Systematic Structure-Based Virtual Screening Approach to Antibody Selection and Design of a Humanized Antibody Against Multiple Addictive Opioids Without Affecting Treatment Agents Naloxone and Naltrexone

Abstract

Opioid drugs, especially heroin, are known as a growing national crisis in America. Heroin itself is a prodrug and is converted to the most active metabolite 6-monoacetylmorphine (6-MAM) responsible for the acute toxicity of heroin and then to a relatively less-active metabolite morphine responsible for the long-term toxicity of heroin. Monoclonal antibody (mAb) is recognized as a potentially promising therapeutic approach to treatment of opioid use disorders (OUDs). Due to the intrinsic challenges of discovering an mAb against multiple ligands, here we describe a general, systematic structure-based virtual screening and design approach which has been used to identify a known anti-morphine antibody 9B1, and a humanized antibody h9B1, capable of binding to multiple addictive opioids (including 6-MAM, morphine, heroin, and hydrocodone) without significant binding with currently available OUD treatment agents naloxone, naltrexone, and buprenorphine. The humanized antibody may serve as a promising candidate for treatment of OUDs. The experimental binding affinities reasonably correlate with the computationally predicted binding free energies. The experimental activity data strongly support the computational predictions, suggesting that the systematic structure-based virtual screening and humanization design protocol is reliable. The general, systematic structure-based virtual screening and design approach will be useful for many other antibody selection and design efforts in the future.

Graphical Abstract

Introduction

Opioid drugs, especially heroin,¹ are known as a growing national crisis in America due to the rapidly increasing overdose deaths,² and “heroin is scarring the next generation.”³ The rapid increase in heroin overdose deaths is related to the fact that heroin is much cheaper and easy to obtain, compared to other opioid drugs. In fact, heroin has become much cheaper than any other drug of abuse, e.g. $10-$20 for a typical single dose (0.1 g) of heroin purchased on the street.⁴ Notably, heroin itself is actually a prodrug and is converted by cholinesterases^{5, 6} to the highly active metabolites 6-monoacetylmorphine (6-MAM) and morphine, as depicted in Figure 1.^7–10 Heroin can cross the blood-brain barrier and be rapidly converted to 6-MAM in brain.¹¹ In addition, both 6-MAM and morphine have higher binding affinities with μ-opioid receptor compared to heroin itself.¹² So, both 6-MAM and morphine are responsible for the toxicity and physiological effects of heroin.^13–19 6-MAM is mainly responsible for the acute toxicity of heroin due to its higher activity and shorter half-life, whereas morphine is mainly responsible for the long-term toxicity of heroin due to its longer half-life.⁶ The opioid drugs exert their main physiological effects through activating μ-opioid receptor in the brain.

Figure 1.

Molecular structures of heroin and its key metabolites, as well as FDA-approved prescription drugs naloxone and naltrexone for treatment of heroin overdose and dependence, respectively.

The well-known connection between heroin abuse and prescription opioid abuse is related to the actual availability and costs, in addition to the common brain protein targets (opioid receptors), of these opioid drugs. In fact, “80% of recent heroin initiates reported that they began their opioid use through the nonmedical use of prescription opioid medications.”⁴ Those who abuse the prescription drugs most often obtain them from friends and family either through sharing or theft. When they are no longer able to get prescription opioid drugs, they start to use illegal opioid heroin, because heroin is easy to obtain on the street and even online and cheap as noted above.

Currently used therapeutic agents for treatment of opioid-induced disorders/toxicity include naloxone (a non-selective and competitive antagonist of opioid receptors) used for overdose treatment and buprenorphine (a partial agonist of μ-opioid receptor and an antagonist/partial agonist of many other receptors), methadone (an agonist of μ-opioid receptor, an antagonist of glutamatergic N-methyl-D-aspartate receptor, and a noncompetitive α3β4 neuronal nicotinic acetylcholine receptor), and naltrexone (a competitive antagonist of μ-opioid receptor and other opioid receptors) used for opioid dependence treatment. These therapeutic agents may be used in various formulations/devices, such as a nasal spray device for naloxone²⁰ for fast toxicity treatment and extended-release naltrexone for relapse-preventing opioid dependence treatment.²¹ All of these therapeutic agents in current clinical use, and most of other therapeutic candidates under preclinical/clinical development, bind to opioid receptors (and/or related receptors) in the brain and, thus, block/regulate the physiological effects of opioid in the body. Indeed, the overdose treatment with naloxone appears to be effective in many cases, but the precondition is that the naloxone treatment can begin soon enough after an opioid overdose. Further, once overdosed, heroin-dependent users may continue to get overdosed again and again until fatal. Some heroin-dependent users survived from one overdose with treatment in hospital, and then died of another overdose the next day.⁴ Even worse, the use of naltrexone or its extended-release formulation Vivitrol actually increased heroin overdose.^22–24 A truly effective heroin treatment should account for not only rescuing heroin users who have already been overdosed, but also preventing the users from overdose again.

