Antibodies to combat viral infections: development strategies and progress

Monoclonal antibodies (mAbs) are appealing as potential therapeutics and prophylactics for viral infections owing to characteristics such as their high specificity and their ability to enhance immune responses. Furthermore, antibody engineering can be used to strengthen effector function and prolong mAb half-life, and advances in structural biology have enabled the selection and optimization of potent neutralizing mAbs through identification of vulnerable regions in viral proteins, which can also be relevant for vaccine design. The COVID-19 pandemic has stimulated extensive efforts to develop neutralizing mAbs against severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), with several mAbs now having received authorization for emergency use, providing not just an important component of strategies to combat COVID-19 but also a boost to efforts to harness mAbs in therapeutic and preventive settings for other infectious diseases. Here, we describe advances in antibody discovery and engineering that have led to the development of mAbs for use against infections caused by viruses including SARS-CoV-2, respiratory syncytial virus (RSV), Ebola virus (EBOV), human cytomegalovirus (HCMV) and influenza. We also discuss the rationale for moving from empirical to structure-guided strategies in vaccine development, based on identifying optimal candidate antigens and vulnerable regions within them that can be targeted by antibodies to result in a strong protective immune response.

Since the development of the hybridoma technology enabling the generation of monoclonal antibodies (mAbs) by Köhler and Milstein in the 1970s 1 , mAbs have become a key class of drugs for cancer and immune disorders, with the 100th antibody-based therapeutic recently gaining US Food and Drug Administration (FDA) approval 2 . However, although harnessing antibodies to combat infectious diseases has a history stretching back more than a century to the applications of serum conferring protection against diphtheria toxin 3 , only a small number of mAb drugs are used to treat or prevent infectious diseases. At the time of writing in 2022, six mAbs targeting pathogens have so far been granted full approval by the FDA (Table 1), for indications including prevention of respiratory syncytial virus (RSV) infection, prevention and treatment of anthrax infection, prevention of recurrence of Clostridioides difficile infection and the treatment of Ebola virus (EBOV) infection.
The rapid spread of COVID-19 in 2020 led to intense efforts to develop neutralizing mAbs that target severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) for the treatment and prevention of COVID-19. As a result, more than 20 mAbs entered clinical develop ment. So far, several of these mAbs have received emergency use authorization (EUA) from the FDA (Table 2) and other regulatory agencies worldwide, and more authorizations are anticipated. Although vaccines have been the mainstay of efforts to tackle COVID-19, mAbs can provide an important contribution for vulnerable populations before or after exposure to SARS-CoV-2, such as people who are immunocompromised or people with mild to moderate COVID-19 who are at high risk of developing severe disease. The mAbs under development providing broadly neutralizing activity against coronaviruses could also be instrumental for preparedness against future pandemics. Furthermore, the momentum built and the knowledge gained from the development of mAbs for SARS-CoV-2 may help accelerate the development of mAbs to combat other infectious diseases.
In this Review, we first provide an overview of the technologies for the discovery and engineering of mAbs to target pathogens, with a focus on viruses. We then describe the progress in the development of mAbs against a range of viruses, including SARS-CoV-2, RSV, Ebola, cytomegalovirus (CMV) and influenza. Efforts to harness mAbs to combat bacterial infections have been reviewed elsewhere 4,5 and are summarized briefly in box 1. Finally, we also discuss how rapid mAb discovery combined with structural vaccinology can support the development of vaccines and therapeutic mAbs.
Emergency use authorization (eUa). a mechanism to facilitate the availability and use of medical countermeasures during a public health emergency. US Food and Drug administration (FDa) issuance of a eUa permits the use of unapproved medical products or unapproved uses of approved medical products when no adequate alternatives are available.

and Laurent Perez 1 ✉
Abstract | Monoclonal antibodies (mAbs) are appealing as potential therapeutics and prophy lactics for viral infections owing to characteristics such as their high specificity and their ability to enhance immune responses. Furthermore, antibody engineering can be used to strengthen effector function and prolong mAb half life, and advances in structural biology have enabled the selection and optimization of potent neutralizing mAbs through identification of vulnerable regions in viral proteins, which can also be relevant for vaccine design. The COVID19 pandemic has stimulated extensive efforts to develop neutralizing mAbs against severe acute respiratory syndrome coronavirus 2 (SARS CoV2), with several mAbs now having received authorization for emergency use, providing not just an important component of strategies to combat COVID19 but also a boost to efforts to harness mAbs in therapeutic and preventive settings for other infec tious diseases. Here, we describe advances in antibody discovery and engineering that have led to the development of mAbs for use against infections caused by viruses including SARS CoV2, respiratory syncytial virus (RSV), Ebola virus (EBOV), human cytomegalovirus (HCMV) and influ enza. We also discuss the rationale for moving from empirical to structure guided strategies in vaccine development, based on identifying optimal candidate antigens and vulnerable regions within them that can be targeted by antibodies to result in a strong protective immune response.

0123456789();:
Antibody characteristics and engineering Antibody structure and function. Antibodies are natural biomolecules generated by plasma cells or stimulated memory B cells after a pathogen infection or vaccination 3 . Structurally, they are Y-shaped heterodimers composed of two light chains of 25 kDa each and two heavy chains of at least 50 kDa, depending on the immunoglobulin isotype. The heavy and light chains are linked by multiple disulfide bridges and noncovalent interactions (Fig. 1a), with variations in the number of interactions and bridges depending on the immunoglobulin isotypes.
Antibodies can also be divided into functional components (Fig. 1b). The two fragment antigen-binding domains (Fabs) bind to and neutralize pathogens. These are linked to the crystallizable fragment (Fc) domain by a hinge region that gives the Fabs a large degree of conformation flexibility relative to the Fc domain, allowing them to strongly interact with any antigen regardless of its orientation. The glycosylated Fc domain binds to other proteins, including Fcγ receptors (FcγRs) on various immune cells and complement protein C1q, to mediate effector functions such as antibody-dependent cellular cytotoxicity (ADCC), complement-dependent cytotoxicity (CDC) and antibody-dependent cellular phagocytosis (Fig. 1c). The Fc domain also influences antibody pharmacokinetics via interaction with the neonatal Fc receptor (FcRn).
Antibodies vary in isotype depending on whether the alpha, mu, gamma, epsilon or delta gene segments recombine with the variable region. In humans, the following genes generate different subclasses of antibodies: two alpha gene segments (IgA1 and IgA2), four gamma gene segments (IgG1, IgG2, IgG3 and IgG4), one mu segment (IgM), one epsilon segment (IgE) and one delta segment (IgD). Each subclass specializes in the elimination of different types of pathogens, except for IgD, for which the function is still poorly characterized 6 . The IgG class is the principal isotype in the blood and extracellular fluid. An important aspect of the different isotypes is that their sequence variation determines their affinities and specificities for FcRn, FcγRs and complement protein C1q. Notably, the IgG1 isotype allows ADCC and CDC (Fig. 1c), IgG2 and IgG4 are poor CDC activators and IgG3 is a potent CDC activator. Most therapeutic mAbs in clinical use or development against infectious diseases are of the human IgG1 isotype, which has affinity for activating FcγRs but also exhibits binding to the inhibitory FcγRIIb, thereby limiting protective Fc effector activities 7 .
Antibodies as antivirals. For more than a century, passive immunization with monoclonal or polyclonal antibodies has been used in the treatment and prevention of infectious diseases, particularly in individuals with immunodeficiencies or individuals for whom vaccination is contraindicated.
Antibodies can combat viral infections through several mechanisms. First, antibodies can prevent viral glycoproteins of enveloped viruses or the protein shell of non-enveloped viruses from binding to the target host cells 8,9 . These viral proteins have two major functions in the viral life cycle: binding to cellular receptors and mediating the fusion of viral and cellular membranes (in the case of enveloped viruses) or penetration into the cytosol (in the case of non-enveloped viruses). For example, the entry of SARS-CoV-2 into host cells is mediated by the interaction between the viral spike (S) glycoprotein and the angiotensin-converting enzyme 2 (ACE2) receptor on the host cell surface. ACE2 is expressed on cells of the respiratory system, gastrointestinal tract and endothelium 10 (Fig. 1b). The spike-ACE2 interaction can be blocked by antibodies targeting the spike receptor-binding domain (RBD) 11 , Antibody-dependent cellular cytotoxicity (aDCC). a mechanism of cell-mediated immune defence whereby an effector cell of the immune system actively lyses a target cell, whose membranesurface antigens have been bound by specific antibodies.

