Virus‐like particles: Next‐generation nanoparticles for targeted therapeutic delivery

Abstract Most drug therapies distribute the agents throughout the entire body, even though the drugs are typically only needed at specific tissues. This often limits dosage and causes discomfort and harmful side‐effects. Significant research has examined nanoparticles (NPs) for use as targeted delivery vehicles for therapeutic cargo, however, major clinical success has been limited. Current work focuses mainly on liposomal and polymer‐based NPs, but emerging research is exploring the engineering of viral capsids as noninfectious protein‐based NPs—termed virus‐like particles (VLPs). This review covers the research that has been performed thus far and outlines the potential for these VLPs to become highly effective delivery vehicles that overcome the many challenges encountered for targeted delivery of therapeutic cargo.


| I N T R O D U C T I O N
Currently, numerous diseases lack adequate treatment, most notably cancer and various genetic disorders. In 2016, the National Cancer Institute estimates that 1,685,210 new cases of cancer will be diagnosed in the United States and 595,690 people will die from the disease-a 35% mortality rate. 1 Typical cancer treatment includes chemotherapy, radiation, and surgery. However, surgery is highly invasive and often fails-especially after metastasis. Chemotherapy and radiation can be effective temporarily, but result in harsh side-effects that drastically reduce quality of life. In particular, systemic administration of chemotherapeutic agents is usually limited by those side-effects and the effective dose at the tumor site is only a small fraction of the administered drugs. 2 In addition to cancer, the development of gene therapies for treatment of genetic disorders, such as mitochondrial disorders and Parkinson's disease has been a major focus of research. By 2012, over 1800 gene therapy clinical trials had been conducted or approved. 3 However, success in clinical trials has been limited because of numerous technical barriers, including difficulty in reaching the targeted tissues. Although the U.S. FDA approved the first oncolytic viral therapy, Imlygic, recently, no virus-derived therapies for gene delivery have been FDAapproved according to the latest information from the FDA's website.
Targeted delivery has long been one of the most promising, but also most challenging, opportunities for improving the treatment of these diseases. Targeted delivery offers three key advantages that systemic delivery lacks: (a) the therapeutic will act primarily at the desired site-of-action, limiting off-target effects such as the harmful sideeffects involved with chemotherapy; (b) the delivery vehicle can provide much higher local concentrations of the therapeutic within the diseased tissues, allowing a more effective treatment; and (c) the delivery vehicle can carry the therapeutic to sites it would not normally be able to reach, improving the efficiency of gene therapies. The first attempts at developing delivery vehicles were antibody-drug conjugates. These (Adcetris)], and many more are in clinical trials. 4 However, they suffer from several limitations including structural heterogeneity, instability, and limited solubility. 4,5 In addition, antibody-drug conjugates typically deliver only a few drug molecules per antibody. 4 In contrast, nanoparticle (NP)-based delivery agents, including liposomal, polymer-based, metal-based, and protein-based NPs, have the potential to provide safer and more effective delivery by encapsulating therapeutic cargo inside the particle with a much higher cargo/carrier ratio. This review will focus on the development of virus-like particles (VLPs), proteinbased NPs derived from viral capsids, as targeted therapeutic delivery agents. Several previously published reviews have covered VLPs. A review from the Bundy lab excellently describes many ways to covalently attach ligands to the surface of VLPs, but lacks further information pertinent to their use as drug delivery vehicles. 6 A 2014 review from the Tullman-Ercek lab covers cargo loading, specific targeting, and application for using VLPs as delivery vehicles, but lacks specific surface modification information and loading small molecule drugs. 7 Another 2014 review from the van Hest lab has an excellent discussion of surface and interior covalent attachment and genetic fusion strategies, but contains no discussion of nonspecific cargo loading or attachment techniques. 8 Lastly, a recent 2016 review covers a large variety of VLPs and other protein-based NPs, but lacks depth for each individual vehicle. 9 This review, while focusing on six of the most used VLPs, attempts to combine and expand on the information within these other reviews while addressing new factors, including particle stability, expression platforms, and purification methods, that are important for the development of these vehicles as therapeutic NPs.