It is highly desirable to explore alternative therapeutic strategies to complement the traditional μ-opioid receptor antagonist approach for treatment of heroin-related opioid overdose and dependence. As alternatives to the traditional μ-opioid receptor antagonist approach, vaccines (that help to elicit antibodies against specific antigens in the body) and monoclonal antibodies (mAbs, for use as the passive vaccination/immunity) have been being developed for treatment of opioid use disorders (OUDs).^25–35 A vaccine could be effective for the dependence treatment, but would require an immune response to be effective and, thus, would not be useful for the overdose treatment. An mAb could be useful for treatment of both the drug dependence and overdose. In particular, an exogenous mAb, which may be used as an exogenous protein therapeutic, is not expected to cross blood-brain barrier (BBB) to interact with any receptors in the brain. Instead, through tightly binding with the opioid drugs in the plasma, the exogenous antibodies are expected to decrease the concentrations of freely available opioids and, thus, attenuate the toxicity and physiological effects of the opioid drugs.

Concerning the feasibility for using an mAb to attenuate the drug toxicity for overdose treatment in addition to the dependence treatment, Janda and associates have demonstrated in mice that an anti-cocaine mAb can be used as an effective antidote to rescue mice after the mice were given a lethal dose of cocaine (post-exposure treatment).³⁶ Based on their encouraging animal data, Janda et al. concluded that “minimal antibody doses were shown to counteract the lethality of a molar excess of circulating cocaine” in the case of the post-exposure treatment and “Passive vaccination with drug-specific antibodies represents a viable treatment strategy for the human condition of cocaine overdose.”³⁶ The work reported by Janda et al. have clearly demonstrated the general concept of utilizing an mAb as a feasible antidote to counteract the drug toxicity post-exposure. It is reasonable to hypothesize that the similar concept can be applied to an anti-opioid mAb as an antidote to counteract opioid toxicity for the overdose treatment, in addition to the dependence treatment.

In fact, there have been various reports of efforts to generate mAbs specifically against morphine^{26–31, 37} or 6-MAM.^32–34 Of the anti-morphine mAbs reported so far, a single chain Fv antibody³¹ can also bind with heroin. There has been no demonstration that any of the reported anti-morphine mAbs^{26–31, 35} can bind with 6-MAM. The reported anti-6-MAM mAbs^32–34 specifically recognize 6-MAM, but not morphine. There has been no demonstration that any of the mAbs reported so far^{26–34, 37} can bind with both 6-MAM (which is mainly responsible for the acute toxicity of heroin) and morphine (which is mainly responsible for the long-term toxicity of heroin), let alone binding with all of the three heroin-related opioids (6-MAM, morphine, and heroin itself). Interestingly, the 6-MAM-specific mAb (known as 6-MAM-214, with an affinity of 0.3 μM or 300 nM for 6-MAM) was indeed able to reduce the acute heroin effects in mice.³² Thus, we may reasonably assume that an mAb capable of binding with all the three heroin-related opioids (6-MAM, morphine, and heroin itself) would be able to more effectively attenuate the toxicity (including both the acute and long-term toxicity) and physiological effects of heroin. An ideal mAb as a new therapeutic candidate should not affect currently available OUD treatment options, particularly naloxone and naltrexone. Hence, our present study was aimed to identify/design an mAb capable of potently binding with all the heroin-related opioids without binding with naloxone and naltrexone.

In general, it is increasingly interesting to develop mAbs as therapeutic proteins. Usually, antibodies may be generated either in vitro, such as the phage and yeast surface display, or in vivo through animal immunization and antibody screening using enzyme-linked immunosorbent assay (ELISA) or Western blot assays, followed by humanization of the identified animal antibody.³⁸ There are a lot of challenges in generating mAbs for therapeutic applications. For example, using the in vivo approach during the antibody discovery stage, immunization affords limited control over antibody affinity and specificity due to the difficulty in controlling antigen presentation to the immune system. Using in vitro methods such as the phage and yeast surface display, a display method is limited by the need of screening a large library etc.³⁸ Hence, it has been challenging to develop an mAb with the ability to bind with all of the three heroin-related opioids (6-MAM, morphine, and heroin itself) or even the ability to bind with both 6-MAM and morphine. On the other hand, as many efforts have been made to discover anti-morphine or anti-6-MAM antibodies, it is also possible to infer that some of these known antibodies might be able to meet the aforementioned criteria of an ideal antibody for heroin overdose and addiction treatment.

Our study started from the development and use of a novel, systematic structure-based virtual screening approach to identifying an mAb (from the currently known antibodies) with a potentially high binding affinity to 6-MAM, morphine, and heroin without significant binding affinity to naloxone and naltrexone. The virtual screening was followed by molecular design for sequence humanization and wet experimental tests. The combined computational and experimental study has led to discovery of a humanized mAb capable of potently binding with multiple addictive opioids (including all the heroin-related opioids and other opioids of abuse) without significant binding with naloxone and naltrexone, demonstrating that the systematic structure-based virtual screening approach is feasible and efficient.