Complement-dependent cytotoxicity
(CDC). The complement system is a network of proteins that form an important part of the immune response by enhancing the opsonization of pathogens, cell lysis and inflammation. CDC is a mechanism of complement-mediated immune defence in which an antibody bound to its antigen activates the complement cascade.

Glycoproteins
Proteins with oligosaccharide chains (glycans) covalently attached to amino acid side chains. Virus surface glycoproteins embedded in the membrane often have a role in interactions with host cells, including receptor binding, and are commonly targeted by host antibodies.
which inhibits viral infection, as discussed later in the Review (Fig. 1b). Antibody effector functions mediated by engagement with complement protein C1q or FcγRs on leukocytes can also be involved in combating viral infections (Fig. 1c). Complement activation by antibodies leads to direct lysis of the virus and/or the infected host cell, and antibodies can also promote or induce phagocytosis, or trigger the release of toxic chemicals, such as cytokines or reactive oxygen species 12 . For instance, it was recently shown that Fc effector functions are required for optimal protection by mAb therapy for SARS-CoV-2; when given after infection, intact mAbs reduced the SARS-CoV-2 burden and lung disease in animals better than loss-of-function Fc variant mAbs 13-15 by mitigating inflammation and improving respiratory mechanics.
Conversely, in rare cases, suboptimal binding of antibodies to virions can facilitate viral pathogenesis through a process known as antibody-dependent enhancement (ADE) 16 , in which recognition of virion-antibody complexes by FcγRs enhances viral entry into host immune cells. ADE was first observed with Dengue virus in the presence of sub-neutralizing antibody concentrations. For example, when the level of maternal antibodies against Dengue virus in newborns wanes, some individuals will experience an interval during which their antibody level will drop below its protective capacity, leading to severe disease following infection 17 . Dengue virus co-circulates as four serotypes, and the increased severity of some secondary infections is thought to be due to enhancement of viral entry by pre-existing antibodies generated following a primary infection with a different serotype that are not able to neutralize the second serotype 18,19 . Enhanced disease was also observed after vaccination with a formalin-inactivated RSV vaccine in infants 20 . The risk of ADE can be reduced by engineering the antibody Fc domain to reduce binding to FcγRs, as noted below.
The first passive immunization approaches used serum derived from animals actively immunized with an antigen such as diphtheria toxin, but such approaches come with the risk of provoking an immune response against non-human antibodies. These risks can be mitigated by using blood from people who have recovered from an illness as the source of antibodies, known as convalescent plasma therapy (CPT) 21 . Although CPT has been historically successful in combating infections 22 , it has proved to be an inconsistent tool, as antibody responses between individuals are highly variable. CPT also has safety risks, including allergic reactions and low risk of infections by other viruses such as HIV, hepatitis B and hepatitis C. Furthermore, widespread use of CPT in the context of a pandemic such as COVID-19 would depend on the availability of a sufficient number of plasma donors and facilities for appropriate processing. Overall, the role of CPT in such cases may be restricted  323 328 All monoclonal antibodies (mAbs) listed are known to target the receptor binding domain (RBD) of the spike (S) glycoprotein of severe acute respiratory syndrome coronavirus 2 (SARS CoV2). Fc, crystallizable fragment; NA, not available; -, no neutralizing activity; +/-, partial neutralizing activity; +++, potent neutralizing activity. a Information on status is with regard to the initial US Food and Drug Administration (FDA) emergency use authorizations (EUAs), unless the product was initially further advanced in other regions or countries. b EUA since withdrawn owing to likely ineffectiveness against the Omicron variant of concern (VOC).
www.nature.com/nrd to the early epidemic phase when therapeutic options are limited. Indeed, although plasma obtained from convalescent donors has previously been used as a therapy for coronavirus infections 23 and was investigated as a potential therapeutic option for treatment of patients severely ill with COVID-19, the results of randomized controlled trials indicated that CPT has no benefit for patients with moderate, severe or critical COVID-19 infection 24,25 . Nevertheless, no adverse effects were reported and a potential beneficial effect in younger patients was observed 26 . Given the limitations of CPT, there has been an increasing focus on the use of neutralizing mAbs for passive immunization for infectious diseases. Neutralizing mAbs with high specificity and potency can be developed and extensively characterized, and lack the risk of blood-borne disease associated with CPT. Furthermore, they can be produced at a large scale in a reasonable time frame with well-established processes.

Strategies to generate human therapeutic antibodies for viral infections.
Most strategies to identify human mAbs to combat pathogens can be classified as either targeted, in which mAbs that bind to a known antigen are directly isolated, or target agnostic, in which functional assays are performed on secreted immunoglobulins obtained from the supernatant of single cell cultures.
The first efficient targeted approach for mAb identification involved panning phage display libraries constructed from the immunoglobulin variable genes of immunized or infected individuals based on binding to a target antigen 27 . Alternatively, random synthetic libraries were also used 28 . Although these methods have led to the isolation of neutralizing antibodies against multiple pathogens (for example, HIV 29 , SARS-CoV-2 (ReF. 30 ) and the anthrax toxin 31 ), the obtained mAbs did not represent the natural antibody repertoire as the antibody fragments were generated from random pairings of immunoglobulin variable heavy (VH) and variable light (VL) regions (Fig. 2a). Indeed, several libraries were based on a single or limited set of V region frameworks, leading to diversification of the CDRH3 only 32 . Moreover, VH/VL pairing is known to be an important diversity factor, and artificial pairings can generate autoreactive molecules 33 , as no negative selection is present 34,35 .
Nevertheless, a naïve human single-chain variable fragment (scFv) phage display library was used to develop a potent antitoxin mAb for anthrax, raxibacumab 36 , which was approved in 2012 (box 1 and Table 1).
A second targeted approach developed subsequently is the direct isolation of antigen-specific memory B cells based on their capacity to bind fluorescent bait antigens, followed by identification of the mAbs they produce 37 . The memory B cells can originate from the plasma of convalescent patients 38 , or from transgenic mice carrying human immunoglobulin loci that produce fully human antibodies in response to immunization with a target antigen 39 . This approach has been particularly successful in the isolation of broadly neutralizing antibodies (bNAbs) targeting the CD4-binding site in the V1/V2 and V3 regions of gp120 and the membrane-proximal external region of gp41 of HIV [40][41][42][43][44][45] . Furthermore, mAbs against the hepatitis B virus viral S antigen (HBsAg) 46 and the SARS-CoV-2 spike protein 47 (Fig. 2b) have also been obtained, as discussed below.
The major limitation of targeted approaches is that target antigens must be known in advance because the selection process is based on binding affinity to the purified antigen rather than neutralization potency. Targetagnostic approaches present a viable alternative when limited information is available on the pathogen to be neutralized 48 . Various methods to obtain single cell cultures of memory B cells or plasma cells have been described 49,50 . Memory B cell immortalization using Epstein-Barr virus 51,52 remains an attractive method because of its limited cost 53 (Fig. 2c). A limitation of the Epstein-Barr virus approach is the suboptimal immortalization of B cells, which plateaus at approximately 35% 51 . However, the development of single cell cultures without the need for B cell immortalization can overcome this limitation [54][55][56] , and has been used for the identification of antibodies against pathogens such as group 1 and group 2 influenza A viruses 57 and BK/JC polyomaviruses 58 . Recently, a functional organotypic system for antibody generation has been reported. The organoid recapitulates germinal centre features in vitro, such as the production of antigen-specific antibodies with affinity maturation and class-switch recombination from human tonsils, and this is a promising step forward for the field of mAb development 59 (Fig. 2d).
In all cases, once identified, the mAb candidates must be sequenced for further recombinant expression. Cloning and expression of individual antibodies was traditionally labour-intensive, and throughput was largely limited to a few hundred clones. However, recent technological advances using nanofluidic devices have considerably increased the throughput of this approach 60 . In addition, advances in next-generation sequencing have enabled high-throughput screening and sequencing of paired antibody repertoires 61 . Currently, isolated B memory cells or plasma cells are injected into microfluidic devices (such as the 10x Genomics platform), generating droplets containing a single cell and lysis buffer with microbeads covered by barcoded primers to generate cDNA encoding VH and VL sequences 60,62-65 . These approaches allow the discovery of antibodies that are potentially useful as therapeutics 66 , in addition to the

