Dual staining of human B cells using intact bacteria identifies 29 novel antibodies against K. pneumoniae
To identify human antibodies that target K. pneumoniae, we selected human B cells that recognized fluorescently labelled K. pneumoniae via their B cell receptor (BCR). Click chemistry was used to couple fluorescent dyes to lipopolysaccharide (LPS), a conserved cell envelope component of Gram-negative bacteria (Fig. 1a, S1a). This was achieved by growing bacteria in the presence of azide-carrying KDO which incorporates into the LPS and provides a handle for ‘clicking’ to cyclooctyne-labeled fluorophores. STORM microscopy (Fig. 1a) and flow cytometry (Fig. S1b) demonstrated successful surface labeling of K. pneumoniae. Next, we examined whether fluorescently labelled K. pneumoniae could be used to identify K. pneumoniae specific B cells, focusing on B cells from healthy individuals. B cells were stained by incubating them with a mix of two fluorescently labelled bacteria preparations, and B cells that bound bacteria were detected via flow cytometry. We identified single-positive B cells that bound either ATTO-488 or Cy5 labelled K. pneumoniae (Fig. 1b,c). In addition, approximately 0.05% of the B cells bound both ATTO-488 as Cy5 labeled bacteria. Adding an excess of unlabeled K. pneumoniae competed primarily with binding of fluorescent K. pneumoniae in the double-positive B cell population (Fig. S1d). This suggests that the dual staining approach helped to enrich for bacterium-specific B cells. Importantly, the double-positive B cell population was strongly enriched for memory B cells (CD27+), whereas no enrichment was observed within the single-positive populations (Fig. 1b). Together, this suggests that dual staining of B cells with fluorescently labelled intact bacteria allows for the enrichment of K. pneumoniae-specific memory B cells.
To proof that the dual staining selects for K. pneumoniae-specific B cells, we single-cell sorted double-positive IgG expressing B cells (Fig. 1c; Q2). Next, the VH/VL regions of the sorted B cells were amplified via RT-PCR and cloned into IgG1 expression vectors. Recombinant IgG1 expression in EXPI293F was performed and expression supernatants were collected. To screen for K. pneumoniae-specific antibodies, the supernatants were incubated with K. pneumoniae, and antibody binding was assessed by flow cytometry (Fig. 1d, S1e). Separate B cell selection rounds were performed with two clinical K. pneumoniae isolates: KpnO2 (O2; KL110) and KpnO1 (O1; KL114), which represent the two most dominant O-types among multidrug-resistant K. pneumoniae17. For KpnO2, a total of 368 double-positive B cells were isolated by single-cell sorting. Screening of expression supernatants revealed 34 positive clones (9.2% success rate) (Fig. 1d). Out of the 34 clones, 12 monoclonal antibodies with a unique CDR3 bound to KpnO2 after large-scale expression and IgG1 purification (Fig. 1e,j, S2a). All 12 clones had undergone somatic hypermutations in both the VH and VL, indicating affinity maturation (Fig. 1f, S2a). For KpnO1, a total of 273 double-positive B cells were sorted, yielding 35 expression supernatants that contained antibodies against KpnO1 (12.8% success rate) (Fig. 1g). Out of these, 17 unique monoclonal antibodies bound to KpnO1 as purified IgG1s (Fig. 1h,j, S2b). Finally, we determined that all clones had undergone somatic hypermutation (Fig. 1i). In conclusion, we developed a B cell staining approach to specifically select B cells recognizing entire K. pneumoniae bacteria. Using this method, we identified 29 unique human antibodies recognizing two clinically relevant K. pneumoniae strains.
Antibodies targeting O2-antigen but not the capsule drive complement activation on KpnO2
Next, we investigated whether the newly identified anti-K. pneumoniae antibodies could induce Fc-mediated C3b deposition, a central step in the complement cascade. We first focused on the antibodies recognizing the KpnO2 strain. Purified antibodies were incubated with KpnO2 bacteria and normal human serum (NHS) as a complement source, and surface deposition of C3b was quantified by flow cytometry. Although most antibodies against KpnO2 triggered C3b deposition in a dose-dependent manner, for three antibodies (UKpn1, UKpn3, UKpn4, grey lines) we did not observe any C3b deposition, even at the highest tested antibody concentration (Fig. 2a). Four antibodies could very potently induce C3b deposition to levels that were > 2-fold over background at 0.1 µg/ml; these ‘active’ antibodies were colored in dark blue (Fig. 2a,b).
