β-lactam antibiotics are the second most consumed drugs and the most frequent ones eliciting allergic reactions (Doña et al., 2012). This poses an important clinical problem since, in the most severe cases, allergic reactions may be life-threatening and reduce the therapeutic options against infections. The frequency of allergic reactions associated with each drug varies over time according to consumption patterns (Martin-Serrano, 2018). Hence, currently, amoxicillin (AX) is the antibiotic most frequently eliciting immediate (IgE-mediated) allergic reactions (Fernandez et al., 2017). In addition, AX is nowadays frequently administered in combination with clavulanic acid (CLV), a β-lactam compound which inhibits β-lactamases activity (Torres et al., 2016). As a consequence of the increase in the frequency of AX-CLV administration, selective reactions to CLV are on the rise, reaching 30% of the allergic reactions induced by this combination (Torres et al., 2010; Blanca-Lopez et al., 2015; Fernandez et al., 2017; Salas et al., 2018).

Diagnosis of a suspected reaction after AX-CLV intake is a challenge, as it involves determining which of the two drugs is the responsible one (Torres et al., 2016). Skin test to CLV is only available in some countries and its sensitivity is not optimal. Therefore, drug provocation tests must be performed indirectly using both AX and AX-CLV to establish a diagnosis of CLV allergy (Doña et al., 2019). However, these in vivo tests are contraindicated in severe life threatening reactions. The ideal alternative is the performance of risk-free in vitro tests (Mayorga et al., 2016), although immunoassays for quantifying IgE specific to CLV are not available and have never been reported. Only in vitro tests such as basophil activation tests or histamine release tests have been used for the evaluation of patients (Torres et al., 2010; Pineda et al., 2015; Salas et al., 2018; Barbero et al., 2019). These functional assays use the CLV molecule to evaluate whether the drug induces cellular activation, but they show suboptimal sensitivity (Ariza et al., 2016b; Doña et al., 2017).

In the context of IgE-mediated reactions to β-lactams, drugs behave as haptens as they are assumed to covalently bind to carrier proteins to induce an immunological response (Ariza et al., 2011; Gonzalez-Morena et al., 2016). Both, the resulting structure of the conjugated drug (antigenic determinant) and part of the protein to which it is attached may be involved in the IgE recognition process (Ariza et al., 2015; Ariza et al., 2016a; Martín-Serrano et al., 2016). Developing new approaches for diagnosing CLV allergy and improving the existing ones requires the inclusion of CLV derivative structures recognized by the immune system, whose identification is much more complex compared with other β-lactam drugs. Complex reactivity of CLV and instability after protein conjugation have delayed the isolation and characterisation of the main CLV antigenic determinants (Barbero et al., 2019), and the lack of monoclonal antibodies against CLV has impeded the identification of proteins involved. Therefore, elucidating the mechanisms and structures involved in the immune system activation by CLV is required to advance in diagnosis.

Protein haptenation by CLV is assumed to occur similarly to other β-lactams (Edwards et al., 1988; Barbero et al., 2019), i.e., benzylpenicillin and AX, by nucleophilic opening of the β-lactam ring by protein amino groups from lysine residues (Batchelor et al., 1965; Yvon et al., 1990; Garzon et al., 2014). However, unlike penicillins, which render stable penicilloyl determinants, the resulting acylated structure of CLV is unstable and degrades, leading to small and heterogeneous epitopes with a very low density in the carrier (Edwards et al., 1988; Torres et al., 2016). We have recently reported the identification of a CLV determinant: N-protein, 3-oxopropanamide, which was addressed through a synthetic approach of its analogues and their ability to activate basophils in a higher proportion of patients compared with the native CLV (Barbero et al., 2019). Moreover, the same CLV fragments bound to protein were identified by proteomic approaches (Barbero et al., 2019). Based on its extraordinary ligand-binding capacity, human serum albumin (HSA) has been traditionally considered the main target protein in the haptenation process for β-lactams, and most studies have focused on characterisation of the penicilloyl-HSA conjugates (Jenkins et al., 2009; Martin et al., 2010; Whitaker et al., 2011; Garzon et al., 2014; Meng et al., 2016; Azoury et al., 2018). Regarding CLV, only a couple of recent studies have reported HSA haptenation, identifying stable N-protein, 3-oxopropanamide determinant on lysine residues in in vitro conjugation at physiological pH (Barbero et al., 2019) and in patients treated with the drug (Meng et al., 2016), whereas direct binding of CLV to lysine residues, subsequent degradation products, pyrazine conjugates and cross-linking conjugates were identified at high concentrations in vitro (Meng et al., 2016).

