Engineering the serpin α₁‐antitrypsin: A diversity of goals and techniques

Abstract

α₁‐Antitrypsin (α₁‐AT) serves as an archetypal example for the serine proteinase inhibitor (serpin) protein family and has been used as a scaffold for protein engineering for >35 years. Techniques used to engineer α₁‐AT include targeted mutagenesis, protein fusions, phage display, glycoengineering, and consensus protein design. The goals of engineering have also been diverse, ranging from understanding serpin structure–function relationships, to the design of more potent or more specific proteinase inhibitors with potential therapeutic relevance. Here we summarize the history of these protein engineering efforts, describing the techniques applied to engineer α₁‐AT, specific mutants of interest, and providing an appended catalog of the >200 α₁‐AT mutants published to date.

1. INTRODUCTION

1.1. The abundant serpin α₁‐antitrypsin

α₁‐Antitrypsin (α₁‐AT), also known as α₁‐proteinase inhibitor (α₁‐PI), is a member of the serine proteinase inhibitor (“serpin”) superfamily of proteins.1, 2, 3 Serpins are primarily inhibitors of serine proteinases in the PA clan (Proteinases of mixed nucleophile, superfamily A) chymotrypsin‐like family,4 and play vital roles in regulating serine proteinase cascades by limiting the amount of active proteinase present.3, 5 Over 1,500 serpin genes organized into 16 clades have been identified in the genomes of all forms of life, including viruses; there are 36 human serpins.6 A serpin nomenclature identifying α₁‐AT as SERPINA1 has been proposed.3 The core structure of serpins includes three β‐sheets and eight to nine α‐helices,3 is highly conserved across phyla, and may have arisen due to the combined effects of gene duplication and pressures to retain functional structures.6 A surface loop of 20–24 residues, called the reactive center loop (RCL), serves as the initial recognition motif for proteinases destined for serpin inhibition and contributes importantly to serpin specificity.

Mature α₁‐AT is a 394‐residue 52 kDa glycoprotein synthesized primarily in the liver, and is the most abundant serpin in human plasma, with typical concentrations between 20 and 53 μmol/L that are further elevated during an acute phase response.7, 8 The principal physiological function of α₁‐AT is to inhibit the serine proteinase neutrophil elastase, which is crucial for controlling the action of this proteinase in the lungs and liver. In individuals with α₁‐AT deficiency, an autosomal co‐dominant disease, circulating levels of α₁‐AT are as low as 10% of normal, contributing to a high risk for chronic obstructive pulmonary disorder (COPD)/emphysema and cirrhosis of the liver.9 α₁‐AT deficiency has been linked to over 30 pathogenic variants,10, 11, 12 with the E264V “S” and E342K “Z” variants accounting for most cases due to a particularly high frequency among people of European decent.13, 14

Unraveling the molecular mechanism underlying α₁‐AT deficiency and the inhibitory mechanism of serpins, in general, have been goals making α₁‐AT the subject of extensive study for over 50 years.15 In the early 1980s, α₁‐AT became the first serpin for which full‐length cDNA was isolated,16 and the first serpin to have a solved X‐ray crystal structure (albeit in a cleaved form).17 At the same time, advances in DNA manipulation and recombinant protein expression methods made it possible to purify large amounts of α₁‐AT, with the initial goal of using it for replacement therapy for individuals with α₁‐AT deficiency.18, 19 Recombinant α₁‐AT has since been expressed in cultured mammalian cells, bacteria, yeast, insect cells, plants, and in transgenic animals (reviewed in Reference 20). Expression in bacteria and yeast is inexpensive and particularly robust, achieving yields of several mg/L up to 1 g/L.21, 22 Optimizing the expression of recombinant α₁‐AT remains a major area of study, as current replacement therapy is dependent on expensive plasma‐derived α₁‐AT.23

Beyond replacement therapy, α₁‐AT has been investigated as a protein therapeutic for many other diseases including diabetes, graft‐versus‐host disease, cystic fibrosis, and arthritis among others.24, 25 This versatility stems from α₁‐AT's ability to inhibit several proteinases tied to inflammation, including neutrophil elastase as well as proteinase 3,26 cathepsin G,27 and several caspases which are cysteine proteinases.28 α₁‐AT has also been investigated as an anti‐infective agent, displaying an ability to suppress HIV‐1 infection,29 and overexpression of α₁‐AT reduces bacterial load in transgenic mice,30 likely by suppressing proteinases required for pathogen proliferation. The production of pro‐inflammatory cytokines is also lowered by α₁‐AT, however, the molecular mechanism contributing to this effect is not yet well‐defined and may be unrelated to proteinase inhibition.31, 32, 33, 34 The globular nature of α₁‐AT may also contribute to its anti‐inflammatory properties, as it has been found to bind pro‐inflammatory cytokines and other proteins outside of the serpin mechanism.24, 35, 36

These potential therapeutic applications and relative ease of recombinant expression, combined with a detailed understanding of α₁‐AT biochemistry, has made α₁‐AT an attractive and versatile scaffold for protein engineering. In 1984, Rosenberg et al. specifically introduced the M358V mutation in an effort to create an oxidation‐resistant α₁‐AT variant for replacement therapy, representing one of the earliest examples of protein engineering.37 Since then, over 200 engineered α₁‐AT variants have been reported, modified using a wide variety of techniques. In this review, we provide an overview of the various methods and rationales for engineering α₁‐AT to have novel and desired properties, highlighting key examples spanning the history of protein engineering as a discipline.