| USING V LPs OVERCOMES THE LIMITATIONS OF CURRENT NP-BASED THERAPEUTICS
Despite many attempts, only a few liposomal and protein-based NPs have been approved for cancer drug delivery, including Doxil and Abraxane-and these are all passive-targeting delivery agents relying on the enhanced permeability and retention (EPR) effect for tumor localization. 5,10,11 Liposomal NPs are limited by particle instability, rapid clearance, and spontaneous membrane fusion with off-target cells. 12,13 The polymer-based NP technologies suffer from structural heterogeneity, particle instability, slow and nonuniform drug release, and potential immunogenicity. 14,15 The more stable metal-based NPs suffer from a lack of specificity and high toxicity. 16 In addition, most of these NPs suffer from clearance mediated by phagocytes and dendritic cells, including Kupffer cells in the liver. Coating NPs with polyethylene glycol (PEG) can help avoid phagocytes and extend the blood circulation time by creating "stealth" brushes. 17,18 However, PEGylation can also reduce NP uptake by the targeted cells and is potentially immunogenic. 17,18 Finally, surface functionalization of these NPs is difficult to control and nonuniform. 19 An alternate type of drug delivery NP that is showing promise is the VLP. 20 VLPs are self-assembled, homogeneous NPs derived from the coat proteins of viral capsids. They typically lack their natural genome and are therefore noninfectious. VLPs are an emerging class of targeted delivery vehicles with the potential to overcome the limitations of other NPs. 7,20 In recent years, several groups have shown that VLPs can pack and deliver therapeutic cargo such as chemotherapeutic drugs, siRNA, RNA aptamers, proteins, and peptides. [21][22][23][24][25][26][27] However, there are still challenges when using VLPs. Similar to other NPs, avoiding phagocyte-mediated clearance is a major challenge, even with PEGylated VLPs. 22,28 In addition, VLP stability can also be an issue. 29 Lastly, recent research has shown that ellipsoid NPs are able to extravasate from the blood vessel more effectively than spherical NPs. 30 This ellipsoid shape is possible for conventional polymeric NPs, but is not feasible for icosahedral VLPs. However, the capability of VLP surface modification allows a variety of functional ligands to be added with the potential to address these limitations. By displaying multiple ligands with high affinity for the tight junctions between endothelial cells, VLPs may be able to efficiently extravasate from the vasculature of the blood vessels.

| C HAL LE NG ES T O TARG ETE D D E L I V E R Y U S I N G N P s
As mentioned previously, targeted drug delivery by NPs must overcome multiple challenges (Table 1). 31 The ability to overcome these challenges must either be intrinsic or be imparted as the VLPs are prepared by: (a) loading the cargo inside the NP, and (b) functionalizing the NP to deliver its cargo primarily to the intended cells. While in the bloodstream and the interstitial space, the NP must remain stable, retain its cargo, and avoid nonspecific engulfment by phagocytes.
Additionally, after extravasation into the extravascular tissue, the NP must specifically target the intended cells while avoiding other healthy   Table 2 summarizes the properties of these VLPs.

| Animal virus-based VLPs
The hepatitis B virus is comprised of an internal protein capsid and a lipid envelope containing other proteins. Two different VLPs can be produced from the virus, using either the core antigen that forms the internal capsid or the surface antigen that spontaneously combines with lipids to form NPs. We will focus on the VLP derived from the hepaitis B core (HBVc) antigen, which is formed from 240 copies of a single protein. 37 These proteins first form dimers, which then assemble with pentameric or pseudo-hexameric junctions in a T 5 4 icosahedral geometry. 37,38,40,48 The VLP has been produced using multiple technologies including Escherichia coli cytosolic accumulation and cell-free protein synthesis. 37,38 The assembled VLPs are typically purified using size-exclusion chromatography or differential centrifugation. 37,38,49 Individual coat proteins have been subsequently obtained by disassembling the VLPs with urea, which allows simultaneous cargo loading and VLP re-assembly. 37,38,49 Unpublished data from the Swartz group indicates that coat proteins with hexahistidine extensions can also be directly purified using immobilized metal affinity chromatography. Purifying the individual coat proteins allows control over cargo loading during VLP assembly, which in the case of HBVc is achieved by increasing the salt concentration to trigger spontaneous self-assembly mediated primarily by hydrophobic interactions. 37 39,43,57 P22 VLPs have been purified using size-exclusion chromatography or differential centrifugation and can also be disassembled using acid to obtain the coat proteins. 41,50,55,56 Addition of scaffold proteins is required to reassemble the P22 VLP, but these can subsequently be removed. 41 These bacteriophage-derived VLPs differ from HBVc VLPs mainly in the assembly stimulus, using additional biomolecules (RNA or proteins) to initiate self-assembly instead of increasing the salt concentration.

