Current state of in vivo panning technologies: designing specificity and affinity into the future of drug targeting

Abstract

Targeting ligands are used in drug delivery to improve drug distribution to desired cells or tissues and to facilitate cellular entry. In vivo biopanning, whereby billions of potential ligand sequences are screened in biologically-relevant and complex conditions, is a powerful method for identification of novel target ligands. This tool has impacted drug delivery technologies and expanded our arsenal of therapeutics and diagnostics. Within this review we will discuss current in vivo panning technologies and ways that these technologies can be improved to advance next-generation drug delivery strategies.

Graphical Abstract

1. Introduction

1.1. Biopanning: A platform to identify novel target ligands

Since its inception in 1985, biopanning has proven to be a powerful method for enhancing the rapid discovery of novel biological ligands for a variety of applications including, but not limited to, drug discovery, delivery and molecular biology[1–5]. The initial biopanning principle harnessed recombinant DNA technology[6] and Sanger sequencing[7] to engineer and develop biological platforms to express and display recombinant peptide-libraries in biological systems, which allowed for high throughput screening of novel ligands to a specified target of interest[5, 8–10]. Since that time, techniques such as directed evolution and next generation sequencing have increased development of high-affinity ligands[5, 11–13]. Furthermore, recombinant libraries have expanded from peptides to include antibodies, RNA, DNA and viral capsids—all providing a fresh and comprehensive approach to identify novel ligands[4, 14–16].

1.2. Targeted Drug Delivery: A need for novel ligand development

Targeted delivery of drugs involves technologies that increase distribution of pharmacologic agents to desired locations in the body. Delivery vehicles can mitigate off-target drug actions that result in significant and dose-limited side effects[17–19]. There are two main strategies for targeted drug delivery: passive or active targeting[18, 20, 21]. Passive targeting strategies rely on the physicochemical properties (e.g. charge, size) of a carrier and the body’s physiology to affect drug distribution[20, 21]. In contrast, active targeting employs specific molecular interactions between the drug carrier and target tissues or cells for delivery[18, 20]. Despite the theoretical advantage of targeting, few FDA approved drug-formulations utilize an active targeting approach (Table 1).

Table 1:

Pharmaceutical products with active drug targeting that are either FDA approved or in late-stage clinical trials

Biopanning method	Native Physiological Function	Clinical Application(s)	Target(s)	Trade Names	Development
Phage Display (Antibody)	Immunoglobulin, pathogen neutralization	Cancer	CD33, CD30, HER2	Mylotarg (Gemtuzumab ozogamicin), Adcetris (Brentuximab vedotin), Kadcyla (Trastuzumab emtansine)	Antibody Conjugated to Chemotherapeutic Drug (Pfizer/Wyeth, Seattle Genetics/Millennium/Takeda, Genentech and Roche)
Phage Display (Kunitz Domain: 90 AA modifications in-between beta sheets)	Protease inhibitor (e.g. APP, TFPI)	Hereditary Angioedema	Kallirkein	Kalbitor (Ecallantide)	Macromolecular active drug discovered through phage display. (Shire)
Phage Display (Monobody: 90 AA modifications in-between beta sheets)	Fibrinectin type III domain	Glioblastoma	VEGFR2	Angiocept (Pegdinetanib)	Macromolecular active drug discovered through phage display. (Bristol-Myers Squibb)
Phage Display (Darpins: 24 ankyrin repeats)	Ankyrin, a membrane protein that anchors proteins to the actin cytoskeleton	Macular degeneration	VEGF-A	Abicipar	Macromolecular active drug discovered through phage display. (Molecular Partners AG)
Phage Display (Peptides, 7-12 residues)	Biologically active small molecules	Cancer	SSTR2, αν integrins, CD13	Lutathera (Lutetium- Lu77-Dotatate), Cilengitide (RGD alone)/ Fluciclatide (RGD in combination with F18), Zafiride (NGR-TNFα)	Peptide radioisotope conjugate, peptide alone/peptide radioisotope conjugate, peptide cytokine conjugate. (Advanced Accelerator Applications/Novartis, Merck/GM Healthcare, Adis Insight)

The most commercially successful class of targeted delivery methods are antibody-drug conjugates or peptide/protein-based systems, wherein cytotoxic drugs are conjugated to a targeting protein via cleavable linkers (Table 1)[19, 22, 23]. These therapeutics have had clinical success due to their ability to recognize target cells with high affinity, to limit off-target toxicity, and to circulate for long periods. However, many protein-based targeting platforms (e.g. antibodies) are limited by their difficulty transporting agents beyond the vasculature due to their size and high binding affinity[24–26]. Therefore, significant interest exists for the development of small protein/peptides and nucleic acid structures that have greater accessibility and tissue penetration properties. Moreover, these small peptide and nucleic acid agents can be designed for oral and transdermal delivery to increases disease state accessibility and improve delivery outcomes[27–30]. Thus, biopanning methods for identifying a variety of targeting ligands are vital for developing and designing effective active targeting-carriers for a broad range of disease applications and administration routes.