1.2. Structural and biochemical studies of α₁ ‐ AT and the serpin mechanism

Crystal structures of intact α₁‐AT revealed that it is composed of nine alpha helices, three beta sheets, and an exposed RCL loop.38 Numerous structural and biochemical studies have shown that α₁‐AT shares both this general structure and a “suicide‐substrate” mechanism of inhibition39 with inhibitory serpins.40, 41 Unlike some serpins, α₁‐AT does not require a cofactor (such as heparin or vitronectin) to maximize its rate of inhibition of target proteinases. Synthesis of α₁‐AT occurs primarily in hepatocytes, where its 24‐residue signal peptide directs the nascent chain to the endoplasmic reticulum for folding and glycosylation prior to maturation in the Golgi apparatus and secretion.42 The protein contains three sites of N‐linked glycosylation (N46, N83, and N247) that help provide quality control of α₁‐AT folding through the glycosylation‐dependent chaperones calreticulin and calnexin,42 and which extend the circulating half‐life of α₁‐AT.43

Detailed biochemical characterization of α₁‐AT has revealed a diverse range of inhibitory targets: neutrophil elastase is inhibited rapidly (with reported rate constants of inhibition of 1.2 × 10⁷ to 7 × 10⁷ L·mol⁻¹·s⁻¹),44, 45 however α₁‐AT also has significant inhibitory activity toward the human serine proteinases chymotrypsin,46 kallikrein 7,47 neutrophil proteinase 3,26 pancreatic elastase,48 kallikrein 14,49 trypsin,45 and cathepsin G27 (listed in descending order of the reported rate of inhibition). Limited inhibitory activity toward other serine proteinases has also been reported, including coagulation factor Xa (FXa),50 coagulation factor XIa (FXIa),51 plasmin,48 thrombin,48 granzyme M,52 and activated protein C (APC).53 Even certain nonserine proteinases such as the metalloproteinase ADAM‐17,35 and the cysteine proteinases calpain‐154 and various caspases28 are reported to be inhibited by α₁‐AT.

The mechanism by which α₁‐AT inhibits all targets has been elucidated by biochemical and structural studies and is thought to be common to all inhibitory serpins.40, 41 Attacking proteinases first form an encounter or Michaelis complex with serpins, in which the RCL contacts the proteinase active site, and a tetrahedral intermediate is formed. Attack of the active site nucleophile (either serine or cysteine) on the serpin reactive center bond (termed P1‐P1′ in the Schechter and Berger nomenclature,55 M358‐S359 in α₁‐AT) releases stored energy and powers a translocation of the proteinase to the opposite pole of the serpin. The proteinase remains tethered to the serpin via an acyl ester bond between the hydroxyl group of the active site serine and the P1 residue's carbonyl group, and the N‐terminal remainder of the RCL is inserted into β‐sheet A as a sixth strand. The rapidity of translocation of the proteinase and RCL loop insertion leaves the proteinase active site distorted, such that it cannot hydrolyze the acyl intermediate in a physiologically meaningful length of time but is instead cleared and degraded via scavenger receptors.56, 57

Studies employing α₁‐AT were critical to solving the serpin mechanism (see Figure 1). Crystal structures of cleaved plasma‐derived α₁‐AT,17 recombinant α₁‐AT M358R non‐covalently bound to S195A‐trypsin,58 recombinant α₁‐AT covalently bound to trypsin,59 and recombinant α₁‐AT M358R covalently bound to pancreatic elastase60 provided static snapshots of different stages of inhibition. Fluorescence resonance experiments demonstrated a major conformational change in α₁‐AT M358R complexed with trypsin but not with proteolytically inactive anhydrotrypsin, and full insertion of the RCL into β‐sheet A.57, 61 These studies helped rationalize earlier results from mutational studies, which identified a critical role for residues in the hinge region of the RCL, defined as stretching from residues P15 to P8. Alterations in this region converted α₁‐AT from an inhibitor to a proteinase substrate or shifted the balance of the branched pathway to favor substrate behavior.44, 62, 63 The full loop insertion revealed by the two covalent α₁‐AT complex structures and these mutagenic studies supported the conclusion, first made using P14 variants of plasminogen activator inhibitor 1 (PAI‐1, SERPINE1), that the rate of loop insertion into β‐sheet A determines whether a stable complex will form, or whether cleaved serpin and regenerated active proteinase will be liberated.64 The α₁‐AT portions of the structures faithfully reproduced the original RCL‐cleaved structure, underlining that the conversion from a stressed to a relaxed state, accompanied by large differences in thermal melting temperatures of 50–60°C,65 is an integral part of the serpin mechanism.

Figure 1.

Branched inhibitory mechanism of α₁‐AT. X‐ray crystal structures of α₁‐AT with PDB codes are shown. In Step 1, α₁‐AT is shown oriented with the RCL (light blue‐dark blue) at the top, forming a noncovalent encounter complex with trypsin (red). The PDB model was modified to convert both proteins to their respective wild type sequences, with the trypsin catalytic triad (H57, D102, and S195) and α₁‐AT P1 and P1′ residues (M358, S359) shown as sticks. The encounter complex can then proceed to form the covalent serpin‐proteinase complex (Step 2a), via reactive center cleavage, RCL insertion into β‐sheet A, and proteinase translocation. In this complex, the proteinase is trapped in the acyl‐enzyme state (trypsin S195, α₁‐AT M358 shown as sticks). Or, the complex decays into released proteinase (not shown) and cleaved serpin (Step 2b)

If the rate of loop insertion determines whether the covalent serpin‐proteinase complex forms, RCL length should be critical. In support of this idea, most serpins have an RCL of the same length, with only a few (e.g., antiplasmin [SERPINF2] and C1 esterase inhibitor, C1INH [SERPING1]) having an RCL one residue shorter. Zhou et al. tested this deduction by inserting one (Add‐1) or two Ala (Add‐2) residues into α₁‐AT M358R between P4 and P3, and by stepwise deletions of P4 (Del‐1) up to P3‐P6 (Del‐4).66 Add‐1 and Add‐2 mutations reduced the rate of FXa inhibition by only twofold, but increased the dissociation rate of α₁‐AT‐FXa complexes by factors of 10⁵ to 10⁶ With thrombin, Add‐1 and Del‐1 were minimally affected, while Add‐2 exhibited complex instability and elevated SI. Del‐2, ‐3, and ‐4 were substrates of either proteinase. Im et al. introduced larger insertions into the RCL, from 3 up to 30 residues, to investigate the underlying strain in the metastable native state.67 Each successive insertion increased the stability of α₁‐AT, by releasing the energy stored by the strained native conformation, which the authors proposed generates the inactive “latent” form which features the uncleaved RCL inserted into β‐sheet A. These insertions also eliminated anti‐proteinase activity. These results provided independent support for the mechanism inferred from the structural studies and imposed a length restriction on RCL engineering.