| Plant virus-based VLPs
The final two commonly-used VLPs to be discussed are derived from plant viruses that infect the cowpea leaf: cowpea chlorotic mottle virus (CCMV) and cowpea mosaic virus (CPMV). Neither virus has a lipid envelope. Both VLPs assemble with icosahedral geometry. 46,47,[58][59][60] The CCMV VLPs are formed from 90 homodimers and can be produced in E. coli or yeast. 46,61 They have been purified using sizeexclusion chromatography or immobilized metal affinity chromatography, using coat proteins with hexahistidine extensions. 46,62 Dimers can be obtained by dialyzing the assembled VLPs against 0.5 M CaCl 2 or by purifying hexahistidine tagged dimers directly. 47,62 Combining the dimers with RNA in a 1:6 mass ratio and lowering the pH to 4-5 induces self-assembly. 47,62-64 CPMV, on the other hand, is formed from 60 copies of the VP60 protein which must first be proteolyzed into the L and S coat proteins (60 copies of each). 59 Unfortunately, the VLP cannot be produced using E. coli or yeast; insect cells or plants must be used. 58,59 The VLPs have been purified using differential centrifugation, but the coat proteins cannot yet be obtained in usable quantities. 58,59 The inability to produce the VLP in E. coli or obtain purified coat proteins adds another challenge for targeted drug delivery; however, CPMV has been actively evaluated for therapeutic use due to the ability to easily display ligands on its surface and load cargo through association with its genome.

| D E S IG N C O N S ID E RA T I O N S I N D E V E LOP I N G V LP s F O R T A R G E T E D D E LI V E R Y
Because of their precise and repeated structures and relatively large cargo capacities, VLPs have many advantages over other types of NPs.
Since they are expressed biologically and formed from multiple copies of the same protein, the VLPs are highly uniform and are easily expressed in bacteria (with some exceptions, such as CPMV  66,[72][73][74] VLPs are much less toxic for parenteral administration than metal NPs, more stable than liposomes, and more uniform than polymer NPs. [75][76][77] Although a significant amount of work is still required to develop the VLPs as delivery vehicles, the current progress shows a great deal of promise.

| Surface functionalization
As discussed, a delivery vehicle must provide several different functionalities. For several of these attributes, the VLP surface must be extensively modified with various biomolecules. These ligands can provide specific cellular targeting, reduce immune responses, and potentially facilitate extravasation. Most approaches require covalent attachment, however, P22 can display ligands through noncovalent interactions. 78 Covalent methods take advantage of either native or nonnatural reactive amino acids ( Figure 2), though genetic fusions to the primary amino acid sequence can also be used to display inserted peptides or proteins. [67][68][69][70][71] Table 3 provides a summary of these surface modifications, including specific references. Although many of the published surface modifications are aimed at vaccines or other uses not related to drug delivery, the same methods can easily be applied. Furthermore, some published studies have described the attachment of ligands for targeting specific cells or avoiding the immune system. 78,84-86 These will be discussed further in later sections.

| Cysteine-based modifications
Arguably the most commonly used reactive amino acid residue, cysteine, can be presented either naturally or by mutation on the VLP surface. Because of its free sulfhydryl group, cysteine will readily and spontaneously form disulfide bonds with other sulfhydryl-containing ligands under oxidative conditions. However, the disulfide bond is also easily reduced and may not be ideal for surface attachments. Alternatively, a series of compounds based on maleimide readily and irreversibly form thioether linkages with cysteine residues at a pH between 6.5

| Lysine-based modifications
Another common amino acid residue that is easily modified is lysine because of its primary amine. Using reactions termed n-hydroxysuccinimide (NHS) ester reactions (because NHS is released as part of the reaction), amide bonds are formed at surface-exposed lysine residues.
The reaction occurs spontaneously between pH 7.2 and 9. This attachment chemistry has been used to display transferrin on MS2, which may allow the VLP to transcytose the blood-brain barrier, a develop-