Box 1 | Monoclonal antibodies as antibacterial agents
Antimicrobial resistance is one of the top ten global public health threats facing humanity according to the World Health organization (WHo). Although traditional small-molecule drug discovery still dominates the landscape for antibacterial solutions to antimicrobial resistance, the prophylactic or therapeutic use of monoclonal antibodies (mAbs) for bacterial infection could have an important role 330 . one of the main advantages of these biologics is their low toxicity, as bacterial virulence proteins are targeted instead of proteins required for survival, therefore avoiding the disruption of the microbiome and, potentially, limiting the development of resistance.
So far, three antibacterial mAbs have been approved for the treatment and prophylaxis of bacterial infections. raxibacumab (Abthrax/Anthrin) 36 and obiltoxaximab (Anthim) 331 are both mAbs targeting the protective antigen (pA) component of the lethal toxin of Bacillus anthracis and are approved to treat inhalation anthrax due to B. anthracis. bezlotoxumab is a mAb that binds to Clostridioides difficile enterotoxin b 332 . this product does not protect against or treat initial C. difficile infection but, rather, is used to reduce the recurrence of infection, which is often seen with C. difficile. possibility of studying the human antibody repertoire at an unprecedented resolution 67 , and could be deployed as emergency response platforms for investigating mAbs from the blood of people who have recovered from emerging viral infections 68 .
Recently, the LIBRA-seq (linking B cell receptor to antigen specificity through sequencing) methodology was developed for high-throughput mapping of paired heavy-chain and light-chain B cell receptor sequences to their cognate antigen specificities 69 . In this approach, Opsonized a state of a pathogen in which antibodies or complement factors are bound to its surface.

Complement-dependent cytotoxicity
FcγRIII binds to antibody-antigen complex, initiating signalling pathways that lead to granzyme and perforin release, killing infected cells

Antibody-dependent cellular cytotoxicity
Macrophage FcγRs recognize antibody-antigen complex, initiating signalling pathways that lead to infected-cell phagocytosis  ) to the angiotensin converting enzyme 2 (ACE2) receptor on host cells is mediated by the viral spike (S) protein, which comprises an S1 subunit (including a receptor binding domain (RBD) and an amino terminal domain (NTD)) and an S2 subunit. Priming of coronavirus spike proteins by host cell proteases such as the transmembrane serine protease TMPRSS2 through cleavage at S1/S2 and S2′ sites (see Fig. 3) is essential for viral entry. Therapeutic antibodies and antibodies elicited by vaccination that bind to the RBD or NTD can block viral binding to ACE2, or block fusion between viral and cellular membranes (see Fig. 3). c | Effector functions of antibodies. mAbs can facilitate target cell death via complement fixation and membrane attack complex (MAC) activation, which is known as complement dependent cytotoxicity (CDC). Antibody dependent cellular cytotoxicity (ADCC) is a mechanism of cell mediated immune defence whereby an effector cell (natural killer cell, macrophage, neutrophil or eosinophil) of the immune system actively lyses a target cell, whose membrane has been bound by specific antibodies. Natural killer cells release cytotoxic factors (perforin and proteases known as granzymes) that cause death of the infected cell. Antibody dependent cellular phagocytosis is the mechanism by which antibody opsonized target cells activate Fcγ receptors (FcγRs) on the surface of macrophages to induce phagocytosis, resulting in internalization and degradation of the target cell through phagosome acidification.  Fig. 2 | Antibody discovery approaches. a | Phage bio panning is based on a library of phages that contain genes coding for variable heavy (VH)/variable light (VL) domains, leading to production of encoded antibodies on phage surfaces. Selection of antibodies produced by phages involves immobilization of the ligand of interest on a solid support (spike (S) protein of severe acute respiratory syndrome coronavirus 2 (SARS CoV2) shown), followed by applying the phage display library to immobi lized ligand to allow binding of specific variants. To eliminate adherent non binders, multiple rounds of washing are usually performed, and remaining bound phages are eluted and re amplified. b | Recombinant antigens conjugated to a fluorescent marker are incubated with class switched memory B cells and sorted according to their capacity to bind the antigen of interest (such as the S protein of SARS CoV2) by fluorescence activated flow cytometry, followed by identification of the antibodies they produce. c | Target agnostic approaches using single B cell culture. Single B cells are seeded on a feeder layer in the pres ence of a cytokine mix and a Toll like receptor (TLR) activator. Culture supernatant is screened for neutralization activity and clones of interest are retrieved and sequenced. d | Workflow for organoid reconstitution from human tonsils to develop an in vitro system that recapitulates key germinal centre features, including production of antigen-specific antibodies, somatic hypermutation and affinity maturation 59 . e | Workflow for single cell immune profiling 69 . B cells are encapsulated with barcoded gel beads in a single partition and undergo reverse transcription followed by PCR. Each cDNA is barcoded from its individual cell of origin and processed for next generation sequencing. Fab, fragment antigen binding domain; mAb, monoclonal antibody; scFv, single chain variable fragment. Panel a is adapted from ReF. 329 , under a Creative Commons license CC BY 3.0.