The α₁‐AT‐trypsin59 and α₁‐AT‐elastase60 complexes remain the only examples of solved crystal structures of covalently linked serpin‐proteinase complexes. In both cases, there is full loop insertion, a tight interface between serpin and proteinase, and enough distortion of the translocated proteinase active site to explain its failure to leave the complex. Where the two structures differ is in the degree of proteinase disordering and proteolytic susceptibility. In the α₁‐AT‐trypsin complex, 37% of the trypsin polypeptide was disordered, while in the α₁‐AT‐elastase complex, less than 10% of the elastase moiety was crystallographically unresolved. The extent to which cognate proteinases are denatured by crushing up against the body of the serpin therefore likely varies in different serpin‐proteinase combinations.40

Mechanistic understanding of how α₁‐AT executes the serpin mechanism of proteinase inhibition is indirectly important for protein engineering efforts and has productively informed multiple efforts to re‐direct the inhibitor's specificity. Gettins and Olson have noted that the complexities of the serpin mechanism place great demands on serpins in general,68 and these demands are increased for engineered serpins that have not benefited from evolutionary selection. Such demands include: metastability; optimal RCL for cognate proteinase recognition; correct length of the RCL for rapid loop insertion; compatibility of RCL residues for integration into β‐sheet A, with the burial of alternate side chains into the hydrophobic interior; and stability of the ensuing complex. There are numerous ways in which a serpin can be inactivated by mutation, either by nature (reviewed in References 11, 12) or in the course of nonproductive protein engineering efforts. Such engineering efforts can be refined by taking advantage of the large literature of loss‐of‐function mutations generated during structure and function studies, such as those involving the proximal44, 62, 63 and distal69 regions of the RCL, changes near the “shutter” region where two strands of β‐sheet A must open to allow RCL insertion,70 RCL length,66 and extensive loop exchange variants in which all or much of the RCL of one serpin has been substituted for the native RCL of another.71, 72, 73, 74 Future efforts may also be aided by advances in molecular modeling, such as that exemplified by a molecular dynamics simulation that predicted and confirmed enhanced inhibition of elastase by α₁‐AT E346K.75

1.3. Measuring the kinetics of inhibition by α₁‐AT

Any effort to enhance α₁‐AT inhibition or re‐direct its specificity requires quantitative tools to characterize variant protein properties. Proteinase inhibition by all inhibitory serpins, including α₁ ‐AT, is a second‐order reaction in which reactants A and B combine to yield product AB, and is generally considered irreversible. The velocity of the reaction is the product of the second‐order rate constant (k₂) and the molar concentrations of the serpin (A) and proteinase (B) reactants. This constant can be determined either continuously under second‐order conditions, using iterative curve‐fitting methods,58 or discontinuously under pseudo‐first‐order conditions.72, 76, 77, 78 In the latter simplification, a large molar excess of serpin to proteinase (typically at least 10‐fold) is employed, and serpin consumption is therefore considered to be negligible. The pseudo‐first‐order reaction velocity becomes the product of the observed pseudo‐first‐order rate constant k_obs and the proteinase concentration, which can be readily determined using amidolysis of chromogenic or fluorogenic substrates to generate product P concentrations at time zero or time t. This constant is readily determined from the slope of a ln(P₀/P_t) versus time in seconds plot; division by the initial serpin concentration then yields k_2. The units of k₂ are L·mol⁻¹·s⁻¹ (or M⁻¹·s⁻¹). Note that some investigators have used different terms from k₂ to describe the rate constant, including k_ass (association) or k_app (apparent) while using the same methodology and units to determine the value of the constant (e.g., 66, 69).

Another parameter of critical importance in characterizing the kinetics of α₁ ‐AT‐proteinase interaction is the stoichiometry of inhibition (SI), defined as the number of moles of serpin required to inhibit a mole of proteinase.76 For plasma‐derived α₁ ‐AT and its arguably most relevant cognate proteinase, neutrophil elastase, this value at physiological ionic strength and pH, should be 1, indicating ideal inhibitor behavior with no substrate activity. Altering the reaction conditions or the serpin's primary structure via mutation can result in an increased stoichiometry as noted above for α₁ ‐AT hinge variants. The SI is determined by incubating varying ratios of α₁ ‐AT with proteinase until the reaction reaches equilibrium (i.e., for several hours) and then determining the residual proteolytic activity versus the serpin/proteinase ratio, extrapolated to zero proteolytic activity.72, 76, 77, 78 SI values are dimensionless.

Although some investigators have employed more complex kinetic models to generate additional rate constants specific for reversible or irreversible steps in the overall serpin pathway,73 k₂ and SI seem to be the most commonly reported parameters. Some investigators have also argued that a true measure of the reaction constant for the branched pathway of serpin inhibition is an apparent association rate constant that is the product of k₂ and SI. While this might be a better description of the combined inhibitory and substrate reactivity of a given serpin and proteinase, readers of this literature must take care to note what kind of constant is being used. For protein engineering purposes, investigators typically attempt to maximize k₂ without elevating SI, to select mutant proteins with the most inhibitory predicted in vivo properties.

