Rational protein design: developing next-generation biological therapeutics and nanobiotechnological tools

Advanced Review

Rational protein design:developing next-generationbiological therapeutics andnanobiotechnological toolsCorey J. Wilson1,2,3∗

Proteins are the most functionally diverse macromolecules observed in nature,participating in a broad array of catalytic, biosensing, transport, scaffolding, andregulatory functions. Fittingly, proteins have become one of the most promisingnanobiotechnological tools to date, and through the use of recombinant DNA andother laboratory methods we have produced a vast number of biological therapeu-tics derived from human genes. Our emerging ability to rationally design proteins(e.g., via computational methods) holds the promise of significantly expanding thenumber and diversity of protein therapies and has opened the gateway to realizingtrue and uncompromised personalized medicine. In the last decade computationalprotein design has been transformed from a set of fundamental strategies to strin-gently test our understanding of the protein structure–function relationship, topractical tools for developing useful biological processes, nano-devices, and noveltherapeutics. As protein design strategies improve (i.e., in terms of accuracy andefficiency) clinicians will be able to leverage individual genetic data and biologi-cal metrics to develop and deliver personalized protein therapeutics with minimaldelay. © 2014 Wiley Periodicals, Inc.

How to cite this article:WIREs Nanomed Nanobiotechnol 2014. doi: 10.1002/wnan.1310

INTRODUCTION

Background—Protein Therapeutics

The emerging field of nanomedicine andnanobiotechnology has transformed our

approach to disease management in which thetreatment of disease—and decisions thereof—areincreasingly being made at the genetic level, whereproteins produce the critical biological functionsnecessary to combat disease. To date, recombinant

∗Correspondence to: [email protected] of Chemical and Environmental Engineering, YaleUniversity, New Haven, CT, USA2Department of Molecular Biochemistry and Biophysics, Yale Uni-versity, New Haven, CT, USA3Department of Biomedical Engineering, Yale University, NewHaven, CT, USA

Conflict of interest: The author has declared no conflicts of interestfor this article.

human proteins make up the majority of federallyapproved nanobiotechnological medicines, whichinclude: (1) therapeutics with enzymatic or regulatoryactivity, (2) therapeutics with special targeting activ-ity, (3) protein vaccines, and (4) protein diagnostics.1

More than 130 proteins are currently approved foruses in the clinical setting. An exhaustive list of pro-tein therapeutics is beyond the scope of the currentreview and is given elsewhere.1 Arguably, one of thebest examples of the impact of protein therapy is thedevelopment and production of recombinant humaninsulin in the treatment of diabetes mellitus, type Iand type II. The production and use of recombinanthuman insulin—opposed to insulin isolated frombovine and porcine pancreas—has dramatically low-ered the cost of production and has resulted in lowerimmunogenicity. Likewise, the advantages of produc-ing recombinant human proteins have been leveragedin other systems. Moreover, recombinant technology(i.e., the modification of DNA via molecular biology

© 2014 Wiley Per iodica ls, Inc.

Advanced Review wires.wiley.com/nanomed

and protein expression typically in Escherichia coli inhigh yields) has set the stage for the development ofnew protein therapeutics with designed properties.

Background—Rational Protein DesignOur ability to confer and control the function of pro-teins via rational design has significantly improvedour understanding of protein structure–function rela-tionships and transformed our ability to controlbiological processes to enhance human health. Therational design of proteins ranges from hypothe-sis driven modification of proteins by inspection toadvanced computer-aided designs, with the latterbeing the primary focus of this review. Current com-putational protein design (CPD) methods are prac-tical tools for identifying amino acid sequences thatenhance protein stability, modify binding specificity,or increase protein solubility.2–12 CPD is analogousto civil engineering—but at the atomic level. Namely,CPD leverages physics-based models to generate ablueprint (prediction) of a protein system. In turn, thedesigned protein system is constructed and any newknowledge obtained during experimental validationcan be used to improve and generate new predictivemodels (see Figure 1). Computational design methodshave advanced to the point that medical and biotech-nological applications are now possible. The dynamicand diverse roles of proteins (e.g., catalyzing biochem-ical reactions, forming receptors, immune responsesand channels in membranes, providing intracellularand extracellular scaffolding support, and transport-ing molecules within a cell or from one organ toanother) combined with our ability to (re)design pro-tein functions promises to revolutionize and personal-ize nanomedicine and nanobiotechnology. This reviewwill highlight some of the notable advances in therational design of protein therapeutics and emergingtechnologies that will promote the next generation oftailored systems.

DESIGNED THERAPEUTICS WITHENZYMATIC OR REGULATORYACTIVITY

Computational enzyme design allows for the intro-duction of new chemical and catalytic traits intopre-existing enzyme scaffolds and thus holds greatpromise as a method to speed progress in the devel-opment of next-generation enzyme therapeutics.Rational design of enzymes is currently accom-plished via single-state (fixed-backbone) modellingand focuses on achieving accurate geometric con-straints of reactive sidechains to stabilize the putative

transition-state,9,15–18 see Figure 1. In principalstabilization of the transition-state will lower theactivation barrier and confer or accelerate catalysisin inert protein scaffolds or existing enzymes, respec-tively. Accordingly, this powerful enabling technologyhas been used to create several therapeutic enzymesand holds the promise of transforming bio-drugdesign. In this section discussion will begin withthree notable case studies in which computationalenzyme design has been leveraged to create putativetherapeutics. Finally, this section will culminate withan outlook on the future direction of enzyme design.

Enzymes Designed to Combat Nerve Agent(Organophosphates) ExposureOrganophosphates (OPs) are degradable organiccompounds used principally as the active componentof pesticides and nerve agents used in chemicalwarfare.19–21 OPs inhibits acetylcholinesterase(AChE), see Figure 2(a); AChE inhibition leads tocontinual acetylcholine stimulus of the muscarinicand nicotinic receptors resulting in cholinergic crisis,acute respiratory failure, and in cases of higher-doseexposure, death. After phosphonylation of the cat-alytic serine in AChE dealkylation may occur (i.e.,aging), resulting in a modified enzyme that is resistantto recovery therapies.21,22 Approximately a quarterof a million deaths are associated with OP pesticidesexposure each year23 and countless other deathsare associated with OP-based chemical weapons.21

Current treatments for nerve agent poisoning includeinjections of atropine (i.e., a competitive inhibitorof acetylcholine at the muscarinic receptor) to turnoff muscle stimulus, and a strong nucleophilicoximesuch as pralidoxime chloride, to dephosphonylatethe AChE catalytic serine before aging.24 Both ofthe compounds offer limited protection because theycannot be administered before exposure and do notaddress any of the long-term side effects associatedwith nerve agent poisoning.25 Thus, these therapeuticcompounds must be administered quickly and donot offer broad-spectrum protection against all OPagents. Accordingly, there is a growing interest indeveloping robust counter measures to acute OPtoxicity.

