Developing Recombinant Antibodies by Phage Display Against Infectious Diseases and Toxins for Diagnostics and Therapy

Antibodies are essential molecules for diagnosis and treatment of diseases caused by pathogens and their toxins. Antibodies were integrated in our medical repertoire against infectious diseases more than hundred years ago by using animal sera to treat tetanus and diphtheria. In these days, most developed therapeutic antibodies target cancer or autoimmune diseases. The COVID-19 pandemic was a reminder about the importance of antibodies for therapy against infectious diseases. While monoclonal antibodies could be generated by hybridoma technology since the 70ies of the former century, nowadays antibody phage display, among other display technologies, is robustly established to discover new human monoclonal antibodies. Phage display is an in vitro technology which confers the potential for generating antibodies from universal libraries against any conceivable molecule of sufficient size and omits the limitations of the immune systems. If convalescent patients or immunized/infected animals are available, it is possible to construct immune phage display libraries to select in vivo affinity-matured antibodies. A further advantage is the availability of the DNA sequence encoding the phage displayed antibody fragment, which is packaged in the phage particles. Therefore, the selected antibody fragments can be rapidly further engineered in any needed antibody format according to the requirements of the final application. In this review, we present an overview of phage display derived recombinant antibodies against bacterial, viral and eukaryotic pathogens, as well as microbial toxins, intended for diagnostic and therapeutic applications.


INTRODUCTION
Antibodies are indispensable tools for basic research (Colwill and Gräslund, 2011), diagnostics (Kourea and Kotoula, 2016) and therapy (Kaplon and Reichert, 2021). In the past -and still todaypolyclonal antibodies (blood serum) are produced in animals like horses (von Behring and Kitasato, 1890). The hybridoma technology was a milestone in antibody generation because it allowed the production of monoclonal antibodies (Köhler and Milstein, 1975). Despite these antibodies are monoclonal, cell secreted antibodies are not always monospecific, as unveiled by the results of a multicentric study on the detrimental effects of genetic diversity in individual hybridoma clones (Bradbury et al., 2018). Of 185 randomly selected hybridomas analyzed in this study, almost one third was found to contain additional productive heavy or light chains, resulting in antibodies with reduced affinity and more than one specificity. Other limitations of the hybridoma technology are the possible instability of the tetraploid, but also often aneuploid hybridoma cell lines (Pauza et al., 1993), the restriction of the immune system which prevents the generation of antibodies against highly conserved or selfantigens and most important the limitations to directly generate human antibodies (Winter and Milstein, 1991). The hybridoma technology most commonly provides murine antibodies for use in research and diagnostics. But, the therapeutic applications of murine antibodies are very limited because repeated administration of murine antibodies can cause a human antimouse antibody reaction (HAMA), which can lead also to severe side effects such as anaphylactic shock (Courtenay-Luck et al., 1986). While various humanization strategies are available to reduce the immunogenicity, these approaches are often laborious (Almagro and Fransson, 2008;. The human hybridoma technology allows the generation of human antibodies, (Cole et al., 1984;Hammond, 2010), but this approach will not be able to deliver antibodies against any targets because specific B-cells from humans would be needed and this is limited by ethical constraints. An alternative for human antibody generation is the use of transgenic animals in which the animal antibody gene repertoire is substituted with the corresponding human gene repertoire, but also this technology is limited by the restrictions of the immune system. This approach was successfully used for several approved therapeutic antibodies (Jakobovits, 1995;Lonberg and Huszar, 1995;Fishwild et al., 1996;Nelson et al., 2010).
A technology which allows to generate fully human antibodies right away, hence avoiding the side effects originating from HAMA, is antibody phage display. Because this approach uses an in vitro selection process, it does not have to rely on immunization and associated limitations, and can use entirely human gene repertoires. Phage display is based on the work of Georg P. Smith on filamentous phage, which infects E. coli and it is the most commonly used antibody display approach nowadays (Smith, 1985). Referring to the pan and the method of gold diggers, the antibody selection process was called "panning" (Parmley and Smith, 1988).
Around 1990/91, the M13 phage display technology was further developed for human antibody generation in parallel at three research sites: DKFZ in Heidelberg (Germany), MRC Laboratory of Molecular Biology in Cambridge (UK) and at the Scripps Research Institute in La Jolla (USA) (McCafferty et al., 1990;Barbas et al., 1991;Breitling et al., 1991;Clackson et al., 1991). For the expression of the antibody::pIII (phage protein III) fusion proteins via phage display, two different genetic approaches have been developed. Initially, the antibody genes were directly inserted into the M13 phage genome and fused upstream of the wildtype pIII gene (gIII) (McCafferty et al., 1990). Due to a better genetic stability and opportunity for independent regulation of phage and antibodies production, today the most successful system is based on phagemids. These are independent plasmids which provide the genes encoding the antibody::pIII fusion proteins and contain a phage morphogenetic signal for packaging of the phagemid into the phage particles (Breitling et al., 1991). In both phage display systems, the antibody fragments are always displayed on the phage surface and the corresponding antibody gene is packaged into the same phage particle. This successful technology was awarded with the Nobel Prize for Chemistry in 2018. For antibody phage display, the most common used antibody formats are the single chain fragment variable (scFv) (Vaughan et al., 1996;Schofield et al., 2007;Hust et al., 2011) or fragment antigen binding (Fabs) (de Haard et al., 1999;Hoet et al., 2005). Other used formats are human VH domains (dAbs), immunoglobulins of sharks (IgNARs) and the variable domains of camel heavy chains (VHHs) (Muyldermans, 2001;Nuttall et al., 2001;Holt et al., 2003;Nuttall et al., 2004;Muyldermans et al., 2009). The use of immune camel, llama or alpaca libraries gained popularity in the last five years. For veterinary research, chicken antibody libraries are often used (Fehrsen et al., 2005;Foord et al., 2007;Wemmer et al., 2010). The diversity of chicken antibody genes is the result of gene conversion. Here, the N-and C-terminal parts of chicken's VH and VL are always identical, which facilitates antibody gene amplification and library cloning (Parvari et al., 1988;Chiliza et al., 2008).
In the antibody selection process (panning), in most cases, the target molecules is immobilized on a solid surface, like plastic surfaces -normally used for ELISA -such as polystyrene tubes or microtiter plate wells, the latter of which is the most simple and commonly used method (Hust et al., 2002;Russo et al., 2018), affinity chromatography column matrixes (Breitling et al., 1991) or magnetic beads (Sawyer et al., 1997). An alternative is the use of biotinylated targets for a selection in solution followed by a "pull-down" with streptavidin beads (Schütte et al., 2009;Wenzel et al., 2020b). After incubating the entire diversity of the phage display library with these immobilized antigens, the tremendous excess of non-or weak-binding antibody phage particles are removed by stringent washing. Afterwards, the bound antibody phage will be eluted, mainly by trypsinization or by pH shift, and used for E. coli infection. Subsequently, the phagemid bearing E. coli will be infected with a helperphage to produce new antibody phage particles with an enriched fraction of antigen specific antibody phage clones. Usually 3-5 panning rounds are performed before screening. Specific binders can be identified in a screening process using monoclonal soluble antibodies or monoclonal antibody phage by e.g. ELISA , WB (Hust et al., 2002) or flow cytometer (Ayriss et al., 2007). The antibody fragment genes can be subcloned to produce antibodies in any desired antibody format, e.g. IgG or scFv-Fc (Hoet et al., 2005;Dübel et al., 2010;Hust et al., 2011;Jäger et al., 2013;Frenzel et al., 2014). Figure 1 shows an overview on the phage display antibody selection process. An overview about phage display in comparison to other display technologies is given by Valldorf et al. (2021).
Depending the origin of the V-gene repertoires, there are two types of libraries: immune libraries and universal libraries. Immune libraries are constructed from convalescent patients or immunized donors. This library type is most used to obtain an antibody against a particular target, e.g. an infectious pathogen like Ebola virus (Maruyama et al., 1999a) or diphtheria toxin (Wenzel et al., 2020a). The advantage of this library type is that the V-genes contain hypermutations and are already affinitymatured for the targeted pathogen.