2. RATIONALE FOR Α₁‐AT MUTAGENESIS AND ENGINEERING OF NOVEL PROPERTIES

Efforts to engineer α₁ ‐AT have spanned a spectrum of different goals, loosely related to general biotechnological aims such as enhancing protein stability, increasing expression yield, and prolonging circulatory half‐life, and more specific objectives such as re‐orienting α₁ ‐AT specificity to inhibit novel proteinase targets. Different regions of α₁ ‐AT have been mutated to advance these different goals, as highlighted in Figure 2. A nonexhaustive list of α₁ ‐AT mutants and their properties reported to date are provided in Table S1.

Figure 2.

Human α₁‐AT labeled with regions of interest, and examples of protein engineering targeted to these regions. Colors of the text boxes correspond with colored residues and regions in the α₁‐AT protein structure. P1 (M358) and P1′ (S359) residues are shown as sticks, to illustrate where the RCL is cleaved. PDB 3NE4 was used to generate the figure

2.1. Altering stability

Increasing the stability of recombinant proteins is a standard biotechnological goal that can be necessary to produce a robust product for pharmaceutical purposes. Serpins depart from this generality because their inhibitory mechanism demands a release of stored energy on reactive center cleavage, to transpose a target proteinase from one pole to the other. The difference in free energy between native α₁ ‐AT and RCL‐inserted α₁ ‐AT is 32.2 kcal mol⁻¹, suggesting significant strain in the protein.65, 67 This obligatory metastability can leave α₁ ‐AT prone to misfolding into nonfunctional conformations, such as the latent conformation or polymerized conformations adopted by the pathogenic Z and S variants.13, 14

It was initially unknown whether stabilizing α₁ ‐AT against thermal or chemical denaturation would impair its function. Kwon et al. screened randomly mutated α₁ ‐AT variants for retention of function after heating of crude Escherichia coli lysates, finding a 13‐fold slower rate of inactivation for F51C but no effect on inhibitory function.79 Similarly, Kim et al. noted that F51L reduced misfolding of the Z variant,80 helping to link thermostability and the propensity for polymerization, a trend that has since been supported by extensive mutagenesis.81 Lee et al. added six additional mutations to F51L to produce “Multi‐7” α₁ ‐AT, which gained exceptional stability to chemical denaturation or conversion to latent form without losing functionality.82, 83, 84 Multi‐7 α₁ ‐AT was used to show proof‐of‐concept for the use of M13 phage display to screen for hyperstable α₁ ‐AT mutants, however, novel mutations were not reported in this study.85 Seo et al. also used random mutagenesis to identify 50 single amino acid substitutions in α₁ ‐AT that increased its thermal stability; only those stabilizing mutations lying near the RCL or β‐sheet A impaired anti‐proteinase function.84 Taken together these observations suggest that maintaining localized strain is necessary for α₁ ‐AT function, but that global strain can be reduced without impairment.

A simpler form of stabilization has also been applied, by replacing the lone cysteine residue in α₁ ‐AT, C232. This residue is highly reactive due to a positively charged local environment, which biases its thiol group toward thiolate formation, which could increase crosslinking with other cysteine residues.86 The C232S mutation ensures α₁ ‐AT cannot form disulfide‐bonded homodimers or heterodimers, but does not affect the rate of proteinase inhibition.87 This mutation is often included when overexpressing α₁ ‐AT, which improves soluble yields,88 and is treated as wild type protein in assays. While few investigators focusing on protein engineering have incorporated stabilizing mutations other than C232S into their strategies, these studies have been indirectly important because they have facilitated α₁ ‐AT crystallization. Multi‐7 mutations combined with M358R and C232S (i.e., “Multi‐9” α₁ ‐AT) were exploited to obtain crystals of α₁ ‐AT complexes with S195A‐trypsin58 and porcine pancreatic elastase.60

Engineering α₁ ‐AT stability is limited by the serpin inhibitory mechanism itself, and the inherent structure of the starting scaffold which has evolved to retain a certain level of strain. To explore these constraints, Porebski employed consensus protein design; using a multiple sequence alignment of serpins to arrive at a synthetic amino acid sequence that contains the most common residue at each position.89 This consensus serpin, conserpin, contained 137 residue differences versus α₁ ‐AT throughout the protein. Impressively, conserpin folded reversibly in response to chemical denaturation, which α₁ ‐AT cannot, by avoiding intermediate states as it folds. Conserpin also remained folded at temperatures up to 110°C, and resisted polymerization even with the addition of the destabilizing Z mutation. Counterintuitively, the crystal structure of conserpin revealed a lower number of intramolecular noncovalent interactions versus other serpins, but had increased electrostatic and hydrogen bonding at regions key for folding.

Although conserpin was able to inhibit proteinases rapidly, the covalent serpin‐proteinase complexes were not stable.89, 90 Replacing conserpin RCL residues P7–P1′ with P7‐P2′ from α₁ ‐AT, so that the RCL length and identity were closer to a natural serpin, improved SI values although complexes continued to dissociate faster than wild type α₁ ‐AT.90, 91 Interestingly, the protein remained unable to efficiently inhibit the primary cognate proteinase neutrophil elastase, which may be due to changes in RCL structural dynamics and the modified electrostatic potential of the serpin body.91 These studies further stress the important balance between the inhibitory properties of α₁ ‐AT and its stability, which remain challenging engineering independent of one another.

2.2. Modifications to increase recombinant protein yield

With the many systems used to express recombinant human α₁‐AT, there have been several strategies used to increase the functional yield. This section will outline specific modifications made to the α₁‐AT DNA sequence to increase yield, however, there are other variables such as the organism and strain selected, plasmid copy number, codon usage, culturing conditions, and purification methods (as reviewed by others20, 21, 92).

