A novel method for producing functionalized vesicles that efficiently deliver oligonucleotides in vitro in cancer cells and in vivo in mice

Nano-based delivery systems have enhanced our ability to administer and target drugs and macromolecules to their targets. Oligonucleotide drugs have great therapeutic potential but often have off-target effects and stability issues. Therefore, they are often encapsulated in vesicles with targeting ligands such as antibodies (Ab) to deliver their cargo. Herein, we describe a novel, scalable and straightforward approach to produce functionalized vesicles called the “Functionalized Lipid Insertion Method.” This method differs from an older approach called the “Detergent-Dialysis Method.” The older method required excess detergent and extensive dialysis over many hours to produce the functionalized vesicles. With our method, only the functionalized lipid is detergent-solubilized during the engineering of the vesicle. The approach reduces the dialysis time, keeps the vesicles intact while orienting the targeting moieties of the functionalized lipid toward the outside of the vesicle. Pilot in vitro and in vivo experiments was performed to show the feasibility of the “Functionalized Lipid Insertion Method.” The dynamic light scattering (DLS) technique suggests that the original vesicular structure was unperturbed. Changes in vesicle size by our method were consistent with the lipid inserted externally into the vesicle. Our approach efficiently delivered oligonucleotides and affected the function of HepG2 cells. Functionalized vesicles achieve targeted delivery of oligonucleotides in mice without inducing a significant immune response through cytokine production or physical signs of the immune response such as inflammation. The industrial and therapeutic significance and implications of functionalized vesicles produced by our method are also discussed.

Introduction cells behave homogenously [54]. For experiments with unmodified and modified vesicles, 1 400 l of them were electroporated with ~4 g miRNA and then they were added wells in 2 the cell-containing 6-well plates. For the DharmaFECT™ 4 transfection reagent 3 experiments, the reagent was added with ~4 g miRNA in a 400 µl solution, and they 4 were added to other cell-containing wells. The final volume for all the wells after adding 5 the solutions was approximately 3 ml. Before RNA extraction, the treated plates were 6 incubated for 72 hours under humidifying conditions at 37 o C with 5% CO2 in a Thermo 7 Fisher Scientific Napco Series 8000 WJ CO2 incubator (Thermo Fisher, Waltham, MA). 8 9 TRIzol Reagent (Invitrogen, Carlsbad, CA) was used to isolate intracellular RNA 10 as per the manufacturer's protocol. All the isolated RNA was stored at -80ºC. 11 Complementary DNA (cDNA) was prepared from intracellular RNA (i.e., microRNA and GraphPad Prism 7 (GraphPad, San Diego, CA). The ΔΔCT values were normalized 1 against the non-coding U6 snRNA expression to indicate the relative amount of miRNA 2 delivered to the cells. 3 In-vitro time course for miRNA uptake 4 Six-well VWR culture-treated plates (Suwanee, GA) were plated with HepG2 cells 5 to a density of 100,000 cells/well. The HepG2 cells were counted using a hemocytometer 6 on a Zeiss Invertoskop 40 C inverted microscope (Zeiss, Oberkochen, Germany). After 7 the cells adhered, the cells were serum-starved for one day. The next day, the cells were 8 treated with functionalized vesicles containing 4 µg of mmu-miR-298-5p. Before 9 extracting RNA, the cells were incubated from 12 to 72 hrs in a humid atmosphere at 10 37°C with 5% CO2 in a Thermo Fisher Scientific Napco Series 8000 WJ CO2 incubator 11 (Waltham, MA). Then, the RNA was extracted and analyzed by qRT-PCR using the 12 procedures described above. 13 MiRNA effect of HepG2 cell wound healing delivered by the functionalized vesicles 14 For quantifying the functional effect of miRNA delivered by mEVs and mLNPs, the 15 functionalized vesicles were loaded with a tumor suppressor miRNA, hsa-miR-26a-5p 16 [47,48]. The HepG2 cells were plated in a 24-well plate at a cell density of 300,000 17 cells/well. A wound was created in each well using a 200 µl pipette tip followed by gentle 18 washing with PBS buffer. The cells were then treated with mEVs and mLNPs containing 19 miR-26a-5p at a dosage of 0.35 mg per well (50 ml/well) in 500 ml of fresh 10% EMEM 20 media and incubated for 72 hours. The controls for this experiment were untreated HepG2 21 cells (abbreviated Cells), cells treated with empty EVs (abbreviated Cells + EVs()), and 22 cells treated with empty mEVs with the ASGR1PAB (abbreviated mEVs(ASGR1PAB)) under 23 similar conditions. The cells were imaged from 0 to 72 hours on an Olympus IX71 inverted 1 Microscope with a TH4-100 power source (Tokyo, Japan). The images were analyzed 2 using the Fiji image processing software and the MRI wound healing tool in ImageJ 3 Software (National Institutes of Health, Rockville, MD). 4 Mouse Studies 5 Animal welfare at the University of Georgia is covered by the NIH Animal Welfare  Table S1 of the Supplementary 13 Information.