The alternatives are universal libraries -in the past also named "single-pot" libraries -made of naïve, semi-synthetic or synthetic gene repertoires. This library type allows to isolate antibody fragments binding to virtually every possible target without the need for an immunization or the availability of patients' blood. (Winter et al., 1994;Dübel et al., 2010;Frenzel et al., 2017). In a pandemic situation, this provides a source for immediate antibody generation against a new arising pathogen, independently from the availability of material from patients (Bertoglio et al., 2021b). The Naïve libraries are cloned from rearranged V genes derived from B cells (IgM) of non-immunized donors. Examples for naïve libraries are the HAL scFv libraries Kügler et al., 2015) and Fab library constructed by de Haard and colleagues (de Haard et al., 1999).
Semi-synthetic libraries are cloned from unrearranged V genes from pre-B cells (germline cells)  or in most cases from one antibody framework (Pini et al., 1998). In this defined framework one or several CDRs, but always the CDR H3, are randomized, e.g. by using DNA-primers covering the CDR H3 section. The semi-synthetic libraries Tomlinson I and J libraries are often used and are the best example for semisynthetic libraries. These libraries have a VH3-23 and Kappa IKV1-39 framework combined with randomized CDR2 and CDR3 (de Wildt et al., 2000). For the FAB310 antibody gene library, a combination of naïve and synthetic repertoire was used. Here, the Fd fragment (VH+CH1) contains randomized CDR1 and CDR2 in the human VH3-23 framework and a naïve CDR3 regions from autoimmune patients. The Fd fragments were combined with light chains also from autoimmune patients (Hoet et al., 2005). The transition of semi-synthetic to synthetic libraries is not strict. In general, complete synthetic libraries are made of human frameworks and here all CDR cassettes are synthetic, the best examples are the HUCAL libraries (Hayashi et al., 1994;Knappik et al., 2000;Tiller et al., 2013). Universal libraries have a theoretical diversity higher than 10 10 independent clones (Schofield et al., 2007;Glanville et al., 2009;Lloyd et al., 2009;Tiller et al., 2013;Kügler et al., 2015).
In vitro display methods, like phage display, differ from classical animal immunization approaches to obtain polyclonals or hybridomas, in the possibility to directly generate sequence defined monoclonal antibodies. Additionally, the phage display antibody selection process occurs entirely in vitro and its customization allows to pre-filter antibodies with the intended binding properties, such as epitope specificity, binding at defined pH and temperatures, or absence of unwanted cross-reactivities. An alternative method that allows to generate sequence defined FIGURE 1 | Schematic illustration of the antibody generation process. For the antibody generation, human phage display libraries or libraries from other species are used. These libraries will be used for the in vitro selection (panning) on a target molecule. In short: the antibody phage particles will be incubated with target molecule (in the illustration, a target molecule was immobilized in a microtitre plate well and a scFv phage display library is used), the non and weak binding antibody phage particles will be washed away and the binding antibody phage will be eluted (the procedure will be repeated 3-5 times) and further analyzed. In the next step, monoclonal binders will be identified, e.g. by ELISA. Finally, the selected antibody fragments can be recloned, because the corresponding gene is packaged in the phage particles, and produced in any antibody format tailored to the final application.
antibodies is the microfluidics-based B-cell selection method (Seah et al., 2018). B-cells secreted antibodies binding to labeled antigen material can be assessed at single cell level in droplets of buffer solution. This method, very laborious and technically demanding, is restricted to the availability of patient or immunized animal B-cells. Additionally, this approach provides a lower degree of flexibility in the number of assays which can be performed in the initial antibody screening phase. The upside of this method is the possibility to discover antibodies already in the IgG format. These features made this method particularly popular for the discovery of therapeutic antibodies from B-cells of disease-recovered patients. This is the case of the FDA approved Neutralizing antibodies to Ebola virus glycoprotein Ebanga (Ansuvimab-zykl) (https://www.fda.gov/ drugs/drug-safety-and-availability/fda-approves-treatment-ebolavirus), the anti-HIV neutralizing antibodies under clinical evaluation (Spencer et al., 2021) or the anti-SARS-CoV-2 antibodies REGN10933 and REGN10987 (Hansen et al., 2020).
To date, 106 monoclonal antibodies and antibody conjugates have been approved by EMA and/or FDA (status March 2021), 87 antibodies are in clinical phase 2 or 3 and 122 antibodies are in phase 1 or IND filed (Source: antibodysociety.org, status April 2020). Most approved antibodies are for the treatment of autoimmune diseases and cancer. The annual sales of therapeutic antibodies exceeded 98 billion US$ in 2017 and the market will be predicted with 137-200 billion US$ in 2022 (Grilo and Mantalaris, 2019). The mode of action of therapeutic antibodies are numerous and include neutralization of substances e.g. cytokines like tumor necrosis factor alpha (TNF-a) (Alonso-Ruiz et al., 2008) or toxins (Rasetti-Escargueil et al., 2017), human cell binding and modulation of the host immune system (Chatenoud and Bluestone, 2007), blocking of receptors which are overexpressed on cancer cells like epidermal growth factor receptor (EGFR) (Peeters et al., 2009) or combinations of these mode of actions (Adams and Weiner, 2005). Currently, eight recombinant antibodies or antibody cocktails are approved for the direct treatment of pathogens or toxins. The first recombinant antibody approved for the treatment of an infectious disease was Palivizumab. It is a classical humanized antibody approved in 1998 for the treatment of Respiratory syncytial virus (RSV) bronchiolitis (Malley et al., 1998). For anthrax treatment, the human antibody Raxibacumab was developed. This antibody is derived from a human phage display library (made by Cambridge Antibody Technology in cooperation with Human Genome Science) and was approved in 2012 (Mazumdar, 2009;Kummerfeldt, 2014). A second antibody for post-exposure treatment of anthrax, the antibody Obiltoxaximab, has been approved in 2020 (Henning et al., 2018). A further approved antibody against toxins is Bezlotoxumab, approved in 2016. This antibody neutralizes the Clostridioides difficile toxin TcdB (Babcock et al., 2006;Johnson and Gerding, 2019). The anti-HIV antibody Ibalizumab has been approved in 2018. The mode of action is very unique for an antiviral antibody, because it blocks cell entry by binding to the human receptor CD4 on the target cells instead of binding to the virus (Beccari et al., 2019). The anti-rabies antibody Rabishield to replace anti-rabies serum therapies has been approved 2018 in India (Wang et al., 2011;Gogtay et al., 2018). End of 2020, two products were approved for the treatment of Ebola: the threeantibody cocktail REGN-EB3 (Atoltivimab/maftivimab/ odesivimab) (Pascal et al., 2018) and the single antibody mAb114 (Ansuvimab-zykl) (Corti et al., 2016). These antibodies showed a good efficiency in clinical trials against Ebola virus especially in comparison to Remdesivir (Mulangu et al., 2019). In the COVID-19 pandemic, the anti-SARS-CoV-2 antibody cocktail Casirivimab (REGN10933)/Imdevimab (REGN10987) Hansen et al., 2020) and the antibody Bamlanivimab (LY-CoV555) (Chen et al., 2021) got a conditional marketing authorization in Europe and an Emergency use authorization (EUA) in the USA for nonhospitalized patients with mild to moderate symptoms. The FDA also authorized the antibody cocktail Bamlanivimab and Etesevimab (JS016, CB6) (Shi et al., 2020) in the USA for the same indication. For the treatment of hospitalized patients with moderate to severe symptoms, phage display derived antibody COR-101 (STE90-C11) against SARS-CoV-2 (Bertoglio et al., 2021a) is in clinical phase 1b/II.
In the next sections, we provide an overview of recombinant antibodies derived from phage display against bacterial and viral pathogens, eukaryotic pathogens (parasites, fungi) and toxins, as well as detailed examples for diagnostics and therapy.

RECOMBINANT ANTIBODIES AGAINST BACTERIA
Toxins are the most frequent antigenic target of therapeutic antibodies used to fight bacterial pathogens. This topic is thoroughly discussed in a dedicated paragraph: "targeting of toxins with recombinant antibodies". Excluding toxin antigens, antibodies raised against bacteria are predominantly used in diagnostics (Skottrup et al., 2011;Ferrara et al., 2012) and to assess the presence of contamination in environmental samples (Griep et al., 1998;Mechaly et al., 2008). As of today, microbiologic bacteria identification by cultivation remains the benchmark approach for the diagnostic of several pathogenic bacterial species, including Mycobacterium tuberculosis (Norbis et al., 2014), Salmonella Typhimurium (Kuhn et al., 2012), or Listeria monocytogenes (Gasanov et al., 2005).