In a recent study, the Redinbo laboratory usedCPD to confer OP nerve agent hydrolysis activity intothe human drug metabolism enzyme carboxylesterase(hCE1).26 hCE1 is homologous in structure and cat-alytic mechanism to AChE27; however, hCE1 does notundergo the dead-end aging process that is observedfor AChE.26 Moreover, hCE1 (wild-type) can be reac-tivated post exposure to the OP sarin; however, reacti-vation does not occur once hCE1 is exposed to the OPs


WIREs Nanomedicine and Nanobiotechnology Rational protein design

Single-state design(canonical enzyme design)

Etotal = Evwd + Eh-bond + Eelec + Eas + Ess + Erp + Eaapp + Erotslf

EtotalEvwdEh-bondEelecEasEssErpEaappErotslf

total energy (for a rotamer at a given residue position)energy of the van der Waals potentialenergy of hydrogen bondingenergy of electrostatic interactionenergy of atomic solvationenergy of secondary structureenergy rotatmer probabilityenergy of amino acid phi-psi angle propensityrotatmer self energy

Multistate design(novel protein design)

Pairwise energies(pre-computed)

Representing interactions of

the combinatorial set

Optimization:

Canonial Design Goal: e.g., Multistate Design Goal: Control TmaxLower Activiation Barrier

Reaction Coordinate

S

PFree

Ene

rgy,

ΔG

AD

P p

rodu

ctio

nA

ctiv

ity

(kca

t s-1

)

Purpose: Identify the global minimum energy conformation (GMEC)

New models for predictingconditional function

Experimental validation: Purpose: characterization of GMEC i.e., experimentally test predictions

Additional optimization:

1400

HOXmax.

Tmax.

TlimitsTlim Tlimits

Tmax.Tmax

HOXlim.

HOXmax

HOXlim.

1200

1000

Act

ivit

y (k

cat

s–1)

AD

P p

rodu

ctio

800

600

400

200

0

1400

1200

1000

800

600

400

200

0

30 40 50 60 70 80 90

Oxidant (molar equivalent)

30 40 50 60 70 80 90 100Temperature (°C)

e.g., Oxidation Resistance: Control HOXmax

1 2

=========

FIGURE 1 | Computational protein design scheme. A given protein can be decomposed into two structural classes: (i) backbone and (ii) aminoacid sidechains. In a single-state enzyme design calculation sidechains are incorporated on to a fixed-backbone scaffold (follow the blue arrow andbox). Each sidechain in the design goal (e.g., enzyme catalytic site) is allowed to assume any number of predefined conformations (i.e., rotamers). Inturn, each amino acid pair is scored based on its interaction with other sidechains (and substrate in the case of enzyme design) producing a totalenergy score (Etotal). All pairwise interactions are recorded in a look-up table and then optimized to identify the putative minimum energy for stabilityand optimal function. The design cycle is conducted in an iterative fashion (green feedback loop and yellow arrows). Innovations brought to bear inrecent studies: To capture the scaffold feedbacks that may influence catalysis the Wilson research group has implemented a provisional discretemultistate design strategy [yellow arrows, also see the Designing Conditional Catalysis (A New Frontier) section below] to confertemperature-adapted function in the adenylate kinase enzyme scaffold. The objective of this study was to implement and improve our ability toconfer temperature-adapted functions13—setting the stage for other conditional functions, e.g., oxidative adaptation,14 see additional optimizationinset. In addition, this work represents a significant attempt to understand enzyme scaffold allosteric feedbacks.

soman or cyclosarin.26 Accordingly, the Redinbo labo-ratory redesigned hCE1 to catalyze the hydrolysis of abroader-spectrum of OPs. The redesign of hCHE1 wasconducted using high-resolution structures of the tar-get enzyme in complex with nerve agent as a guide. Inturn, histidine and glutamic acid residues were intro-duced proximal to the hCHE1’s native catalytic triad,redesigning the active site. The resultant enzyme vari-ant demonstrated significantly increased rates of reac-tivation following exposure to select G-agents (i.e.,soman, cyclosarin, and sarin), see Figure 2(a). Impor-tantly, the addition of these residues did not affectthe high affinity binding of the selected nerve agentsto the enzyme’s active site. Thus, using two aminoacid substitutions, a novel enzyme was created thatefficiently converted a group of hemisubstrates (i.e.,small molecules that can initiate a reaction, but arenot able to complete turnover) into true fully reac-tive substrates. Accordingly, the authors believe thisnovel enzyme may lead to new and effective counter-measures for nerve agent exposure.

In addition to the above, the Tawfik andBaker laboratories collaborated to engineerBrevundimonas diminuta phosphotriesterase’s(PTEs) for broad-spectrum OP-based nerve agentdetoxification.28 To this end, this team developeda direct screen for the detoxification of the entirerange of known OP nerve agents combined witha computational method to guide protein librarydesign. Analogous to the Redinbo study, Tawfik andBaker aim to intercept (via hydrolysis) OPs prior tointeraction with AChE. Ideally, enzymatic counter-measures should display not only broad-spectrumhydrolysis of OPs but catalytic efficiencies (kcat/KM)that exceeds 107 M−1 min−1.29 Paraoxon (an OPagent used as an insecticide) is the natural substrateof PTE (i.e., Paraoxon is enzymatically hydrolyzedwith kcat/KM of 2.0× 109 M−1 min−1); thus, in prin-ciple PTE can be modified (i.e., redesigned) to confercatalytic promiscuity to expand the number of OPcompounds the PTE can hydrolyze. After severalrounds of iterating between protein library design and



O(a) (b) (c)O

P

P P

P

F

(G-agents)

Gluten protein

Chromosomal DNA

Putativetherapeutic vector

crossover

Partically digested gluten(immunogenic peptides)