Besides HSA, other proteins can be target of haptenation with β-lactams. Transferrin (Magi et al., 1995; Ariza et al., 2012; Garzon et al., 2014) and immunoglobulins (light and heavy chains) (Ariza et al., 2012; Garzon et al., 2014) have been reported to be target serum proteins to be modified in vitro by ampicillin and/or AX. Besides serum proteins, surface and intracellular proteins have been reported to form antigenic determinants with β-lactams (Binderup and Arrigoni-Martelli, 1979; Watanabe et al., 1986; Watanabe et al., 1987; O´Donnell et al., 1991; Gonzalez-Morena et al., 2016; Sánchez-Gómez et al., 2017) and the intracellular haptenation process has been suggested to be cell type-dependent (Ariza et al., 2014). The identification of novel drugs or reactive metabolite conjugates with proteins is extremely challenging and relevant due to the role of these structures in the activation of the immune system (Labenski et al., 2009). The gold standard to detect and identify drug-protein conjugates is high resolution mass spectrometry (Garzon et al., 2014); nevertheless this approach requires sophisticated equipment that is not available to every laboratory. Alternatively, specific antibodies can be used through immunoassays (ELISA, immunoblotting) for the detection of these conjugates (Martin-Serrano, 2018); however, the lack of specific antibodies for some epitopes and the lack of sensitivity for the detection of low concentrations of conjugates or low degree of protein modification is the major drawback of these detection techniques (Tailor et al., 2016). The use of labeled drugs or reactive metabolites to enrich the drug modified fraction in complex samples and to improve the detection is an option to be considered (Koizumi et al., 2011; Ariza et al., 2014; Gonzalez-Morena et al., 2016). The avidin-biotin interaction provides great affinity and sensitivity, as well as the possibility of coupling modification of proteins by biotinylated compounds with several methods for detection, purification, and imaging (Ariza et al., 2014; Martin-Serrano, 2018). Previous studies have combined biotin labeling with proteomic techniques for the identification of potential protein targets for haptenation (Garzón et al., 2010; Ariza et al., 2014; Sánchez-Gómez et al., 2017; Shan et al., 2017) or modified protein residues (Havelund et al., 2017). The use of biotinylated AX (AX-B) has been shown to increase the detection sensitivity of AX modified serum proteins by immunological methods (Ariza et al., 2014) and has allowed the detection of novel cellular targets (Sánchez-Gómez et al., 2017).

Unlike AX, no antibody specifically targeting CLV is available, which has hampered the immunological detection of protein-CLV conjugates. To overcome this issue, in this work we envisaged an alternative strategy consisting in the design of appropriate biotinylated derivatives of CLV as highly sensitive and straightforward tools to study haptenation and developing methods to identify CLV target proteins. Two different structures, CLV-Biotin (CLV-B) and CLV-tetraethylenglycol-Biotin (CLV-TEG-B), were synthesized as probes for detecting haptenated serum proteins through streptavidin-based amplification technique. Both compounds bear biotin moiety linked through CLV carboxylic group without altering β-lactam group, which is reactive against proteins. The reactivity of biotinylated analogs of CLV and their ability to form conjugates was studied with a simple nitrogen nucleophile, as well as with HSA as protein model for haptenation. In addition, conjugation to a simple model peptide was analyzed, choosing ^182–195HSA peptide due to its previous identification as target of AX (Garzon et al., 2014) and CLV (Barbero et al., 2019). The biotinylated derivative was demonstrated to be a straightforward tool to identify serum proteins target of modification.