2. In vivo biopanning platforms for developing novel targeting ligands

The immune system has evolved a tremendously complex and efficient means to produce antibodies that recognize new and recurrent foreign antigens either by expanding clonal populations of B-cells or by processing and presenting new antigens for antibody production. Antibodies against the new infectious agent are then selected as part of the adaptive immune response and improved through affinity maturation[31]. Thus, the diversity of unique antibodies in humans is in the billions[31, 32]. Biopanning is a process that mimics antibody selection by screening large libraries for potential binding ligands and then engineering and refining the ligands to increase binding affinity and targeting[8, 33, 34]. Through biopanning, novel ligands can be identified for target receptors for use as molecular targeting agents in drug delivery. In this review, we will discuss the current active clinically approved biopanned agents, the literature that outlines in vivo biopanning technologies, and strategies to design novel panning platforms to improve ligand discovery and validation.

2.1. In vivo biopanning as a tool to discover novel ligands

The success of any biopanning strategy relies heavily on the design and stringency of a selection scheme to maximize stringency and eliminate non-specific binders. All biopanning paradigms utilize a ligand library that contains an inherent coding method (e.g. affiliated nucleic acid or chromophores) for moiety identification and reproduction after recovery of potential binders. Libraries are panned against a series or combination of positive- or negative-selection “schemes” to enrich the capture of specific binders over weak non-specific binders[35, 36].For example, bacteriophage libraries can be incubated initially with a non-specific cell-type to remove off-target phage-binding (i.e. negative selection strategies). After removal of non-specific binders, the recovered phage population is then incubated with the target cell-type and washed to ensure stringent selection of high-affinity binders (i.e. positive selection)[9, 37, 38]. Multiple cycles and combinations of positive- or negative-selection can be performed to enrich for binding-phage further. After recovery of the high-affinity binders, the bacteriophage DNA is sequenced to identify each binding-motif[37, 39]. Validation of specific binders from the recovered pool is generally the most challenging and time-consuming step in biopanning[11, 13, 40, 41]. In in vivo biopanning, negative selection occurs naturally and with significant stringency as off-target tissue and protein interactions eliminate non-specific ligands to enrich the recovery of target-specific ligands and cellular receptors[42–44] (Figure 1). In theory, in vivo panning harnesses the heterogeneity of an intact biological system and improves the specificity of the identified ligand[43, 44]. Over the last 20 years, several in vivo panning display technologies and strategies have been developed. This section summarizes and highlights lessons learned.

Figure 1.

In vivo phage panning schematic

2.2. In vivo Phage Display

The success of in vivo biopanning display is dependent on the library diversity and the stability of the panning platform. Increases in diversity increases overall hit rates and enhances novel outputs. Enhanced stability increases hit recovery and allows diversification of biological applications. We will outline within this review the current panning platforms each with their own current diversification and stability issues (Table 2).

Table 2:

Current in vivo Panning Platforms

Panning platform	Library type	Diversity	Typical number of selection rounds	Size	In vivo stability
Bacteriophage	Peptide	~10e9 (M13) to ~10e11 (T7)	3–5 rounds	60 nm for T7, 880 nm x 9 nm for M13	~4 hrs for unmodified and ~20 mins for peptide displayed M13[45], ~10 mins T7[46], increases after modification to reduce immune recognition [47]
	Antibody	~10e8	2–4 rounds
Aptamer libraries	RNA and DNA	10e10- to 10e60 (depends on length, 4^n)	> 15 rounds (without PEG modification)	~10 kDa [48]	Low without modification
	Somamer	N/A (added post-SELEX)	N/A (added post-SELEX)	Depends on modification. I.e. PEG --> 40 kDa, NFkB-aptamer ~110 kDa[48]	High, but depends on modification
Virus	Virus	10e6 – 10e8	2–5 rounds	25 nm diam.	High
	Displayed peptide	Theoretical 10e8	3–5 rounds	~28 nm diam.	High but slightly reduced compared to unmodified[49]