The many drawbacks of these methods are the low throughput and high costs caused by their inherently long experimental duration, the higher risk of exposure to pathogenic organisms and the need of highly trained personnel. Real-time PCR and other molecular diagnostics approaches have been introduced to reduce the time needed for the identification of many bacteria (Liang et al., 2018), while offering high detection sensitivity and specificity. Nevertheless, such methods most often depend on the availability of laboratory infrastructure and, if portable, are less amenable to high throughput parallelized measurements . Similarly, mass spectrometry is able to provide high sensitivity, however, this technology requires expensive devices and trained personnel (Dixon et al., 2015). On the other hand, antibody-based detection methods such as ELISA and immunochromatography, are faster, simpler, and well-suited for high-throughput. These features, together with the lower costs compared to other techniques, contributed to the success of these diagnostic approaches world-wide, including developing countries, where resources may be scarce and laboratories sparsely distributed over the territory.
Up to date, a vast plethora of antibodies for diagnostic applications has been generated via phage-display against different bacterial proteins or carbohydrates (Table 1). Interestingly, research on this field seems to be in constant rise, as 44% of these antibodies have been generated only in the past five years.
In the following paragraphs, we offer detailed examples on antibodies against different bacterial pathogens that were generated using phage display.
Mycobacterium tuberculosis (Mtb) is the bacteria responsible for a lung infection known as Tuberculosis (TB). This pathogen specifically attacks and replicates inside the macrophage cells in the alveoli, causing fatigue, fever and coughing (Fu and Fu-Liu, 2002). As of 2021, this disease affected approximately 10 million people and caused 1.5 million deaths worldwide per year (World Health Organization, 2021). TB early diagnosis is essential to improve treatment and limit disease spread. The protein secreted by Mtb in the highest amounts, Antigen 85 (Ag85), represents an ideal target for diagnostics . Ag85 is an oligomeric protein composed of three relatively small (30-32 kDa) homologue proteins (Ag85A, Ag85B and Ag85C) . In a novel approach described by Ferrara et al. (2012), a combination of antibody display techniques was successfully adopted to select Ag85 complex-specific antibodies. A universal phage display naïve scFv-library with a nominal diversity of~10 11 unique antibodies was used for initial selection on Ag85, obtaining a sub-library of 10 5 pre-selected antibodies. Second, the selected antigen-specific scFv genes were used for the generation of yeast-display antibody sub-library. Upon completion of yeast based antibody generation, each clone was tested in FACS for its ability to specifically recognize the antigen target resulting in the discovery of 192 clones having the desired properties. Sequence analysis revealed that 111 of these antibodies were unique and could be used in further Ab characterization. Chessboard antibody combination testing resulted in the identification of 7 antibody pairs that detect as little as 22.7 nM Ag85 in patient sera or 6.1 nM in serum freeconditions. The uniqueness of this work relies in the fact that each validated antibody was specific for the Ag85-complex only. Other approaches were aimed at discovering subunit-specific antibodies, as in the case of Fuchs et al. Here, the naïve HAL7/8 phage display libraries were incubated with recombinant Ag85B subunit (Fuchs et al., 2014). The resulting antibodies were shown to have, in sandwich ELISA or lateral flow test, an antigen protein limit of detection of respectively 10 ng/mL and <5 ng/ mL. Nevertheless, when using bacterial cell extracts or filtered culture material, immunodetection could be only performed in ELISA or western blot. The authors concluded that the absence of an antibody affinity improvement step may have been the reason for their results. This supposition found indirect confirmation in a different study by Sixholo and colleagues (Sixholo et al., 2011) describing the selection of antibodies against the Mtb 16 kDa heat shock protein X (HspX) from chicken semi-synthetic Nkuku antibody gene library. By the repetition of four panning rounds against recombinant antigen material, three unique scFv's were confirmed to bind HspX in ELISA. On purpose, the clone with the weakest binding was chosen for in vitro affinity maturation by different methods. In the first approach, a mutant library of the parental antibody gene was generated by error-prone PCR, resulting in a diversity of~3 × 10 7 unique scFv clones. Repeating the antibody selection using this library under more stringent conditions, led to the discovery of three mutant scFv's with increased binding affinity. Direct binding comparison in ELISA showed that the newly generated antibodies produced a signal~11 times higher than the parental clone. The result was confirmed by the improved association and dissociation kinetics in SPR analysis. All affinity-improved antibody genes contained mutated amino acids in both CDR and framework regions. In the second approach, a glycine residue alone was replaced to the common 15-mer linker connecting the VH to VL, resulting in tetramer formation (Schmiedl et al., 2006). The avidity effect leading to cooperative binding resulted in a binding signal increase of~14 times in ELISA. Antibodies against Mtb were also generated by Hamidon and colleagues (Hamidon et al., 2018) utilizing a different approach. A patient immune TB library with a nominal diversity of 10 9 different scFv antibodies could be generated starting from the B-cell material from six individuals infected with TB. This immune library was used to select antibodies against Mtb a-crystalline, resulting in three unique antigen-specific clones validated for antigen binding in ELISA and Western Blot assays.
Periodontitis is a disease where inflammation of the dental gum results in its progressive retraction and, if not cured on time, loss of teeth. One of the major pathogens involved in gum infection is Porphyromonas gingivalis (Socransky et al., 1998). The RgpB cysteine proteases secreted by this pathogen is one of the causing factors of the gum inflammation and constitutes a valid biomarker candidate (Imamura, 2003). Skottrup et al. (2011) generated antibodies against immobilized RgpB from a naïve library of camelid antibodies with a diversity of~5 x 10 7 different clones. Such antibodies consist of a monomeric VHH domain (~15 kDa), which offer multiple advantages for diagnostics, in particular high yields in E. coli production systems, good stability and small epitope accessibility. This library yielded an elite antibody clone with 362 pM affinity towards the recombinant RgpB, also specific to the antigen presented on cells. While not inhibiting RgpB catalytic activity, this antibody showed in subtractive inhibition ELISA a limit of detection equal to~8 x 10 6 cells/mL of saliva.
In the field of foodborne gastrointestinal infection causing pathogens, Salmonella Typhimurium is probably the most studied bacteria (Jacobsen et al., 2011). In the attempt to generate antibodies for the detection of this pathogen, Meyer et al. (2012) utilized ORFeome phage display for biomarker  (Postel et al., 2016) discovery. Starting from bacterial genomic material, a random protein fragments library was screened using sera from Salmonella infected animals. In a second step, the antigens corresponding to the newly discovered immunogenic proteins were used to select human antibodies from the naïve HAL gene libraries. Antibodies are also potent tools for the monitoring of a vaccination campaign and discrimination of infected from vaccinated subjects. For the development of DIVA (Differentiating Infected from Vaccinated Animals) vaccines against S. Typhimurium, Selke and colleagues (Selke et al., 2007) generated an engineered bacterial strain lacking the marker protein OmpD (outer membrane protein D). A serological test using OmpD-specific antibodies constitutes a tool to discriminate infected animal from vaccinated ones. Meyer et al. (2011) generated scFv antibodies against OmpD for antigen detection in competitive ELISA on swine serum. Another very prominent foodborne pathogen is Listeria monocytogenes. This bacterial infection, listeriosis, can cause severe illness, including meningitis and sepsis when reaching the systemic circulation. As of today, 13 different L. monocytogenes serotypes have been discovered, of which type 4b, 1/2a and 1/2b have shown to be the most virulent (Datta and Burall, 2018). Contaminated food also contains non-pathogenic Listeria species, making the diagnosis more difficult. Indeed, their faster growth rate during the enrichment step required for bacteria detection is the major cause of false-negative results in Listeria monocytogenes food diagnostics (Besse et al., 2010). Hence, species-specific immunodiagnostic tools constitute a necessary resource to improve both L. monocytogenes as well as Listeria spp. detection. Gene and colleagues (Gene et al., 2015) used naive and immune camelid antibody libraries for the selection via phage display of diagnostic nanobodies against internalin B (InIB), a protein which is strongly involved in host-cell invasion. Two of the five generated nanobodies, R3-03 and R3-30, showed picomolar binding affinities to InIB, representing a valuable resource for bacteria immunodetection in poorly concentrated samples. A recent study from Moreira et al. (2020) described the use in combination of different phage display techniques to identify new Listeria spp. protein biomarkers and rise antibodies against them. Initially, the human naïve antibody display phage library HAL9/10 was used to generate antibodies against different cell fractions from L. monocytogenes (cell wall, membrane, and cytoplasm). This way, four binders were isolated, which allowed the identification of a novel target for Listeria spp. detection named pyruvate dehydrogenase complex-enzyme 2 (PDC-E2) by using both immunomagnetic separation/mass spectrometry (IMS-MS) and ORFeome phage display. One of these antibodies was also used in immunofluorescence on non-permeabilized cells confirming the surface localization of the target. Antibody selection against PDC-E2 recombinant antigen yielded 16 additional antibodies. These 20 antibodies were finally tested in indirect ELISA against a panel of 17 Listeria species (including the most virulent L. monocytogenes serotypes 4b, 1/2a, and 1/2b) and 16 non-Listeria species. Two antibodies, GSM313-E9 and GSM313-H8, provided 100% sensitivity and specificity for Listeria spp. detection. Moreover, the binding region (epitope) of 18 out of the 20 antibodies was identified via single-gene phage display for epitope mapping, enabling to define the lipoyl domains of PDC-E2 as the major contributors for the detection through this target.