Initial attempts to express α₁‐AT in E. coli, using a DNA sequence lacking the endogenous secretion signal peptide, resulted in functional protein but with relatively low yields.18, 93 Tessier et al. noted that the sequence immediately following this new start codon likely forms a stem‐loop secondary structure, obstructing ribosome initiation in E. coli.94 Modifications to this region by either adding a N‐terminal peptide, introducing silent mutations, or by deleting up to the initial 15 amino acids, resulted in a 10‐ to 200‐fold increase in expression.94, 95, 96 Deleting N‐terminal residues was subsequently found to be crucial for the expression of α₁‐AT RCL hinge mutants in E. coli, which otherwise had negligible expression.62

α₁‐AT sequences containing a truncated N‐terminus, the addition of an N‐terminal affinity tag, or containing various silent mutations are the most commonly used sequences for expression in E. coli, which escape the issue of an obstructed or improper start codon.21 However, the overexpression of recombinant proteins in E. coli often results in the formation of inclusion bodies, insoluble aggregates of protein that accumulate in the cytoplasm, which significantly reduces the recoverable yield of α₁‐AT. Schulze et al. investigated the importance of RCL mutations on α₁‐AT solubility in E. coli,97 noting that that the need for the RCL to insert into β‐sheet A may lead to dimers forming between unstable intermediate molecules of α₁‐AT as they fold, akin to a molecular mechanism proposed for the pathogenicity of the Z E342K mutation.14, 98 The M351E mutation at position P8 in the RCL, significantly reduced aggregation in E. coli, increasing soluble protein yield by over twofold.97 This mutation likely prevents the formation of stabilizing hydrophobic interactions with β‐sheet A, but importantly the M351E mutation did not disrupt inhibitory function.99

The M351E mutation has also been used to increase expression in insect cells, with the additional modification of an N‐terminal fusion protein to facilitate α₁‐AT secretion.100, 101 The addition of a signal peptide or fusion protein is a common technique used to increase recombinant protein expression, and has been used to improve α₁‐AT yields from E. coli,102 yeast,103 and plants.104, 105 In the first study to overexpress α₁‐AT using cultured mammalian cells, modification to the 5′ untranslated region and endogenous signal peptide improved expression up to fivefold,106 however, more recent studies using the wild type endogenous signal peptide achieved superior yields.107, 108 There are therefore a variety of protein and DNA modifications that can be made to α₁‐AT to improve the yield, which is largely dependent on the recombinant expression system used.

2.3. Oxidation resistance

Individuals with α₁‐AT deficiency are predisposed to COPD and emphysema, which is worsened by cigarette smoking, due to the smoke causing oxidation of the remaining α₁‐AT.9, 109 This effect highlights α₁‐AT's susceptibility to oxidation, which may underlay the pathogenesis of emphysema even in individuals without α₁‐AT deficiency.110 α₁‐AT oxidation may additionally be exacerbated by neutrophil‐mediated generation of free radicals during lung inflammation, leading to a further imbalance in the elastase/α₁‐AT stoichiometry in the lungs.8, 111 α₁‐AT's susceptibility to oxidation results from the protein containing nine methionine residues, two of which occur in the RCL: M358 (being the crucial P1 residue recognized by target proteinases); and M351 close to the hinge region of the RCL. That both methionines remain intact, without being oxidized to methionine sulfoxide, is a requirement for efficient inhibition of elastase.111

As mentioned previously, M358V was the first mutation deliberately introduced into α₁‐AT to engineer its properties.37 M358V has since been characterized by many others, revealing that although this mutation modestly reduces the rate of neutrophil elastase inhibition (1.2‐fold to 2.7‐fold lower), it retains inhibitory function after being exposed to chemical oxidation while its wild type counterpart does not.112, 113 M358V also improves the rate of pancreatic elastase inhibition by ≈10‐fold, likely due to this proteinase's preference for valine at P1.4, 112 The double mutant M351V M358V significantly improves oxidation resistance, and maintains neutrophil elastase inhibition.111

Substituting M358 with other hydrophobic residues has also been explored as a way of improving oxidation resistance. Cysteine at this position is not resistant to oxidation, due to the exposed terminal sulfur, while other hydrophobic residues (A/F/I/L) are resistant but with modestly reduced rates of elastase inhibition compared to M358V.113 This finding is somewhat surprising, given neutrophil elastase's preference for A/I/T/V at P1,4 which may suggest a fundamental requirement for methionine for efficient inhibition via the serpin mechanism.

Silberstein et al. recently reported the high‐yield expression of α₁‐AT M358V in Nicotiana benthamiana, and argued that this mutant could be used as an improved biologic drug, a “biobetter,” to replace therapeutic preparations of wild type α₁‐AT currently obtained from plasma.114 The plant‐derived M358V inhibited neutrophil elastase in vitro as well as a current therapeutic preparation of α₁‐AT, tradename Prolastin‐C®, with the additional property of being oxidation resistant.114 Similarly, Zhu et al. reported combining efforts to create a more thermostable α₁‐AT with oxidation resistance, also to be used as a bio‐better.115 The expression of F51L/M351V/M358V in E. coli was robust, with the F51L stabilizing mutation increasing the denaturation temperature by 6°C, in addition to being oxidation resistant.115 These developments represent steps toward engineered α₁‐AT being used for replacement therapy, benefiting from the extensive work understanding α₁‐AT inactivation by oxidation, and efforts to improve the overall stability of the protein.