2.2.1. Bacteriophage library platforms

Bacteriophage, or phage, libraries are generated through integration of a variable peptide sequences or proteins/antibodies onto a coat protein of bacteriophages[4, 50]. The phage platform incorporates both the displayed motif and the DNA encoding information into a single compact construct, allowing for simultaneous panning and subsequent sequencing of massively diverse libraries[8, 9]. For peptide libraries, the most commonly used phage vector is the M13 phage; the foreign peptide sequence, generally 7–12 amino acids in length, can be fused to minor pIII (3–5 copies of the peptide sequence) or major pVIII (2700 copies of the peptide) coat proteins[51]. It was discovered early on in phage biopanning that fusion to the pVIII reduces specificity of the isolated peptide sequence due to the selection of low affinity peptide binders that benefit from the increased avidity that results from high copy display [51]. In contrast, libraries displayed on pIII at 3–5 copies per phage generally yield peptide binders with higher affinity. Thus, commercial libraries are generally constructed with a large diversity (~10⁹) of peptide sequences fused to the pIII coat protein of a filamentous M13 phage. The pIII display technology has been used to fuse antibody and small protein fragments [3, 5, 9]. There are a few examples of in vivo selection phage displaying antibody or protein fragments[52–55]. Other phage platforms include T7, T4 and λ bacteriophage; however, M13 phage is most commonly used due to its single coat fusion site, and ease of purification[51]. However, it has been demonstrated that the complex purification systems with T7- and T4-lytic phage allow for increased library diversity[56]. Moreover, T7 and T4-phage are small (~60 nm diameter) compared to M13 (~900 nm, length) and have been shown to have greater tissue penetration in vivo, enhancing delivery to M13-inaccessible sites such as the liver and brain [47, 57]. Lytic phage is also thought to have increased retention and reduced degradation in lysosomes, which might increase sequence yields in in vivo panning[56].. Despite these advantages the majority of in vivo panning studies have utilized recombinant, commercially available M13 peptide phage libraries.

2.2.2. Applications of in vivo phage display in humans

Initial in vivo biopanning efforts utilized phage libraries administered intravascularly and resulted in identification of a wide array of peptide “vascular zip codes” that target vasculature-specific ligands[42, 58] (Table 3). Further, peptide zip codes were shown to target agents directly to an organ or disease-associated vasculature[27, 59–66]. Following the success of biopanning in rodents, a similar phage-display method was applied in human biopanning screens. The first in vivo human panning experiment to map the human vasculature was conducted on a patient declared dead via brain-based determinations and results published in 2002 [67]. Subsequently, similar studies were performed with cancer patients whereby phage was recovered from biopsy specimens after intravenous administration of libraries [68, 69]. These later human studies recovered peptides recognizing targets commonly associated with tumor-related vascular targets such as interleukins, annexin, cathepsins, and prostaglandins[67–69]. Considering the vast amount of data generated from these early human vascular screens, one could suggest that it may be worth revisiting the human in vivo biopanning strategies to take full advantage of next-generation sequencing technologies and bioinformatics tools, as well as other routes of administration.

Table 3.

Vascular targeting ligand zip codes discovered by in vivo phage display.

Target	Disease Target	Sequence	Reference
Atherosclerosis	Vascular accessible atherosclerotic plaque	CLVEAYPGLSVRSC	Houston 2001
Cancer Vasculature	Lung Cancer vasculature	SVSVGMKPSPRP	Lee 2007
Obesity	Adipose vascular	CKGGRAKDC	Kolonin 2004
Cancer Vasculature	Gastric Cancer Vasculature	CGNSNPKSC	Zhi 2004
Atherosclerosis	Vascular smooth muscle cells	CNIWGVVLSWIGVFPEC	Michon 2002
Cancer Vasculature	Prostate Cancer	IAGLATPGWSHWLAL	Newton 2006
Cancer Vasculature	Breast Cancer	CDCRGDCFC	Arup 1998
Cancer Vasculature	Pancreatic cancer	CRGRRST	Joyce 2003
Cancer Vasculature	Squamous cell carcinoma.	CSRPRRSEC	Hoffman 2003

2.2.3. Limitations to in vivo phage display: tissue distribution and circulation half-life

As discussed, intravenous administration of M13 phage libraries in animals and humans led to the discovery and characterization of peptide zip codes that target organ-specific vasculature and tumor-binding peptides for drug delivery applications[27, 42, 58, 67]. Bacteriophage are inherently stable in biological milieus, a useful trait for productive in vivo selection. Still, in vivo phage-display biopanning methods are limited in the reach of the application to target sites outside of the vasculature or vasculature-leaky disease sites due to limited extravasation and penetration of the M13 phage. While alternative display technologies with smaller size such as covalent display technologies such as polyosome, ribosome or RNA-peptide display, could in theory distribute more broadly into tissues, these agents have not been successfully designed to withstand the harsh conditions of an intact biological system[70, 71]. The smaller T7 phage systems might reach more potential targets in vivo through increased accessibility, but trends in biopanning strategies (away from lytic phage) would suggest other process-related limitations complicate the use of T7- and T4-pahge.