RECOMBINANT ANTIBODIES AGAINST VIRUSES
Viruses are in every kind of environment and are able to infect any sort of host. New viral species are continuously discovered and therefore much research has been devoted during the recent  (Lillo et al., 2011) decades to generate vast sets of antibodies against viruses, both from universal and immune libraries using phage display technology. Target for selection of antibodies ranged from peptides and recombinant viral proteins to complete viral or pseudoviral particles, leading to the isolation of antibodies against human pathogenic viruses such as Sin nombre virus (Velappan et al., 2007), Dengue virus (Cabezas et al., 2008;Saokaew et al., 2014), Influenza virus (Lim et al., 2008;Sui et al., 2009), VEEV (Kirsch et al., 2008), Norovirus (Higo-Moriguchi et al., 2014), SARS coronavirus (Sui et al., 2004), SARS-CoV-2 (Bertoglio et al., 2021b) or Hepatitis C (Songsivilai and Dharakul, 1998) from naïve antibody gene libraries.
The predominant format harnessed for initial selection of antibodies against viruses has been scFv format (Gould et al., 2005;Wu et al., 2014), although Fab (Wyrzucki et al., 2014;Zhao et al., 2014) and VHH libraries (Fatima et al., 2014) were often used. In a peculiar case, Xiao et al. used the antibody CH2 domain as scaffold to generate binders against gp120 of HIV (Xiao et al., 2009). In the case of HIV, some unique aspects could be highlighted as for example that the neutralizing anti-gp41 antibody HK20 has a higher neutralization potency as scFv or Fab compared to IgG, indicating that the IgG format has more difficult access to the epitope, due to the formats bigger steric hindrance (Sabin et al., 2010). Another noteworthy example is the anti-gp41 VHH 2H10, where a CDR3 tryptophan is not involved in epitope binding, but resulted essential for virus neutralization (Lutje Hulsik et al., 2013).
In the following paragraphs, we present detailed examples for antibody generation and engineering using phage display against different viral taxa.
Vaccinia virus is the arketypal virus within the genus of Orthopoxvirus. This virus is with 300x 230 nM a large DNA virus with a genome of about 0,2 Mb (Malkin et al., 2003). Among the species included in Orthopoxvirus genus, there are monkeypox virus, cowpox virus and especially variola virus, the causative agent of smallpox in humans. Occurrence of natural smallpox has been eradicated in 1977 thanks to a massive WHO vaccination program, which started ten years earlier. However, since no vaccination of the general population is nowadays conducted and the use of variola virus as potential threat following intentional release has renewed the research interest for safe and effective smallpox vaccines, due to case fatality rates of 30% or more among unvaccinated subjects (Henderson et al., 1999). It is assumed that immunity against one poxvirus provides broad immunity against most of the viral family members given the high similarity shared among orthiopoxviruses (Earl et al., 2004;Berhanu et al., 2008). Using an immune scFv phage display library constructed from vaccinia virus immunized patients, human vaccinia specific antibodies were isolated. In vitro plaque-reduction neutralization tests proved that seven of these antibodies were able to neutralize both, vaccinia and cowpox virus. Five of them, additionally neutralized monkeypox virus (Tikunova et al., 2012). Another set of antibodies were selected from a Fab immune library derived from chimpanzee, vaccinated with vaccinia virus. After conversion into a chimeric chimpanzee/human IgG format, antibody 8AH8AL, binding to vaccinia protein B5, was neutralizing in vitro for vaccinia and smallpox virus. Its protective effect was assessed in mice challenged with vaccinia virus even when the monoclonal antibody was administered 2 days after challenge. In this model, 8AH8AL provided significantly greater protection than that of the previously isolated rat anti-B5 antibody 19C2 . Since experimental work with smallpox virus is extremely restricted and essentially not allowed, vaccinia had to be used as model.
Ebola Virus and Marburg Virus, two members of the Filoviridae family, cause severe hemorrhagic fevers and extremely high mortality of up to 90% in infected humans. Besides the obvious public health concern associated with its natural outbreaks, Ebola virus might be considered a potential agent of biological warfare and bio-terrorism (Goodchild et al., 2011). Phage display-derived human antibodies targeting Ebola virus were selected from an immune library originated from survivors of the 1995 Ebola virus outbreak in Kikwit, Democratic Republic of Congo (Maruyama et al., 1999b). Nucleoprotein, envelope glycoprotein and secreted envelope glycoprotein binders have been isolated in this study. Antibody (KZ52), recognizing the envelope glycoprotein, neutralized in vitro both, as Fab (50% neutralization at 0.4 µg/mL) as well as full IgG (90% neutralization at 2.6 µg/mL) (Maruyama et al., 1999a). Follow-up studies demonstrated effective protection in vivo in a guinea pig Ebola challenge model, upon administration of the antibody up to one hour post viral challenge . Unfortunately, KZ52 showed no protection in macaques challenged with Ebola, even though the antibody was given in a two-dose regimen with the first dose one day prior viral challenge and the second after three days the former dose (Oswald et al., 2007). Further examples of antibodies selected through phage display against Filoviridae family members encompass a murine scFv and two shark IgNAR V immune libraries generated against inactivated Zaire Ebola virus to yield several antibodies specific for the viral matrix protein VP40 and the viral nucleoprotein (Goodchild et al., 2011). Of note, this work is the first example of a successful targeted IgNAR V isolation from a shark immune library. The antibody combination X10B1/X10H2 generated against the glycoprotein of Ebola Sudan virus (SUDV) showed 100% in vivo protection in scFv-Fc format in a mice challenge model. After 35 days post infection, the mice were challenged again with SUDV and all mice survived showing the development of an own protective immune answer against SUDV (Froude et al., 2017). These antibodies were also tested as IgG and the X10H2 antibody was successfully employed in a cocktail with the non-phage display derived antibody 16F6 (Herbert et al., 2020). Recombinant antibodies against Marburg virus were generated using human synthetic libraries (Amatya et al., 2019) or macaque immune libraries (Froude et al., 2017). The macaque antibodies showed in vitro neutralization and up to 100% in vivo protection against wt Ci67 Marburg virus in a prophylactic mice challenge model using 100 µg scFv-Fc/mice.