| Aspartate-or glutamate-based modifications
Although not as commonly used, the last class of reactive natural amino acid residues includes the carboxylic acids aspartate and glutamate. Unlike strategies involving cysteine and lysine, coupling to these residues requires multiple steps. First, the carboxylic acid must be activated using 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide hydrochloride (EDC). Once activated, it will react with NHS to form an NHS ester. Now that the carboxylic acid side-chain has essentially become an NHS ester, previously described with regard to lysine modifications, we can use a ligand with an exposed primary amine to form a stable amide bond. This strategy has been used primarily with CCMV to display peptides or fluorescent probes. 72

| Nonnatural amino acid-based modifications
Beyond the 20 natural amino acids, many nonnatural amino acids have been used for site-specific protein conjugation reactions. The two nonnatural amino acids most frequently incorporated into VLP coat proteins are azidohomoalanine (AHA) and p-amino-phenylalanine (pAF).
These amino acids are incorporated into proteins in two ways: global methionine replacement and amber stop codon suppression. Because AHA is very similar to methionine, AHA will be incorporated at each AUG codon if the methionine supply is rate-limiting; this is termed global methionine replacement. 79 Bacteria auxotrophic for methionine or cell-free protein synthesis can be used to limit methionine availability. 38 The protein yield using global methionine replacement can be rather high from optimized procedures, but this approach will not work

| Genetic modifications
The final covalent attachment method we will discuss is genetic modification, in which the gene for the desired surface ligand is fused to the gene for the coat protein of the VLP. While the added peptide can inhibit protein folding as well as VLP assembly, this approach has been shown to work for most VLPs, but not CCMV. [67][68][69][70][71] Most of the work has fused proteins to either termini of the coat protein, but the HBVc VLP has also been shown to accept heterologous protein domains within the sequence of the coat protein itself. 67 As shown in Figure 1, the HBVc VLP possesses 120 "spikes" on its surface. Protein domains have been inserted such that they are displayed at each spike, allowing optimal surface presentation. These genetic fusion methods have been used to display various heterologous proteins including antibody fragments for specific cellular targeting. 67

| Affinity-based noncovalent modifications
P22 is unique compared to the other VLPs because of the existence of the decoration (or "dec") protein. This protein has high affinity for the surface of P22. 78 By fusing ligands to the "dec" protein, an affinitybased noncovalent system was developed for surface display on the P22 VLP without requiring alteration of the coat proteins. This approach was used to display CD40L, derived from T cells, and the CD47 "self-peptide," developed by the Discher lab, which shows promise in avoiding phagocyte engulfment. 106

| Efficient cargo loading and retention
VLPs have been used to load a range of molecules, including small molecules (chemotherapeutics, fluorescent probes, polymers), nucleic acids, peptides, proteins, and even other NPs. 21 The approaches include both covalent and noncovalent methods. Noncovalent methods are ideal as they do not require modification of the cargo, however, covalent methods typically have the advantage of more efficient encapsidation and retention of the cargo. As with surface modifications, most covalent methods for cargo loading take advantage of reactive amino acids and use the same chemistries described above (Figure 2), though some use genetic fusions to the primary amino acid sequence. 25,83,107 Both methods will be discussed for the different types of cargo. See Table 4 for a list of cargo and loading methods, including specific references. shown to covalently conjugate methacrylate to the coat proteins using the "click" reaction and encapsidate them during VLP assembly. 22,79 Lastly, the P22 VLP has been shown to covalently load nickel ions (through iodo-phen linkages) and derivatives of biotin, fluorescein, and gadopentetic acid (through maleimide-based initiators) by conjugating to interior cysteines. 45,124,125 The plant virus-based VLPs are unique compared to the others because they allow noncovalent loading of small molecules. CCMV has been shown to load polystyrene sulfonate through noncovalent electrostatic interactions between the cargo and the coat proteins. 98,126 CPMV has been used to covalently load chemotherapeutics (doxorubicin) through conjugation to aspartates and glutamates (EDC/NHS reactions followed by esterification) and maleimide derivatives of fluorescent probes by attachment to cysteine residues. 108,109 Additionally, fluorescent probes and an antibiotic (proflavin) have been noncovalently incorporated within CPMV with the small molecules electrostatically adsorbed by CPMV's RNA genome. 110 Although these molecules may not all be therapeutically relevant,