In addition to the size-based tissue exclusion, M13 biopanning is limited by a relatively short circulation half-life (4 hrs), which is reduced significantly for recombinant M13-based libraries (<20 min), likely due to a combination of reduced in vivo stability, increased biological interactions, and increased internalization rate to cells[45, 72]. Furthermore, after internalization, M13 phage are degraded within lysosomes in 30 mins[45, 72]. T7-phage have a faster clearance, with half-lives less than 20 min for both wildtype and peptide-modified phage [46, 73]. Indeed, all phage are recognized readily by the immune system and sequestered within immunological relevant organs[46]. It is worth noting that in animals with a compromised immune system, phage exhibit an increased circulation potential, like due to the reduced recognition of phage through B-cell interactions [46, 72]. Nonetheless, engineering phage or other peptide display technologies to improve the circulation half-life and to limit degradation by mutating immunogenic properties of the macromolecular structure would result in a more robust platform for in vivo selection.

2.2.4. In vivo phage panning from alternative administration routes

As intravenous administration of phage libraries resulted in rich identification of binding motifs within the vasculature, biopanning from alternative administration routes would be expected to provide access to alternative targets. Indeed, non-systemic administration of phage libraries have yielded targeting ligands for a wide variety of diseases[29, 30, 44, 74–77](Table 4). Thus, the route of administration provides access to alternative targets that would normally be restricted by intravenous routes. For example, Wan et al., biopanned for phage that could transport to the brain after applying phage to the nasal epithelium of rats[74]. In a similar fashion, Sellers et al. administered phage by peripheral injection into the gastrocnemius, of mice, to identify ligands that targeted axonal import for delivery into the spinal cord 24 hours post-injection[44]. In these studies, peripheral administration with recovery at distal sites at time that eclipse several circulation half-lives (i.e. ≥ 1 hour), provides selection pressure for recombinant phage-sequences that demonstrate strong targeting potential.

Table 4.

Alternative paradigms for non-systemic biopanning strategies.

In vivo Phage Panning Delivery Mechanism	Disease Target	Citation
Intranasal	Neurological Disorders	Wan 2009
Intramuscular	Neurological Disorders	Sellers 2016
Transdermal delivery	Diabetes	Chen 2006
Transdermal delivery	Obesity	Lee 2011
Intratracheal	Respiratory Disorders	Wu 2003
Oral	Intestinal Disorders	Duerr 2004
Intraperitoneal	Gastric metastases	Akita 2006

2.3. In vivo viral capsid panning

2.3.1. Adeno-associated virus (AAV) panning

Viral capsids are intrinsically robust particles well-suited to in vivo library selection. Viral particles exhibit considerable stability in vivo, have diameters on the order of nanometers allowing for broad circulation, are easily cloned and produced, and have naturally evolved to encapsulate genetic information to allow for high-throughput viral clone identification[15, 78–80]. In vivo viral capsid panning has primarily been performed with adeno-associated viruses (AAVs), which are icosahedral, nonenveloped viruses belonging to the Parvoviridae family. AAVs package single-stranded DNA and are 20–25 nm in diameter allowing for effective transport and delivery. Recombinant AAVs (rAAVs) are of particular interest as vehicles for gene therapy due to their favorable safety profiles[81]. Moreover, rAAVs can be engineered to eliminate most viral DNA, as well as to target certain tissues and cell types[82]. While much of biopanning has focused on AAVs, it is worth noting that CCMV, CPMV, and TMV are attractive viral particles because they do not naturally infect humans, their surfaces may be functionalized with peptides or small molecules, and their cargo may include imaging contrast agents or small molecule drugs[79, 83, 84]. Thus, future design of panning strategies might benefit from re-engineering these vector platforms.

Recombinant AAV particles with increased tissue tropisms can be engineered by generating highly diverse capsid libraries (>10⁶ – 10⁸ surface protein variants) and performing directed evolution in vivo to select for variants with desired biodistribution properties. There are several different approaches to generating viral capsid libraries as discussed below. As with phage display, in vivo AAV panning involves injecting the capsid library into laboratory animals and recovering the viral DNA from harvested tissues. Library variants that are recovered at a high frequency in tissues of interest compared to off-target tissues are enriched and re-injected for another round of in vivo panning. In addition to selecting for tissue tropism, rAAV biopanning has been used to discover clones that exhibit tissue-specific transduction by utilizing a Cre-based reporter to reveal tissue-specific increases in rAAV reporter protein expression[85–87]. In this strategy, rAAV particles carrying Cre-sensitive fluorescent reporter genes are subjected to in vivo selection in transgenic mice that express Cre recombinase in specific cell-types. A higher fluorescent signal correlates to increased transduction efficiency of a particular rAAV clone in a particular cell type. After multiple rounds of in vivo selection, rAAV variants with optimized characteristics including tissue tropism and transduction efficiency may be identified and further analyzed.