Flaviviridae family member Dengue virus (DENV) causes at least 100 million symptomatic infections each year and became a major health and economic burden in over 50 countries worldwide (Moreland et al., 2010;de Alwis et al., 2012;Moreland et al., 2012). Its genome encompasses a single open reading frame, contained in the~11 kb positive strand RNA. The four circulating serotypes of DENV show approximately 70% sequence homology (Moreland et al., 2012;Zhao et al., 2014). DENV infection can result in clinical symptoms that include dengue fever, dengue hemorrhagic fever (DHF) and dengue shock syndrome (DSS) (Yamanaka et al., 2021), which is presumably promoted by antibody-dependent enhancement (ADE) (Taylor et al., 2015). ADE results from partially or nonneutralizing anti-dengue IgG antibodies from a former infection mainly of another dengue serotype. The presence of ADE is associated with the onset of DHF that can be fatal (Halstead, 2014). Two Fabs, namely 5H2 and 5D9, were isolated from a chimpanzee immune library and neutralized dengue type 4 virus in a plaque reduction neutralization test (Men et al., 2004). In another example, human scFv antibodies specific to dengue virus envelope protein were selected by panning against recombinant full length envelope protein and its domain III . Because DENV envelope protein plays an essential role in virion assembly and viral entry, scFvs selected in this study were shown to block DENV infection in vitro . Dengue nonstructural protein 5 (NS5) is involved in viral replication and host immune response modulation. A naïve human Fab-phage library was screened for NS5-specific antibodies using various NS5 variants from Dengue Virus serotypes 1-4 as antigens for panning and characterization (Zhao et al., 2014). Alternating the use of NS5 derived from different dengue serotypes for each selection round, this strategy allowed the identification of two monoclonals that are crossreactive against all four dengue serotypes. Another study presented antibodies selected against antigen Dengue virus particles directly captured from supernatant of infected Vero cells. Here, highly serotype specific antibodies were generated. Seven out of nine antibodies were specific to only one DENV serotype. One Dengue-3 selected clone cross-reacted with Dengue 1, whereas another clone showed pancross-reactivity despite being selected on Dengue 2 particles. Interestingly, all of the obtained antibodies recognized several strains of distinct genotypes within the corresponding serotype (Cabezas et al., 2008). Antibody C9 was isolated after panning against dengue envelope protein using a mouse/human chimeric Fab library. The antibody crossreacts with DENV1-3 and neutralizes DENV2 in cell-based assays after conversion into full length IgG (Moreland et al., 2012). Besides scFv and Fab, also variable domain heavy-chain antibodies (VHH antibodies) have been selected using phage display technology to target DENV2 NS1 protein. After four rounds of panning, 20 hits were selected (Fatima et al., 2014). A VHH antibody combination was developed which allows the detection of NS1 of all four Dengue strains . For the therapy of dengue, the application of full-length antibodies like IgG may not provide the best solution for dengue therapy in the context of ADE. Antibodies with a silent Fc part or antibody fragments would be the preferred option for the development of therapeutic antibodies against dengue.
Venezuelan equine encephalitis virus (VEEV), an alphavirus of the Togoviridae family, causes both equine epidemics and also encephalitis in humans (Johnson and Martin, 1974;Weaver et al., 2004). Much research addressed the generation of neutralizing antibodies against VEEV since it is classified as Category B agent by the Centers for Disease Control and Prevention (CDC). An immune library from human donors was used as source for selection of mAbs targeting both VEEV envelope glycoproteins E1 and E2 (Hunt et al., 2010). The isolated Fabs L1A7 and F5 were neutralizing in vitro, with F5 being 300 times more effective than L1A7. Subsequently, F5 in full IgG format was employed to generate neutralization-escape variants of VEEV for epitope mapping purposes. Another study harnessed an immune macaque library to generate human-like antibodies. Among these antibodies, scFv-Fc ToR67-3B4 was protective in mice when administered 6 h post viral challenge with VEEV Trinidiad strain, allowing 80 to 100% survival after a challenge with 100-fold LD 50 . Counterintuitively, scFv-Fc ToR67-3B4 was able to neutralize other VEEV strains in vitro, but not Trinidad one. This tests proved that in vitro neutralization is not mandatory for an antibody to be actually protective in more complex infection models (Rülker et al., 2012). A further study described the antibody selection from a human naïve scFv gene library using complete, active VEEV particles as target. Specific detection of the VEEV strains TC83, H12/93, and 230 by the isolated antibodies was verified. Remarkably, none of the selected scFv phage clones showed cross-reactivity with Alphavirus species of the Eastern equine encephalitis virus (EEEV) and Western equine encephalitis virus (WEEV) antigenic complex or with Chikungunya virus (CHIKV), making them ideal tools for immunological detection and diagnostic of Alphavirus species (Kirsch et al., 2008). From WEEV immunized macaques, two different scFv antibody gene libraries were constructed. Reformatted as scFv-Fc, three antibodies originating from these libraries specifically bound WEEV in ELISA with little or absent cross-reactivity with other alphaviruses. Additionally, their neutralization potency was assessed in vitro. This was indeed the first study in which in vitro neutralizing antibodies against WEEV were developed. About 1 ng/mL of the best antibody (ToR69-3A2) neutralized 50% of 5 x 10 4 TCID50/mL WEEV (Hülseweh et al., 2014). Four anti-WEEV antibodies from this study were also tested in an in vivo WEEV aerosol challenge approach. A dose dependent protection was measured. Interestingly, not only the best neutralizing antibody ToR69-3A2 was protective in vivo, but also the non-neutralizing antibody ToR68-2C3. The mode of action of the non-neutralizing antibodies cannot be the blocking of the virus host cell receptor interaction, but protection may results from complement-mediated lysis, antibody-dependent cell-mediated lysis of infected cells or opsonization followed by uptake by phagocytes (Burke et al., 2018). Single domain antibodies generated against E3E2 of WEEV from a llama immune library allowed the development of a MagPlex sandwich immunoassay to detect specifically WEEV and discriminate from other alphaviruses .
Influenza viruses are the etiologic agents of seasonal flu, a worldwide spread disease. In recent years, the "bird flu" (H5N1) and the pandemic "swine flu" (a variant of H1N1) moved into the spotlight of public attention and also of research focus, given the global spread of these variants. Due to many genetic events, such as e.g. antigenic drift and shift, new influenza variants will keep occurring in the future and will challenge vaccine and diagnostic development (Webster and Govorkova, 2014;Trombetta et al., 2015;Keilman, 2019). Sui et al. (2009) selected antibodies from a naïve scFv library against the H5 hemagglutinin ectodomain. Hemagglutinin is a trimer and the extracelluar part consist of a stalk domain and a globular head domain (Krammer and Palese, 2013). They identified ten antibodies binding to the trimeric H5, nine of which shared the same germline framework (VH1-69). These antibodies were converted into IgG1 and were protective in mice at dose of 10 or 15 mg/kg in both a prophylactic and therapeutic challenge models, respectively. Very remarkably, some antibodies crossneutralized H1, H2, H5, H6, H8, and H9 influenza strains. These phage display derived antibodies are good candidates for broadspectrum influenza immunotherapy (Sui et al., 2009).
Rabies is caused by the rabies virus which infects the central nervous system, always resulting in fatal disease, before Louis Pasteur developed the rabies vaccination. The current postexposure therapy is based both on vaccination and polyclonal anti-rabies sera (Goudsmit et al., 2006;Briggs, 2012;Crowcroft and Thampi, 2015). From two different immune scFv libraries, 147 unique recombinant antibodies were selected against the rabies glycoprotein . The neutralization of the selected mAbs was tested in vitro using 27 different street rabies virus isolates and the best neutralizing antibodies were further investigated in vivo in a hamster rabies . This antibody was further analyzed with another human antibody, termed CR57, derived by somatic cell hybridization technique, in combination (Dietzschold et al., 1990), both in in vitro and in vivo models (Goudsmit et al., 2006). In a clinical phase 1 study, the safety and efficacy of the mAb cocktail (named CL184) was tested (Bakker et al., 2008) and subsequently in phase 2 studies. The antibodies were renamed Rafivirumab (CR57) and Foravirumab (CR4098). The in vivo efficacy in a hamster model was demonstrated with several rabies isolates (Franka et al., 2017). The antibody Rabishield, which is not derived by phage display, has been approved in 2018 (Wang et al., 2011;Gogtay et al., 2018).
The emergence of new viral species able to leak from their animal reservoirs and lastly infect the human host has been described several times in the last decades and has led to either local epidemics or global pandemics. In this regard, Henipavirus and Betacoronavirus genera, respectively, are emblematic examples of zoonotic viral diseases that posed serious threats to human health. The generation of monoclonal antibodies through phage display targeting these pathogens has been mainly investigated towards the development of new therapeutic agents, even though in the case of SARS-CoV-2 monoclonal antibodies are of foremost interest also for diagnostic purposes.