| Small molecules
the results indicate that the methods can successfully load small molecules into many of the VLPs and the approaches can easily be extended to load other small molecule drugs. However, the fact that the only noncovalent loading of small molecules uses a polymerized cargo (that is bigger than most chemotherapeutics) or adsorbs the molecule within nucleic acids shows how difficult it is to load and retain these small cargoes. This is due to the presence of pores throughout the VLP structures, as seen in Figure 1. The development of nonporous VLPs would allow more efficient noncovalent loading and retention of drugs and will be beneficial for future drug delivery strategies using VLPs.

| Nucleic acids
MS2 VLPs are particularly suited to loading RNA. They require a short stem-loop RNA hairpin, which is typically part of their genomic RNA, to assemble into capsids. 42

| Peptides and proteins
There are four main ways to load peptides or proteins into VLPs: (a) fusing the peptide or protein sequence to the amino acid sequence of the coat protein; (b) conjugating the peptide or protein to the genome; (c) engineering electrostatic interactions between the cargo and the coat protein; and (d) passive encapsidation. The first method loads  load proteins and peptides by genetically fusing them to the scaffold protein, which in these cases is not removed from the VLP after assembly. 24,25,117,120,122,123,127 CCMV loading has been accomplished using passive encapsidation, genetic fusions, and leucine zippers added to both the cargo and the coat proteins. 46,62,83,128 Given the difficulty in production and purification of CPMV, it is not surprising that it has not been used to load proteins yet.

| Nanoparticles
A significant body of work has studied the use of VLPs for the development of improved contrast agents. By loading the standard NP-based contrast agents within VLPs, the new NPs gain improved relaxivities which then give higher resolution images. Additionally, if the VLPs are further modified to target specific cells, the signal-to-noise ratio is increased even further giving clear images of, for example, tumors. To that end, HBVc and CPMV VLPs have been loaded with iron oxide NPs through coordination to the coat proteins or through passive encapsidation. 49,129 CPMV has also been shown to load iron oxide and gadolinium NPs through coordination to the genomic RNA. 103,104 CCMV has been used to load gadolinium derivatives through electrostatic interactions with the coat proteins or "click" chemistry. 104,130 Lastly, unrelated to MRI, MS2 was loaded with quantum dot 585 for particle tracking. 21 While not immediately therapeutically relevant, using these VLPs for diagnostics could also greatly improve patient quality-of-life by detecting the disease at an earlier stage and more accurately assessing therapeutic efficacy. Furthermore, iron oxide NPs have the possibility of being used for radio frequency ablation to actively destroy targeted tumor cells. 131

| NP uniformity and stability
Unlike the metal-based, liposomal, and polymer-based NPs, VLPs are highly uniform. VLPs, produced with an exact number of coat proteins and arranged in a consistent geometry, will have significantly lower lotto-lot variability and identical cargo release profiles. Additionally, once inside the targeted cells, the VLPs should degrade and release all of the therapeutic cargo at once-unlike polymer NPs which slowly degrade and release the cargo over time. 75 While slow cargo release may occasionally be beneficial, immediate release is likely to be more effective in most cases and especially for cancer treatment.
At the same time, protein-based design means that the VLPs are not as stable as polymer NPs. Fortunately, this drawback is known and has been studied in the hopes of making better VLPs. These studies focused on HBVc, MS2, and Qb VLPs. The HBVc VLP forms intradimer disulfide bonds that stabilize the 120 dimers, and Qb forms disulfide bonds that link the pentameric and hexameric subunits at the 5-and 3fold axes of symmetry. 29 A mutant MS2 VLP was also designed to form disulfide bonds within the pentamers and hexamers similar to Qb. 29,132 Upon formation of the disulfide bond networks within these VLPs, the dissociation temperatures increased: HBVc from 72-93 to 97, MS2 from 55-70 to 73, and Qb from 40 to 85-1008C. Furthermore, a mutant HBVc designed with an additional 240 disulfide bonds that covalently link every coat protein was engineered and shown to be stable in PBS and over multiple freeze/thaw cycles, but to disassemble in reducing conditions mimicking the cytosol. 40 This mutant VLP shows great promise for use as a delivery vehicle.