Antibody-dependent phagocytosis
B cells are mixed with a panel of DNA-barcoded antigens and both the antigen barcode(s) and B cell receptor sequences are recovered via single cell next-generation sequencing. This enabled the antigen specificity of thousands of B cells from two subjects infected with HIV to be mapped, and the predicted specificities were confirmed for numerous HIV, influenza and SARS-CoV-2-specific antibodies, including known and unknown bNAbs (Fig. 2e). Indeed, large antibody data sets can be analysed computationally to infer antibody sequences and binding modality, and this has emerged as a powerful method for the identification of structurally related antibodies from sequence databases [70][71][72] . However, a limitation of the LIBRA-seq methodology, which is inherent to targeted approaches, is the need for a recombinant bait antigen that requires extensive validation 73 .
Antibody engineering. Several regions of mAbs can be engineered to improve their therapeutic characteristics.
In addition to the variability induced by isotype usage, several mutations in the Fc domain have been identified to increase or decrease the effectiveness of ADCC and/or CDC. For example, etesevimab, an anti-SARS-CoV-2 mAb developed by Eli Lilly that has received an EUA for COVID-19, has been engineered to lack FcγRI and FcγRII-binding activity with L234A and L235A mutations (the LALA modification) 74 in the Fc domain to reduce safety concerns over the potential to exacerbate disease through ADE mechanisms 75 . Another example is provided by AZD7442, a cocktail of the anti-SARS-CoV-2 mAbs tixagevimab (AZD8895) and cilgavimab (AZD1061) 76 developed by AstraZeneca, which has received an EUA for COVID-19. Both mAbs in the combination have engineered Fc domains including L234F/L235/P331S substitutions 77 (the TM modification), resulting in little or no binding to various FcγRs or complement protein C1q, and little or no effector function in vitro 76 . Other mAbs against COVID-19 and their modifications are described in Table 2.
Engineering efforts have also focused on improving the mAb half-life in vivo by reducing IgG catabolism. This is regulated by mAb interaction with FcRn, which functions as a recycling or transcytosis receptor and is responsible for maintaining IgG and albumin in circulation and bidirectional transport across polarized cellular barriers. FcRn binds to IgG at the CH2-CH3 junction in a pH-dependent manner. IgG tightly binds at acidic pH (pH 6.0) but not at physiological pH (pH 7.4). Moreover, hydrophobic interactions between FcRn and Fc are stabilized by salt bridges formed between anionic residues on FcRn and protonated histidine or glutamic acid residues of the IgG Fc region in positions 117, 132 and 137 or 310, 435 and 436, respectively. Therefore, mutagenesis of Fc region residues at the FcRn-Fc interface is used to increase the half-life of IgG in the circulation.
Multiple antiviral mAbs with engineered Fc regions to extend their half-life have entered clinical development. Among the furthest advanced is sotrovimab (also known as VIR-7831 and GSK4182136), a mAb developed by Vir Biotechnology and GlaxoSmithKline against SARS-CoV-2 that received a EUA from the FDA in 2021. Sotrovimab was developed with an Fc domain that includes M428L and N434S amino acid substitutions (the LS modification) to extend antibody half-life 78 . Vir Biotechnology also incorporated the LS modification into VIR-3434, as well as G236A/A330L/I332E amino acid substitutions (the GAALIE modification) 79 for enhanced FcγRIIIa binding. This mAb was designed to prevent chronic infections of hepatocytes by all ten hepatitis B virus genotypes and is in a phase II trial (NCT04856085).
Further antiviral mAbs that use the LS modification include VRC01LS, VRC07-523LS and elipovimab (GS-9722) for HIV 80 . VRC01LS, a broadly neutralizing mAb that was developed by the National Institutes of Health (NIH), shows an approximately fourfold longer serum half-life than the parent antibody with a wild-type Fc domain (VRC01) and showed similar neutralizing activity in serum to VRC01 during 48 weeks of a phase I trial 81 . It has also been studied in a phase I trial (NCT02256631) in combination with VRC01 and another variant of VRC01, VRC07-532LS (ReF. 82 ), that also has the LS modification. Elipovimab, developed by Gilead, is derived from the HIV-neutralizing antibody PGT121 (ReF. 83 ) and has an engineered Fab region to lower the immunogenicity and improve the stability at low pH, as well as the LS modification in the Fc domain to extend its half-life.
The triple amino acid mutation at M252Y/S254T/ T256E (the YTE modification) was shown to promote a fourfold increase in serum half-life of mAbs due to increased binding to FcRn 84 , and was used in tixagevimab and cilgavimab on top of the TM modification (see above) to extend their half-life 76 . Another example using the YTE modification is nirsevimab, a mAb targeting the RSV fusion (F) glycoprotein developed by AstraZeneca and Sanofi Pasteur 85 (see below). In a phase III clinical trial in healthy preterm infants, nirsevimab showed an extended half-life, offering protection from RSV for a typical 5-month season with a single intramuscular dose (50 mg) 86,87 .
Antibody engineering can also be used to generate novel antibody formats, such as bispecific antibodies (bsAbs) designed to recognize two different epitopes or antigens. A single bsAb can therefore bind to two different proteins or two different sites on the same protein.
A wide range of bsAb formats have been developed, particularly for oncology applications 88 , but there have also been a few bsAbs investigated for infectious diseases. For example, a bsAb targeting both the receptor-binding site (RBS) of the Niemann-Pick C1 (NPC1) protein and a conserved surface-exposed epitope on the EBOV glycoprotein was shown to neutralize all known EBOVs by co-opting viral particles for endosomal delivery and conferred post-exposure protection against multiple EBOVs in mice 89  Extending the multi-specificity concept further, trispecific antibodies engineered to interact with three independent HIV envelope determinants (the CD4-binding site, the membrane-proximal external region and the V1/V2 glycan) conferred complete immunity against a mixture of simian-human immunodeficiency viruses in non-human primates (NHPs), and showed higher potency and breadth than any previously described single broadly neutralizing mAb 91 . One such agent, SAR441236, has entered phase I development (NCT03705169).
Other novel engineered antibody formats that could be applied in antiviral agents include small camelid VHHs (15 kDa), known as nanobodies, that retain full antigen specificity, in contrast to mouse and human antibody-binding domains (50 kDa). Furthermore, nanobodies possess extended complementarity-determining regions, enabling binding of epitopes that are not normally accessible to conventional antibodies 92 , such as conserved viral domains that are often masked by glycan shields. Clinical trials demonstrated that they are safe and possess low immunogenicity 93 . Interestingly, transgenic mice encoding 18 alpaca, 7 dromedary and 5 Bactrian camel VHH genes were shown to generate potent neutralizing nanobodies against SARS-CoV-2 (ReF. 94 ).
Finally, antibody mimetics such as designed ankyrin repeat proteins (DARPins) 95 can provide high affinity and offer multi-specificity. For instance, a multi-DARPin (ensovibep) that binds simultaneously to all three units of the SARS-CoV-2 spike RBD 96 developed by Molecular Partners and Novartis 97,98 is in clinical trials (NCT04828161).