Henipaviruses known to be pathogenic to the human host are Hendra virus (HeV) and Nipah virus (NiV), that caused small outbreaks in Australia and Malaysia, respectively, first recognized during 1990 (Weatherman et al., 2018). Bats are the animal reservoir of these viruses and their transmission to human beings is either mediated through amplifying hosts, like pigs or horses, or direct, as documented in NiV Bangladesh 2001 outbreak, through contact between same contaminated food (Halpin et al., 2011;Weatherman et al., 2018). The infection of these negative-stranded RNA viruses causes mainly severe respiratory and neurological symptoms and sequelae with high mortality rate (Wong and Tan, 2012). The initial contact of these virus to host cells is mediated by attachment to glycoprotein G (also known as RBP -Receptor Binding Protein-), that binds the cellular surface proteins ephrinB2 or ephrinB3 (Bonaparte et al., 2005;Negrete et al., 2005;Negrete et al., 2006;Navaratnarajah et al., 2020). The soluble forms of HeV G protein was used to isolate 17 antibodies from a phage display human Fab naïve library (Zhu et al., 2006), to date the only report of phage displayderived anti-NiV/-HeV antibodies. Seven of them were further investigated and given the 80% sequence similarity between G proteins of HeV and NiV, some of them resulted in effective cross-recognition of the two viral species. Fab m102 was further improved through light chain shuffling, VH random mutagenesis and IgG1 conversion demonstrating cross-neutralization of both HeV and NiV and normal serum half-life in ferrets (Zhu et al., 2008). Further testing of this monoclonal antibody demonstrated its efficacy in NiV-infected ferrets (Bossart et al., 2009). For African green monkeys challenged with HeV (Bossart et al., 2011) and NiV (Geisbert et al., 2014), further evidence suggests that the therapeutic window in African green monkeys is different based on the NiV viral strain used (Mire et al., 2016). Its safety was also recently demonstrated in a Phase I first-inhuman clinical trial (Playford et al., 2020), establishing the security criteria necessary to proceed to verify its effectiveness in infected humans in further clinical trials.
The Betacoronavirus genus (single-stranded, positive-sense RNA viruses) has clearly demonstrated its potential as source of worldwide health threats during the beginning of the new millennium, with SARS-CoV and SARS-CoV-2, both emerging from China in late 2002 and late 2019, respectively (Ksiazek et al., 2003;Lu et al., 2020). Furthermore, smaller outbreaks were caused by MERS virus in Saudi Arabia, during 2012 as it was first isolated, and South Korea (2015) (Zaki et al., 2012;Memish et al., 2020). Unlike the latter virus that keeps circulating in camelids (Sabir et al., 2016;Kandeil et al., 2019), SARS-CoV and SARS-CoV-2 originated from bats (Guan et al., 2003;Zhou et al., 2020) and spread worldwide. If the first SARS-CoV-caused pandemic was contained by summer 2003 (Cherry, 2004), COVID-19 pandemics is still lashing the entire world with recurring waves of infection. Phage display-derived antibodies are among the multiple approaches quickly deployed to generate monoclonal antibodies directed to SARS-CoV-2 (Jiang et al., 2020). In this emergence scenario, anti-nucleocapsid protein antibodies have been generated through phage display for diagnostic purposes (Kim et al., 2021). However, the most efforts have been focused on spike protein, and specifically to the receptor binding domain (RBD), to directly block the interaction with the host receptor, aiming to select neutralizing antibodies for novel therapeutics development. Various formats, e.g. IgG, Fab, VHH and bispecifics, from mainly human, either naïve or immune, libraries have been investigated and their neutralization efficiency was mostly studied in vitro, with some mAbs that were also tested in animal models (Li et al., 2020a;Li et al., 2020b;Noy-Porat et al., 2020;Zeng et al., 2020;Bertoglio et al., 2021a;Bertoglio et al., 2021b;Bracken et al., 2021;Ku et al., 2021;Lim et al., 2021). At the moment of writing, the only anti-SARS-CoV-2 phage-display derived antibody known to enter clinical trials is STE90-C11 (COR-101) (ClinicalTrials.gov ID: NCT04674566) (Bertoglio et al., 2021a).
A summary of antibodies generated by phage display against viruses is presented in Table 2.
The majority of the antibodies to these targets are derived from human antibody gene libraries, but libraries derived also from mouse (Fu et al., 1997), chicken (Hu et al., 2012), camel (Saerens et al., 2008) or macaque (Schütte et al., 2009) have been successfully applied to select recombinant antibodies against eukaryotic pathogens.
Detailed examples for the recombinant antibodies against several eukaryotic pathogens derived from phage display are given the following paragraphs.
Allergic bronchopulmonary aspergillosis, saprophytic aspergilloma, chronic necrotizing aspergillosis and the invasive aspergillosis which is highly lethal are caused by Aspergillus fumigatus and are the most important Aspergillus-related diseases (Broderick et al., 1996;Latge, 1999). As one possible consequence of a hematopoietic stem cell transplantation or solid organ transplantation, an invasive aspergillosis can occur in these immunocompromised patients (Rüping et al., 2008). Therefore, an early diagnosis of aspergillosis is crucial for a successful treatment. The development of a panel of human antibodies binding to Crf2 was described by Schütte et al. (2009). The glycosyol hydrolase Crf2 is located in the cell wall of the growing hyphae. Two kind of antibody gene libraries were used: a macaque immune library and the human naïve antibody libraries HAL4/7 . In addition, two different antibody selection strategies were performed: A) a panning on recombinant antigen immobilized on immunostrips. B) a panning in solution using biotinylated antigen. Six antibodies were selected from the human naïve libraries and ten from the macaque immune library. It was observed, that all antibodies generated by panning in solution bound to conformational epitopes and the antibodies generated on directly immobilized Crf2 bound to linear epitopes. Seven antibodies bound to the native antigen Crf2 on growing hyphae of Aspergillus fumigatus shown by histopathological immunofluorescence microscopy. These antibodies didn't show cross reactions with other Aspergillus species or Candida albicans. It was shown that anti-Crf2 antibody MS112-IIB1 was able to neutralize the enzymatic activity of Crf2, but was not able to reduce the fungal growth in a rat model of invasive pulmonary aspergillosis (Chauvin et al., 2019).
Besides the diagnosis in patients, the preventive diagnosis of Aspergillus pathogens in agricultural products is of importance. Therefore, Wang et al. developed a nanobody-polyclonal antibody Sandwich-ELISA (Wang T. et al., 2017). To obtain polyclonal antibodies and for the construction of a camelid immune library, rabbits or alpacas, respectively, were immunized with extracellular antigens and mycelia lysate of A. flavus, a strain with high levels of aflatoxin. Panning resulted in one VHH nanobody against the extracellular antigens and two nanobodies against the mycelia lysate. The developed Sandwich-ELISA is detecting aflatoxin producing strains of the Aspergillus spp. in agricultural products before the fungi can produce high levels of aflatoxin and is therefore suited for early detection.