Results and Discussion

Prediction from virtual screening

According to our virtual screening of available antibodies, including all of those (listed in SI Table S1) with known experimental binding affinity to any of the heroin-related opioids, 6-MAM may potentially bind with the 26 known antibodies listed in SI (Table S1), but with different binding free energies ranging from −16.6 kcal/mol (for the most potent one, i.e. Ab13) to −2.0 kcal/mol (which represents a negligible binding affinity) (see SI Table S2). Hence, we further computationally estimated the binding free energies of these 26 antibodies with morphine, heroin, naloxone, and naltrexone. All the calculated binding free energies are summarized in SI (Table S2) in comparison with available experimental data.

Interestingly, for many of these antibodies listed in Table S2, experimental binding free energies (converted from the previously reported K_d, K_i, or IC₅₀ values according to the well-known thermodynamic equation) have already been known for their binding with morphine and naloxone. According to the experimental binding affinities summarized in Table S2, the most potent anti-morphine antibodies (Ab1 to 9), Ab15, and Ab16, can also potently bind to naloxone with nanomolar K_d or IC₅₀ values (ranging from 33 nM to 145 nM). Ab15 even can potently bind with naltrexone (IC₅₀ = 310 nM). A therapeutic antibody capable of binding with naloxone or naltrexone would be problematic because it would block the favorable pharmacologic action of naloxone (the currently available therapeutic agent for opioid overdose treatment) or naltrexone (the currently available therapeutic agent for opioid dependence treatment). For this reason, these potent anti-morphine antibodies would not be suitable for consideration as the desirable therapeutic candidates. Nevertheless, the available experimental data for enough number of antibodies binding with morphine and naloxone allowed us to analyze the potential correlation between and computational and experimental data and, thus, validate our computational protocol. In comparison, only six antibodies have experimental data available for their binding with 6-MAM, with a very narrow range of the K_d values (100 to 300 nM) that had limited our correlation analysis for 6-MAM before we obtained further experimental data to expand the range (see below).

As showed in Figure 2, the predicted binding free energies correlate with the corresponding experimental data very well, with the correlation coefficient being 0.6938 and 0.7635 for morphine and naloxone, respectively. The good correlation between the computational and experimental data suggests that our computational protocol for predicting the binding structures and relative binding free energies is reasonable. Thus, we were confident to predict the potentially most potent antibody for 6-MAM based on the computational data obtained from the use of the same computational protocol.

Figure 2.

The linear correlation between the predicted binding free energies with the experimentally derived binding free energies (converted from the reported binding affinity data listed in SI Table S2 by using the well-known thermodynamic equation) for morphine (A) and naloxone (B).

Within all antibodies listed in Table S2, Ab13 was predicted to have a relative low binding free energy with morphine, 6-MAM, and heroin (corresponding to the calculated binding free energies of −15.1, −15.6, and −16.6 kcal/mol, respectively); a lower binding free energy means a higher binding affinity. Interestingly, Ab13 was predicted to have a relative high binding free energy (i.e. the lowest binding affinity) with naloxone and naltrexone (corresponding to the calculated binding free energies of −1.8 and −3.2 kcal/mol, respectively). In other words, Ab13 was predicted to be the most promising antibody capable of potently binding with 6-MAM, morphine, and heroin without significant binding affinity with naloxone and naltrexone. Notably, Ab13 was reported as 9B1 in literature.³⁵ Although the original report³⁵ did not report the determination of the exact binding affinity of 9B1 with morphine, the binding constant of 9B1 with morphine was estimated to be “~10⁹ M⁻¹” (i.e. the low nanomolar range), which suggests that the computational binding free energies are reasonable. In addition, the high-resolution X-ray crystal structure (up to 1.6 Å) of Ab13 was reported.³⁵ Therefore, we selected Ab13 (i.e. 9B1) for further studies. For convenience, we will use the original name of Ab13, namely 9B1, in the discussion below.