Antibodies to combat viral infections Coronaviruses, including SARS-CoV-2.
Coronaviruses are enveloped positive-sense single-stranded RNA viruses belonging to the Coronaviridae family 99 . They can infect a wide variety of mammalian and avian species, causing respiratory and/or intestinal tract diseases. Human coronaviruses are major causes of the common cold and are responsible for up to 30% of mild respiratory tract infections and atypical pneumonia in humans 100 . Four different coronaviruses usually circulate in the human population: HCoV-OC43, HCoV-HKU1, HCoV-NL63 and HCoV-229E 101 .
In the past two decades, three coronaviruses with the potential to cause life-threatening disease in humans have emerged. Severe acute respiratory syndrome coronavirus (SARS-CoV) emerged in 2002 in China and spread, resulting in 8,100 infections and nearly 800 deaths in 37 countries 102,103 . Ten years later, the Middle East respiratory syndrome coronavirus (MERS-CoV) emerged in the Arabian Peninsula and spread to 21 countries, causing outbreaks in humans and infecting around 2,500 individuals, with a fatality rate of 35% 104,105 . In 2019, infections by a coronavirus now known as SARS-CoV-2 that can cause fever, severe respiratory illness, pneumonia, diarrhoea, dyspnoea and multiple organ failure were identified in China 106,107 . As of February 2022, more than 414 million cases have been confirmed, leading to at least 5.8 million deaths, according to the World Health Organization (WHO) 108 .
All coronaviruses enter the host cells using a trimeric spike (S) transmembrane glycoprotein 109 . The S protein is a type I membrane class I fusion protein 110 and is organized into two functional subunits, which remain non-covalently bound in the pre-fusion conformation of the protein (Fig. 3a). The amino-terminal S1 subunit is responsible for binding to the host cell receptors, whereas the carboxy-terminal S2 subunit is responsible for fusion of the viral and cellular membranes 111 . The S1 subunit is further divided into an N-terminal domain (NTD) and a RBD (Fig. 3a).
SARS-CoV-2 is phylogenetically closely related to SARS-CoV, sharing approximately 79.6% genomic sequence identity 112 , and similar to SARS-CoV uses the S1 RBD to bind to ACE2 receptors on host cell types such as pneumocytes and enterocytes 113 . After host cell binding, a conformational change in the S2 subunit results in virus fusion and entry into the target cell 114 (Fig. 3b).
The S glycoprotein has been the primary focus of efforts to develop mAbs to target SARS-CoV-2, as it was already known to be a target of potent neutralizing mAbs against SARS-CoV 115 and MERS-CoV 116 . More than 20 mAbs that target the S glycoprotein, originating either from the B cells of convalescent patients with COVID-19 or immunization of humanized mice, have been tested in clinical trials, and some have received an EUA from the FDA for the treatment of patients with mild to moderate COVID-19 or for pre-exposure prophylaxis ( Table 2). These include sotrovimab 117 , the combination of casirivimab and imdevimab [118][119][120][121] , bamlanivimab 122-124 (used as a monotherapy or in combination with etesevimab 125 ), the combination of cilgavimab 126 and tixagevimab 76,127 , regdanvimab [128][129][130] and bebtelovimab 131 .
However, a major limitation of most mAbs evaluated so far against COVID-19 has been the rapid appearance of SARS-CoV-2 variants of concern (VOCs) that can escape both single mAbs and cocktails of mAbs 132 (Table 2). VOCs such as Alpha, Beta, Gamma and Delta have 9-12 mutations in regions of the S glycoprotein, which typically have only a partial impact on the effectiveness of therapeutic mAbs. However, Omicron variants have accumulated more than 35 mutations in the S glycoprotein, of which 15 occur in the RBD, which is not only the site that binds to the host receptor ACE2 but also the key target of therapeutic mAbs, as well as neutralizing antibodies produced by the natural and vaccine-induced immune response. The emergence of Omicron VOCs (BA.1, BA.1.1 and BA.2) has rendered numerous mAbs with EUAs and/or in advanced clinical development partially or almost ineffective. These include the combination of casirivimab and imdevimab, the combination of bamlanivimab and etesevimab, and sotrovimab (Table 2). These mAbs have all been developed from patients infected with SARS-CoV-2, with the exception of sotrovimab, which was isolated from an individual infected with SARS-CoV. Nevertheless, two recently developed mAbs, bebtelovimab and P2G3, retain full activity against the Omicron VOCs 131,133 .
Additional, ultrapotent neutralizing antibodies binding the RBD have been identified, and some are currently in clinical development (Table 2). These mAbs are categorized into five groups based on their clustering and binding to the RBD (Fig. 3c)  Y449, G485 and F486; this group includes P5C3 (ReF. 138 ), S2M11 (ReF. 135 ), COVA2-39 (ReF. 139 ) and mAb 2-4 (ReF. 140 ). Group 3 mAbs bind on the right side of the ridge opposite from group 1 and target Y449, E484 and F490; this group includes BD-368-2 (ReF. 134 ), CVO7-270 (ReF. 141 ) and P2B-2F6 (ReF. 142 ). The group 4 mAbs bind the lower half of the left side of the RBD, targeting Y369, C379, P384 and T385, and include mAbs binding the CR3022 cryptic site 143 ; mAbs of this group are thought to act by destabilizing the pre-fusion conformation of the trimeric S protein 144 . Finally, group 5 mAbs bind the rear right side of the protein and include S309 (ReF. 145 ) and REGN10987 (ReF. 146 ) (Fig. 3d).
Although the RBD is immunodominant, additional regions of the S protein are immunogenic, most notably the NTD 140,147,148 (Fig. 3a). Structural characterization of NTD-specific antibodies 4A8 (ReF. 149 ) and 4-8 (ReF. 140 ) showed that mAbs targeted the upper side of the protruding area of the NTD. Epitope mapping of 41 NTD-specific mAbs led to the identification of six antigenic sites, one of which is recognized by all known NTD-specific neutralizing antibodies and has been termed the 'NTD supersite' , consisting of residues 14-20, 140-158 and 245-264 (ReF. 147 ). The mechanism of neutralization by which NTD-specific antibodies act remains to be fully determined, although it may involve the inhibition of conformational changes.
Cross-reactive conformational S2 epitopes have also been described 150 . Moreover, five mAbs cross-reacting with the stem helix of multiple betacoronavirus S proteins were recently identified in individuals convalescing after COVID-19 infection 151 . The biological significance of these different mAbs is still under investigation.
It is crucial to establish the target populations for treatment with mAbs and to define what should be the optimal timing for their use. It appears that mAbs may play a prophylactic role in individuals deemed to be at high risk of severe COVID-19, such as older people and/or individuals with polymorbidities, and immunocompromised individuals with poor or no response to vaccination. Several reports suggested that mAbs prevent COVID-19 in high-risk individuals potentially exposed to SARS-CoV-2 in nursing homes or within households 152 . Another question to address is how to increase the duration of action of mAbs, as they usually only allow a temporary window of protection.
Finally, although there were initial concerns about the risk of anti-SARS-CoV-2 mAbs causing ADE, there is currently no evidence to show ADE occurs with any of the mAbs tested in clinical trials. For a review of this issue, please see ReFS 75,153 .
Human respiratory syncytial virus. RSV is an enveloped negative-stranded RNA virus belonging to the Pneumoviridae family 154 . RSV infections are extremely common and typically result in mild respiratory symptoms. However, infection in infants and older adults accounts for a substantial hospitalization burden in both age groups.
High levels of RSV-neutralizing mAb titres correlate with protection in children and adults, including older people 155,156 . The first RSV intravenous immuno globulin infusion preparation, named RespiGam, was used prophy lactically from the late 1990s to the early 2000s to prevent severe RSV-associated lower respiratory tract disease in young children with bronchopulmonary dysplasia or premature birth 157,158 . The use of RespiGam was discontinued in 2003 and replaced by prophylaxis with palivizumab, the first neutralizing mAb developed to treat severe RSV infection in high-risk infants 159 , which was approved in 1998.
Of the three RSV surface proteins (F, G and SH), F-specific antibodies account for the majority of neutralizing activity in the sera of infected humans 160,161 , and so the F glycoprotein has been the focus of mAb development for RSV. F is a trimeric type I fusion glycoprotein responsible for merging the viral membrane with cellular membranes, and similar to many other viral fusion glycoproteins it undergoes major structural rearrangements during the transition from the pre-fusion to the post-fusion state 110 . Palivizumab is a humanized mAb that binds to antigenic site II of the F glycoprotein. Importantly, site II as well as sites I, III and IV are present in both pre-fusion and post-fusion conformations of the F glycoprotein (Fig. 4a). Therefore, palivizumab does not prevent triggering of conformational changes in F, and presumably blocks entry and membrane fusion by preventing the pre-hairpin to hairpin or the hairpin to post-fusion conformational change 162 . Motavizumab, an affinity-matured derivative of palivizumab that binds to the F glycoprotein with tenfold greater potency 163 , was shown to generate a relative decrease of 26% in RSV hospitalization compared with palivizumab. However, results from phase III clinical trials in 2010 showed only a marginal improvement in comparison with palivizumab 164 and an increase of adverse events 164 , leading to termination of its development.
Novel mAbs under evaluation for RSV such as nirsevimab (MEDI8897) and suptavumab (REGN2222) target antigenic sites that are present only in the pre-fusion conformation of the F glycoprotein. Nirsevimab, which is derived from a mAb called D25 isolated by AIMM Therapeutics using a target-agnostic approach 165 , targets the antigenic site Ø 166 and is more potent than palivizumab in vitro 167 . It has also been engineered for extended half-life through 'YTE' substitutions in the Fc region (see Antibody engineering section), which means that only one dose of nirsevimab may be required to cover a typical 5-month RSV season, rather than the five once-monthly doses that would be required for palivizumab. Indeed, nirsevimab showed an extended half-life offering protection from RSV for a typical 5-month season with a single intramuscular dose in a phase III trial 86 . However, suptavumab, which targets the antigenic site V, failed to meet the clinical end points in a phase III study due to the appearance of resistant RSV B strains 168 The glycoprotein of EBOV is a trimeric class I fusion protein formed by three disulfide-linked GP1-GP2 hetero dimers forming a chalice-shaped trimer on the viral surface 171 . The GP1 subunit binds to the EBOV receptor, NPC1, allowing GP2-mediated fusion of the viral and host cell membranes 172 (Fig. 4b). GP1 bears the RBS, glycan cap and mucin-like domain. GP2 contains an N-terminal peptide, internal fusion loop, stalk and transmembrane domain. Of note, the GP1 subunit contains a core domain which is shielded by a 'glycan cap' , made by the heavily glycosylated mucin-like domain. The mucin-like domain is dispensable for viral entry, but is a decoy target for host antibody responses 173 .
Given the seriousness of the Ebola virus disease (EVD) and potential challenges associated with a large outbreak, there is an urgent need for therapies. The success in NHPs of ZMapp, which is a combination of three chimeric mAbs, 13C6 from MB-003 and 4G7 with 2G4 from ZMab 174 , illustrated the potential use of mAb therapies against EVD. The ZMapp cocktail was evaluated in humans during the 2014-2016 Ebola outbreak in West Africa, although efforts in NHPs to simplify the ZMapp regimen to contain fewer mAbs have not been successful 175 . Therefore, the possibility of obtaining fully human mAbs from individuals who survived EVD was investigated.
Two mAbs (mAb100 and mAb114) were isolated from an individual who survived Ebola in the Democratic Republic of Congo in 1995 (ReF. 176 ) (Fig. 4b). mAb114 binds to the GP1 head on an epitope at the physical intersection of the glycoprotein subunits 177 . It demonstrated neutralization and Fc-dependent cell-targeting activities in vitro, and potently activated phagocytosis and natural killer cells 176 . REGN-EB3 (marketed as Inmazeb, Regeneron Pharmaceuticals) is a combination of three fully human mAbs (REGN3470, REGN3479 and REGN3471) targeting the EBOV glycoprotein. These antibodies were obtained from humanized transgenic mice that had been immunized with DNA constructs encoding the EBOV glycoprotein and/or recombinant purified virus glycoprotein 178 . REGN3479 (now known as maftivimab) recognizes the conserved GP2 fusion loop and provides neutralizing activity, whereas REGN3471 (now known as odesivimab) recognizes the outer glycan cap and has cell-targeting functions. REGN3470 (now known as atoltivimab) binds to the GP1 head and offers both neutralization and cell-targeting activities, including FcγRIIIa and other FcγR-related functions.
During the EVD outbreak that occurred in the Democratic Republic of Congo in 2018, the triple mAb cocktail ZMapp, the monotherapy mAb114 (ansuvinmab, also known as VRC 608; developed by Ridgeback Biotherapeutics) and the triple mAb combination REGN-EB3 (maftivimab, odesivimab and atoltivimab) were evaluated in an umbrella trial. After an interim analysis, mAb114 monotherapy and REGN-EB3 were both found to be superior to ZMapp with respect to the primary outcome and patient mortality 179 , and were approved by the FDA for the treatment of EVD in 2020 (ReFS 180,181 ).