Malaria is one of the most common mosquito-borne diseases caused by five different species of Plasmodium, P. falciparum, P. vivax, P. malariae, P. ovale and P. knowlesi (Cox-Singh et al., 2008). In humans P. falciparum is responsible for most malariarelated deaths globally, while P. vivax is the most widespread parasite (Popovici and Meńard, 2015). The parasites infect erythrocytes and destroy these cells at the end of the     replication cycle. In tropical and subtropical countries there are approximately 3.2 billion people at risk of malaria (Tolle, 2009;Muerhoff et al., 2010). Because the life cycle of Plasmodium consists of different development stages, different targets of P. falciparum were addressed for antibody generation. Roeffen et al. (2001) targeted Pfs48/45, a surface protein of P. falciparum which is expressed during macrogamete and zygote stages. Pfs48/ 45 is also a potential vaccine candidate because transmissionblocking anti-Pfs48/45 antibodies taken up by the mosquito during the blood meal and block the oocyte development within the mosquito's intestinal tract (Rener et al., 1983;Kumar and Carter, 1984;Vermeulen et al., 1985). To generate anti-Pfs48/45 antibodies, two human immune libraries were constructed from B-lymphocytes derived from P. falciparum patients with transmission-blocking immunity. The antibody selection was performed in immunotubes with immobilized extract of gametocytes. To specifically generate antibodies against Pfs48/45, the scFv-phage were eluted by competition using a cocktail of four rat monoclonal Abs recognizing distinct epitopes on Pfs48/45 (epitopes I, IIb, III and V). Remarkably, one epitope, the epitope III of Pfs48/45, was bound by all of the generated antibodies. Human monoclonal antibodies against the Block 2 region of Plasmodium falciparum merozoite surface protein-1 (PfMSP-1) were isolated by phage display from a scFv immune library of a malaria patient (Sowa et al., 2001). A Fab-immune library was constructed using lymphocytes of thirteen adults with acquired immunity to malaria. Finally, three Fabs (RAM1, 2 and 3) were isolated by panning on the merozite surface protein 3 (MSP-3 194-257 ) (Lundquist et al., 2006). This protein is involved in the binding of heme and the antibodies promote the depletion of Plasmodium by monocytes (Oeuvray et al., 1994;Imam et al., 2014). Therefore, the N-terminal fragment of MSP-3 was also tested in clinical vaccines trials (Audran et al., 2005). Binding of the antibodies to native parasite protein was demonstrated for all three antibodies in immunofluorescence microscopy and immunoblot. The native MSP-3 was bound by RAM1 and RAM2 in fixed and permeabilized cells. The anti-MSP-3 antibodies were produced as IgG1 and also in the IgG3 format and tested in an antibody-dependent cellular inhibition assay (ADCI). The IgG3 format of RAM1 showed an inhibition rate that is comparable to affinity-purified polyclonal anti-MSP-3 211-237 antibodies derived from human immune donors. In contrast, the IgG1 format also showed a lower inhibition in this assay compared to IgG3 (Lundquist et al., 2006). In 2021, Seidel-Greven et al. (2021) presented a monoclonal antibody directed against AMA-1, the apical membrane antigen 1. As anti-AMA-1 antibodies show a high prevalence in immune individuals it is not only an interesting target for therapeutics but also one of the major blood-stage vaccine candidates. A human malaria Fabimmune library was used to select antibodies against a set of three AMA-1 variants. To improve the affinity and functional activity a light chain shuffling was performed. After subcloning to IgG1/ϰ, affinity and competition studies were performed by surface plasmon resonance (SPR) spectroscopy. In addition to an immunofluorescence assay (IFA), the antibodies were tested in an in vitro growth assay. Here, all antibodies were able to inhibit the growth of four different P. falciparum strains including 3D7. Therefore, those antibodies might be the fundament for the development of an antibody cocktail for therapeutic or prophylactic anti-malaria approaches. In 2014, Leow et al. (2014) aimed to find new antibodies with higher thermal stability. As degradation of antibodies due to high temperatures in the endemic countries are affecting the shelf life of malaria rapid diagnostic tests (RDTs). Therefore, they performed a panning with a naïve human scFv library on the histidine-rich protein 2 (PfHRP2) of P. falciparum. PfHRP2 is a biomarker used in most RDTs, as it is abundantly expressed in erythrocytes across all asexual stages and is released during the rupture of the red blood cells. Additionally, it remains detectable in the blood for up to 28 days. Isolated binders were cloned into the IgG format, produced in CHO cells and were compared to a commercial antibody. The selected antibodies showed comparable specificity and sensitivity, with higher thermal stability, making them candidates for a field test in a prototype test kit. To further cope with the degradation issues, Leow et al. (2018) created a shark immune library consisting of VNARs. The variable domain of the shark antibodies is known to have a high thermal stability and therefore is a potential alternative for the RDTs. A wobbegong shark (Orectolobus ornatus) was immunized with different malaria proteins (PfHRP2, PfpLDH and Pvaldolase), used for the construction of a VNAR single domain library and antibodies were selected against all three proteins. Pvaldolase and pLDH are used for diagnosis of non P. falciparum malaria infections and P. vivax infections. Specificity and sensitivity of the anti-PfHRP2 antibodies were demonstrated by Sandwich-ELISA and dot blot analysis. Further engineering could lead to a new generation of RDTs. Table 3 gives an overview of antibodies generated by phage display against eukaryotic pathogens.

ANTIBODIES AGAINST TOXINS
The Center for Disease Control and Prevention (CDC) classifies several bacterial toxins and/or the pathogens producing these toxins as category A or B agent as they are a high risk to national security and public health (see: https://www.niaid.nih.gov/ research/emerging-infectious-diseases-pathogens). Thus, these toxins are a relevant target for development of diagnostics and therapeutics (Froude et al., 2011). Here, antibody phage display technology is a powerful tool to select diagnostic as well as neutralizing antibodies against complete active toxins or certain domains by using human naïve antibody libraries with high diversity (Indrawattana et al., 2010;Amaro et al., 2011;Neri et al., 2011). The aim of such neutralizing antibodies is in most cases to block the interaction of the toxin to its cellular target, by binding to the cell binding domain of the toxin. However, neutralization of the toxicity is also possible by antibodies directed against other domains like the translocation domain or the enzymatic domain Wenzel et al., 2020a). Immunization of animals with toxoids, non-toxic subunits or selected toxin domains allows for the isolation of high-affinity antibodies with immune antibody libraries.
Especially well-suited for the construction of such immune antibody libraries are macaques as there V-genes are very similar to their human counterparts (Chassagne et al., 2004;Pelat et al., 2007;Pelat et al., 2009;Rülker et al., 2012;Miethe et al., 2014;Miethe et al., 2015). Alternatively, also B-cells from vaccinated humans can be used in cases where an approved vaccine is available (Wenzel et al., 2020a). So far, neutralizing antibodies have been successfully selected against several toxins, classified as category A agents, such as botulinum toxins causing botulism from Clostridium botulinum (Amersdorfer et al., 2002;Conway et al., 2010;Hu et al., 2010;Miethe et al., 2014) and anthrax toxins from Bacillus anthracis (Pelat et al., 2007) and also against different category B agents, such as staphylococcal enterotoxin B (Larkin et al., 2010;Chen et al., 2019). We focused on toxins produced by microorganisms and did not include e.g. plant toxins like the potential biowarfare agent ricin  or anti-venom, e.g. anti-snake toxin, antibodies (Laustsen et al., 2018).
One microorganism with the high-risk to be potential used as a bioweapon is the Gram-positive, anaerobic, spore-forming bacterium Clostridium botulinum and other Clostridium spp. that produce the most toxic substances known: eight different serotypes (A-H) of botulinum neurotoxin (BoNT). Five of these serotypes (A, B, E, rarely F, only one case of H) cause human botulism. Human botulism results in flaccid muscle paralysis what requires intensive hospital care and passive immunization (Arnon et al., 2001;Barash and Arnon, 2014). The most toxic substance is serotype A with a LD 50 of about 1 ng/kg by intravenous route, about 10 ng/kg by the pulmonary route and about 1 µg/kg for the oral route (Dembek et al., 2007). BoNTs consist of a 50 kDa light chain and a 100 kDa heavy chain linked by a disulfide bound. Receptor-mediated endocytosis followed by translocation of the light chain across the membrane into the neuronal cytosol is facilitated by two functional domains (Hc and Hn) of the heavy chain. The light chain contains the catalytic domain responsible for the BoNT toxicity. Currently botulism is treated by the application of anti-botulism immunoglobulins, such equine serum or the human serum BabyBIG against type A and B. Yet the availability of BabyBIG is limited while equine serum may cause serum sickness and hypersensitivity. Here, antibody phage display technology has been used to generate toxin-neutralizing antibodies against each serotype. One single domain antibodies (VHH) was selected by phage display of a llama immunized with a cocktail of seven BoNT toxoids (A-F) (Conway et al., 2010;Shriver-Lake et al., 2017). Other VHH have been isolated from immunized alpacas and were produced as VHH-Fc fusion. These antibodies were protective against LD50 BoNT/A and showed protection over 14 days after antibody administration (Godakova et al., 2019). Furthermore, a human antibody gene library could be generated after in vitro immunization with BoNT/A (Hu et al., 2010). Macaque immune libraries were used to isolate neutralizing scFv against the light chains (Lc) and heavy chains (Hc) of BoNT/A, -B and -E with nM affinities against by the EU AntibotABE project (Chahboun et al., 2011;Miethe et al., 2014;Avril et al., 2015;Rasetti-Escargueil et al., 2015). These antibodies were germlinehumanized and the in vivo protection was analyzed in non-lethal and lethal mice challenge experiments. For BoNT/A and BoNT/ B neutralizing anti-Lc and anti-Hc antibody showed only limited protection, but the combination of anti-LC and anti-HC antibodies showed a synergistic effect and 100% protection when using 2.5 µg/mice . Interestingly, no neutralizing antibodies against the Hc of BoNT/E were selected, but the neutralizing anti-Lc antibody alone showed full in vivo protection using only 2.5 ng IgG/mice (Derman et al., 2016).