2.4. Thrombin inhibition

Over 40 years ago, the rare α₁‐AT “Pittsburgh” variant was found to cause a severe and ultimately fatal bleeding disorder by converting the protein into a potent thrombin inhibitor.116 The discovery that a single α₁‐AT mutation was responsible, M358R, confirmed the suspected location of the reactive site and indicated that mutations in this region could significantly change α₁‐AT's inhibitory properties.15, 117 It was subsequently noted that α₁‐AT M358R also inhibited the contact factor enzymes FXIa, FXIIa, and kallikrein.51, 118 This profile encouraged investigators to express α₁‐AT M358R in yeast and to test in a piglet model of Pseudomonas aeruginosa septicemia, where it reduced mortality and prevented consumption of antithrombin (SERPINC1) and FXI.119 However, yeast‐derived α₁‐AT M358R failed to prevent mortality in an E. coli septicemia model in baboons and may have exacerbated coagulopathy.120 The negative outcome may have arisen due to the use of recombinant α₁‐AT M358R preparations containing cleaved serpin and yeast‐specific glycans, and/or due to additional properties of α₁‐AT M358R such as the ability to inhibit APC121 and plasmin.

The difficulties encountered in these initial attempts at preclinical use of α₁‐AT M358R prompted several investigators to hone α₁‐AT M358R to target thrombin more specifically, in hopes of developing a heparin‐independent thrombin inhibitor and antithrombotic agent. Single and double substitutions of the P2 and P1' residues of α₁‐AT for the corresponding residues of antithrombin or C1INH, both very poor inhibitors of APC, yielded no increase in selectivity (defined quantitatively as the quotient of the k₂ for thrombin and the k₂ for APC).122 In contrast, Hopkins et al. found that substituting the P7‐P2 residues of antithrombin for those of α₁‐AT improved selectivity; the most selective variant found in this study, α₁‐AT LS, replaced α₁‐AT M358R's P7‐P2 and P2′‐P3′ residues with those of antithrombin (FLEAIPRSIP to AVVIAGRSLN).122 The LS variant exhibited a 10,000‐fold reduction in its rate of APC inhibition, at the cost of a twofold reduction in the rate of thrombin inhibition, for a selectivity factor of ≈8,000. A follow‐up study demonstrated faster inhibition of thrombin for a less extensive variant (lacking the I361N mutation) but greater residual anti‐APC activity, for a selectivity factor of ≈3,200.123

Our group initially attempted to transfer the thrombin specificity of heparin cofactor II (HCII; SERPIND1) by substituting its RCL for that of α₁‐AT from P16 to P3′, excepting the P1 Arg. This mutagenic effort reduced the rate of thrombin inhibition by 3.3‐fold, but that of APC by 68‐fold, with a selectivity factor of ≈160; the SI was also elevated threefold by this manipulation.72 Appending residues 1–75 of HCII, a negatively charged extension that binds thrombin exosite 1, to the N‐terminus of α₁‐AT M358R enhanced the rate of thrombin inhibition by α₁‐AT M358R by 21‐fold.77 The fusion protein was termed HAPI M358R for HCII(1–75)‐ α₁‐PI M358R. Our group subsequently combined the HCII fusion approach with additional RCL mutations: HAPI RCL4 (incorporating P16‐P3′ HCII residues and M358R); and HAPI RCL5 (incorporating the LS variant RCL from122).124 HAPI RCL5 inhibited thrombin 13‐fold more rapidly than α₁‐AT M358R, and APC 108‐fold less rapidly, for an overall selectivity factor of ≈10,000. The chimera inhibited FXa and FXIa 2‐ to threefold less rapidly than thrombin, and FXIIa 14‐fold less rapidly than α₁‐AT M358R.125 At a dose of 7.2 mg/kg body weight, E. coli‐derived HAPI RCL5 was superior to α₁‐AT M358R in reducing thrombus weight in the vena cava of mice treated with ferric chloride and in minimizing fibrin deposition in murine endotoxemia.125 Grafting another thrombin exosite 1‐binding domain derived from the leech inhibitor hirudin onto α₁‐AT M358R also enhanced its rate of thrombin inhibition, but to a lesser extent than HCII 1–75.126

Our group also used a T7 phage display system to fully randomize the P7‐P3 residues of α₁‐AT M358R, and to select two novel variants that enhanced the rate of thrombin inhibition by twofold via directed evolution: DITMA and AAFVS (compared to wild type FLEAI residues).78 This system can be exploited to vary different portions of the RCL and/or to probe with different proteinases via positive or negative selection.127

2.5. Inhibition of factors IXa and Xa

While most α₁‐AT protein engineering efforts have focused on the RCL, alterations to strand 3 of β‐sheet C have been shown to create an exosite facilitating its interactions with FXa and FIXa.128 Building on their previous demonstration of such an exosite in antithrombin,129 Izaguirre et al. altered two residues in α₁‐AT M358R (K222Y and L224E) lying under the RCL in the body of the serpin, and recreated the exosite. The rate of inhibition of FXa was increased 14‐fold versus α₁‐AT M358R in the variant, and 11‐fold with respect to FIXa, without altering the reaction stoichiometry. Substituting a FXa recognition site IEG for the P4‐P2 sites of α₁‐AT M358R increased the k₂ for FXa 10‐fold, and combining this motif with K222Y and L224E effected an overall 20‐fold and 26‐fold increased rate of Xa and IXa inhibition versus α₁‐AT M358R, respectively.128 Much of the increased rate of inhibition could be ascribed to the increased affinity of the reactants and the resulting increased ease of forming a Michaelis complex.