Human cytomegalovirus. Human cytomegalovirus (HCMV) is an enveloped double-stranded DNA virus
with a genome size of more than 235 kb, making it the largest known genome of human herpesviruses. HCMV is a member of the betaherpesvirus family and can establish lifelong latency in healthy individuals. Primary infection is generally asymptomatic; however, viral reactivation in immunocompromised hosts can be a life-threatening disease and vertical virus transmission during pregnancy is one of the leading causes of congenital birth defects 182,183 .
Several mAbs targeting the gH/gL or gB complexes have been isolated and have shown modest efficacy in an in vitro model of infection. To identify the most potent HCMV neutralizing antibodies, the Lanzavechia group isolated a large panel of mAbs from memory B cells in naturally infected donors and found that the pentameric complex represented the main target of neutralization against HCMV, eliciting neutralizing antibodies with a potency several orders of magnitude greater than any other HCMV complex 8,48 . This discovery shed light on the importance of the antigen selection to identify potent neutralizing mAbs and to design efficient vaccines.
The use of HCMV-specific mAbs for the prevention of HCMV infection and disease after allogeneic haematopoietic stem cell transplantation 195 or solid organ transplant has been studied extensively 196 . Among the mAbs that have been evaluated in clinical trials, MSL-109, a human mAb targeting HCMV surface glycoprotein gH 197 , was tested as a supplementary treatment for patients with AIDS with HCMV-induced retinitis, but development was halted during phase II/III trials owing to lack of efficacy 198 . RG7667, a mixture of two mAbs binding the gH/gL and pentamer complexes 199 , could potently neutralize HCMV infections of all the cell types tested. However, when evaluated in a phase II trial for recipients of kidney transplants 200 , RG7667 did not meet the primary end point within 12 weeks post-transplant. Next, CSJ148, which consists of two anti-HCMV mAbs, an anti-gB mAb (LJP538) and an anti-pentamer mAb (LJP539) 201 , was evaluated in a phase II trial for prophylaxis of HCMV in patients undergoing haematopoietic stem cell transplantation 202 and also did not achieve the primary end point. Nevertheless, in NHPs, the presence of durable and potently neutralizing antibodies at the time of primary infection was shown to prevent transmission of systemically replicating maternal rhesus CMV to the developing fetus 203 .
So, despite these significant development efforts, no anti-HCMV mAb has yet been FDA-approved. The only clinically available antibody-based therapy is still Cytogam, a preparation of CMV immunoglobulin used for intravenous injection; however, the neutralization capacity of this preparation is suboptimal 204,205 .
Influenza. Influenza, a member of the Orthomyxoviridae family, has four types -influenza A, B, C and D -all of which have a segmented, negative-sense, single-stranded RNA genome. Influenza A and B are responsible for severe infections in humans, whereas influenza C causes only mild symptoms and influenza D is not known to infect humans 206 . The viral genome consists of eight segments that encode at least 12 proteins: haemagglutinin (HA), neuraminidase (NA), PB2, PB1, PB1-F2, PA, PA-X, NP, M1, M2, NS1 and NS2 (ReF. 207 ).
The HA head is the major target of the antibody response, providing only limited breadth due to its high sequence diversity and changes in glycosylation sites 210 . Although the structural architecture of HA is conserved overall, the sequence as well as glycosylation patterns differ among different subtypes 211 . Due to its high sequence variability, the elicited antibody response against the immunodominant head is strain-specific and provides only short-lived immunity 212 . The head contains the RBS that is responsible for viral attachment to the host cells through binding to sialic acid receptors. The RBS forms a shallow pocket and consists of four segments: the 190 helix and the 130, 150 and 220 loops 213 . The RBS itself is relatively conserved except for the 220 loop, whereas the remaining head is highly diverse in sequence.
Antibodies generally target five major antigenic sites, Ca1, Ca2, Cb, Sa and Sb for H1, and sites A-E for H3, which are located around the RBS 206 . Antibodies to the RBS site are potent as they block viral attachment or prevent receptor-mediated endocytosis, and therefore neutralize the virus, rendering it unable to infect cells 214 . However, these antibodies are generally strain-specific and thus not capable of providing immunity against drifted strains. Some exceptions have been observed, such as the bNAbs C05 (ReF. 215 ), S139/1 (ReF. 216 ) and F045-092 (ReF. 217 ), which target the RBS and are able to neutralize within their group. Furthermore, recently identified bNAbs target hidden epitopes at the HA trimer interfaces near the HA head domain that become accessible during a 'breathing motion' of the subunits 218 . FluA-20 has shown broad reactivity to most subtypes by recognizing a conserved epitope at the trimer interface and functions by inhibiting cell to cell spread of the virus 219 .
On the other hand, the HA stem is highly conserved mainly within subtypes and, to some extent, across subtypes, evolving at a much slower rate than the head domain 220 (Fig. 4d). Several bNAbs against this region have recently been characterized in humans after infection or vaccination, making it a highly interesting vaccine design target 221 . These bNAbs generally target a hydrophobic pocket around the Trp21 HA2 residue on the HA stem in close proximity to the fusion peptide and block membrane fusion by retaining the HA in its pre-fusion state 222 .
Other mechanisms of action for stem-directed antibodies involve inhibition of proteolytic cleavage, reduction of viral egress by blocking NA activity through steric hindrance, ADCC and antibody-dependent cellular phagocytosis 223 . These antibodies tend to be less prevalent in humans and often demonstrate little or no neutralizing activity, but are protective in challenge studies 221 . Many of these antibodies are group-specific due to differing glycosylation patterns between group 1 and group 2 influenza HA at position N38, which is only present in the group 2 HA adjacent to the major antigenic site on the HA stem and, thus, can interfere with antibody binding. The F10 (ReF. 224 ) and CR6261 (ReF. 225 ) mAbs are specific for group 1 strains, whereas CR8020 (ReF. 226 ) and CR8043 (ReF. 227 ) mAbs are group 2specific. CR8020 and CR8043 are encoded by VH1-18 and VH1-3 germline regions, whereas group 1 stemspecific neutralizing antibodies have been shown to generally derive from the VH1-69 germline gene 228 (Fig. 4d). Recognition of both group 1 and group 2 has been shown to commonly involve VH1-18-derived, VH6-1-derived, VH3-23-derived or VH3-30-derived antibodies. Some of the broadest neutralizing antibodies, such as FI6v3 (ReF. 229 ), CR9114 (ReF. 230 ), 39.29 (ReF. 231 ) and MEDI8852 (ReF. 232 ), are able to engage HAs from both group 1 and group 2. The exact mechanisms driving differences in immunogenicity of the head versus the stem are not fully understood. However, several explanations have been proposed, including the restricted spatial availability of the stem domain to www.nature.com/nrd B cell receptors because of its close proximity to the viral membrane 233 .
MEDI8852 is a human IgG1 mAb isolated from a patient with uncomplicated influenza A infection. MEDI8852 was evaluated in phase IIa trial and is still in development by MedImmune 234 . Vir Biotechnology is developing VIR-2482, an influenza A neutralizing mAb that binds to the conserved region of HA and neutralizes all major strains since the Spanish flu in 1918 (H1N1) 232 . It is therefore tempting to speculate that this type of antibody could be used as a universal prophylactic agent, overcoming the limitations of current influenza vaccines, for which the antibody response is dependent on individual seasonal antigens. In addition, because the serum half-life of VIR-2482 has been increased by Fc domain engineering, a single dose can last the entire influenza season of around 5-6 months.