Another serious infectious disease is anthrax caused by toxins of Bacillus anthracis, an aerobic, Gram-positive, spore-forming bacterium found in soils all over the world. Bacillus anthracis secretes the lethal toxin (LT) consisting of the lethal factor (LF) and the protective antigen (PA); while the edema toxin (ET) is formed by the edema factor (EF) and PA . In the pathogenesis of anthrax only LT is essential (Inglesby et al., 2002). Production and dissemination as aerosol of anthrax toxin is easily possible highlighting its potential as a bioweapon (Oncü et al., 2003). Vaccines are currently based on the subunit PA to induce the generation of neutralizing antibodies. For treatment, commercial monoclonal antibodies against PA, such as Raxibacumab, are commonly used in combination with antibiotics (Kummerfeldt, 2014). Raxibacumab was approved by the FDA in 2012 to treat inhalational anthrax. However, as PA could be mutated and modified to escape neutralizing PA antibodies while still maintaining biological activity, the use of anti-PA antibodies alone presents a security issue. An alternative and additional target presents the LF domain. Such an antibody against the LF domain is 2LF, isolated from an immune library via antibody phage display technology (Pelat et al., 2007). Also a synergistic effect of the combination of an anti-PA together with an anti-LF antibody is to be expected and would improve the efficacy of the therapy. Staphylococcal enterotoxin B (SEB) from Staphylococcus aureus is an example for a bacterial toxin classified as category B agent, that could potentially cause foodborne illness. Staphylococcus aureus produces in total twenty-one types of staphylococcal enterotoxins leading to food poisoning with abdominal cramps, diarrhea and vomiting (Thomas et al., 2006;Ono et al., 2008). The most potent toxin secreted by S. aureus is SEB, a single 28 kDa polypeptide. As a superantigen, it stimulates T cells leading to an overproduction of cytokines, resulting in fever, hypertension and in some cases death. Phage display was employed to identify the epitope of a SEB specific monoclonal antibody using a peptide phage library (Urushibata et al., 2010) and to generate recombinant antibodies from a murine immune library (Singh et al., 2010). Additionally, synthetic human Fab libraries was used to generate human monoclonal antibodies blocking SEB binding to MHC-II (Larkin et al., 2010) and neutralizing SEB in vitro and in vivo . In contrast to the therapeutic use, phage display was applied to generate nanobodies against SEB from camel immune and naive libraries useful as a diagnostic tool. These nanobodies were either directly coupled with alkaline phosphatase in a sandwich-ELISA (Sun T. et al., 2020), or used for western-blot and ELISA in an indirect detection system (Zanganeh et al., 2019).
In the late 1970ies, Clostridioides difficile was discovered as the pathogen causing antibiotic treatment associated diarrhea (CDAD) (Bartlett et al., 1978). Only strains expressing at least Toxin B (TcdB) mostly together with Toxin A (TcdA) result in disease and typical symptoms of CDAD (Bartlett et al., 1978). Only strains expressing at least Toxin B (TcdB), mostly together with Toxin A (TcdA), result in disease and typical symptoms of CDAD (Natarajan et al., 2013). TcdA and TcdB are homologous multidomain single-chain proteins of 308 and 270 kDa, respectively. Both toxins share the same domain architecture. The N-terminus harbors the catalytic domain, a glucosyltransferase (GTD) acting on small Rho-GTPases (Just et al., 1995;Busch et al., 1998), followed by a cysteine protease domain (CPD) catalyzing the proteolytic autoprocessing and upon translocation releasing the GTD into the cytosol (Egerer et al., 2007;Reineke et al., 2007). Pore formation and translocation of the N-terminal portion of the toxin is facilitated by the translocation domain (TLD) (Genisyuerek et al., 2011). The C-terminus of TcdA and TcdB composes repetitive elements that combine long and short repeats in so called CROPs (Combined Repetitive Oligo Peptides). The CROPs are responsible for the first contact of the toxin with the target cells by interaction with carbohydrate structures on the cell surface (von Eichel-Streiber et al., 1992).
Due to increasing CDAD case numbers and epidemic outbreaks (Rupnik et al., 2009;Depestel and Aronoff, 2013), these two toxins were targets of several antibody-generationcampaigns (Babcock et al., 2006;Marozsan et al., 2012). In 2011, Hussack and coworkers were the first to report the successful generation of anti TcdA antibodies by phage display using a llama derived immune library. Four VHH antibodies (A4.2, A5.1, A20.1, and A26.8), showed potent neutralization of the cytopathic effects of TcdA on fibroblast cells in an in vitro assay (Hussack et al., 2011). Similarly, Yang et al. used an alpaca derived immune library to generate six and eleven unique VHH antibodies against TcdA and TcdB, respectively. The single domain antibodies showing the most potent neutralizing activity and highest affinity to nonoverlapping epitopes were combined in a tetravalent bispecific antibody format designated as "ABA". The heterotetramer ABA consists of two copies of the anti TcdB VHH E3 as well as the one copy of each of the anti TcdA VHHs AH3 and AA6 in the following order: AH3/E3/E3/AA6. ABA was shown to bind to TcdA and TcdB simultaneously and one single injection of ABA was enough to reverse fulminant CDAD in mice. In another antibody generation campaign, the naïve scFv libraries HAL9/10 (Kügler et al., 2015) were used to isolate the first fully human, phage display derived antibodies against TcdB. Using various TcdB fragments for the panning process, partially combined with preincubation steps, 31 unique antibodies were generated directed against a broad range of epitopes. The antibodies were tested in an in vitro neutralization assay using Vero cells. Interestingly, the two most potent antibodies bound to the same, formerly unknown, epitope within the glycosyltransferase domain of TcdB .
Diphtheria toxin (DT) is a typical A-B-toxin with an A subunit (catalytic domain) and a B subunit that contains the cell binding domain and the translocation domain. DT is produced by a corynephage b infected pathogenic Corynebacterium stain, mainly Corynebacterium diphtheriae (Murphy, 1996). To date, the only approved therapeutic is a serum from horses immunized with diphtheria toxin, as invented by Emil von Behring and Shibasaburo Kitasato over 100 years ago (von Behring and Kitasato, 1890). Antibodies were generated by phage display from immune libraries of boost vaccinated donors. Neutralizing antibodies against all three domains of DT were selected. The best antibodies showed an in vitro neutralizing antibody with a relative potency of 454 IU/mg at 4xMCDt (minimal cytopathic dose). When using higher toxin concentration, the neutralization efficacy was highly reduced, but when using two or three antibodies in combinations the neutralization was restored. An in vivo potency of 79.4 IU/mg was achieved in an intradermal challenge assay (Wenzel et al., 2020a).
An Overview of recombinant antibodies against toxins are given in Table 4.

CONCLUSION
A large number of recombinant antibodies for application in diagnostics and therapy were already generated by phage display against viral, bacterial and eukaryotic pathogens as well as toxins. Antibody phage display allows the generation of antibodies originating from several species, including human, camel, llama, alpaca, chimpanzee, macaque, pig, mice, chicken or shark. These antibodies are derived from two types of library  The COVID-19 pandemic should teach us to be prepared for the next pandemic, which beyond any doubt will come. We need permanent preparedness platforms which will react on novel pathogens arising worldwide and provide antibodies for diagnostics and as potential therapeutic lead candidates. This preparedness platforms should also have a budget for immediate start of GMP production and clinical phase I/II studies. In the most cases, new pathogens may have only a local impact, but this cannot be assured. Also these antibodies are always valuable tools for research and diagnostics of the pathogens.
We do not build a fire department when the house is already burning, we need to establish it in advance and maintain this service also in times without emergency. We have to think in the same way about infectious diseases and potential pandemics.

AUTHOR CONTRIBUTIONS
All authors listed have made a substantial, direct, and intellectual contribution to the work, and approved it for publication.

FUNDING
We acknowledge funding from the European Community's Seventh Framework Program (FP7/2007(FP7/ -2013 under agreement no. 241832 granted to the AntiBotABE project (http://www.antibotabe.com) and funding received from the European Union's Horizon 2020 research and innovation programme under grant agreement No 101003650, funding from Federal State of Lower Saxony, Niedersächsisches Vorab (VWZN2889) and financial support of MWK Niedersachsen (14-76103-184 CORONA-2/20). We thank CNPq for supporting the scholarship of GMSGM. This review is an updated and revised version of a formerly published review (Kuhn et al., 2016).  (Hsu et al., 2011) ELISA, enzyme linked immunosorbent assays; MS, mass spectrometry; WB, western blot.