CLV is a potent inhibitor of β-lactamase enzyme of increasing interest. Due to the increasingly worrying problem of antibiotic resistance, its consumption is on the rise and, as a consequence, the number of reported selective allergic responses induced by CLV after AX-CLV intake has increased significantly (Fernandez et al., 2017; Montañez et al., 2017). Unfortunately, the diagnostic work-up aiming to identify which drug of the AX-CLV combination is responsible for a reaction is not trivial. Risky in vivo tests show suboptimal sensitivity for CLV due to the fact that testing must be performed with AX-CLV combination, in which CLV is always in a lower ratio. This impedes both discriminating the drug eliciting the reaction and reaching CLV concentrations high enough to trigger reactions (Fernandez et al., 2017). On the other hand, in vitro tests to diagnose IgE-mediated reactions to CLV are limited to the basophil activation test using native CLV, with 40–50% of sensitivity (Mayorga et al., 2019). However, a recent study has reported that the inclusion of synthetic determinants of CLV increased the sensitivity of this assay up to 69%. Interestingly, only determinants with a N-protein, 3-oxopropanamide structure and ability for protein conjugation (spontaneous reactivity against amino groups) were found to induce basophil activation (Barbero et al., 2019). This finding indicates that drug-protein conjugates play a crucial role in the induction of allergies. Therefore, after getting insight into potential determinants responsible for allergies to CLV, herein we investigate potential proteins targets of CLV haptenation. A convenient strategy would be to follow a procedure similar to that reported for immunological detection of AX–protein conjugates with antibodies recognizing the lateral chain of the AX molecule, which successfully allowed the identification of serum proteins coupled to AX (Ariza et al., 2012). However, the lack of a suitable antibody against CLV has prevented the use of immunological detection approaches for our objective. Therefore, we considered label-drug techniques. Despite the fact that label-dependent techniques cannot be used to study conjugate formation in patients and usually do not provide information on the site of modification and/or structure of the conjugates, they are quite useful to visualize conjugates and to identify the modified proteins (Gonzalez-Morena et al., 2016). This is a big step to gain insight into conjugates formation in cases in which there are no specific antibodies against the drug or the conjugation takes place in such a low extension that it hampers detection using label-free approaches.

With the aim of detecting serum proteins that conjugate to CLV, biotin was chosen as a tag due to its extremely high affinity for streptavidin and the lack of impact on the properties of its substrate in most cases (Diamandis and Christopoulos, 1991). The strength of the biotin-streptavidin interaction makes this approach very sensitive. However, in order to retain their proper biological activity, different factors should be taken into account to design tagged molecules, such as the introduction of a biotin moiety into the parent molecule that could result in steric hindrances of its interaction with certain targets (site of protein recognition), the preferential solubility in aqueous media, the presence of intact functional groups that are involved in the protein conjugation, and the stability of resulting biotinylated drug coupled to protein after conjugation. In search of all these aspects is a challenge in the case of CLV, especially due to its complex reactivity and instability after β-lactam opening (Finn et al., 1984; Martin et al., 1989; Baggaley et al., 1997; Brethauer et al., 2008). This probably has hampered any success in the detection of specific IgE antibodies through immunoassays as well as the production of antibodies against CLV.

CLV was successfully labeled with a biotin moiety in the carboxylic group at C3, through two different approaches that keep the β-lactam ring intact. Direct coupling rendered CLV-B, which was not soluble in water, needing proportions of DMSO to solubilize in aqueous media for performing protein incubation experiments. An improved design consisted in CLV-TEG-B, which included an extending TEG linker between CLV and the biotin moiety. Such hydrophilic linker increases the hydrophilicity of the compound, which provides solubility in aqueous media, besides the flexibility and length of the spacer would allow enough distance between CLV molecule and the biotin tag to potentially result in less steric hindrance interactions, both in the protein-CLV conjugation process, and making biotin moiety more available to interaction with streptavidin, leading to higher detection efficiency. Due to these advantages we focused mainly on CLV-TEG-B conjugation studies.

Since CLV degrades at basic pH (Martin et al., 1989), physiological pH conditions were used for the biotinylated derivative conjugation as previously optimized for CLV (Barbero et al., 2019). This differs from optimized AX or other penicillins conjugations which have been reported to be performed in basic pH and which ensure lysine amino group deprotonation and favor nucleophilic attack to β-lactam carbonyl (Ariza et al., 2012; Pajares et al., 2020). Here we observed that CLV is very stable at physiological pH whereas 50% of CLV-TEG-B suffers hydrolysis in its β-lactam at 16 h, which would still permit 50% of CLV-TEG-B to conjugate during incubation time. We also have observed that both CLV and CLV-TEG-B display reactivity toward simple amines, through β-lactam ring opening. In spite of their similar reactivity in presence of nitrogen nucleophiles, they differ in terms of stability subsequent to conjugation. After amide formation the resulting conjugate for CLV breaks down into multiple compounds, whereas the conjugate formed with CLV-TEG-B does not suffer degradation in its biotinylated drug linked (Figure 6). Similar reactivity behavior was desired for both, parent and biotinylated drug, in order the tagged drug can emulate conjugation process occurring with the native drug (Ariza et al., 2014; Gonzalez-Morena et al., 2016). Nevertheless, the stability of the conjugated biotinylated analogue was a critical aspect for keeping the biotinylated moiety attached to the drug, and therefore their usefulness for detecting haptenated proteins.