To understand these differences in complement activation, we studied dose-dependent antibody binding to KpnO2 (Fig. 2c). Intriguingly, we observed that the binding of antibodies to the surface of Klebsiella did not directly correlate with their capacity to activate complement (Fig. 2b,d). For example, we observed that two of the inactive antibodies could potently bind to the Klebsiella surface, up to levels that are equal to most active antibodies (Fig. 2b,d). These data indicate that the binding of IgG1 to the surface of K. pneumoniae does not automatically lead to Fc-mediated activation of the complement system.
We wondered whether the functional differences between the antibodies could be explained by differences in their target specificity. To identify the bacterial surface targets of the antibodies, we generated a transposon library in KpnO2 using barcoded transposons28,29 We hypothesized that antibody binding would be lost for mutants where the transposon disrupts a gene that is involved in surface expression of the antibody target. The transposon library was exposed to either a complement-activating (UKpn2) or a non-activating (UKpn1) antibody, and transposon mutants that were no longer bound by the antibodies were isolated using single-cell sorting of live bacterial cells. We found that approximately 0.5-1% of the transposon mutants were no longer recognized by the antibodies (Fig. S3a). Barcode sequencing of the sorted UKpn1-negative mutants revealed that sixteen unique transposons had inserted in six different genes involved in capsule synthesis (Fig. 2e). Furthermore, sequencing of UKpn2-negative mutants identified twenty unique transposon insertions in five different genes in the rfb locus involved in biosynthesis of the O-antigen polysaccharide 26 (Fig. 2f). To confirm these results, we used the lambda-red recombination system to specifically delete the wbaP (capsule glycosyltransferase) and wbbO (encoding the O-antigen synthesis galactosyltransferase) genes. Indeed, we observed that KpnO2 ∆wbaP and ∆wbbO deletion mutants were no longer bound by UKpn1 and UKpn2, respectively (Fig. 2g). We analyzed the binding of the other KpnO2-specific antibodies to the ∆wbaP and ∆wbbO mutants. This revealed that all complement activating antibodies recognized the O-antigen, whereas the inactive antibodies all target the capsule (Fig. 2g). Additional binding experiments revealed that anti-capsule antibodies recognize another K. pneumoniae strain containing the KL110 capsule locus (Fig. S3b). Furthermore, the anti-O-antigen antibodies specifically recognized the O2-antigen, except for antibody UKpn14, which was also cross-reacted with an OL104-antigen expressing strain (Fig. S3c).
Various O2-antigen subtypes can be expressed, based on genetic variation within the O locus, as well as the presence of accessory genes. The most common are the O2a and O2afg16. O2afg strains have, in addition to the O2a locus, a gmlABC gene cluster involved in modifying the O2-antigen16,30. While all O2-specific antibodies bound O2a strains, we found that UKpn6 and UKpn7 did not recognize O2afg strains (Fig. S3d). This suggests that the epitopes of UKpn6 and UKpn7 in the O2a antigen are shielded by the O2afg modification. To further specify the difference in the epitopes of UKpn2 and UKpn6, we exposed the KpnO2 transposon library to UKpn2 and UKpn6, and sorted UKpn2-negative mutants that were still bound by UKpn6 (Fig. S3e). Transposon sequencing revealed five unique transposon insertions in the gene orf731 that is located directly downstream of the rfb operon (Fig. S3f). This suggests that the galactosyltransferase encoded by orf7 has a role in O2a biosynthesis that is crucial for binding of UKpn2, but not for UKpn6.
In summary, while the identified antibodies bind to the surface of K. pneumoniae, this alone does not result in complement activation. Instead, the specific antigenic target seems to play a crucial role in initiating this process.