2.3.2. Directed evolution of randomly, semi-randomly, and rationally designed AAV libraries

Directed evolution of viral capsids for tissue targeting and transduction potency involves imposing selective pressures on large, random or semi-random capsid libraries[88]. This tool allows for rapid affinity and selectivity improvements, enhancing AAV-mediated delivery outcomes. One successful library approach utilizing this technique, has been to randomly shuffle Cap (capsid) genes from different AAV serotypes[89]. DNA shuffling has yielded rAAV particles with up to 13-fold increased gene transfer in heart muscle cells compared to preexisting AAVs, and particles with over 400-fold increased resistance to neutralizing antibodies[89, 90]. Libraries may also be generated in a semi-random fashion by targeted mutagenesis of Cap genes. For example, a rAAV with up to 15-fold increased transduction of astrocytes was developed by mutagenizing the capsid at loop regions performing in vivo library selection[91]. Peptide insertion, whereby peptides of random lengths and sequences are incorporated into the viral capsid surface, has also been used to engineer improved AAV tropism[16, 92]. Korbelin and colleagues recently used this library approach to engineer pulmonary tissue-specific rAAVs with high gene delivery efficiency[93]. Library diversity is a critical feature in increasing likelihood of successful selection. Thus, another approach to viral capsid panning is to utilize several different types of libraries in the same experiment. By combining four different libraries, a rAAV with 100-fold increased access to projection neurons compared to preexisting AAVs was engineered[85]. Future libraries may combine libraries multiplicatively rather than additively, allowing for even higher theoretical complexity. For example, after randomly shuffling Cap genes, random peptide insertions may be introduced into the same library.

Random library generation approaches yield many rAAV particles with reduced transduction efficiency or non-functional particles, reducing the effective library diversity. To address this limitation, rational design of viral capsid libraries can be performed utilizing known viral capsid structures and functional regions. The Schaffer group recently developed the SCHEMA library approach that was successfully used for in vivo panning[87]. In this approach, the SCHEMA algorithm[94] and recombination as a shortest path (RASPP) method[95] were applied to maximize both library diversity and structural conservation. After generating and screening a library of over 1.6 million rAAV variants, a novel variant was isolated that exhibited 60% transduction efficiency in adult neural stem cells, 24-fold higher GFP expression than AAV9, and an increased ability to evade neutralizing antibodies. This example highlights how rational modifications to libraries to selectively modify diversity can improve delivery and therapeutic outcomes and should be considered in in vivo panning design.

2.4. In vivo Aptamer Panning

2.4.1. Aptamers and aptamer selection

Nucleic acid aptamers are single-stranded oligonucleotides, typically 15–100 base pairs in length, that form three-dimensional structures capable of binding to targets with high specificity and affinity. Aptamer libraries used in biopanning generally contain 10¹³-10¹⁵ unique sequences, consisting of a 20–60 base pair random region flanked by fixed primer regions on the 5’ and 3’ ends. From this starting pool, DNA or RNA aptamers are selected through the iterative process of Systematic Evolution of Ligands by Exponential Enrichment (SELEX), involving cycles of incubation with the target and subsequent amplification and recovery of binding sequences[2, 96]. With successive cycles, high affinity aptamers are enriched until they dominate the library population[14, 36, 97]. The population of aptamers present after selection can be characterized by next generation sequencing (NGS), and individual aptamers can then be synthesized and investigated for binding behavior.

Aptamer ligands offer several advantages over conventional targeting peptides or antibodies. Aptamers can form a variety of complex secondary structures with comparable or superior binding affinity of antibodies (micromolar to picomolar), and they can bind to a wide breadth of targets, such as ions, small molecules, and proteins. Since aptamers are nucleic-based, they exhibit reduced immunogenicity, increased tissue penetration, facile manufacturing, and high shelf-life stability compared to peptides and proteins. Thus, aptamers are an attractive class of targeting agents for drug delivery. However, aptamers are highly susceptible to nuclease degradation and kidney filtration, resulting in limited circulation time and half-life in vivo. Multiple studies have demonstrate that these challenges can be mitigated by engineering chemical modification into the aptamer backbone or side groups, non-natural nucleotide substitutions, or end capping, and reducing susceptibility to endo- and exo-nucleases[98, 99], which have aided the development of in vivo aptamer panning strategies[36].

2.4.2. Library design for in vivo SELEX

There are several considerations involved in the design aptamer libraries for in vivo SELEX, including nucleotide type, library length, chemical modification, and length. Aptamer library lengths are typically between 20–80 base pair, while aptamers short or long random regions can successfully bind, those with longer random regions exhibit greater structure complexity[100] and increased library diversity at the expense of cost for production. Both RNA and DNA aptamer libraries have yielded binding sequences from in vivo panning. RNA libraries tend to increase library diversity and improve panning outcomes but usually at the expensive of stability. Aptamers based on RNA, which is naturally single-stranded have a greater range of three dimensional structures compared with DNA, which can result in a higher binding-affinity [101]. However, RNA aptamers are also more expensive and require an extra step of reverse transcription prior to amplification.