Because published data identifies HSA as the main target for haptenation with drugs, we used this protein as a model for our study (Ariza et al., 2014). We observed that addition of CLV-TEG-B to HSA takes place in vitro, as deduced by MALDI-TOF MS, and that this modification is concentration-dependent, consistent with recent studies involving CLV (Meng et al., 2016; Barbero et al., 2019). These results are in agreement with the SDS-PAGE analysis of these conjugates, that showed that the extent of biotinylation was concentration-dependent of CLV-TEG-B and CLV-B used during incubation. MALDI-TOF MS analysis provided not a single mass but a molecular weight distribution containing protein species modified to a different extent, including unmodified protein, thus yielding only an estimate of the mass increment that may be related to the degree of haptenation. Moreover, MS comparison between conjugates with CLV-TEG-B and CLV (Barbero et al., 2019), and taken into account that the molecular weight of identified determinants of native CLV is 70 Da, indicates that haptenation taking place with CLV-TEG-B occurs in a lower extent than with CLV, which could be explained by the steric hindrances of the biotinylated moiety. The identification of HSA sites modified by CLV-TEG-B was only carried out for the conjugate formed using 1:10 protein:drug ratio, as performed previously with the native CLV (Barbero et al., 2019), although this concentration is higher than that of the drug in vivo under therapeutic conditions (Nagarajan et al., 2013). In the case of the biotinylated CLV, to get an exact mass increment of the fragment that links to the protein, enrichment of the biotinylated fraction was required to get any result by LC-MS/MS. We searched a 601.2 Da mass increase, related to the incorporation of a 602.2 Da with loss of the hydrogen atom of the HSA lysine fragment, compatible with the haptenation by one CLV-TEG-B thorough β-lactam ring opening without any other fragmentation. However, only a mass increment of 600.2 was found in the ^404–430HSA peptide. An identical mass increment was found in a CLV-TEG-B treated-HSA peptide, used in a controlled experiment as a simpler model, which confirms that CLV-TEG-B adds to the protein with a 600.2 Da mass addition, although a low level of haptenation was obtained and the attached structure could not be elucidated. This ^182–195HSA peptide has been previously reported to be modified in vitro by CLV-derived structures, consisting of N-protein, 3-oxopropanamide identified determinant of CLV (Barbero et al., 2019) or others bearing higher molecular weight (Meng et al., 2016), and by a series of penicillins, such as benzylpenicillin (Meng et al., 2011), AX (Garzon et al., 2014), flucloxacillin (Jenkins et al., 2009), and piperacillin (Whitaker et al., 2011).

Besides lysine residues, the modified ^404–430HSA peptide contains other amino acid nucleophiles as arginine, which could be modified by acylation of the β-lactam. For instance, HSA haptenation with native CLV has been reported to occur also through histidine residues that are modified with the open β-lactam of the complete CLV molecule or with its pyrazine metabolite (Meng et al., 2016). However, modification of residues different from lysine was not observed in our previous study with native CLV (Barbero et al., 2019). In the present study, the modified ^404–430HSA peptide seems to be adducted on the ^410–419HSA sequence (RYTKKVPQVS). From these residues, we assume lysine as the residue with the highest reactivity toward this compound, due to its lower pka and previous literature regarding CLV protein binding (Meng et al., 2016; Barbero et al., 2019); however, at this point, there is no experimental evidence about which aminoacid forms the adduct with CLV-TEG-B. In any case, the modified peptide with CLV-TEG-B differ from those modified by original CLV, attached as 70 Da antigenic determinant, observed at two different residues, Lys 195 and Lys 475 (Barbero et al., 2019). This was somewhat expected, and could be explained by the presence of the biotin moiety that may impose steric impediments for binding to some targets or it may shield part of the molecule (Ariza et al., 2014). In spite of this, both compounds seem to compete for binding to proteins. This is in agreement with a previous study suggesting that AX and its biotinylated derivative could bind to protein common sites (Ariza et al., 2014). Besides competition for common site binding, other interferences in the protein could explain this behavior such as a change in the conformation structure of the protein and a binding/addition to closed sites that involves changes in steric and electronic effects of target lysine residues.

The chance of using biotinylated derivatives of CLV for the identification of modified serum proteins was studied by SDS-PAGE and blot followed by biotin detection since biotinylated CLV could be detected with a high sensibility using streptavidin conjugated with peroxidase. The analysis of HSA modification by biotinylated-CLV showed that protein extent modification was drug concentration-dependent. Importantly, the biotin moiety remained linked to the protein after conjugation to provide detection, and the method showed a great sensitivity for this application. Moreover, a semiquantitative estimation of biotinylation indicated that at 1:100 protein/drug molar ratio, only a 9% of the protein would form conjugate, if only one site were modified. This confirm the remarkable sensitivity of the method and may suggest a low protein modification extension, in agreement with the required enrichment of biotinylated fractions for high resolution MS techniques.