AChE

Nerve agent

Acetylcholine

Enzyme

Aged

Inhibit acetlcholinesterase (AChE) Cholinergic crisis

Designed enzymefor hydrolysis of nerve agent

(V-agents)

SCH2CH2N( )2CH3 CH3

Organophosphatesnerve agents

Ner

ve te

rmin

al

Ner

ve te

rmin

al

Pos

tsyn

aptic

cel

l

Pos

tsyn

aptic

cel

l

OR ORR

Designed peptidases...digest immunogenic peptides

(Therapy for Celiac Desease)

Gene replacementvia homologous recombination

Designer nucleases(Simulate DNA break repair)

FIGURE 2 | Designed enzyme therapeutics. (a) Designed catalysis to intercept organophosphate nerve agents. Nerve agents inhibitacetylcholinesterase via blocking native chemical signals (e.g., acetylcholine), resulting in cholinergic crisis. Prolong exposure to nerve agents canresult in a phosphonylated enzyme (aged) that is resistant to treatment. Novel enzyme therapeutics aim to intercept and hydrolyse nerve agents,rendering the OP compounds impotent. (b) Individuals with Celiac disease consume foods containing gluten their immune system will produceantibodies to the immunogenic peptides produced for partially digested gluten proteins. Designed proteases have been produced to aid with thedigestion of gluten proteins. (c) Designed nucleases that stimulate the process of homologous recombination, putative therapeutics for genereplacement. Engineered endonuclease promote rapid cleavage kinetics and production of 3′ overhangs upon cleavage, which can increase rates ofhomologous recombination.

experimental library screening, a collection of PTEvariants were developed that: (1) hydrolyze the toxicSP isomers of OP V-agents (i.e., with kcat/KM valuesof up to 5×106 M−1 min−1) and (2) also efficientlydetoxify OP G-agents (see Figure 2(a)). Tawfik andBaker have provided convincing evidence that thisnew class of PTE-based OP catalysts provide the basisfor broad-spectrum nerve agent detoxification.

Designing Enzyme Therapeutics for CeliacDiseaseCeliac disease (CD) is an autoimmune disorder ofthe small intestine. When, individuals with CD con-sume foods containing gluten (e.g., wheat, barley, orrye) their immune system will produce antibodies togluten proteins.30 Anti-gluten immunoglobulins, sub-sequently, attack the intestinal lining causing inflam-mation of the small intestine, ultimately damagingthe villi. Overtime, intestinal inflammation associatedwith the CD immunogenic response will reduced the

absorption of nutrients and can result in malnutrition.Currently, there is no cure for CD; however, treatmentof CD is accomplished via a strict diet that limits oreliminates the consumption of gluten. Gluten consistsof two major components, gliadins and glutenins, andthere are three main types of gliadin (𝛼, Υ, and 𝜔), towhich the body is intolerant in CD. In a recent study,15

the Seigel laboratory used computational methods toredesigned an endopeptidase (i.e., kumamolisin-As,from the bacterium Alicyclobacillus sendaiensisto) toaid in the digestion of 𝛼-gliadin (Figure 2(b)). Thecomputationally designed enzyme (designated Kuma-Max) exhibits a 116-fold greater proteolytic activityfor a model gluten tetrapeptide than the native scaffoldenzyme, and greater than 800-fold switch in substratespecificity toward immunogenic portions of glutenpeptides. In addition, KumaMax degrades over 95%of an immunogenic peptides implicated in CD in underan hour and is resistant to proteolysis by digestiveproteases. Thus, through identification of a naturalenzyme (kumamolisin-As) with pre-existing qualities



(i.e., high activity at low pH), the Siegel laboratorywas able to computationally redesign the enzyme scaf-fold to produce a novel variant (KumaMax) that canpotentially serve as an oral therapeutic for CD.

Site-Specific DNA Repair (Gene-therapyreagents)Gene targeting (i.e., the process of gene replace-ment by homologous recombination) is a techniquethat can be used to modify or correct genomicDNA (Figure 2(c)). Model experiments in Saccha-romyces cerevisiae,31 Drosophlla,32 Caenortiabdltlselegans,33 and stem cells34,35 have demonstrated thata double-strand break in the chromosomal targetenhances the frequency of localized recombinationevents and, targeted gene corrections (or modifica-tions) requires site-specific DNA cleavage to initiatesubsequent cellular processes for DNA break repair.36

Homologous recombination pathways to correct dele-terious genetic mutations or insert a desired DNAsequence can be stimulated via nuclease mediated dou-ble stranded breaks—i.e., via a template with DNAsequences (arms) that match the sequence surround-ing the target cleavage locations. Accordingly, genetargeting shows great potential as a tool for humangene therapy and proof of principle has already beendemonstrated in mouse models.37 Targeted genomicbreaks to activate gene conversion can be achievedvia a number of nucleases.38,39 However, each ofthese nucleases differs in its enzymatic properties (i.e.,specificity and efficiency). Our ability to design (orredesign) nucleases will set the stage for developmentof novel tools for gene delivery.

LAGLIDADG homing endonuclease (LHE)has several characteristics that make this particularenzyme an attractive gene-therapy tool—e.g., (1)rapid cleavage kinetics, (2) production of 3′ over-hangs upon cleavage, which can increase rates ofhomologous recombination, and (3) small encodingreading frames for efficient nuclease delivery. Accord-ingly, our ability to engineer (redesign) the specificityof LHE holds great promise as a general tool forgene targeting (Figure 2(c)). Toward this end, Bakerand colleagues recently reprogrammed HE specificityby iterating between computational modeling andexperimental selection,38 creating an extensive set ofHE variants with novel DNA cleavage specificities.To generate these enzymes, the Baker laboratoryused a high-throughput bacterial selection system toidentify high specificity endonuclease produced viaCPD. In general, high specificity involved an aminoacid sidechain that makes hydrogen bonds with thetarget base; such motif interactions were identified via

computational screening in this study and this infor-mation was subsequently used to bias library design.The combination of a highly efficient selection systemwith input from computational models in librarydesign is an advance over previous approaches usingrandom selection or computation alone. This studyis also significant because rationally designing thisparticular class of homing nucleases to cleave noveltarget sites compared to counterparts,40 is particularlydifficult. While these results are promising, addi-tional knowledge regarding the context-dependencebetween bases must be mechanistically understoodfor reprogramming of nuclease specificity to succeedgenerically.