In another example of extra‐RCL engineering, Yang et al. substituted the heparin‐binding D helix of antithrombin for the native residues in α₁‐AT. The α₁‐AT/D helix chimera acquired a 100‐fold acceleration of thrombin inhibition in the presence of heparin, as well as in vivo anti‐inflammatory properties thought to arise from its new ability to bind to heparan sulfate proteoglycans of the endothelium and mediate vascular protective effects.130

2.6. Inhibition of activated protein C

Polderdijk et al. sought to enhance the inhibition of APC by α₁‐AT, in hopes of developing a new treatment agent for hemophilia.131 Extrapolating from clinical observations that the severity of hemophilia A or B was lessened in patients whose factor V (Leiden) was resistant to inactivation by APC, these investigators reasoned that effectively inhibiting APC could preserve enough prothrombinase activity to remedy either bleeding diathesis. Using structural comparisons of the active sites of thrombin and APC, they designed an α₁‐AT M358R variant with Lys substitutions at P2 and P1′ (α₁‐AT KRK). This variant exhibited no detectable inhibition of thrombin, a sevenfold slower inhibition of APC than α₁‐AT M358R, a 400‐fold reduced rate of FXa inhibition, and a 1,000‐fold reduction in the rate of FXIa inhibition. In hemophilia B fIX knockout mice, α₁‐AT KRK administered subcutaneously at 15 mg/kg body weight eliminated the bleeding diathesis and reduced bleeding to levels seen in wild type mice, using a tail clip model. If these promising results hold in further preclinical development, α₁‐AT KRK could become a drug candidate for hemophilia A and B treatment, one of a new class of “inhibitors of coagulation inhibitors.”132 The same group, in a follow‐up study, randomized the P2 and P1' residues and screened a library of α₁‐AT M358R variants, but did not find one superior to α₁‐AT KRK for APC inhibition.133

2.7. Inhibition of contact pathway proteinases

Researchers recently engineered α₁‐AT M358R to inhibit plasma kallikrein (PKa).134 de Maat et al. sought a novel therapeutic agent to control disorders of excessive bradykinin generation, which arises from PKa digestion of high molecular weight kininogen. One such disorder is hereditary angioedema, which results from C1 esterase inhibitor (C1INH, SERPING1) deficiency, which releases FXIIa, FXIa, and PKa from C1INH control. Double mutant α₁‐AT P357A/M358R exhibited a fivefold increase in the rate of PKa inhibition, and a similar decrease in thrombin inhibition.135 Similarly, synthetic peptide cleavage studies were leveraged to alter α₁‐AT to PFR at P3‐P1, which retained much of α₁‐AT M358R's activity against PKa, but not complement C1s, FXIIa, or APC.136 Recently, investigators selected P4‐P1′ SMTRS and SLLRS, from the activation peptide of FXII and peptide substrate libraries, respectively, to re‐orient α₁‐AT specificity. These variants inhibited PKa, FXIIa, and plasmin, but also retained residual anti‐thrombin and anti‐APC activities. The latter were eliminated in α₁‐AT‐SLLRV, which exhibited k₂ values of 1.88 × 10⁵ M⁻¹ s⁻¹ for PKa and 1.34 × 10⁴ M⁻¹ s⁻¹ for FXIIa. This variant, when given to mice at 8 mg/kg body weight, reduced ferric chloride‐induced arterial thrombosis, prevented carrageenan‐induced paw swelling, and reduced epithelial leakage of fluorescent dextran in the dextran sulfate colitis model.134

2.8. Inhibition of other proteinases

The well‐characterized structure and function of α₁‐AT, the lack of a requirement for a GAG cofactor for optimal activity, and its promiscuous activity, has also made this serpin a useful scaffold for engineering inhibition of proteinases beyond the coagulation cascade. The human serine proteinase furin, a proprotein convertase, was an early target for engineered inhibition, as furin‐mediated cleavage of viral glycoproteins (gp) facilitates HIV‐1, measles, and influenza infection.137 By introducing a furin consensus sequence into the α₁‐AT RCL via two mutations (A355R M358R), Anderson et al. demonstrated that α₁‐AT could be converted to a furin inhibitor, which blocked the processing of HIV‐1 gp160 required for cellular invasion.138 This engineered “α₁‐PDX” protein has since been shown reduce the infectivity of HIV‐1 and measles in vitro,139, 140 as well as reduce tumor growth and atherosclerosis in mice.141, 142 The latter findings likely stem from furin's role in processing the pro‐peptide of human metalloproteinases involved in tumor cell migration and endothelial remodeling.141

Efforts to improve the inhibition of furin and other proprotein convertases, by adding additional positively charged residues to the RCL, have had mixed results.143, 144 However, transplanting the P6‐P1 RCL residues from the natural furin inhibitor Serpin B8 into α₁‐AT improved the rate of furin inhibition twofold versus α₁‐PDX, and surprisingly 10‐fold versus the original Serpin B8.145 Selectivity toward other proprotein convertases could be tuned via additional mutations, by introducing P′ RCL residues from Serpin B8 or Serpin B8 exosite residues.146 α₁‐AT, therefore, appears to be an attractive and tunable scaffold for engineering inhibitors of a variety of proprotein convertases.

Serpins are also capable of inhibiting cysteine proteinases, which are not primary targets of α₁‐AT. To understand the basis for cysteine proteinase inhibition, Irving et al. introduced RCL residues from the serpin SCCA‐1, which inhibits papain‐like cysteine proteinases, into α₁‐AT.147 The resulting chimeric serpins inhibited cathepsins L, V, and K rapidly (k₂ up to 1.32 × 10⁶ M⁻¹ s⁻¹) with low SI, provided that the RCL length remained similar to α₁‐AT.

To engineer α₁‐AT to inhibit kallikrein‐14, rather than borrowing RCL residues from known serpins, Felber et al. used an alternative approach.49 Peptides first identified as optimal kallikrein‐14 substrates by phage display were inserted into the α₁‐AT RCL, yielding novel serpins with inhibitory activity toward several human kallikreins. However, the serpin anti‐chymotrypsin was ultimately found to be a better scaffold than α₁‐AT for engineering kallikrein‐14 inhibition, illustrating the importance of the original serpin sequence.

