducted for each material.
Cytotoxicity analysis. Cytotoxicity of cells treated with the polymers was determined by a 3-(4,5-Dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay according to well-established protocols. Generally, HeLa cells were plated in 96-well plates at a density around 10 4 cells per well and cultured overnight. The cells were incubated with the materials at concentrations equal to those used in transfection experiments with or without 50 nM siRNA for 24 h. After that, 100 μ L MTT containing medium (0.5 mg/mL) was added to each well. After incubation for 2 h at 37 °C, the medium was removed and 150 μ L DMSO was added to dissolve the insoluble purple formazan. Absorbance of the solution in each well was detected at 490 nm by a microplate reader (Thermo Fisher Scientific, U.S.). Five repeats were conducted for each sample and the data were given as means ± S.E.M.