Implications for vaccine development Antigen-antibody interactions and structural vaccinology for vaccine design.
Vaccines have proven to be the most effective prophylactic strategy for infectious diseases 235 . However, traditional vaccine development approaches 236 , which rely on three categories of vaccines (live-attenuated, inactivated and dissociated pathogens), have failed for viruses such as HIV, RSV, influenza, hepatitis C virus, HCMV or EBOV.
For most vaccines, the antibody response is crucial and, thus, the identification of antibodies that can potently neutralize a pathogen is a key factor for accelerating vaccine development. Reverse vaccinology 2.0, also known as antibody-based vaccinology, aims to overcome the limitations of traditional approaches by engineering novel vaccines based on the structural characterization of antigens in complex with their cognate antibodies, with the antigen-specific antibody response acting as a correlate of protection 237 (Fig. 5).
Antigens are generally identified by proteomic analysis of crude homogenates of infected cells and, then, chosen based on immunogenicity and their ability to stimulate an immune response. However, this approach is time-consuming, and structural integrity or immunogenicity is not guaranteed upon expression of a newly identified protein. Experimentally driven isolation of neutralizing mAbs and identification of their target is still the best approach to identify vaccine candidates. One of the main advantages of antibody-driven vaccinology is isolation of potent neutralizing mAbs. These mAbs will be useful for passive immunization of immuno compromised patients, and/or as therapeutic agents during the acute phase of infection. An additional advantage of the antibody-driven approach is that mAbs can be instrumental in identifying the optimal vaccine antigen for which mAb binding can block virus transmission.
This identification step can be done by immunoprecipitation coupled to mass spectrometry. In addition, once identified, both the antigen and the mAb can be used for structural vaccinology. The latter approach aims at elucidating the atomic structures of the viral antigens with a neutralizing Fab (Fig. 5). Structural vaccinology is a valuable source of information to engineer antigens for stabilization purposes. The combination of antibody-driven and structure-based antigen design strategies is particularly efficient in developing rapid responses to emerging infectious disease threats.
Facilitated by the improvements in high-throughput B cell technologies, the structural insights into human mAbs have been instrumental in directing the immune response to conserved antigenic sites 72,[238][239][240][241] . Complementary to this approach, structure-based design has been employed to completely remove domains containing sites targeted by non-neutralizing antibodies or to identify possible positions for the introduction of glycosylation sites to mask such epitopes 242 .
Respiratory syncytial virus. RSV provides a case study for the reverse vaccinology 2.0 concept. For RSV, neutralizing antibodies mainly target the head domain of the fusion (F) protein 243 (Fig. 4a). To focus the immune response against antigenic sites in the F head, Boyington et al. 244 designed truncated F immunogens containing only the head region. These immunogens were successful in eliciting neutralization titres comparable with full-length pre-fusion F protein. These proteins are especially vulnerable in their pre-fusion state, and thus have been the target of structure-guided stabilization efforts to lock antigens in their pre-fusion conformation in order to promote a neutralizing antibody response that prevents membrane fusion 160 . This structural information can be used to identify stabilizing mutations such as disulfide bonds or proline mutations to rigidify the protein backbone and mutations to fill cavities to impede transition from pre-fusion to post-fusion states. Additionally, structure-based design has been applied to fuse domains for multimerization or to remove potential unstable regions 242 .
A prime example for the successful stabilization of a pre-fusion antigen is the F protein, where the elicitation of a potent neutralizing antibody response relies on targeting the pre-fusion state 160 . Multiple structure-based design strategies were applied to stabilize the F protein in its pre-fusion conformation. McLellan et al. designed intra-protomer disulfide bonds coupled with additional mutations to fill cavities in the native antigen to achieve a stabilized pre-fusion antigen (Ds-Cav1) 166 . A different design approach by Krarup et al. focused on introducing proline residues to prevent structural rearrangements of the F protein from occurring during the adoption of the post-fusion conformation 245 . Immunization studies in animals confirmed the improved neutralization potential of pre-fusion F constructs over post-fusion antigens, further supported by a subsequent study by Joyce et al. which highlighted that stability improvements of the DS-Cav1 immunogen in the pre-fusion state increased the neutralizing antibody response fourfold as compared with Ds-Cav1 (ReF. 246 ).
A comparatively young field in structure-based vaccine design is epitope scaffolding, which relies on the transplantation of viral epitopes onto unrelated carrier proteins, so-called scaffolds, to focus the immune response against conserved, functional sites that are known to be targeted by neutralizing antibodies 247 . This strategy has first been applied to the design of novel immunogens for HIV [248][249][250][251] , although one limitation in antigen transplant is the stabilization of distant discontinuous epitopes 252 . For RSV, computational design of immunogens was first achieved by McLellan et al., who designed immunogens for antigenic site II of the F protein by grafting of the epitope of motavizumab onto an unrelated protein scaffold 253 . However, immunization of mice with this epitope scaffold did not elicit neutralizing antibodies, although it did elicit sera with F-binding activity 253 . Subsequent design efforts focusing on the same antigenic site resulted in a novel epitope scaffold that engaged site-specific antibodies with high affinity and boosted subdominant antibodies with enhanced neutralization 254,255 . A recent study by Sesterhenn et al. demonstrated that a cocktail formulation of three computationally designed immunogens displaying RSV F sites Ø, II and IV elicited a neutralizing antibody response and allowed for the focusing of the antibody response against specific antigenic sites upon immunization of mice and NHPs 256 . Using a strategy termed motif-centric design, the authors computationally designed de novo topologies around the extracted F antigenic sites to improve epitope stabilization and accurate display of the antigenic site 256 .
Although epitope scaffolding approaches have facilitated the design of novel immunogens that display neutralization epitopes, these designed immunogens have mostly been restricted to small, continuous epitopes 248,249,254 with few exceptions 256,257 . In contrast, observed epitopes often consist of multiple segments 258,259 that are challenging for structure-based design. As an example, Marcandalli et al. demonstrated that the use of structure-based design of a self-assembling protein nanoparticle presenting a pre-fusion-stabilized DS-Cav1 in a repetitive array on the nanoparticle exterior induced neutralizing antibody responses up to tenfold higher than trimeric DS-Cav1 alone 260 . The same nanoparticle was used to derive promising vaccines against SARS-CoV-2 (ReF. 261  Ervebo is a live, attenuated recombinant vesicular stomatitis virus-based vector expressing the envelope glycoprotein gene of Zaire EBOV instead of the VSV-G gene (Fig. 4b). NHP studies demonstrated rVSV-ZEBOV efficacy in stringent conditions, where a single inoculation at ~1 × 10 7 pfu was shown to protect against illness, viraemia and death after challenge with a high dose of EBOV (1,000 pfu, generally thought to represent 100-1,000 times the lethal dose (median lethal dose) in experimental animal studies) [268][269][270] . Zabdeno/Mvabea each contain a monovalent replication-incompetent virus. Zabdeno, an adenoviral vector of serotype 26 that encodes the EBOV-GP Mayinga variant (Ad26.ZEBOV), is used for the priming injection, and Mvabea, a modified vaccinia virus Ankara-Bavarian Nordic Filo-vector encoding the same glycoprotein (MVA-BN-Filo) 271 , is used for the booster.
Other vaccines in development include cAd3-EBO Z, an attenuated version of a chimpanzee adenovirus (cAd3) encoding the EBOV-GP glycoprotein 272 . Finally, a recombinant nanoparticle vaccine with the EBOV-GP Makona strain was found to induce a potent immune response 273 .
Human cytomegalovirus. The development of HCMV vaccines began in the early 1970s, and two attenuated virus strains were isolated for laboratory work: AD169 and Towne 274,275 . The AD169 attenuated strain was quickly abandoned whereas the Towne attenuated strain progressed to extensive testing in recipients of solid organ transplants and healthy volunteers 276 . Recipients of kidney transplants were shown to be highly protected against serious CMV disease and graft rejection. However, protection against viral infection was not statistically significant. The next development in the quest for CMV vaccines was the identification of a surface protein of CMV called glycoprotein B or gB (Fig. 4c). When combined with the MF59 oil-in-water adjuvant, www.nature.com/nrd the vaccine was safe and partially effective 277 , but the levels of neutralizing antibodies were weak in humans after three injections [277][278][279] . The subunit gB protein was also combined with the AS01 adjuvant to stimulate Toll-like receptor 4 (TLR4), which elicited higher and more prolonged levels of anti-gB antibodies in humans. Unfortunately, the adjuvanted vaccine was never tested for efficacy.
The gB antigen is a class III trimeric fusion protein and is used in the post-fusion conformation 110 . Therefore, it was suggested that gB in the pre-fusion conformation could generate a higher neutralizing response, but recent results have, surprisingly, indicated that this might not be the case 280 . The pentameric complex of proteins present on the surface of CMV consists of glycoprotein H (gH), glycoprotein L (gL) and the products of genes UL128, 130 and 131A. The complex has been shown to generate far higher titres of neutralizing antibodies than gB 48,281 (Fig. 4c). This discovery has since driven much of the HCMV vaccine field and vaccine trials are ongoing. However, if neutralizing antibodies are necessary to prevent infection and spread of HCMV, a strong T cell response is also needed to suppress reactivation of the virus in patients who are seropositive. Forthcoming clinical trials using either a recombinant protein subunit (NCT05089630) or mRNA (NCT05105048) will indicate whether the pentamer can be used alone or whether it needs to be injected together with gB 195,282 .
Influenza. One of the major obstacles for effective influenza vaccines is the immunodominance of the highly diverse HA head that is responsible for viral attachment. Structure-based design methods have been focused on the conserved but immunorecessive HA stem region (Fig. 4d).
To shift the antibody response towards the stem domain, chimeric HA molecules have been engineered, consisting of a common HA stem paired with different HA heads [283][284][285] . Repeated immunizations of mice demonstrated that a stem-specific antibody response can be mounted, resulting in heterologous and hetero-subtypic immunity 284 . In ferrets, these constructs were shown to reduce viral loads after influenza virus challenge 286,287 . Chimeric HA has been tested in a phase I clinical trial and was found to be safe, and able to induce a broad, strong, durable and functional immune response 288 .
Although Kanekiyo et al. demonstrated that the isolation of RBDs from different HA strains and multimerization on ferritin nanoparticles resulted in the elicitation of a B cell response against conserved epitopes 289 , the most promising design strategies are based on isolated HA stem-only antigens created by removing the HA head 290,291 . Two such design strategies are headless HAs from Yassine et al. 220 and the mini-HAs from Impagliazzo et al. 233 .
The headless HA immunogens were developed by removing the HA head from an H1 strain, followed by multiple rounds of structure-based design that yielded stabilized stem immunogens 220 . Lethal influenza challenge with a hetero-subtypic H5N1 strain showed complete protection in mice and partial protection in ferrets; however, no cross-group reactivity with group 2 viruses was observed 220 . Applying the same structure-based design strategy to H3 and H7 HA resulted in two group 2 headless immunogens that were able to elicit protective, homo-subtypic antibodies in mice 292 . Immunization of NHPs with headless H3 generated neutralizing antibodies against divergent H3N2 strains and selected H10N8 as well as H7N9 strains 293 .
The mini-HA molecules developed by Impagliazzo et al. follow a similar design approach. A combination of rational and library-based design approaches was applied to generate stabilized HA stem molecules that lack the immunodominant head and the transmembrane region 233 . The designs were based on an H1 subtype and were shown to be protective in lethal heterologous and hetero-subtypic challenge in mouse models 233 . In pre-exposed NHPs, the mini-HAs elicited an expanded influenza-specific humoral immune response when compared with trivalent inactivated influenza vaccine 294 . Together, these results demonstrate that immunogens lacking the immunodominant head domain can elicit a group-specific, hetero-subtypic immune response 293,294 . Recently, a ferritin nanoparticle-based vaccine incorporating the ectodomain of HA from an H2N2 pandemic strain was demonstrated to be safe and immunogenic in a phase I clinical trial, supporting its potential application in pandemic preparedness and universal influenza vaccine development 295,296 .
A different approach to focus the immune response against conserved antigenic sites that is structurally less demanding relies on the masking of non-neutralizing antigenic sites in the HA head through hyper-glycosylation and removal of glycans from the HA stem 297,298 . Eggink et al. applied this strategy to hyper-glycosylated immunodominant epitopes in the HA head, resulting in the silencing of immunodominant sites. Immunization of mice confirmed a shift of the antibody response towards the immunorecessive stem domain and improved protection of mice after viral challenge 299 . Beyond focusing the immune response against the HA stem, this strategy can be also used to improve the immune response against subdominant epitopes in the HA head 218 .