2.4.3. Recent progress in in vivo SELEX

For drug delivery applications, in vivo biopanning has an inherent advantage over in vitro panning by providing the stringent degradation, transport, and complex competitive binding pressures present in vivo during the selection process. In vitro SELEX has been performed successfully against purified target proteins or even whole cells; however, these in vitro selected aptamers may fail to bind to the same target in vivo. This may be attributed to differences in selection conditions, as aptamer conformation and binding has been demonstrated to depend on pH, temperature and ionic conditions[102–105]. To address this shortcoming, in vivo SELEX can be performed in living animal models to identify binding constructs in the context of a representative disease environment.

Mi et al. published the first RNA aptamer identified via in vivo SELEX in 2010[106]. In this work, a random library of 2’-fluoropyrimidine modified RNA sequences was injected intravenously into hepatic colorectal cancer-bearing mice. This chemical modification increases aptamer stability through increased nuclease resistance, thus increasing potential target binding by enhancing aptamer circulation time. Fourteen cycles of SELEX were performed and an aptamer that bound to p68, an RNA helicase, with nanomolar affinity was identified[14, 106]. Since this original publication, this process has been adapted for use with human intrahepatic xenografts to identify RNA aptamers against human helicase DHX9[107].

In vivo SELEX has also been used to evolve aptamers capable of penetrating the blood brain barrier (BBB). Penetration of therapies across the BBB remains a challenge due to tight cell junctions and highly selective transport of molecules. Cheng et al. injected intravenously a 2’-fluoropyrimidine modified RNA library to identify an aptamer that could penetrate the BBB after being, one of which represented 50% of all sequences. Further characterization of the most highly enriched aptamer from round 22 revealed that it internalized by brain-vascular endothelial cells. The authors did note high levels of RNA aptamer sequences in clearance tissues (liver and kidney) which was attributed to the reticuloendothelial system. Additionally, since the in vivo SELEX strategy required over 20 rounds of selection, the authors recommended a combinatorial in vitro and in vivo selection if screening for specific disease phenotypes[108].

Library stability and circulation time can drastically affect selection efficiency. In vivo SELEX was carried out successfully using a polyethylene glycol (PEG)-modified 2’fluoropyrimidine RNA library to identify an aptamer that binds to and inhibits proliferation of non-small-cell lung cancer (NSCLC) cells[109]. Libraries were conjugated with PEG in order to increase aptamer circulation time and reduce renal clearance by increasing molecular weight. A binding aptamer was emerged at cycle 8 of selection and accounted for 90% of the enriched RNA by cycle 11, thus requiring much fewer rounds of selection compared with other in vivo SELEX reports[106, 108]. The authors attributed this early emergence to the addition of PEG within the RNA pool and the increased circulation time, which eliminated non-specific binding[109]. This aptamer was further used to facilitate delivery of a chemotherapeutic drug. In another recent example, Shen and coworkers used a DNA thiophosphate-backbone-substituted[110] library in vivo SELEX in a breast cancer bone metastasis model to identify a DNA aptamer that accumulates in the tumor microenvironment[111]. After 10 rounds of selection from tumor tissue, the top occurring aptamer was tested and shown to bind to cancer and myeloid-derived tumor suppressor cells. Furthermore, aptamer-conjugated liposomal doxorubicin formulations were effective in inhibiting tumor growth compared to non-targeted liposomes[36, 106].

For successful in vivo aptamer panning, it is critical that aptamer libraries are able to infiltrate the target area for selection, extraction, and amplification. However, unmodified aptamers are highly susceptible to nuclease degradation, resulting in a short half-life and circulation time in vivo. Additionally, their small size contributes to their rapid renal filtration. To prevent degradation, aptamer libraries have been modified chemically to replace the 2’ position with a fluoro (F), amino (NH₂), or O-methyl (OCH₃) group, and by capping the 3’ end with inverted thymidine[99, 112]. While chemical modifications can impart improved stability and circulation, chemical modification to the aptamer backbone or side-chains can affect aptamer folding and binding. Nonetheless, these changes can been utilized in-SELEX or post-SELEX biopanning strategies. For the in-SELEX biopanning, sequences with the chosen modifications are used in the starting DNA or RNA library and undergo subsequent rounds of selection. However, modifications are limited to 2’-aminopyrimides, 2’-fluoropyrimidines, 2’-O-methyl nucleotides, and locked nucleic acids (LNA). Mi, Cheng, and Wang all utilized in-SELEX modifications for their in vivo panning strategies[107–109, 111]. Recently, SOMAmer (slow off-rate modificed aptamer) reagents have emerged. They contain modified nucleotides that have conjugated functional moieties, creating stable structures with high nuclease resistance and binding affinities[113]. For the post-SELEX strategy, chosen modifications are implemented within pre-selected aptamers at specific locations during chemical synthesis. However, these modifications can affect aptamer structure and folding and thus impact its binding behavior.