Designing Conditional Catalysis (A NewFrontier)Our ability to control the function of enzymes viarational design will not only improve our under-standing of protein structure–function relationships,CPD will allow practitioners to generate novel andtailored biologics that will transform our abilityto control biological processes to enhance humanhealth. Toward this end, computational designmethods have advanced to the point that catalyticproperties have been conferred to previously inertprotein scaffolds.9,17,18 That is, these design strategieshave leveraged our limited understanding of enzymetransition-state theory and shown that rational designis an invaluable tool for challenging and driving ourunderstanding of these complex structure–functionrelationships. Yet, rational enzyme design alonetypically confers modest enzymatic function andusually requires subsequent rounds of refinementvia arbitrary mutagenesis (remote from the activesite) to realize catalytic efficiencies on par with thoseobserved in naturally occurring enzymes. Our currentinability to predict the role of the protein scaffoldstructure and corresponding feedback mechanismsupon mutation reflects a significant gap in our under-standing of enzyme function beyond the catalytic-sitestructure. Toward this end the Wilson laboratoryaims to advance our fundamental understandingof enzyme structure–function relationships and totranslate this new knowledge of protein catalysts intonovel predictive tools that will exploit new reactiondetails to create conditional enzymatic catalysis (i.e.,enzymes that function at prescribed temperatures,pH, or oxidative conditions). Such proteins will be ofuse in a broad range of medical and biotechnologicalapplications.

A careful balance between structural-stabilityand flexibility41,42 is a hallmark of enzymatic func-tion and temperature can affect both properties.43–46



Canonical (fixed-backbone) enzyme design strategiescurrently do not consider the role of these prop-erties, see Figure 1. In a recent study13 the Wil-son Laboratory described the rational design of 100temperature-adapted adenylate kinase enzymes via anovel multistate design strategy that takes into consid-eration the impact of conformational changes to thebackbone structure and stability, in addition to exper-imental analysis of thermostability and function, seeFigure 1. A significant feature of this work is that con-ferred adaptation was achieved without any modifica-tion to the active site. Rather, protein designs focusedon scaffold redesign (i.e., redesigning allosteric feed-back). Comparison of the experimental temperature ofmaximum activity to the melting-temperature acrossall 100 variants reveals a strong correlation betweenthese two parameters. In turn, experimental stabilitydata was used to produce accurate predictions of ther-mostability, providing the requisite complement for denovo temperature-adapted enzyme design. In princi-ple, this level of design-based analysis can be appliedto any protein, paving the way toward identifying andunderstanding the hallmarks of the thermodynamicand structural limits of function.

In addition to the above, the Wilson labora-tory has also developed the metrics to design novelproteins that are resistant or adaptive to oxidativestress.14,47 Specifically, the Wilson group used com-putational design to identify the role of certain oxi-dizable residues and other physicochemical propertiesin protein oxidation and in a follow-up study thesemetrics where applied to produce enzymes that areresistant to oxidative damage, see Figure 1. Release ofthe disinfecting oxidants, hypochlorous acid (HOCl)and hypobromous acid (HOBr), from neutrophils and

eosinophils, respectively, plays a critical role in theimmune response to potential pathogen infections(e.g., mechanisms related to wound healing), as wellas host tissue damage associated with inflammatorydiseases (e.g., arthritis, atherosclerosis, cystic fibrosis,and Alzheimer’s disease). Owing to their abundance,high reactivity, and integral cellular function, proteinshave been identified as particularly important in vivotargets for oxidation via hypohalous acid. Thus ourability to confer resistance to oxidative stress for cer-tain proteins will allow us to design novel proteintherapies for inflammatory diseases, diabetic woundhealing, and other oxidative milieu.

Designer AntibodiesAntibodies or immunoglobulins (Ig) are multi-chainproteins that are important components of theimmune system. Antibodies are used to identifyand neutralize foreign substances (e.g., bacteria andviruses). Antibodies recognize foreign materials viaunique molecular tags called an epitope, usually onthe surface of the objective that complements theantibodies paratope. Structurally, an antibody is com-posed of four fundamental parts, two identical heavychains and two identical light chains, cross-linked bydisulfide bonds, see Figure 3. The Ig antigen-bindingsite consisting of hypervariable regions (i.e., regionsof extensive amino-acid diversity) located in roughlythe first 110 amino acids of the light and heavy chainsof the antibody. Combinatorial association of differ-ent light and heavy chains generate a minimum of108 different antibody molecules, and our ability toengineer and redesign immunoglobulins has expandedthe diversity further. This section will discuss several

(a) (b)

Var

iabl

e re

gion

Antibody structure

Reaction coordinate

S

H3C

OCH3

H3CN

O

OOCH3

OH

CO2H

Extractedsplenocytes Tumor cells

Antibody producinghybridomas

Mouse Inoculation with Haptedto simulate antibody production

NO

Designed transition-state analogues

H3CN

O

OCH3

OHO O–O O

P

Con

stan

t reg

ion

Free

ene

rgy,

ΔGH

eavy chain

Light chain

Antigen

-binding site

FIGURE 3 | (a) Antibody schematic. (b) Flow chart for the production of catalytic antibodies. The process begins with the design of atransition-state analog. Next the analog is typically injected into and animal to illicit antibody production; in principal, extraction of animal wholeserum will contain polyclonal antibodies. To isolate monoclonal antibodies requires the extraction of spleen cells and merging them with an immortalcell line to produce hybridomas and antibodies can be subsequently isolated. Finally, generated antibodies can be improved via rational design.



(a)

(b)

Antigen-presenting cell Antigen-presenting cellCD28

CD80/CD86

TCR

Granules

Inhibit degranulationSignal degranulation

Mast cellMast cell

Antigen

FcγRIIB

IgE

Engineeredimmunosuppressant

Preformed mediators

FcεRI

MHC

Belatacept,abatacept,Xpro9523

Naive T cell Naive T cell

ProliferationsEffector Function

ProliferationsEffector Function

FIGURE 4 | Redesigned antibodies. (a) Immunosuppressants belatacept, abatacept, and novel analog Xpro9523 with enhanced binding to humanCD80 and CD86, prevents T cell proliferation. (b) Designed antibody inhibitors for allergic and inflammatory reactions mediated via mast cells andbasophils. Stimulation of mast cells and basophils via crosslinking of Fc𝜀RI promotes degranulation, which leads to allergic and inflammatoryreactions. Engineered fusion proteins mitigate Fc𝜀RI crosslinking via the coengagement of Fc𝛾RIIB and Fc𝜀RI.

advances in the rational design of antibodies and thereimpact on the state-of-the-art.