RNA extraction and quantitative RT-PCR (qRT-PCR)
14 In-vivo Quantification of miRNA 15 After 72 hours, the mice were euthanized using carbon dioxide followed by a 16 necropsy as described [59]. About 100 mg of tissue were obtained to assess the amount 17 of miRNA delivered in the tissue. The sections were suspended in 1 ml of TRIzol reagent 18 (Invitrogen, Carlsbad, CA, USA) in a 1.5 ml microcentrifuge tube. The tissues were 19 homogenized with a 1000 µl digital pipette and a Bel-Art TM Pro Culture Cordless 20 Homogenizer Unit (Thermo Fisher, Waltham, MA). The samples were then centrifuged at (μg/ml) −1 cm −1 , and the RNA purity (>99%) was estimated by the 260 nm/280 nm ratio. 5 The miRNA and U6 snRNA were amplified for quantification using the TaqMan™ 6 microRNA reverse transcription and the TaqMan™ Universal PCR MasterMix (Thermo 7 Fisher Scientific, Waltham, MA) as explained above. 8 Cytokine assay to probe immunogenicity 9 The LegendPlex™ 8-panel Th1/Th2 Bio-plex cytokine assay kit was used to 10 quantitatively determine the immunogenic response of mice after 72 hours of 11 functionalized vesicle treatment [61]. About a milliliter of blood was withdrawn 12 immediately after euthanizing the animals in BD Microtainer ® tubes containing serum 13 separator (SST TM ) (Becton, Dickinson and Company, Franklin Lakes, NJ). The blood was 14 allowed to clot at room temperature for 30 minutes. The tubes were then centrifuged at 15 1000 g for 15 minutes at 4°C using a Microfuge™ 22R (Beckman Coulter, Brea, CA), and 16 the serum was then stored at -80°C. Before performing the assay, the serum was 17 centrifuge at 10,000 g for 10 minutes at 4°C using a Microfuge™ 22R. The kit has 18 immunoassays coupled to magnetic beads for detecting the following eight inflammatory 19 factors, including the granulocyte-macrophage colony-stimulating factor (GM-CSF), 20 interferon γ (IFN-g), tumor necrosis factor α (TNF-a), and several interleukins (IL), IL-2, 21 IL-4, IL-5, IL-10, IL-12 (p70). The plate was then read using a Luminex Magpix system  One-way analysis of variance (ANOVA) tests was used to determine statistical 2 significance between groups comparing relative miRNA expression. A confidence interval 3 of 95% with all p-values less than 0.05 was considered significant (*). Student's T-Test 4 was also used to compare two groups, with a 95% confidence interval. Data were 5 analyzed with Microsoft Excel (Microsoft, Redmond, WA) and GraphPad Prism 7 6 (GraphPad, San Diego, CA).