Additionally, unmodified aptamers are within the size range for renal excretion and are thus rapidly excreted by the kidneys. To mitigate this, aptamers are often modified by the addition of a bulky side group, such as PEG or cholesterol. This strategy increased the plasma circulation half-life of Macugen, an aptamer used to treat macular degeneration, from 9 hours to 12 hours[99]. Wang utilized a combinatorial approach of 2’-fluoropyrimide and PEGylation in order to increase in vivo stability[109]. Aptamers can also be multimerized to create multivalent constructs above the renal molecular weight threshold. Borbas et al. designed a tetrameric aptamer against the MUC1 tumor marker. This construct had improved stability, tumor-retention, and pharmacokinetic properties compared to the monovalent construct[114]. As such, a library particle system displaying multiple copies of each aptamer could be another approach for in vivo aptamer panning.

3. Future of panning

In vivo panning for novel ligands has enormous potential for targeted drug delivery. The platform inherently allows for simultaneous selection of molecules that are able to withstand harsh biological environments, and home to targets in a complex and heterogeneous millieu. However, the design of successful in vivo biopanning requires consideration of multiple design principles. Success in panning is dependent on both the diversity and stability of the library. Here we will discuss ways to both enhance these parameters as well as manipulate panning procedures to artificially augment current panning systems.

3.1. Increasing in vivo library stability

Library stability is necessary to recover bound ligands. Maintaining the stability of a library in a harsh biological environment has challenges and in vivo panning requires library exposure to multiple biological milieu, potentially impacting the diversity of the final library at the target site. Novel targets might be lost as a result of degradation. Ultimately the perfect library display platform would be engineered to withstand harsh biological milieu. The conditions a display platform must withstand during biological transport include enzymatic degradation, harsh shear force, dramatic pH changes including acidic endosome/lysosomal conditions following platform internalization into cells, etc. Some groups have begun to address some these challenges in in vitro panning systems. Rangel et al. incorporated a cell penetrating peptide, penetratin, on the on the pVIII major coat protein of M13 phage, enhancing intracellular release for subcellular organelle ligand selection[39]. It’s possible strategies like these could be applied in vivo, allowing the library platform to avoid lysosomal degradation and improve ligand recovery. However, modulation to increase lysosomal stability likely does not eliminate loss of phage simply due to inherent physical transport limitations. Overcoming transport challenges is likely overcome through panning design.

3.1.1. Design of clever panning strategies to improve library recovery.

In order to increase the likelihood of completing a successful in vivo biopanning, careful consideration of the inherent biology of the target and/or disease state must be given to each strategy. The most successful targeting ligands have been discovered by in vivo panning that carefully considered the inherent disease state and a specific route of administration that would allow for increased exposure to the target. Incorporation of the disease state into the panning strategy inherently increases affinity and specificity of a ligand, through the incorporation of heterogeneity of diseased and “normal” target sites. Thoughtful routes of administration to exploit routes of delivery to disease tissue or the intended target could improve ligand selection through enhanced ligand presentation, by decreasing library degradation during transport to the target site, or by improving negative selection by passage through competitive non-specific binding targets. Such examples include Sellers et al. recent neurological panning strategy where we were able to harness biological transport within the peripheral nervous system from muscles to the spinal cord[44]. Through intramuscular injections we were able to isolate and validate a novel peptide that helps to engage this transport pathway (Figure 2). This study highlights how biological processes can be integrated into panning design strategies. While strategies like these allow for increased target exposure it does not enhance the diversity of the starting library nor does it improve ligand affinity.

Figure 2.

Improve panning strategies to take advantage of inherent biology. Here, Sellers et al. capitalized on the transport from muscles to the spinal cord to bypass the blood brain barrier [44]

3.2. Increased in vivo library diversity

In vivo panning success achieved when a ligand with high affinity and specificity to a target site is discovered. While panning strategies can be modified to improve outcomes, in general increasing the number of panned ligands will inherently increase the statistical probability of a hit. As shown in Table 2, current in vivo display platforms have significant differences in absolute diversity. Many researchers have probed these platforms in vitro to increase diversity through synthetic modification, such as artificial nucleic or amino acids, as well as, the introduction of chemical functionality. Such examples include SOMAmers, which add non-natural functional groups to aptamers to more closely mimic protein interactions, and one-bead-one-compound libraries for peptide selection which can include non-natural amino acids which can reverse chirality[113, 115]}. These platforms have yielded high affinity motifs not necessarily found in nature, and can be combined with natural ligand libraries to improve diversity. These diversification techniques have been widely applied to in vitro selection but have not been either optimized or applied for in vivo panning. Increasing the synthetic tool sets could dramatically enhance in vivo panning target identification and outcome. These days, panning products are often assessed through next generation sequencing methods where ligand hits from each panning round are screened for sequence similarities and alignment[11, 116, 117]. After even the first round of in vivo panning, clear trends generally emerge within this analysis. This pattern recognition can be reinvested into subsequent rounds of panning to enhance library affinity, specifying diversity, a technique generally termed directed evolution.