Designing Improved ImmunosuppressantsBelatacept and abatacept are FDA approved immuno-suppressants that are used for prophylaxis ofkidney transplant and rheumatoid arthritis, respec-tively. Belatacept and abatacept are fusion proteins

composed of cytotoxic T lymphocyte-associatedantigen 4 (CTLA4) and human Fc IgG1 that inhibitsT-cell activation via binding to CD80 and CD86 onantigen-presenting cells (see Figure 4(a)). Belataceptand abatacept are nearly identical—i.e., belataceptdiffers from abatacept by two amino acids.48 How-ever, abatacept is less effective in transplant models asthis immunosuppressant does not completely blockCD86 mediated costimulation.48 Accordingly, it



was reasoned that more potent immunosuppressiveproperties could be achieved via an abatacept variantwith greater avidity for both CD80 and CD86 anti-gens (i.e., the development of Belatacept). In otherwords, the enhanced avidity of belatacept for CD80and CD86 (i.e., improve binding to CD80 and CD86by twofold and fourfold, respectively) translated intogreater immunosuppressive activity when comparedto abatacept.

To be effective both belatacept and abataceptmust be chronically infused, accordingly there is apressing need to develop next-generation immunosup-pressants that can be administer more conveniently.Toward this end, researches at Xencor used structurebased protein engineering to design belatacept andabatacept derivatives (i.e., Xpro9523) that enhancedbinding to human CD80 and CD86, coupled with twoIgG1 Fc substitutions that enhanced binding to humanFcRn.49 The rationale for the additional enhance-ment is immunoglobulin half-life (in vivo) is regu-lated via a pH-dependent Fc domain / FcRn receptorinteraction, which in principle increases the relativeserum level of therapeutic antibodies.50 Accordingly,scientists at Xencor reasoned that similar Fc substitu-tions shown to increase persistence of antibodies (inother work51) would also extend the in vivo half-lifeof next-generation belatacept/abatacept immunosup-pressant variant Xpro9523. Relative to abatacept andbelatacept, Xpro9523 has an increased binding affin-ity to CD80, CD86, and FcRn—i.e., (respectively)5.9, 23, and 12-fold compared to abatacept; 1.5,7.7, and 11-fold compared to belatacept. Accord-ingly, Xpro9523 suppressed T-cell proliferation betterthan abatacept and also suppressed inflammation inarthritic mouse models. Finally, Xpro9523 saturatedCD80 and CD86 more effectively than both abata-cept and belatacept with longer half-life in cynomol-gus monkeys. These data suggest that Xpro9523 maydemonstrate superior efficacy and dosing conveniencecompared with previously developed immunosuppres-sants (i.e., abatacept and belatacept) when adminis-tered in humans.

Engineered Mast Cell and BasophilInhibitorsMast cells and basophils (i.e., specific granulocytes)play a critical role in allergic and inflammatoryreactions. Both granulocytes are activated via theinteraction of IgE (bound to an antigen) with Fc𝜀RIreceptors on the surface of a mast cell or basophil.Upon crosslinking of Fc𝜀RI (via an IgE-antigen com-plex) a signaling cascade rapidly leads to the releaseof preformed mediators via degranulation52—i.e.,

promoting the release of inflammatory mediators(e.g., prostaglandins, vasoactive peptides, proteases,cytokines, heparin, and chemokines) from secre-tory vesicles, see Figure 4(b). In the short term(immediate-phase) the release of certain mediatorsproduce an allergic reaction and result in vasodilation,increased vascular permeability, upregulation of vas-cular adhesion molecules and bronchoconstriction.Prolonged stimulation (late-phase allergic reaction)leads to cytokine and chemokine production, whichinduces the recruitment of inflammatory cells andT-cell activation. Accordingly, ample effort has beenmade to understand mechanisms leading to mast celland basophil activation, in addition to, reconcilingmeans to inhibit pathways involved in the regulationof degranulation.

The detailed study of Fc𝜀RI stimulation and inhi-bition has led to the development of several promisingtherapeutic approaches for suppressing activation ofmast cells and basophils. The most straightforwardapproach involves the development of monoclonalantibodies specific for IgE to prevent antigen recogni-tion via a reduction of free IgE serum levels of aller-gic individuals. Clinical studies reveal encouragingresults using IgE-specific antibodies as a treatment forasthma,53 peanut allergies,54 and rhinitis.55 However,IgE-specific antibody therapy targets IgE of any speci-ficity and have broad-spectrum function, which limitsefficacy. Another approach involves suppressing acti-vation (i.e., degranulation) of mast cells and basophilsvia coengagement of Fc𝛾RIIB and Fc𝜀RI receptors viabispecific antibody conjugates56,57 (see Figure 4(b)).Fc𝛾RIIB and Fc𝜀RI coengagement via bispecific anti-body fusion proteins shows tremendous potential asa novel therapeutic and proof of principle has beendemonstrated for cat and peanut allergies. However,bispecific antibodies do not necessarily prevent Fc𝜀RIcrosslinking (i.e., mediated via the IgE-antigen com-plex) resulting in a less effective suppression. In addi-tion, this class of bispecific conjugates can also bind(activate) Fc𝛾RIIA over Fc𝛾RIIB, and are believed tohave low affinity for Fc𝛾RIIB, consequently these ther-apeutics lack the needed receptor specificity to makethem fully effective.58

To investigate the therapeutic potential ofdirected coengagement of Fc𝜀RI and Fc𝛾RIIB in theabsence of Fc𝜀RI crosslinking, Cemerski et al. devel-oped a novel bispecific antibody fusion with greateraffinity for Fc𝛾RIIB.58 In other words the goal wasto create a selective (mast cell and basophil) inhibitorthat avoids unwanted stimulatory effects associatedwith crosslinking of Fc𝜀RI. In brief, Cemerski andcolleagues engineered a bispecific fusion protein com-posed of the coupled Fc domains of human IgG1 and



murine IgE (i.e., Fc𝜀–Fc𝛾). To increase the specificityof the novel antibody Fc𝜀–Fc𝛾 conjugate, researchersdesigned the IgG1 Fc domain to bind to humanFc𝛾RIIB with 100-fold greater affinity, compared tonative IgG1 Fc. As a proof of principle, it was shownthat the Fc𝜀–Fc𝛾 conjugate binds with high affinityto murine Fc𝜀RI and human Fc𝛾RIIB on mast cells(i.e., in transgenic mice), and triggers phosphorylationof Fc𝛾RIIB, and inhibits Fc𝜀RI-dependent calciummobilization. Thus it is reasonable to postulate thata fully human Fc𝜀–Fc𝛾 biologic, with high-affinitycoengagement of Fc𝜀RI and Fc𝛾RIIB has great poten-tial as a novel therapy for allergy and possibly othergranulocyte-mediated pathologies.