Methods for producing functionalized vesicles
2 This work describes a novel method for producing functionalized vesicles involving 3 detergent and dialysis that we called the "Functionalized Lipid Insertion Method." This 4 method is distinct from the "Detergent-Dialysis Method" used for decades and described 5 in the literature since the early 1980s [17-24]. 6 The two approaches are shown schematically in Fig. 1. In the "Detergent-Dialysis must be extensively dialyzed over many hours or days (Step 2). As dialysis removes the 10 detergent, the vesicles randomly form functionalized vesicles from the solubilized 11 components. The composition of the original components will influence the functionalized 12 vesicle size. During the process, the functionalized lipids in the mixture will randomly 13 orient toward the inside and outside the vesicle (red arrows near Step 3). However, even 14 the long dialysis period is not enough to remove all the unintegrated components. 15 Therefore, column chromatography is often needed in addition to dialysis to remove 16 contaminants that interfere with the functionalized vesicle (Step 3) [e.g., 17]. natural vesicle such as an EV must be broken up into its components to be detergent-19 solubilized (Step 1) [19]. This process will disrupt the vesicular structure and possibly its 20 natural functions. Next, the natural vesicle's components, including lipids and proteins, 21 and the functionalized lipid are dissolved at a high detergent concentration (Step 2). This 22 detergent solubilization stage is followed by a long period of dialysis (Step 3). Because 23 the detergent concentration is so high, column chromatography is often used to remove 1 any remaining contaminating detergent (Step 4). Because of the detergent solubilization, 2 the original EV because its natural components randomly reassemble to form it. Both 3 integral membrane proteins (green arrows) and functionalized lipids (red arrows) assume 4 random orientations within the mEV. Some of the outward-facing proteins are now inward-5 facing proteins by this approach (green arrows). Some functionalized lipids face inside 6 the vesicle cavity, where they cannot perform their intended function (red arrows). solubilized by this approach is the functionalized lipid. In the diagram, the reactive lipid 10 (i.e., DPSE-PEG2000-Maliemide) is solubilized by detergent and dialyzed for two hours 11 to remove excess detergent (Step 1). The purpose of eliminating the excess detergent is 12 to prevent it from dissolving the target preformed vesicle. Then the detergent-solubilized 13 reactive lipid is incubated for 1 hour with an excess of a targeting component like an 14 antibody (Ab) to allow cross-linking between them (Step 2). Afterward, the functionalized 15 lipid is incubated for another hour with a preformed artificial vesicle or a natural vesicle 16 (Step 3). This incubation period allows the components to slowly combine without external 17 perturbation that might disrupt the vesicles like sonication. Gentle mixing can be 18 performed during this stage. Finally, the detergent bound to the functionalized lipid is 19 removed by 2 hours of dialysis (Step 4). A lot less detergent is present, so the dialysis 20 period is significantly shortened. Also, the relatively low detergent concentration reduces 21 the risk that excess detergent will disrupt the target vesicle. Detergent removal from the 22 functionalized lipid exposes hydrophobic parts of the FA tail. The functionalized lipid 23 inserts itself into the target vesicle as an entropic process to reduce exposure to these 1 hydrophobic surfaces. The FA tail of the functionalized lipid will insert into the target 2 vesicle from the outside. Naturally, the functionalized part of the lipid on the other side of 3 the molecule (red arrowhead) will be facing outward (Step 4). We have already 4 experimentally demonstrated that we can orient integral membrane proteins in lipid 5 bilayers using a similar principle [52,53]. 6 Possible mechanisms for functionalized vesicle endocytosis and miRNA delivery 7 The general theory for delivering molecules to a cell is that a targeting molecule 8 on the functionalized vesicle will bind to a surface receptor or excreted receptor ligand.