Binding modes of 9B1 with heroin-related opioids

We further analysed the predicted binding mode of 9B1 with 6-MAM, morphine, and heroin. As the predicted binding mode of 9B1 with morphine was nearly identical to its X-ray crystal structure complex (PDB ID: 1Q0Y) with RMSD of 0.078, therefore, the X-ray crystal structure complex together with the predicted binding structures of 6-MAM and heroin, were analysed and presented in Figure 3. As shown in Figure 3A, I59, W33, L101, and Y104 of the heavy chain and Y34, W93, and L98 of the light chain form a hydrophobic pocket and encase 6-MAM. E50 of the heavy chain has strong ionic interactions with the cationic nitrogen of 6-MAM, and this cationic nitrogen also forms π-cation interactions with W93 of the light chain and W99 of the heavy chain. The 6-acetyl group of 6-MAM forms a hydrogen bond with the W33 side chain of the heavy chain. In addition, as shown in Figure 3A, the methyl substituent on the nitrogen of 6-MAM has favorable Van der Waals interaction with W99 side chain of the heavy chain and L98 side chain of the light chain; replacing this methyl group with allyl group (naloxone) or cyclopropylmethyl group (naltrexone) would lead to clash with the W99 and L98 residues, which is likely the reason why both naloxone and naltrexone were predicted to have negligible binding affinity with 9B1. Compared to 6-MAM, morphine and heroin employ a similar binding mode in binding with 9B1 (see Figure 3B and 3C). Besides the interactions of 6-MAM with 9B1, the 3-acetyl group of heroin has favorable hydrophobic interactions with L101 and Y104 side chains of the heavy chain, whereas morphine has lost the hydrogen bond between 6-MAM with W33 of the heavy chain. These binding features further explain the computational insights from the above-mentioned binding free energy calculations that 9B1 is expected to have favorable binding with 6-MAM, heroin, and morphine without favorable binding with naloxone and naltrexone. Here it should be noted that one might also be able to predict that 6-MAM and heroin could fit into the paratope pocket of 9B1 by manually checking its X-ray crystal structure complex with morphine. However, it is the virtual screening approach which allowed us to independently select 9B1 as our final choice in comparison with other antibodies even without using the X-ray crystal structure of 9B1.

Figure 3.

The predicted binding modes of 9B1 with 6-MAM (A) and heroin (C), as well as the X-ray crystal structure of morphine with 9B1 (B) (PDB ID: 1Q0Y). Yellow dashed lines indicate the hydrogen bonds.

9B1 humanization design

9B1 was acquired from mouse by using the hybridoma technique.³⁵ To make use of this antibody for future use in human, 9B1 must be humanized to reduce immune response. Currently, many methods have been used in antibody humanization, the most wildly used methods are complementarity determining regions (CDRs) graft and antibody resurfacing.³⁹ As CDRs graft change the framework regions (FRs) of the humanized antibody, certain key residues in FRs of 9B1 play an important role in binding with 6-MAM, heroin, and morphine according to the modelled binding structures (see Figure 3). These critical residues should remain unchanged during the process of humanization. Therefore, our 9B1 humanization design was carried out by using an MD simulation and bioinformatics analysis-based resurfacing humanization method. Initially, the amino-acid sequences of the variable domains of the heavy (VH) and light (VL) chains of 9B1 were analyzed using the IMGT/DomainGapAlign tools,⁴⁰ and the closest human germline V and J genes were identified. As shown in Figure 4, the top-ranked human germline sequences for VH are IGHV1–46*01, IGHV1–69*02, and IGHJ4*01, and the top-ranked human germline sequences for VL are IGLV7–46*01, IGLV8–61*01, and IGLJ3*01. In order to maintain the binding pocket for 6-MAM, heroin, and morphine, murine amino-acid residues within the binding pocket were retained. The framework residues that are buried in the structure were considered potentially important for maintaining the conformation of the variable region and, thus, were also retained in the humanized antibody. Other murine residues that differ from the human counterpart were replaced. Following these criteria, one variable heavy chain (H1) and three possible light chains (L1, L2, and L3) of the humanized antibody were proposed (see Figure 4).

Figure 4.

Alignment of amino-acid sequences of 9B1, top-ranked human antibodies, and the humanized antibodies. The heavy chain and light chain are presented in panels A and B, respectively. Highly similar residues are colored in red and framed in blue. The arrow, “TT” and helix indicate the β-sheets, turns and α-helixes domain of 9B1. “1” in green indicate disulfide linkage within the heavy chain or light chain.

Further, for assessment of the structural stability of the humanized antibody within the above three possible choices of the humanized antibody, we carried out MD simulations to examine the fluctuation of the residues in the binding pocket and the backbone atoms of the three possible humanized antibodies (denoted as H1L1, H1L2, and H1L3). For comparison, MD simulations were also performed on the original structures of 9B1 and its complex with morphine, and the time-dependent root-mean-square deviations (RMSDs) for all the simulated structures are provided in SI Figure S2. Based on the time-dependent RMSD data (Figure S2), the structures of 9B1 and its complex with morphine were stable during the MD simulations, as the RMSDs of the pocket residues and backbone atoms were only ~0.5 and ~1.1 Å, respectively. The ligand (morphine) in the complex had a very low fluctuation, with the RMSD being ~0.25 during the simulation, showing the stability of the complex.

In comparison, the pocket residues of H1L1 were stable during the MD simulation on H1L1 without morphine bound, but less stable during the MD simulation on the H1L1-morphine complex. On the contrary, the pocket residues of H1L2 were less stable during the MD simulation on H1L2 without morphine bound, as the RMSDs continually increased during the simulation, although the simulated H1L2-morphine complex structure was relatively more stable. Within the three humanized antibodies, H1L3 has the most stable structure, as reflected by the low RMSDs (for both the pocket residues and backbone atoms, as well as morphine bound) that were stable during the MD simulations on H1L3 with or without morphine bound. So, H1L3 was our final choice of the humanized antibody, denoted as h9B1 for convenience.