Catalytic AntibodiesWhat are catalytic antibodies? Given the high affin-ity and specificity of the antigen-binding site, cat-alytic groups can be introduced directly into theantigen-binding site of an immunoglobulin (e.g., bychemical modification, site-directed mutagenesis orgenetic selection) and this conferred function, whichis facilitated by the immunoglobulin scaffold, repre-sents the simplest form of a catalytic antibody, seeFigure 3(b). The classic approach for engineering cat-alytic antibodies is to design transition-state analog(rather than forward protein scaffold design) to serveas haptens, which will elicit antibodies with desiredcatalytic activities. The key advantage of catalytic anti-bodies over canonical enzymes resides in the elegantsimplicity of rapidly generating catalysis for difficultand exotic reactions. Using the above strategies severalputative therapeutic catalytic antibodies have beendeveloped.

Barbas et al. develop prodrug chemistrydesigned to take advantage of the broad scopeand mechanism of a certain catalytic antibody, inwhich an enamine mechanism of natural aldolaseswas imprinted within the antibody-binding site.59

Namely, the aldolase catalytic antibody was used toactivate cancer prodrugs (i.e., masked doxorubicinand camptothecin). The benefit of prodrug deliveryis the substantial reduction in toxicity. The dox-orubicin and camptothecin prodrugs are selectivelyactivated by the aldolase catalytic antibody whenadministered at therapeutic concentrations. More-over, Barbas and colleagues demonstrated the efficacyof this antibody-directed enzyme prodrug therapy aseffective chemotherapeutic strategies via their abilityto kill human colon and prostate cancer cell lines.

In addition to the above, the Landry laboratoryhas produced a series of catalytic antibodies that treat

cocaine addiction.60–62 Notably, catalytic antibodiesare ideal for treating cocaine addiction and overdose,as no antagonist to cocaine is known. This is due inlarge part to the lack of selectivity of direct therapeu-tics (i.e., dopamine receptor agonists and antagonists),in addition to the central role of these receptors in thecentral nervous system and the cardiovascular system,which cocaine acts upon. Accordingly, the difficultiesassociated with cocaine antagonist development havehighlighted the need to produce alternate strategiesto intercept cocaine prior to receptor binding. Whilea direct interception with standard antibodies seemsappealing, binding would be stoichiometric and wouldrapidly deplete the therapeutic biologic. Whereas, acatalytic antibody directed toward cocaine mitigatesthis problem because the protein is not consumed. Inprinciple, other addiction and toxicities can be treatedin a similar fashion.

Structural analysis of several catalytic antibod-ies shows significantly similar physicochemical prop-erties relative to canonical enzymes. That is to say,catalytic antibodies and enzymes have comparableactive sites and utilize similar ligands for substratebinding and catalysis. Accordingly, it is reasonable toassume CPD (i.e., forward protein design) will con-tribute to improving the rational design and accel-erate discovery of this important class of biologicaltherapeutics.

CONCLUSION

The de novo design and redesign of proteins haschanged the landscape of drug design and has broughtus one step closer to truly personalized medicine. Inthe upcoming years improvement in force-field devel-opment (i.e., the principal means of protein mod-elling) will improve both the accuracy and efficiencyof CPD strategies. Notably, our ability to reconcileand model multistate properties (e.g., allosteric com-munication and cooperative communication13,63) willlikely improve nearly every aspect of conferring orimproving function in a given protein scaffold. Ourability to identify and exploit those structural feed-back mechanisms between the scaffold and active site(or binding site) will also facilitate the rational designof other functions that are useful in sensitizing macro-molecules to elicit function, such as energy and elec-tron transfer.64 In the interim the current state ofthe art will continue to enable the development ofnovel therapeutic proteins and improve the functionof existing therapeutics—albeit at a relatively modestpace.



REFERENCES1. Leader B, Baca QJ, Golan DE. Protein therapeutics: a

summary and pharmacological classification. Nat RevDrug Discov 2008, 7:21–39.

2. Bradley P, Misura KMS, Baker D. Towardhigh-resolution de novo structure prediction forsmall proteins. Science 2005, 309:1868–1871.

3. Dahiyat BI, Gordon DB, Mayo SL. Automated designof the surface positions of protein helices. Protein Sci1997, 6:1333–1337.

4. Dahiyat BI, Mayo SL. Protein design automation. Pro-tein Sci 1996, 5:895–903.

5. Dahiyat BI, Mayo SL. De novo protein design:fully automated sequence selection. Science 1997,278:82–87.

6. Dahiyat BI, Mayo SL. Probing the role of packingspecificity in protein design. Proc Natl Acad Sci USA1997, 94:10172–10177.

7. Dahiyat BI, Meade TJ, Mayo SL. Site-specific modifi-cation of 𝛼-helical peptides with electron donors andacceptors. Inorg Chim Acta 1996, 243:207–212.

8. Dahiyat BI, Sarisky CA, Mayo SL. De novo proteindesign: towards fully automated sequence selection. JMol Biol 1997, 273:789–796.

9. Jiang L, Althoff EA, Clemente FR, Doyle L, Rothlis-berger D, Zanghellini A, Gallaher JL, Betker JL, TanakaF, Barbas CF, et al. De novo computational design ofretro-aldol enzymes. Science 2008, 319:1387–1391.

10. Korkegian A, Black ME, Baker D, Stoddard BL. Com-putational thermostabilization of an enzyme. Science2005, 308:857–860.