Antibodies that target the O1-antigen differ in their potential to drive complement activation
For antibodies targeting the KpnO1 strain, we also assessed the correlation between antibody binding and complement activity. Measuring antibody-dependent C3b deposition revealed that 13 of the 17 antibodies activated the complement system on KpnO1 (Fig. 3a,b). Some antibodies efficiently induce C3b deposition at 0.01 µg/ml, whereas others required a 100-fold higher concentration. Similar to what was observed for KpnO2, we observed that antibody binding (Fig. 3c,d) to KpnO1 did not always correlate with the antibody’s capacity to induce C3b deposition. For example, antibodies UKpn80 and UKpn82 showed potent binding to the surface of KpnO1 while hardly inducing complement activation (Fig. 3b,d). Next, we aimed to identify the target of the anti-KpnO1 antibodies. Since the anti-KpnO2 antibodies were all directed against either the O-antigen or the capsule polysaccharide, capsule (∆wbaP) and O-antigen (∆wbbO) deficient mutants of KpnO1 were generated. We observed that all antibodies still bound to KpnO1 mutants lacking the wbaP capsule gene (Fig. S4). In contrast, antibody binding to the ∆wbbO mutant was abrogated, suggesting that all antibodies against KpnO1 targeted the O-antigen (Fig. 3e). Genes involved in the synthesis of the O-antigen are encoded by same locus in O1-antigen and O2-antigen strains16. However, O1-antigen strains express two additional genes (wbbYZ), which are involved in attaching D-galactan-II (O1) on top of a D-galactan-I (O2) segment at the distal part of the O-antigen to form a O1-antigen cap (Fig. 3f)16. Upon specific deletion wbbY, we observed that 16 out of 17 antibodies could no longer bind to KpnO1 (Fig. 3e). Strikingly, we found that UKpn69 could still bind to KpnO1 ∆wbbY strain, although with reduced capacity, suggesting suggests that UKpn69 has some affinity for D-galactan-I as well (Fig. 3e). This shows that all the identified anti-KpnO1 antibodies targeted the O1-antigen polysaccharide on KpnO1. We next analyzed whether anti-KpnO1 antibodies could also recognize other O1 strains than KpnO1. Although most antibodies showed a preference for KpnO1, four antibodies (UKpn69, UKpn72, UKpn76 and UKpn77) could cross-react with all tested O1 strains (Fig. 3g). Interestingly, these four broadly reactive antibodies all belong to the group of potent complement-activing antibodies. Analysis of antibody binding to bacterial lysates showed that only the broadly reactive antibodies could recognize the O1-antigen outside the context of the bacterial cell membrane (Fig. 3h). This suggests that these broadly reactive antibodies recognize different epitopes within the O1-antigen than the other KpnO1 recognizing antibodies. Altogether, these data show that identified O1-antigen specific antibodies differ in their capacity to drive complement activation.
Hexamerization-enhancing mutations strongly improve complement activation by anti-capsule antibodies
To understand differences between our complement-active and -inactive antibodies we studied the role of IgG hexamer formation. We hypothesized that inactive antibodies have a poor capacity to form Fc-dependent hexamers. To study this, we introduced the E430G mutation in the IgG1-Fc domain that can improve Fc-mediated hexamer formation2. For KpnO2, we focused on two ‘active’ antibodies targeting the O2-antigen (UKpn2 and UKpn6) and two ‘inactive’ antibodies targeting the capsule (UKpn1, UKpn3). For the antibodies targeting the O2-antigen, we observed that hexamerization-enhancing mutations could not enhance C3b deposition (Fig. 4a). In contrast, the same mutations had a very strong impact on the anti-capsule antibodies. Whereas wild-type anti-capsule antibodies do not activate complement at the tested concentrations (< 3 µg/ml), introducing the E430G mutation allowed the antibodies to induce C3b deposition at concentrations of 0.3 µg/ml and higher. (Fig. 4b). For Kpn01, we found that hexamerization-enhancing mutations showed a moderate enhancement of the active UKpn72 antibody, only at lower antibody concentrations (Fig. 4c). For the inactive UKpn62 and UKpn82 we observed that the E430G mutation did not improve C3b deposition. Altogether, these data demonstrate poor IgG clustering by anti-capsule antibodies which can be enhanced by hexamerization-enhancing mutations.