| Pharmacokinetics and pharmacodynamics
Although there have not been in vivo biodistribution studies for HBVc and P22 VLPs to our knowledge, in-depth studies have been performed for MS2, Qb, CCMV, and CPMV. We focused on studies using intravenous administration into mice or rats as model systems, which are the systems likely to be studied for initial evaluation of VLP-based targeted therapeutics.
The distribution of MS2 VLPs, labeled internally with 64 Cu or 18 F, was determined in mice at 24 hr and rats at 3 hr after intravenous administration. In both cases, MS2 accumulated primarily in the liver and the spleen. 114,121 PEGylation of MS2 was also studied since PEG has been shown to act as a "stealth agent" to avoid immune clearance. 121 PEGylated MS2 VLPs behaved similarly, except retention in the spleen was significantly reduced. 121 This ability to avoid the immune system is extremely valuable as it will likely increase the effective dose that reaches the targeted tissue. Furthermore, work has shown that the CD47 ectodomain or the CD47 "self-peptide," which has been displayed on VLPs, can also be used to avoid the immune system. 106 Qb, labeled externally with gadolinium, was also studied in mice at 4-5 hr after intravenous administration. 97 Qb VLPs accumulated in the liver, but unlike MS2, accumulated at lower levels in the spleen. 97 The biodistribution of the plant virus-based VLPs, CCMV, and CPMV, intravenously injected in mice at various times, are mostly similar. They primarily accumulate in the liver, spleen, kidney, and GI tract. 61,103,[133][134][135] CCMV, labeled with 125I, also showed significant retention by the thyroid, probably due to the iodine. 61 PEGylation of CPMV VLPs greatly reduced accumulation in the liver and spleen, which suggests CCMV and CPMV could also benefit from the CD47 ectodomain displayed on the surface to avoid the immune system. 106,134 Because developing VLP-based targeted therapies for cancer is a primary application, biodistribution studies in mice possessing tumor xenografts were also conducted. MS2 or PEGylated CPMV VLPs were injected intravenously and partially accumulated in the tumors after 24 hr. This was hypothesized to be because of the EPR effect. 121,135 We suggest that the selective accumulation in these tumors could be greatly improved using cellular targeting ligands displayed on the VLPs, which was described previously, in addition to PEG or the CD47 ectodomain to avoid phagocyte engulfment. 106