9
The vesicular uptake can occur by fusion or by several potential endocytosis mechanisms 10 [62]. The specific mechanism of vesicular uptake has not been entirely resolved but 11 appears to be cell-dependent [62].    Insertion Method" on the vesicular size distributions. Because functionalized vesicles are assembled all at once with the "Detergent-Dialysis Method," these experiments will test 1 the hypothesis that the two approaches will influence the vesicular size distribution. 2 Fig. 3 shows DLS measurements of functionalized vesicles produced by the two 3 methods. The predominant particle sizes were determined in each solution by comparing 4 the particle size distribution (PSD) to the particle diameter. Fig. 3A shows the predominate 5 DDM micelle size (solid line) and a detergent-solubilized NBD-DSPE or DPSE-PEG2000- 6 maleimide micelle size (dotted line) present in solution. The DDM micelle was 6.8 ± 1.6 7 nm in diameter, while the detergent-solubilized FAs were slightly smaller at 5.7 ± 1.4 nm, 8 consistent with the sizes you would expect for these molecules.    Fig. 3E (solid line) were 21.2 ± 5.9 2 nm, which is similar to the EV sizes produced with the "Detergent-Dialysis Method" (Fig.   3 3C, solid line). In Fig. 3E (dotted line), the EVs increased only 17% or 5 nm to 24.7 ± 6.9 4 nm. Thus, the size increase for mEVs was relatively modest, suggesting that their original 5 structures were intact, but less functionalized lipid was inserted per EV than LNPs. One 6 possibility for the size difference between mLNPs and mEVs is that the plasma 7 membranes of EVs are crowded with many intrinsic membrane proteins that interfere with are smaller than mLNPs, their surface areas will be much larger within a given volume 10 than mLNPs. In other words, a lot more functionalized lipids will be required to saturate 11 the membrane surface of mEVs than mLNPs. implanted in nude mice [77]. Finally, ASOs have been delivered to HepG2 cells by linking 2 them to the ASGR1 receptor substrate N-acetylgalactosamine [78]. As far as we know, 3 ASGR1PAB or any ASGR1 Abs have never been used to facilitate molecular uptake 4 through the ASGR1 receptor with HepG2 cells. In this study, functionalized vesicles were 5 made with ASGR1PAB, which targets the EC domain of the ASGR1 receptor. 6 The focus of my laboratory is on drug transporters called P-glycoprotein (P-gp), 7 which plays a major role in drug disposition, anti-cancer drug resistance, and biological . 17 We were hopeful that oligonucleotides targeting P-gp and delivered by 18 functionalized vesicles could ameliorate some of these challenges. Our former 19 collaborator, who is a miRNA expert, had us investigate P-gp with mouse miR-298 based . As far as we know, mouse miRNA-298 is not known to affect P-gp expression in any species. That revelation was one of the 1 primary reasons that the collaboration was terminated. Despite this setback, we felt the 2 described functionalized vesicle technology could eventually be used to investigate P-gp. 3 The mouse miRNA-298 could be used to test the feasibility of functionalized vesicles 4 since it would theoretically not have any physiological effects on the human HepG2 cells. 5 MiRNA nomenclature can be confusing for a novice. The formal names for mouse 6 and human miR-298 are mmu-miR-298-5p and hsa-miR-298-5p, respectively [87]. The HepG2 cells to mmu-miR-298-5p increased exponentially by treatment with mLNPs or 22 mEVs. After incubating the cells for 24 hours, the uptake efficiency doubled every ~12 23 hours for the mEVs and ~8 hours for the mLNPs. The uptake at 72 hours for mLNPs and 1 mEVs was 600 and 250-fold higher, respectively, above untreated HepG2 cells versus 2 the U6 snRNA. After 72 hours in these experiments, the relatively high miRNA uptake 3 was deemed an appropriate time frame to test mEVs and mLNPs in mice. 4 Modified vesicles reduce HepG2 cell proliferation by enhancing hsa-miR-26a-5p 5 uptake 6 The functional effects of treating HepG2 cells were tested with the 7 mEVs(ASGR1PAB), and mLNPs(ASGR1PAB) loaded with miRNA hsa-miR-26a-5p 8 (abbreviated mEV(ASGR1PAB, miRNA) and mLNP(ASGR1PAB, miRNA)). The miRNA hsa-9 miR-26a-5p was chosen to inhibit HepG2 cell proliferation and promote HepG2 cell 10 apoptosis [48]. In other words, the miRNA will slow the closure gap of wounds made of a 11 layer of HepG2 cells.