3.2.1. Directed evolution to drive improved library hits.

Directed evolution techniques have been employed in viral and phage display in an in vitro context[5, 87]. These platforms significantly enhance specificity and affinity of ligands identified. However the leap has not yet made a dramatic impact on in vivo panning[5, 38, 87, 116, 118]. In directed evolution, following rounds of selection, patterns emerge. These motifs can be incorporated into subsequent panning rounds and affinity can be honed utilizing a deconvolution approach. The libraries can be generated either through iterative single positional scanning and/or multi-site mutational sublibrary scanning[51]. In both strategies, motif modifications are generally accomplished through site-specific mutagenesis or through random generation whereby motifs are fragmented and reassembled[51]. When individuals are refining library design for subsequent panning rounds, it might also be useful to scan bioinformatic databases to see if the motif identified have sequence homology with known native ligand binders. This could be particularly useful in phage display technology as sequence specificity of clinically translated sequences have been shown through these techniques to hone to three amino acid positions [68, 119, 120]. The beauty of panning is that multiple mutations can be generated in several libraries and be explored in a parallel fashion such that non-specific binders are rapidly removed and affinity honed simultaneously (Figure 3). Unlike in vitro selection, incorporating this technique into in vivo selection allows for refinements to be made within the context of the real world application reducing time later for application optimization.

Figure 3.

Directed evolution to improve affinity and specificity. Here, high diversity libraries are injected in vivo, selected and amplified. Selected variants are used to develop re-diversified libraries via fragmentation, reassembly and mutation. These libraries are used to improve affinity in subsequent panning rounds.

3.3. Novel display platforms.

The incorporation of both diversity and stability into a single, easy to use platform would be beneficial. The reason the vast majority of in vivo panning literature has used the M13 platform is because of its experimental ease of use. New protein and viral based platform technologies are also arising on the in vivo panning horizon. Synthetic nucleocapsids that were recently developed by the Baker and King groups might also be applied to in vivo panning[121]. These synthetic protein nanocages encapsulate their own genomes and are of comparable size to AAVs, allowing high-throughput production, broad circulation in vivo, and nucleic acid sequence-based identification. Furthermore, directed evolution strategies could be easily applied to this technology. Current generation synthetic nucleocapsids possess circulation half-lives comparable to M13 bacteriophage and have been engineered to display peptides for tagging and purification. With diameters < 30 nm, synthetic nucleocapsids are therefore an attractive future platform for in vivo panning.

4. Conclusions

In vivo panning has been relatively unexplored relative to in vitro panning since its inception in 1996. However, sequences identified by in vivo panning have moved quickly toward clinical use. The most well-known recovered in vivo panned sequences RGD[27] and NGR[122] have multiple on-going clinical trials both alone and in combination with therapeutics/diagnostics (Table 1) [123–125]. Advancements in in vivo panning strategies, through clever selection design, increased in vivo stability, and innovative directed evolution strategies, will launch our discoveries and improve delivery into the next 20 years

Table 5:

In vivo SELEX

Aptamer library	Disease Model	Target	Affinity	Library Modification	Random Region (bp)	Library Circulation Time
RNA (Mi 2010, In vivo selection of tumor-targeting RNA Motifs)	Colorectal cancer	p68 RNA helicase	13.8 nM	2’-fluropyrimidine modified RNA	40	20 minutes
RNA (Mi 2016, In vivo selection against human colorectal cancer xenografts identifies an aptamer that targets RNA helicase protein DHX9)	Colorectal cancer	DHX9 RNA helicase		2’-fluropyrimidine modified RNA	40	20–30 minutes
RNA (Cheng 2010, In vivo SELEX for Identification of Brain-penetrating aptamers)	CNS delivery	Brain endothelial cells		2’-fluropyrimidine modified RNA	40	1–3 hours
RNA (Wang 2018, In vivo SELEX of an inhibitory NSCLC-specific RNA aptamer from PEGylated RNA library)	NSCLC	NSCLC cells	9 nM	2’-fluropyrimidine modified RNA and PEG modification	40	6 hours
DNA (Liu 2018, A novel DNA aptamer for dual targeting of polymorphonuclear myeloid-derived suppressor cells and tumor cells)	Breast cancer	Breast cancer cells + myeloid derived suppressor cells	2.47 nM	Thiophosphate backbone substitution	30	4 hours

Abbreviations

(AAVs)

Adeno-associated viruses

(RASPP)

Recombination as a shortest path

(SELEX)

Systematic Evolution of Ligands by Exponential Enrichment

(NGS)

Next generation sequencing

(BBB)

Blood brain barrier

(SOMAmer)

Slow off-rate modified aptamer