2.9. Increasing circulation half‐life

Pharmacokinetic studies of plasma‐derived α₁‐AT concentrates administered to α₁‐AT‐deficient patients have established a long circulatory half‐life (t_1/2) of 7.7 to 8.9 days.148 While nonglycosylated recombinant α₁‐AT produced in E. coli demonstrates highly similar kinetics of enzyme inhibition in vitro,20 its t_1/2 is much shorter.149 This accelerated clearance can be ameliorated by chemical modification. Cantin et al. exploited the lone free thiol (C232) and conjugated polyethylene glycol (PEG) chains of varying lengths to this residue.149 Although these investigators did not calculate pharmacokinetic parameters, they found that the residual amount of recombinant α₁‐AT conjugated to 20 or 40 kDa PEG, but not 5 kDa PEG, did not differ from that of plasma‐derived α₁‐AT in mice up to 24 hr post‐injection.

Chinese Hamster Ovary (CHO) cells and more recently human embryonic kidney (HEK293) cells have been used to express recombinant human proteins that have been approved by regulators for clinical use. Functional expression of wild type recombinant α₁‐AT in CHO and HEK293 cells has been reported, although without pharmacokinetic information.90, 91 Amann et al. engineered CHO cells, combining CRISPR/Cas9‐mediated disruption of 10 rodent glycosylation‐related genes and overexpression of human St6 β galactoside α‐2,6‐sialyltransferase 1 (ST6GAL1), to produce recombinant α₁‐AT with N‐linked glycan structures more closely resembling those in plasma‐derived α₁‐AT than in standard CHO‐derived α₁‐AT.108

Several groups have attempted to prolong α₁‐AT t_1/2 via glycoengineering, which is the addition or modification of glycans on the protein. By modifying α₁‐AT's existing sugars with sialylation enzymes, Lindhout et al. increased the t_1/2 from 5 to 27 hr in mice.150 Lusch et al. introduced additional glycosylation consensus motifs N‐x‐S/T at residues N90, N123, and N201, finding that that additional glycosylation at only N123 was superior, which increased the serum t_1/2 of HEK293‐derived α₁‐AT by 63%.151 Similarly, Chung et al. tested the introduction of seven different N‐x‐S/T motifs, finding that the fourth glycan at Q9N or D12N/S14T increased the t_1/2 of CHO‐derived α₁‐AT by 2.2‐fold versus the wild type in rats.152

The constant portion of human IgG1 (Fc) has been fused to proteins of interest to prolong their t_1/2 by imparting the property of recycling via the neonatal Fc receptor.153 An α₁‐AT‐Fc fusion protein has been reported to lack anti‐elastase activity154, 155; presumably, this arises either from steric hindrance due to the dimerization of the α₁‐AT‐Fc protein engendered by two disulfide bonds in the Fc hinge region near the fusion point, or an indirect conformational distortion of the α₁‐AT moiety. The strategy would, therefore, be unlikely to be effective for recombinant α₁‐AT proteins being developed as novel inhibitors of any proteinase. The fusion protein is instead being developed by those interested in anti‐inflammatory effects of α₁‐AT that are independent of anti‐proteinase function, such as regulation of tissue necrosis factor and neutrophil activation.34, 154, 155 Pharmacokinetic studies have not been reported, although the α₁‐AT‐Fc fusion exhibited prolonged anti‐inflammatory effects in mouse models of gouty arthritis and ischemia–reperfusion.156, 157

3. CONCLUDING REMARKS

Considerable research effort has gone into engineering the function and specific properties of the serpin α₁‐AT. While these efforts have reaped many dividends, including an expanded understanding of serpin structure and function, the field is not yet at the point where α₁‐AT variants can be designed a priori and then confirmed to function as desired. Attaining such an ideal state will require, at a minimum, new advances in protein structural modeling and molecular dynamics simulations in the long term. Even in the case of the synthetic thermostable conserpin, it remains difficult to alter specificity productively due to the many demands of the serpin mechanism.91 In the medium term, it is likely that combining different approaches, such as phage display or serpin‐proteinase docking data to select α₁‐AT variants and then reconfiguring them into long‐lasting injectable protein drugs, will be undertaken. In the shorter term, it is likely that protein engineering modifications improving the pharmacokinetics of wild type α₁‐AT will make feasible clinical trials for new indications. With respect to engineered serpins, Serpin PC (also called α₁‐AT KRK),131, 158 and the α₁‐AT‐Fc fusion protein are poised at this writing to enter early clinical trials.159 In so doing they will join only one other engineered serpin that has taken these steps to the clinic, the viral serpin Serp1, which showed promise in a small phase II randomized clinical trial in patients with acute coronary syndromes.160

In spite of the intensity of mutagenesis undertaken to date (partially summarized in Table S1), there remain areas of the protein in which more investigation should be focused: the salt bridge between P5 and the body of the protein, which is key to RCL flexibility161; the “south pole” of α₁‐AT, an interface that could be modified to facilitate proteinase migration and retention; and the β‐sheet A strands into which the RCL must insert to form complexes. Few investigators have considered the possibility of immunogenicity of engineered serpins, and this possibility should be addressed in silico and in animal studies to see if this is a likely scenario, and one for which immune‐camouflage approaches like PEGylation might be appropriate.

The abundance of α₁‐AT, its ease of purification, and its early association with human disease made it an archetype for serpins. Research to date has identified its value as an adaptable scaffold for serpin investigation, and in the future, this value will likely be extended into clinical use along the lines outlined in this review.

CONFLICT OF INTEREST

The authors have no conflicts of interest to declare.

Supporting information

Supplemental Table 1 An Excel file containing a descriptive list of over 200 published engineered α₁‐AT variants (Scott_Sheffield_SupTable1_191112.xlsx). The list is non‐exhaustive, meant to provide an overview of the diversity of α1‐AT mutants published. The most recent version is available via Figshare https://doi.org/10.6084/m9.figshare.10294889

Scott BM, Sheffield WP. Engineering the serpin α₁‐antitrypsin: A diversity of goals and techniques. Protein Science. 2020;29:856–871. 10.1002/pro.3794