11. Kuhlman B, Dantas G, Ireton GC, Varani G, Stod-dard BL, Baker D. Design of a novel globular pro-tein fold with atomic-level accuracy. Science 2003,302:1364–1368.

12. Schueler-Furman O, Wang C, Bradley P, Misura K,Baker D. Progress on modeling of protein structures andinteractions. Science 2005, 310:638–642.

13. Howell SC, Inampudi KK, Bean DP, Wilson CJ. Under-standing thermal adaptation of enzymes through themultistate rational design and stability prediction of 100adenylate kinases. Structure 2014, 22:218–229.

14. Sivey JD, Howell SC, Bean DJ, McCurry DL, Mitch WA,Wilson CJ. Role of lysine during protein modification byHOCl and HOBr: halogen-transfer agent or sacrificialantioxidant? Biochemistry 2013, 52:1260–1271.

15. Gordon SR, Stanley EJ, Wolf S, Toland A, Wu SJ, HadidiD, Mills JH, Baker D, Pultz IS, Siegel JB. Computationaldesign of an 𝛼-gliadin peptidase. J Am Chem Soc 2012,134:20513–20520.

16. Privett HK, Kiss G, Lee TM, Blomberg R, Chica RA,Thomas LM, Hilvert D, Houk KN, Mayo SL. Iterativeapproach to computational enzyme design. Proc NatlAcad Sci USA 2012, 109:3790–3795.

17. Rothlisberger D, Khersonsky O, Wollacott AM, JiangL, DeChancie J, Betker J, Gallaher JL, Althoff EA,Zanghellini A, Dym O, et al. Kemp elimination cata-lysts by computational enzyme design. Nature 2008,453:190–194.

18. Siegel JB, Zanghellini A, Lovick HM, Kiss G, LambertAR, Clair JLS, Gallaher JL, Hilvert D, Gelb MH,Stoddard BL, et al. Computational design of an enzymecatalyst for a stereoselective bimolecular diels-alderreaction. Science 2010, 329:309–313.

19. Ohbu S, Yamashina A, Takasu N, Yamaguchi T, MuraiT, Nakano K, Matsui Y, Mikami R, Sakurai K, Hino-hara S. Sarin poisoning on Tokyo subway. South Med J1997, 90:587–593.

20. Yokoyama K, Araki S, Murata K, Nishikitani M,Okumura T, Ishimatsu S, Takasu N. A preliminarystudy on delayed vestibulocerebellar effects of Tokyosubway sarin poisoning in relation to gender difference:frequency analysis of postural sway. J Occup EnvironMed 1998, 40:17–21.

21. Newmark J. Nerve agents. Neurologist 2007, 13:20–32.

22. Millard CB, Lockridge O, Broomfield CA. Organophos-phorus acid anhydride hydrolase activity in humanbutyrylcholinesterase: synergy results in a somanase.Biochemistry 1998, 37:237–247.

23. Gunnell D, Eddleston M, Phillips MR, Konradsen F.The global distribution of fatal pesticide self-poisoning:systematic review. BMC Public Health 2007, 7:357.

24. Gray AP. Design and structure-activity-relationshipsof antidotes to organo-phosphorus anticholinesteraseagents. Drug Metab Rev 1984, 15:557–589.

25. Doctor BP, Saxena A. Bioscavengers for the protectionof humans against organophosphate toxicity. ChemBiol Interact 2005, 157:167–171.

26. Hemmert AC, Otto TC, Chica RA, Wierdl M, EdwardsJS, Lewis SM, Edwards CC, Tsurkan L, CadieuxCL, Kasten SA, et al. Nerve agent hydrolysis activitydesigned into a human drug metabolism enzyme. PLoSOne 2011, 6:e17441.

27. Redinbo MR, Potter PM. Mammalian car-boxylesterases: from drug targets to protein thera-peutics. Drug Discov Today 2005, 10:313–325.

28. Cherny I, Greisen P Jr, Ashani Y, Khare SD, Ober-dorfer G, Leader H, Baker D, Tawfik DS. EngineeringV-type nerve agents detoxifying enzymes using com-putationally focused libraries. ACS Chem Biol 2013,8:2394–2403.

29. Ashani Y, Goldsmith M, Leader H, Silman I, SussmanJL, Tawfik DS. In vitro detoxification of cyclosarin inhuman blood pre-incubated ex vivo with recombinantserum paraoxonases. Toxicol Lett 2011, 206:24–28.

30. Sollid LM. Coeliac disease: dissecting a complex inflam-matory disorder. Nat Rev Immunol 2002, 2:647–655.



31. Choulika A, Perrin A, Dujon B, Nicolas JF. Inductionof homologous recombination in mammalian chromo-somes by using the I-SceI system of Saccharomyces cere-visiae. Mol Cell Biol 1995, 15:1968–1973.

32. Gloor GB, Nassif NA, Johnson-Schlitz DM, Pre-ston CR, Engels WR. Targeted gene replacement inDrosophila via P element-induced gap repair. Science1991, 253:1110–1117.

33. Donoho G, Jasin M, Berg P. Analysis of gene tar-geting and intrachromosomal homologous recombi-nation stimulated by genomic double-strand breaksin mouse embryonic stem cells. Mol Cell Biol 1998,18:4070–4078.

34. Smih F, Rouet P, Romanienko PJ, Jasin M.Double-strand breaks at the target locus stimulategene targeting in embryonic stem cells. Nucleic AcidsRes 1995, 23:5012–5019.

35. Cohen-Tannoudji M, Robine S, Choulika A, PintoD, El Marjou F, Babinet C, Louvard D, Jaisser F.I-SceI-induced gene replacement at a natural locusin embryonic stem cells. Mol Cell Biol 1998, 18:1444–1448.

36. Bibikova M, Beumer K, Trautman JK, Carroll D.Enhancing gene targeting with designed zinc fingernucleases. Science 2003, 300:764.

37. Perez EE, Wang J, Miller JC, Jouvenot Y, Kim KA,Liu O, Wang N, Lee G, Bartsevich VV, Lee YL, et al.Establishment of HIV-1 resistance in CD4+ T cellsby genome editing using zinc-finger nucleases. NatBiotechnol 2008, 26:808–816.

38. Thyme SB, Boissel SJ, Arshiya Quadri S, Nolan T,Baker DA, Park RU, Kusak L, Ashworth J, BakerD. Reprogramming homing endonuclease specificitythrough computational design and directed evolution.Nucleic Acids Res 2013, 42:2564–2576.