Serum proteins identified as candidate targets of CLV-TEG-B by 2D-electrophoresis and mass spectrometry were HSA, haptoglobin, and immunoglobulin heavy and light chains. In addition, the haptenation of transferrin could be proposed on the bases of its molecular weight and isoelectric point in 2D gels. This finding represents a great progress in our understanding of the mechanisms driving CLV allergy. Furthermore, these CLV derivatives could be useful for complex systems study, in which modified proteins could be purified with streptavidin columns. Previous studies of protein haptenation by AX have allowed to confirm the results obtained with biotinylated drug with those obtained with the native drug. By assaying serum protein conjugates with AX or biotinylated AX by 1D and 2D-electrophoresis and following Western blot using anti-AX antibodies for the AX detection or streptavidin for the biotin detection, it was observed that both compounds bind to the same serum targets (HSA, transferrin and IgE light and heavy chains), with the only difference of a weak binding to haptoglobin, which is undetectable using immunological AX detection, but is detectable with biotinylated AX due to the higher sensitivity of this method (Ariza et al., 2014; Martin-Serrano, 2018). The lack of antibodies against CLV impedes the immunological evaluation of protein-CLV conjugates and, therefore, we could not assess if the proteins haptenated by CLV are the same as the modified by its biotinylated derivative. However, this comparison study in the case of AX (Ariza et al., 2012; Ariza et al., 2014) could be somewhat extrapolated to the case of CLV.

We have set up a model that may shed light into the process of protein haptenation by CLV through the use of highly sensitive approaches, such as labeling with biotinylated analogues, which allow the detection of its target serum proteins. These results strongly suggest that both, CLV and biotinylated CLV, are able to bind proteins through nucleophilic attack of the β-lactam carbonyl group by the protein amino nitrogen, process leading to the opening of the β-lactam ring. Unlike CLV protein conjugation, that promotes the decomposition of clavulanate fragment, the protein conjugates obtained with the CLV-TEG-B are stable enough to allow detection at great sensitivity. The results herein reported are of great interest since, for the first time, serum proteins that may act as carriers in allergic reactions to CLV are identified. Other alternative approaches for these studies are hampered by the complex reactivity of CLV and instability after conjugation. Further structural information on the binding sites on various targets would provide potential antigenic determinants to be used in diagnostic procedures and in studies on the mechanisms of CLV induced allergy.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found in the article/Supplementary Material.

The studies involving human participants were reviewed and approved by Comité de Ética de la Investigación Provincial. The patients/participants provided their written informed consent to participate in this study.

DP-S, EP-I, MJT and MIM conceived and designed the study. NB optimized synthetic approach and performed characterization of compounds. AM-S prepared compound in higher scale and performed reactivity studies. AS, JMM and FJSG performed proteomic experiments. AM-S, AA, DP-S and MIM analyzed all data, prepared figures, and wrote the manuscript, with input from MJT and EP-I.

Work at MJT and MIM laboratory was supported by Instituto de Salud Carlos III (ISCIII) of MICINN (grants cofunded by ERDF: “Una manera de hacer Europa” (PI17/01237, PI18/00095, RETIC ARADYAL RD16/0006/0001 and Euronanomed Program AC19/00082), Miguel Servet I program (CP15/00103) and Sara Borrell program (CD17/00146)), Andalusian Regional Ministry of Health (PI-0179-2014, PE-0172-2018). Work at EP-I laboratory was supported by the Spanish Ministerio de Economía, Industria y Competitividad (CTQ 2016-75870-P), Ministerio de Ciencia y Educación (PID 2019-104293GB-I00), Ministerio de Ciencia e Innovación [Proyectos de I+D+I Programación Conjunta Internacional, EuroNanoMed 2019 (PCI 2019-111825-2)], ISCIII RETIC ARADYAL RD16/0006/0012 and Junta de Andalucía (UMA18-FEDERJA-007). Work at DP-S laboratory was supported by Grants from Agencia Estatal de Investigación, Ministerio de Ciencia e Innovación (MICINN, Spain) and European Regional Development Fund, SAF 2015-68590-R and RTI 2018-097624-B-I00, ISCIII RETIC ARADyAL RD16/0006/0021.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.