Experimental validation of the designed antibodies

To prepare the identified antibody 9B1 and the humanized version (h9B1), we assembled the sequences of heavy and light chains with the immunoglobulin heavy constant gamma 1 (IgG1, P01857) and human immunoglobulin kappa constant (IGKC, P01834), respectively. The use of internal ribosome entry site (IRES) allowed for the co-expression of a light chain and its corresponding heavy chain under the control of the same promoter. Both 9B1 and h9B1 proteins were expressed in the CHO-S cells and purified. A major band of approximately 75 kDa was observed, corresponding to the integrity of heavy and light chains. Two other major bands with molecular masses of approximately 50 kDa (heavy chain) and 25 kDa (light chain) were also observed (Figure 5A).

Figure 5.

(A) SDS-PAGE analysis of the purified 9B1 and h9B1 under the reducing condition. (B) Saturation binding of 9B1 with 2 nM [H³]-morphine. (C) Saturation binding of h9B1 with 2 nM [H³]-morphine. (D-F) Competitive binding of 2 nM [H³]-morphine against 6-MAM (D), heroin (E) and hydrocodone (F) with 9B1 (20 nM). (G-I) Competitive binding of 2 nM [H³]-morphine against 6-MAM (G), heroin (H) and hydrocodone (I) with h9B1 (60 nM). (J) Screening of 9B1 and h9B1 against other drugs at 10 μM. (K) The correlation of the predicted binding free energies with the experimentally derived binding free energies (converted from the experimental binding affinities using the well-known thermodynamic equation) with 6-MAM. The concentrations of the antibody, [H³]-morphine, and the competing drug were 60 nM, 2 nM, and 10 μM, respectively. 10 μM morphine and deionized water were set as positive and negative controls and normalized to the 100 % and 0 % inhibition rates, respectively. All experiments were performed in triplicate. The experimental data were analyzed using the GraphPad Prism 7 software to determine the K_d or IC₅₀, then the IC₅₀ values were converted to the K_i values for the purpose of comparison with the K_d values.

The binding affinities of 9B1 and h9B1 with various ligands were assessed by the liquid scintillation counting method. According to the experimental data, 9B1 was able to potently bind with morphine (K_d = 33.7 nM (Figure 5B), which is reasonably close to (but not exactly the same as) the previously estimated binding affinity (~10⁹ M⁻¹).³⁵ Interestingly, our experimental data depicted in Figure 5 reveal for the first time that 9B1 can also potently bind with 6-MAM (K_d = 4.8 nM) (Figure 5D) and heroin (K_d = 1.3 nM) (Figure 5E). Further, as expected, h9B1 was also able to potently bind to morphine (K_d = 127 nM) (Figure 5C), 6-MAM (K_d = 11.9 nM) (Figure 5G), and heroin (K_d = 11.8 nM) (Figure 5H). The binding affinities of h9B1 with these heroin-related opioids are only slightly weaker than the 9B1.

In addition, we also examined whether 9B1 and h9B1 can potently bind to other drugs including naloxone, naltrexone, buprenorphine, hydrocodone, meperidine, oxycodone, methadone, amethaphetamine, ketamine, cocaine, and fentanyl (Figure 5J). According to the initial screening at 10 μM, most of these drugs including naloxone, naltrexone, and buprenorphine did not show significant binding affinities with 9B1 and h9B1. However, both 9B1 and h9B1 can potently bind to hydrocodone (a widely abused prescription opioid), with K_i = 8.2 nM (Figure 5F) and 39.3 nM (Figure 5I), respectively. Besides, 9B1 and h9B1 also showed relatively lower binding affinities with hydrocodone, oxycodone, and fentanyl (see Table 1 for a summary of the experimental binding affinities). The experimentally measured high binding affinities of the humanized antibodies with these heroin-related opioids without significant binding with naloxone and naltrexone are consistent with our computational predictions from the structure-based virtual screening.

Table 1.

Experimental binding affinities of 9B1 and h9B1 with multiple drugs.

Drugs	Kd or Ki (nM)a
	9B1	h9B1
morphine	33.7 (Kd)	127 (Kd)
6-MAM	4.8	11.9
heroin	1.3	11.8
hydrocodone	8.2	39.3
meperidine	8,890	15,700
oxycodone	2,670	6,910
fentanyl	32,900	9,730
naloxone	>10,000	>10,000
naltrexone	>10,000	>10,000
buprenorphine	>10,000	>10,000

^aThe data are K_i values unless indicated explicitly otherwise.

Further, with the newly obtained binding affinity of 6-MAM with 9B1, we were also able to show that the computationally predicted binding free energies with 6-MAM excellently correlate with the corresponding experimental data (Figure 5K), with a correlation coefficient of 0.9468.