39. Boissel S, Jarjour J, Astrakhan A, Adey A, Gou-ble A, Duchateau P, Shendure J, Stoddard BL, CertoMT, Baker D, et al. megaTALs: a rare-cleaving nucle-ase architecture for therapeutic genome engineering.Nucleic Acids Res 2013, 42:2591–2601.

40. Li T, Huang S, Zhao X, Wright DA, Carpenter S, Spald-ing MH, Weeks DP, Yang B. Modularly assembleddesigner TAL effector nucleases for targeted gene knock-out and gene replacement in eukaryotes. Nucleic AcidsRes 2011, 39:6315–6325.

41. Bandaria JN, Dutta S, Hill SE, Kohen A, CheatumCM. Fast enzyme dynamics at the active site of formatedehydrogenase. J Am Chem Soc 2008, 130:22–23.

42. Bandaria JN, Cheatum CM, Kohen A. Examination ofenzymatic H-tunneling through kinetics and dynamics.J Am Chem Soc 2009, 131:10151–10155.

43. Fields PA. Review: protein function at thermal extremes:balancing stability and flexibility. Comp Biochem PhysA 2001, 129:417–431.

44. Henzler-Wildman KA, Lei M, Thai V, Kerns SJ, KarplusM, Kern D. A hierarchy of timescales in protein

dynamics is linked to enzyme catalysis. Nature 2007,450:913–927.

45. Pena MI, Davlieva M, Bennett MR, Olson JS, ShamooY. Evolutionary fates within a microbial populationhighlight an essential role for protein folding duringnatural selection. Mol Syst Biol 2010, 6:1–11.

46. Wolf-Watz M, Thai V, Henzler-Wildman K, Hadji-pavlou G, Eisenmesser EZ, Kern D. Linkage betweendynamics and catalysis in a thermophilic-mesophilicenzyme pair. Nat Struct Mol Biol 2004, 11:945–949.

47. Zeng T, Wilson CJ, Mitch WA. Effect of chemicaloxidation on the sorption tendency of dissolved organicmatter to a model hydrophobic surface. Environ SciTechnol 2014, 48:5118–5126.

48. Vincenti F, Dritselis A, Kirkpatrick P. Belatacept. NatRev Drug Discov 2011, 10:655–656.

49. Bernett MJ, Chu SY, Leung I, Moore GL, Lee SH,Pong E, Chen H, Phung S, Muchhal US, Horton HM,et al. Immune suppression in cynomolgus monkeysby XPro9523: an improved CTLA4-Ig fusion withenhanced binding to CD80, CD86 and neonatal Fcreceptor FcRn. MAbs 2013, 5:384–396.

50. Roopenian DC, Akilesh S. FcRn: the neonatal Fc recep-tor comes of age. Nat Rev Immunology 2007, 7:715–725.

51. Zalevsky J, Chamberlain AK, Horton HM, Karki S,Leung IW, Sproule TJ, Lazar GA, Roopenian DC,Desjarlais JR. Enhanced antibody half-life improves invivo activity. Nat Biotechnol 2010, 28:157–159.

52. Kraft S, Kinet JP. New developments in FcepsilonRIregulation, function and inhibition. Nat Rev Immunol2007, 7:365–378.

53. Soler M, Matz J, Townley R, Buhl R, O’Brien J,Fox H, Thirlwell J, Gupta N, Della Cioppa G. Theanti-IgE antibody omalizumab reduces exacerbationsand steroid requirement in allergic asthmatics. EurRespir J 2001, 18:254–261.

54. Leung DY, Sampson HA, Yunginger JW, Burks AWJr, Schneider LC, Wortel CH, Davis FM, Hyun JD,Shanahan WR Jr, for the TNX-901 Peanut AllergyStudy Group. Effect of anti-IgE therapy in patients withpeanut allergy. N Engl J Med 2003, 348:986–993.

55. Casale TB, Condemi J, LaForce C, Nayak A, Rowe M,Watrous M, McAlary M, Fowler-Taylor A, Racine A,Gupta N, et al. Effect of omalizumab on symptoms ofseasonal allergic rhinitis: a randomized controlled trial.JAMA 2001, 286:2956–2967.

56. Kinet JP. A new strategy to counter allergy. N Engl JMed 2005, 353:310–312.

57. Zhu D, Kepley CL, Zhang K, Terada T, Yamada T,Saxon A. A chimeric human-cat fusion protein blockscat-induced allergy. Nat Med 2005, 11:446–449.

58. Cemerski S, Chu SY, Moore GL, Muchhal US, Des-jarlais JR, Szymkowski DE. Suppression of mast celldegranulation through a dual-targeting tandem IgE-IgG



Fc domain biologic engineered to bind with high affinityto Fc𝛾RIIb. Immunol Lett 2012, 143:34–43.

59. Shabat D, Rader C, List B, Lerner RA, Barbas CF 3rd..Multiple event activation of a generic prodrug triggerby antibody catalysis. Proc Natl Acad Sci USA 1999,96:6925–6930.

60. Mets B, Winger G, Cabrera C, Seo S, Jamdar S, YangG, Zhao K, Briscoe RJ, Almonte R, Woods JH, et al.A catalytic antibody against cocaine prevents cocaine’sreinforcing and toxic effects in rats. Proc Natl Acad SciUSA 1998, 95:10176–10181.

61. Yang G, Chun J, ArakawaUramoto H, Wang X,Gawinowicz MA, Zhao K, Landry DW. Anti-cocaine

catalytic antibodies: a synthetic approach to improvedantibody diversity. J Am Chem Soc 1996, 118:5881–5890.

62. Tramontano A, Janda KD, Lerner RA. Catalytic anti-bodies. Science 1986, 234:1566–1570.

63. Meyer S, Ramot R, Kishore Inampudi K, Luo B,Lin C, Amere S, Wilson CJ. Engineering alternatecooperative-communications in the lactose repres-sor protein scaffold. Protein Eng Des Sel 2013, 26:433–443.

64. Tobin PH, Wilson CJ. Examining photoinduced energytransfer in Pseudomonas aeruginosa azurin. J Am ChemSoc 2014, 136:1793–1802.


Documents

Rational protein design: developing next-generation biological therapeutics and nanobiotechnological tools