| Specific cellular targeting and cargo delivery
While many different cell targeting ligands have been evaluated, ranging from glycans to specific receptor-ligands such as folate and transferrin, the most common targeting ligand is the antibody fragment, although recently RNA and DNA aptamers have been used more frequently. [136][137][138][139] Most research has focused on developing ligands to was also shown to selectively internalize into HeLa cells through receptor-mediated endocytosis and to deliver functional siRNA. 66 Moreover, the MS2 surface has been functionalized with a peptide (SP94) that has high affinity to human hepatocellular carcinoma cells. 21 These SP94-MS2 VLPs delivered their cargo, ricin toxin A-chain, to the targeted cells and specifically killed those cells without affecting the control cells. 21 Antibody fragments also have been displayed on the MS2 surface, although they have not been tested using cell models. 79 Notably, the M.G. Finn group functionalized Qb with human transferrin and observed cellular uptake and internalization of the VLPs through clathrin-mediated endocytosis in BSC1 cells. 141 Furthermore, they displayed glycan ligands on the Qb surface for specific targeting of cells expressing human CD22 receptors. 85 Those VLPs were then loaded with either green fluorescent protein or porphyrin (for photodynamic therapy) and selectively delivered to CHO cells stably expressing human CD22. 58,142 Human epidermal growth factor (EGF) as well as a fluorescent dye were displayed on the Qb surface, and those functionalized VLPs induced autophosphorylation of the EGF receptor and apoptosis of A431 cells. 69 In addition, as with MS2, antibody fragments have been displayed on Qb, though no cell targeting data have been reported. 79 Although there has not been a specific targeting study using CCMV, CCMV VLPs containing EYFP RNA were transfected into mammalian BHK cells. 143 Those VLPs were shown to protect the RNA cargo from RNases, and EYFP expression was observed in the BHK cells. 143 The Finn group displayed folic acid on CPMV, and showed the specific binding and endocytosis of the functionalized CPMV VLPs by KB cells expressing folic acid receptors. 86 They also produced fluorescent dye-labeled CPMV displaying cyclic RGD ligands to target specific integrins, and those VLPs were selectively endocytosed by several different cells overexpressing the integrins (SW480, A549, and HeLa  Table 1) cancer cells and HEK293 cells). 96 Although lacking actual targeting data, the Finn group also displayed transferrin on CPMV. 144 In addition, CPMV was functionalized with intron 8, a receptor-binding module derived from Herstatin, to target HER2 receptors. 93 For tumor imaging, NIR dye-labeled CPMV VLPs were also conjugated to a bombesin analog, and their uptake by PC-3 prostate cancer cells was observed. 58 Tumor homing was further demonstrated using human prostate tumor xenografts on the chicken chorioallantoic membrane model. 58 Lastly, CPMV was functionalized with a fluorescent peptide and a VEGFR-1 specific peptide, F56, to target endothelial cells. 81  peptide, P22 has also been functionalized with the HIV-Tat peptide, and CPMV has been functionalized with arginine-rich R5 peptides. 21,24,80,90,95,102 One proposed mechanism of cationic cellpenetrating peptides (HIV-Tat and the arginine-rich R5 peptides) is through a direct electrostatic interaction with the negatively charged phospholipids that form the endosomal membrane. This is postulated to result in membrane destabilization and endosome lysis. 31,145 Cellpenetrating peptides containing protonatable secondary and/or tertiary amine groups (histidine-rich H5WYG peptide) can absorb protons across the endosomal membrane, resulting in a swelling from an influx of water and/or ions and leading to rupture of the endosomal vesicle. This is known as the "proton sponge effect." 31 Although there are some working examples of these peptides, further research is needed.

| C ONC LUSION S A ND PE RSPE CTIVES
Although VLP-based targeted drug delivery remains a nascent technology that requires further studies to prove its clinical efficacy, significant progress has been made. Many of the initial disadvantages of using VLPs have been remedied, as shown in Table 1, and the previous studies explored in this article have laid excellent groundwork for addressing the remaining challenges. Although each VLP has advantages and disadvantages relative to each other, we believe that HBVc, Qb, and MS2 show the most promise. The advantages of these VLPs are that they: 1. can be produced using cell-free protein synthesis 29,38 2. can load small molecules, nucleic acids, and proteins 21,22,52,65,113,126 3. can be stabilized with disulfide bonds 29,40,132 4. can incorporate nonnatural amino acids for ease of surface functionalization through the "click" reaction 67,79 5. can be functionalized to display antibody fragments for specific cellular targeting 67,79 6. can be functionalized to display PEG to avoid the immune system (not shown for HBVc VLPs) 79 7. will disassemble in the reducing conditions of the cytosol to release their cargo (not shown for MS2 VLPs) 29,40 Although it has not been experimentally proven, the disulfide bonded mutant of MS2 should behave similarly to Qb and the disulfide bonded mutant of HBVc and disassemble in cytosolic conditions. 29,40,132 Likewise, although HBVc has not been functionalized with PEG to our knowledge, the ease of nonnatural amino acid presentation and "click" conjugation will facilitate such experiments. 79,101 Additionally, all three can be functionalized with the CD47 ectodomain or the CD47 "self-peptide" for potential avoidance of phagocytic clearance. 106 MS2 and Qb have also been functionalized with transferrin which may allow transcytosis across the blood-brain barrier, allowing the VLPs to be used for neurological disorders. 66,141 While P22, CCMV, and CPMV do not currently have the same advantages as the other VLPs, we believe the same technology can be applied for them in the future.
It is also suggested that additional work focus on fully overcoming the challenges listed in Table 1 reducing off-target organ accumulation, mainly in the liver, kidney, and spleen. We suggest that the advances summarized here, and the suggested future directions, indicate a bright and important future for VLP-mediated targeted drug delivery.