12
Wound healing assays with bright field microscopy has been effectively used to  The treated mice were euthanized after three days using carbon dioxide followed 11 by cervical dislocation, and their organs were harvested (Fig. 6B). The tissue was 12 analyzed using RNA extraction protocols described in the Materials and Methods and snRNA. Finally, the Ct values allowed us to estimate the relative-fold miRNA uptake in 17 the mouse organs by the ∆∆Ct method (Fig. 6E). 18 In vivo studies with mmu-miR-298-5p were originally intended to modulate P-gp 19 expression as recommended and suggested by our former collaborator, which would fit 20 well within Specific Aim #1 of the supporting grant, but did not have that function  Eventually, our laboratory wants to use functionalized vesicles to modulate in vitro 11 and in vivo P-gp transporter expression. Luckily, siRNAs have already been shown to 12 affect mouse P-gp expression [53,107]. Therefore, functionalized vesicles by this 13 approach loaded with the P-gp targeting siRNA may be a practical way to investigate the 14 P-gp transporter expression and effects in the future.  For these pilot studies, the relative miRNA uptake was normalized against the non-21 coding U6 snRNA [109]. Normalizing against U6 snRNA demonstrates the feasibility of 22 our approach but is not ideal for the following reasons. U6 snRNA expression can vary in 23 mouse tissues (even within the same organ), differ significantly between genders, and be 1 affected by a mouse's age [108,[110][111][112][113]. U6 snRNA expression also likely varies greatly 2 between mouse strains and sub-strains [114,115]. Even the miRNA of interest can 3 theoretically affect U6 snRNA levels, although no significant changes in U6 snRNA 4 expression were observed in these pilot studies by functionalized vesicle treatment (data 5 not shown). 6 A table of mice used in these pilot studies is provided in Table S1   In vivo miRNA delivery by functionalized vesicles with a control antibody 13 As a negative control for in-vivo targeting ability of mEVs and mLNPs, young 14 female C57BL6/J mice were treated with functionalized vesicles loaded with mmu-miR- 15 298-5p and GFPMAB as a control antibody. Fig. 7 shows the uptake of mmu-miR-298 16 treated with modified vesicles with the GFPMAB and containing mmu-miR-298 17 (abbreviated mEVs(GFPMAB, mmu-miR-298) and mLNPs(GFPMAB, mmu-miR-298-5p)). 18 The highest miRNA uptake was observed in the mouse spleen after treatment with the 19 functionalized vesicles. The IP injection puts the functionalized vesicle into the lymphatic  to the organs was very similar (Fig. 8A). The normalized distribution of mice treated with 23 mLNPs(NPHS2PAB, mmu-miR-298) and mLNPs(GFPMAB, mmu-miR-298) in Fig. 8D was   1 virtually identical to the distribution in Fig. 8B, implying that both types of functionalized 2 vesicles engineered with NPHS2PAB had similar targeting mechanisms in mice. 3 In vivo targeted miRNA delivery to the liver by functionalized vesicles with the 4 ASGR1 antibody 5 Male and female Nu/Nu mice were treated with mEVs(ASGR1PAB, mmu-miR-298) 6 and mLNPs(ASGR1PAB, mmu-miR-298), respectively. Figs. 9A and C show relative with ACE2MAB will be relatively non-specific but more evenly distributed among the organs 18 than mEVs and mLNPs with GFPPAB, which primarily targeted the spleen (Fig. 7). 19 Fig. 10 shows the uptake of mmu-miR-298 and their effects on the cytokine levels 20 after treatment of mice with miRNA-loaded mEVs and mLNPs engineered with ACE2MAB.

21
In Fig. 10A, from high to low fold miRNA uptake by mEV treated mice versus mice treated 22 with SFM only, the miRNA uptake in mice by mEVs was ~280 fold in the liver, ~250 fold 23 in the kidneys, ~200 fold in the spleen, ~200 fold in the lungs, ~90 fold in the small 1 intestine, ~70 fold in the heart and ~25 fold in the brain.

2
In Fig. 10B, the relative miRNA uptake was analyzed with mice treated with 3 mLNPs(ACE2MAB, mmu-miR-298) versus SFM only treated mice (controls). The 4 distribution of miRNA delivery to organs by mLNPs differed significantly from mEVs. The 5 uptake of miRNA by treatment of mice with mLNPs(ACE2MAB, mmu-miR-298) was highest 6 for the small intestine (~830 fold) followed by the kidney (~530 fold), spleen (~400 fold), 7 liver (~400 fold), heart (~388 fold), and lungs (~100 fold). Virtually no miRNA uptake was 8 observed in the brains of mice. Except for the spleen, these relative miRNA uptake levels relatively high miRNA uptake after treatment with functionalized vesicles is likely due to 12 IP administration rather than specific targeting by ACE2MAB. 13 The immunogenicity of the functionalized vesicles was assessed by LegendPlex™ 14 8-panel Th1/Th2 Bio-plex assay, which tests for eight major anti-and pro-inflammatory 15 cytokines from two types of CD4+ T helper lymphocytes [129]. Cytokine profiling has been  In this work, we present the "Functionalized Lipid Insertion Method" to bioengineer 6 targeting functionalized vesicles with a surface coated with PEG-linked Abs that we refer 7 to as mEVs or mLNPs. We explain how our approach differs significantly from the  (Fig. 9). The least specific miRNA delivery to mouse organs was from functionalized 19 vesicles with the ACE2MAB (Fig. 10). At least over the 72 hour study period, the 20 functionalized vesicles also did not appear to be non-immunogenic (Fig. 10).    In addition to cationic liposomes, liposomes can also be conjugated with target-1 specific vectors to design cell-specific vehicles. Targeted liposomes have been prepared 2 by conjugating Aptamer (AS1411), a target-specific single-stranded oligonucleotide, 3 using thioether linkage between DSPE-PEG2000 maleimide on the liposomal surface. 4 These Aptamer functionalized liposomes were used to deliver miR-29b to A2780 cells,  liposomes, gives more room to play with designing or modifying liposomes as desired.