Conclusion

The systematic structure-based virtual screening of available monoclonal antibodies and computational design of antibody humanization has led to identification of a promising antibody (9B1) from the know anti-morphine antibodies and a humanized antibody (h9B1) that can potently bind to multiple addictive opioids (including 6-MAM, morphine, heroin, and hydrocodone) without significant binding with currently available opioid overdose/dependence treatment agents naloxone, naltrexone, and buprenorphine. Specific for 9B1, we have determined that K_d = 4.8 nM for 6-MAM, 1.3 nM for heroin, 33.7 nM for morphine, 8.2 nM for hydrocodone, and 2.7 – 32.9 μM for oxycodone, meperidine, and fentanyl, without significant binding affinity to other drugs tested. For h9B1, we have determined that K_d = 11.9 nM for 6-MAM, 11.8 nM for heroin, 127 nM for morphine, 39.3 nM for hydrocodone, and 6.9 – 15.7 μM for oxycodone, fentanyl, and meperidine, without significant binding affinity to other drugs tested. The humanized antibody h9B1 may serve as a promising candidate for treatment of OUDs. The experimental binding affinities reasonably correlate with the computationally predicted binding free energies. The experimental activity data strongly support the computational predictions, suggesting that the systematic structure-based virtual screening and humanization design protocol is reliable. The general, systematic structure-based virtual screening and design approach will be useful for any other antibody selection and design efforts in the future.

Materials and Methods

General strategy of the systematic structure-based virtual screening for mAb selection

Antibodies, including a total of five classes, i.e. immunoglobulin (Ig)A, IgD, IgE, IgG, and IgM, are known as affinity proteins that are a key component of the adaptive immune system.³⁸ The ability of antibodies to bind to foreign invaders with high affinity and specificity is central to their functions. IgGs, including four subclasses, i.e. IgG1 to 4, are the most abundant class of antibodies, constituting approximately 75% of the serum immunoglobulin repertoire. Notably, all IgGs share the same overall architecture, with differences only in six loops known as the complementarity-determining regions (CDRs) in the antigen-binding site. Utilizing these remarkable structural features, our general strategy of the systematic structure-based virtual screening approach to antibody selection and identification starts from collection and structural modeling of all mAbs whose amino-acid sequences (particularly the six CDRs) are available, with the hope that some of the available mAbs can meet the above-mentioned goal – capable of potently binding with all the heroin-related opioids without binding with naloxone and naltrexone. Using the modeled structure of each mAb available, one can dock each of the interested ligands (e.g. 6-MAM, heroin, morphine, naloxone, and naltrexone in the current study) to the antigen-binding site and carry out further simulations and calculations to computationally estimate the binding free energy with each ligand. Based on the computationally estimated binding free energies, one can predict which mAb most likely can meet the goal of the potency and selectivity.

Specific for the need of binding with multiple opioids, the computationally estimated binding free energies clearly indicated which mAb should have the highest overall potency for binding with 6-MAM, morphine, and heroin without significant binding affinity to naloxone and naltrexone. The computationally selected most promising mAb was modeled further for humanization, and the predicted humanized mAb was prepared and tested in vitro for its actual binding affinities with various ligands including 6-MAM, heroin, morphine, naloxone, and naltrexone etc. Described below are the detailed procedures of the used computational methods.

Modeling of the antibody structures

The amino-acid sequences of variable domains of the antibodies were obtained from the IMGT/LIGM-DB database (http://www.imgt.org)⁴⁰ or manually collected from the references cited (see SI – Tables S1 and S2). The homology models of antibodies were built using the PIGSpro software⁴¹ which includes known X-ray crystal structures of a number of antibodies. Using the software, for homology modeling of each antibody structure, the template (used for the framework structure modeling) was selected based on the sequence alignment with those of the known X-ray crystal structure (see Tables S1), the loops were kept in the similar canonical structure of template for the loop modeling method, and all other parameters were set as the default of the software. Then the initial homology models were refined by performing a series of energy minimization processes. Specifically, the Amber14SB force field⁴² was applied for the proteins in vacuo using AmberTools18.⁴³ The nonbonded cutoff for the real-space interactions was set to 10 Å. Two stages of energy minimization were conducted using a hybrid protocol of 8000 steps of steepest descent minimization followed by a conjugate gradient minimization until the convergence criterion (the root-mean-square of the energy gradient is less than 1.0 × 10^–4 kcal/mol·Å) was satisfied or the maximum of 2000 iteration steps was reached. During the energy minimization, a force constant of 10 kcal/mol·Å² was applied on the antibody backbone atoms. Then the final conformations were used for virtual screening described below.