COVID-19.
Acquisition of structural data for the SARS-CoV-2 S protein played an important role in the development of multiple COVID-19 vaccines, including the extraordinarily successful mRNA vaccines. Another critical aspect in the development of effective vaccines was the intensive experience gained from studies of the S protein of other coronaviruses such as HCoV-HKU1 (ReF. 300 ) and MERS-CoV 301,302 led by the McLellan and Veesler laboratories. This allowed the design of stabilizing mutations in the versions of the S protein encoded in the mRNA vaccines (BNT162b2 mRNA from Pfizer-BioNTech and mRNA-1273 from Moderna) and some vectored vaccines (Ad26.COV2.S from Janssen) 303 , such as deletion of the polybasic cleavage site, inclusion of stabilizing mutations 304 and inclusion of trimerization domains 305,306 .
HIV. Research on the development of an HIV vaccine has strongly benefited from the isolation of human bNAbs, reviewed in depth elsewhere [307][308][309][310] . The use of bNAb-instructed stabilization of HIV Env trimers is the basis for many promising immunogens 311 . In addition, the most promising strategies use stabilized trimers with the aims of activating particular bNAbproducing cell precursors and guiding their affinity maturation to generate mature bNAbs -a strategy called germline targeting 312 . Recently, 97% seroconversion was reported 313 in healthy subjects vaccinated with the eOD-GT8 immunogen 314,315 , demonstrating the strong potential of the germline-targeting strategy. Molecular analysis of the B cells induced during the clinical trial (NCT03547245) will provide a road map to accelerate progress towards an HIV vaccine. A phase I clinical trial in which an mRNA vaccine encoding the eOD-GT8 immunogen is used for the priming injection has recently been initiated (NCT05001373).

Closing perspectives
The emergence of SARS-CoV-2 and the devastating COVID-19 pandemic have emphasized the severity of the threat of emerging infectious diseases, especially those of zoonotic origin. In response, there has been unprecedented success in the discovery and development, manufacturing and regulatory evaluation of several anti-COVID-19 vaccines and mAb therapeutics in a very short time, achieved through exceptional mobilization of public and private resources. These advances have been based on the enormous scientific progress made in immunology and vaccinology over the last few decades, and also the crucial previous knowledge on the biology of coronaviruses, such as the S protein as a target for neutralizing antibodies.
Efforts to harness antibodies to combat COVID-19 have also benefited from technological advances and expertise gained particularly in the area of human bNAbs from the HIV field. It is expected that the generation of anti-infective mAbs that are urgently needed will benefit from the recent clinical successes and case studies highlighted in this article and elsewhere 316 . However, mAbs still have limitations. The potential for mutations in the viral targets of antibodies to allow viruses to escape neutralization has recently been highlighted with SARS-CoV-2 Omicron VOCs, and also by the development of resistant variants when selective pressure is applied in the setting of drug treatment, which has been observed in immunocompromised patients treated with bamlanivimab and etesevimab 317 , sotrovimab 318 or REGN-COV2 (ReF. 319 ).
With the recent advances in machine-learning algorithms, and the dramatic increases in the repertoire of available antibodies, further research should provide insight into the structural properties required for bNAbs 320 . Moreover, the progress made with gene-editing technology opens the possibility of engineering functionalities into human cells. Indeed, B cells engineered to carry broadly neutralizing B cell receptors are likely to represent a milestone to address pathogens with no vaccines or inefficient vaccines 321 . Vaccination by edited B cells able to differentiate into memory B cells, plasmablasts and long-lived plasma cells could be a valuable option for HIV and influenza prevention. Finally, given that a key current limitation of mAbs is still their limited distribution in tissues, novel formats or ways to specifically direct mAbs to the targeted tissues may render mAbs even more powerful tools in fighting viral infections.
Published online 20 June 2022