21
The Detergent-dialysis method has been shown to be as effective as if not more than the increase temperature stability through chemical modification [160]. MiRNA also has the potential for slowly leaking from functionalized vesicles [161].

2
Going beyond these pilot studies 3 In these pilot studies, oligonucleotide delivery through our functionalized 4 engineering approach in HepG2 cells and its function was demonstrated. 5 Oligonucleotides were also delivered to mice from different strains, different genders, and 6 different ages with these bioengineered functionalized vesicles (Table S1). Over the 7 narrow 72 hour time frame of these experiments, the functionalized vesicles were non-8 immunogenic. These pilot studies gave us a tantalizing look at the potential of our novel 9 approach to bioengineer functionalized vesicles to deliver macromolecules both in vitro 10 and in vivo. However, more in-depth studies will be required for use as a potential 11 therapeutic and for studying the biological functions of oligonucleotides such as miRNA. [87], a wound-healing assay of HepG2 was used to determine the functional effects of the 2 hsa-miR-26 (Fig. 5). To examine the functional effects of specific miRNA on proteins, a 3 Western blot should be used to quantify the protein of interest in addition to the relative 4 protein mRNA level. 5 In Fig. 2B, we discuss the theoretical possibility of targeting secreted receptor  In vivo experiments using our functionalized vesicles will require careful selection 13 of non-coding RNAs (ncRNAs) for normalization of miRNA uptake and mice. The mice 14 should also be strain, gender, and age-matched. Mouse studies should include both 15 genders, and the numbers of mice used in the study should be statistically high enough. 16 Choosing appropriate non-coding RNA controls (ncRNAs) is essential for accurate 17 miRNA uptake quantitation. Therefore, several ncRNAs from each mouse tissue should 18 be examined for expression stability and variability. Several different approaches have 19 been developed for this purpose [168][169][170]. A tool called BestKeeper estimates the 20 stability by the standard deviation (SD) of mean cycle quantitation (Cq) values between 21 candidate ncRNAs [168]. The stability of ncRNAs can also be determined using the ∆Cq 22 method, where the SD is calculated by the ∆Cq of ncRNA gene pairs [169]. The GeNorm algorithm is another approach to gauge ncRNA stability with a calculated gene stability M 1 value [170]. Finally, the calculation of variation between groups of candidate control 2 genes can be accomplished with the NormFinder algorithm [171]. For the best results, 3 these four methods can be used together to determine the optimum reference genes for  Step 1, functionalized lipids containing a targeting molecule such as an Ab are mixed with 5 lipids and high concentrations of detergent, leading to complete solubilization of all the 6 components. In Step 2, the high concentrations of detergent are dialyzed over time to 7 remove the detergent. Once the detergent reaches a specific concentration, the mLNPs 8 will spontaneously form, but the orientations of the functionalized lipids become 9 randomized (red arrows) within the lipid bilayer. Even after this long dialysis period, 10 column chromatography is often used to remove detergent-solubilized components in 11 Step 3. B) Shows the same process of producing functionalized vesicles from natural 12 vesicles like EVs. Because EVs are intact vesicles, they are lysed into their individual 13 components in Step 1 prior to detergent solubilization and dialysis in Steps 2 and 3. In

14
Step 3, because proteins are likely to be embedded in the natural vesicle lipid bilayer, 15 their orientations will be randomly oriented (green arrows) like the functionalized lipids 16 (red arrows). Column chromatography can be used in Step 4 to remove detergent-17 solubilized contaminants. C) Shows the process of producing mLNPs using the     Engineered exosomes for targeted co-delivery of miR-21 inhibitor and 4 chemotherapeutics to reverse drug resistance in colon cancer, Journal of

Figure Strain
Common