Virtual selection of an antibody capable of selectively binding with heroin-related opioids

The computationally refined antibody structures were superimposed together using the PyMol software⁴⁴ in order to transform the atomic coordinates of all proteins into the same coordinate system with a commonly defined center of box required to define for systematic molecular docking, and the MglTools⁴⁵ software was used to prepare the protein and ligand pdbqt files for the docking. During the docking, the binding site was determined based on the antibody-morphine complex in 9B1 (PDB: 1Q0Y), the geometric center of the co-crystallized antibody-morphine complex was indicated as the active center of the docking box (the size_x, size_y, and size_z were set to 30, 30, and 30, respectively) which was large enough to cover the entire region of the binding site for all antibodies. The docking calculations were conducted by using the AutoDock Vina software,⁴⁵ and all the default parameters were adopted. For each antibody-ligand complex, docking was repeated 3 times and the top-4 ranked binding poses of each time were selected for further computational evaluation in multiple steps. First, the selected complex structures were energy-minimized using the same approach as described above for the energy-minimization of antibody structures without a ligand. The general Amber force field (gaff)⁴⁶ was used for the ligands. Second, the energy-minimized complex structure was relaxed by performing a short (20 ps) molecular dynamics (MD) simulation using the SANDER module of the AmberTools18 software⁴³ in vacuo with a constant temperature (T = 300 K). A restrain (with a force constant of 2 kcal/mol·Å²) was applied on the backbone of antibody. The SHAKE algorithm⁴⁷ was used to restrain the covalent bonds with hydrogen atoms, and the time step for the MD simulation was set to 2 fs. The long-range electrostatic interactions were treated by using the particle mesh Ewald (PME) algorithm,⁴⁸ and the nonbonded cutoff for the real-space interactions was set to 12 Å. Third, the last snapshots of the MD simulations were energy-minimized again using the same method described above for the energy-minimization of the free antibody structures without a ligand. Finally, the MMPBSA module of the AmberTools18 software was used to calculate the binding free energy of each antibody-ligand binding pose, leading to the identification of the antibody-ligand binding pose associated with the lowest binding free energy for each antibody-ligand complex. With the lowest binding free energy pose for each antibody-ligand pair and the corresponding binding free energy determined, we were able to select the antibody with lowest possible binding free energy (i.e. the highest possible binding affinity) with 6-MAM and with the best possible overall binding affinities with other heroin-related opioids as well as the desirable selectivity over naloxone and naltrexone.

Antibody humanization design

Sequence alignment and modification.

Humanization of the computationally selected antibody Ab13 (9B1) was performed by using its X-ray crystal structure complexed with morphine³⁵ (PDB: 1Q0Y) from PDB database (http://www.rcsb.org) and the immunoinformatic modelling tools made available by the IMGT database.⁴⁰ Briefly, the variable heavy and light chain sequences (VH and VL) of the murine antibody were compared with human germline sequences using the IMGT/DomainGapAlign tools.⁴⁰ The top-ranked human germline sequence was used as the template. Murine antibody residues that differ from the human sequences on the surface area were replaced, excluding the residues near the binding pocket and anchor residues. Individual residues that are clearly not involved in the binding with the ligands in the murine antibody were changed to the corresponding residues of the human antibody. The image of the aligned sequences was created using the ESPript 3.0 software.⁴⁹

MD simulation and final sequence selection.

The above-mentioned sequence modification based on the sequence alignment and simple structural modelling led to multiple (three) possible choices of the sequence of the humanized antibody, i.e. three possible humanized antibodies denoted as H1L1, H1L2, and H1L3. The initial structures of the antibodies (generated by using the PIGSpro software⁴¹ as described above) were refined further by performing a series of energy minimization processes and restrained MD simulations in order to know which one of the three choices is most reasonable. Concerning the computational details, the Amber14SB force field⁴² and the generalized Amber force field (gaff)⁴⁶ were used for the proteins and ligands, respectively. The TIP3P water molecules⁵⁰ were added as the solvent and the solute atoms were at least 10 Å away from the boundary of the water box using AmberTools18.⁴³ The counterions (i.e. Na⁺ ions for murine antibodies or Cl⁻ ions for humanized antibodies) were added to neutralize the system. The long-range electrostatic interactions were handled by the particle mesh Ewald (PME) algorithm,⁴⁸ and the nonbonded cutoff for the real-space interactions was set to 10 Å. Energy minimization was performed using a hybrid protocol of 8000 steps of the steepest descent energy-minimization followed by the conjugate gradient energy-minimization until the convergence criterion (the root-mean-square of the energy gradient is less than 1.0 × 10^–4 kcal/mol·Å) was satisfied or the maximum of 2000 iteration steps was reached. During the energy minimization, a force constant of 100 kcal/mol·Å² was applied on the ligand and protein backbone atoms. Then the systems were heated up from 0 to 303.15 K linearly over a time period of 50 ps with the restraint (force constant of 10 kcal/mol·Å²) on all heavy atoms in the NVT ensemble, followed by equilibrating for 325 ps with a Langevin thermostat⁵¹ in the NPT (P = 1 atm and T = 303.15 K) ensemble by gradually decreasing the force constant from 10 to 0.2 kcal/mol·Å². Finally, the 5-ns production run was carried out with the PMEMD module of the Amber12 in the NPT (P = 1 atm and T = 303.15 K) ensemble. The SHAKE algorithm was used to restrain the covalent bonds with hydrogen atoms, and the time step was set to 2 fs, the snapshots were saved every 2 ps. The RMSD values were calculated by CPPTRAJ module of AmberTools18 using the energy-minimized conformations as the references.

Plasmid construction

To prepare antibodies 9B1 and h9B1, the amino acid sequences of heavy and light chains of variable domains of 9B1 and h9B1 were linked with the immunoglobulin heavy constant gamma 1 (IgG1, P01857, heavy chain) and immunoglobulin kappa constant (IGKC, P01834, light chain), respectively. The heavy and light chains were translated to human gene sequences by using the Backtranseq provided by the EMBL-EBI,⁵² and the codon was optimized using the COOL.⁵³ The Kozak sequence and signal peptide sequences for heavy chain or light chain were added to optimized genes. Then the genes for heavy chain and light chain were linked by inserting an Internal Ribosome Entry Site (IRES) between them. The designed genes were synthesized by GeneArt (Invitrogen, Carlsbad, CA), followed by cloning the genes into the pCMV-MCS vector at the BamHI and SalI sites for h9B1, and at the BamHI and XhoI sites for 9B1. The oligonucleotides were synthesized by the Eurofins Genomics (Louisville, KY), restriction enzymes and the KLD Enzyme Mix used for ligation were purchased from New England Biolabs (Ipswich, MA). The final plasmids used for transfection were verified by sequencing services provided by Eurofins Genomics (Louisville, KY).

Protein expression and purification

CHO-S cells were grown under the condition of 37 °C and 8% CO₂ in a humidified atmosphere. The constructed expression vectors for 9B1 and h9B1 were transfected into CHO-S cells using Mirus TransIT-PRO® Transfection Kit. 400 mL cells at the density of 2 × 10⁶ cells/mL were transfected with 400 μg of expression vector and 400 μL of transfection reagent. Culture supernatants were harvested 5 days after transfection by centrifugation with 10000 rpm for 15 min at 4 °C. Antibodies were purified by using a protein A resin (MabSelect SuRe^™ from GE Healthcare, Chicago, IL), as we used previously.^54–56 Briefly, 5 mL resin was packed in a column, equilibrated with 20 mM Tris-Cl (pH = 7.4), loaded the culture supernatants with flow rate of 1–2 mL/min, washed with wash buffer (20 mM Tris-Cl, 300 mM NaCl, pH = 7.4), and eluted with elution buffer (50 mM citric acid, 300 mM NaCl, pH = 4). Then the eluate was concentrated and stored in PB buffer. Purified proteins were analysed by SDS-PAGE (Invitrogen, Carlsbad, CA).

Antibody binding assays

The binding constant of the antibody with [H³]-morphine were tested using liquid scintillation counting. Briefly, 2 nM [H³]-morphine was incubated with different concentration of antibody at room temperature for 60 minutes. The total volume of mixture was 100 μL, and pH was 7.4. Following filtration with EMD Millipore Amicon^™ Ultra-0.5 Centrifugal Filter (30 kD) and EMD Millipore Amicon^™ Ultra 0.5 mL vials, 50 μL of the filtrate was added to 3 mL of 3a70BTM complete counting cocktail (RPI Research Products, Mount Prospect, IL). After vortex, the radioactive value of the cocktail was read, the K_d value was calculated using the GraphPad Prism 7 software.

The binding affinities of each antibody with other drugs were calculated from its binding constant with [H³]-morphine (K_d) and IC₅₀ values against corresponding drugs, using the IC₅₀-to-K_i converter software (http://www.umich.edu/~shaomengwanglab/software/calc_ki/index.html). The IC₅₀ values of corresponding drugs were measured as follows. 100 μL of mixture (pH 7.4) containing 2 nM [H³]-morphine, 20 or 60 nM of the antibody, and a varying concentration of drug was incubated at room temperature for 60 minutes. Then the mixture was filtered with EMD Millipore Amicon^™ Ultra-0.5 Centrifugal Filter (30 kD) and EMD Millipore Amicon^™ Ultra 0.5 mL Vials, 50 μL of the filtrate was transferred to 3 mL of 3a70BTM complete counting cocktail (RPI Research Products, Mount Prospect, IL), then the radioactive value of the cocktail was read, and the IC₅₀ value was calculated using the GraphPad Prism 7 software.

The binding energies of the antibodies with the corresponding drugs were calculated via the well-known thermodynamic equation:

$$\Delta G=\mathit{\text{RT}}\text{ln}\frac{K_d}{\text{C}}$$

in which R is the ideal gas constant (0.0019872), T is the temperature (298.15), and the standard reference concentration C = 1 mol/L. K_i values were assumed equal to K_d.