Antibody-driven C3b deposition results in effective phagocytosis and killing of Kpn by human neutrophils
Next, we studied whether antibodies driving C3b deposition on K. pneumoniae also induce downstream bacterial killing. Since deposited C3b molecules can bind complement receptors on phagocytic cells, we tested whether the antibodies stimulate phagocytosis of GFP-labeled K. pneumoniae by human neutrophils. As expected, anti-O2- and anti-O1-antibodies that stimulated C3b deposition also induced phagocytosis in a dose-dependent manner (Fig. 5a,b). In line with the C3b deposition results, phagocytosis of KpnO2 was only observed for the anti-O2-antigen antibodies UKpn2 and UKpn6, but not for the capsule targeting antibodies, UKpn1 and UKpn3 (Fig. 5a). As expected, introduction of the hexamer-enhancing mutations in UKpn1 and UKpn3 strongly enhanced phagocytosis (Fig. S5a). For KpnO1, we found that only the complement activating antibody UKpn72 induced phagocytic uptake of KpnO1 by human neutrophils (Fig. 5b). To test if anti-K. pneumoniae antibodies induced neutrophil-mediated killing, we analyzed the number of surviving bacteria and found that the KpnO2 recognizing antibodies UKpn2 and 6 also reduced the number viable bacteria in the presence of neutrophils (Fig. 5c). For KpnO1, UKpn72 reduced bacterial viability in the presence of neutrophils (Fig. 5d). Next to phagocytosis, C3b deposition could also lead to downstream formation of MAC pores. However, when KpnO2 and KpnO1 were incubated with antibodies and serum in the absence of neutrophils, we observed no decrease in bacterial viability (Fig. S5b,c). This indicates that the membrane attack complex plays no role in complement-mediated killing of these strains but that neutrophils are required. To summarize, we showed that antibody-dependent complement deposition leads to phagocytic uptake and killing of K. pneumoniae.
Anti-capsular antibodies act synergistically in binding and complement activation
Finally, we were curious whether combining different antibodies targeting the same bacterium would affect complement activation on K. pneumoniae. We focused on KpnO2, as we have antibodies targeting two different antigens on the same strain (O2-antigen and the capsule). Combining an anti-O2-antigen antibody with an anti-capsule antibody had no effect on complement activation (Fig. S6a). Similarly, mixing two antibodies recognizing the O2 antigen (UKpn2 and UKpn6) did not affect their capacity to drive C3b deposition (Fig. 6a). To our surprise, combining two anti-capsule antibodies (UKpn1 and UKpn3) had a strong impact on complement activation. Whereas the individual anti-capsule antibodies induced limited C3b deposition, mixing the two antibodies potently boosted complement activation with a more than 30-fold increase in efficiency (Fig. 6b). Similar results were obtained using another Klebsiella strain that expresses the same capsule type (Fig. S6b). Enhanced C3b deposition by the anti-capsule antibody mixture also led to increased phagocytosis (Fig. 6c) and killing (Fig. 6d) of Klebsiella by human neutrophils. Thus, we show that mixing two inactive anti-capsule antibodies strongly potentiates their capacity to drive complement activation.
To understand why anti-capsule antibodies can act together, we analyzed whether the antibodies influence each other’s binding to the surface. For this purpose, anti-capsule antibodies and anti-O2-antigen antibodies were fluorescently labelled and KpnO2 was incubated with a fixed concentration of fluorescent antibody in combination with increasing concentrations of unlabeled antibody. For anti-O2 antibodies, we observed that UKpn2 and UKpn6 compete, indicating they bind a similar epitope, or epitopes that are in proximity of each other (Fig. S6c). Interestingly, mixing the two anti-capsule antibodies did not lead to competition, but showed synergistic binding, as binding of fluorescent UKpn3 was strongly enhanced in the presence of UKpn1 (Fig. 6e). Vice versa, unlabeled UKpn3 also enhanced binding of fluorescent UKpn1 (Fig S6d).
We wondered whether the synergistic binding could be a result of antibodies interacting with each other via the IgG Fc-tails. We hypothesized that Fc-mediated hexamer formation could potentially lead to a more stable binding of the anti-capsule antibodies to the surface. To test this, we used the SpA-B protein, which binds to IgG-Fc domains and thereby prevents IgG1 antibodies to form hexamers32. To first verify the activity of SpA-B in the context of Klebsiella, we showed that SpA-B fully blocks C3b-depositon induced by an anti-O2 antibody (UKpn2, Fig. 6f). This agrees with previous studies, showing that hexamer formation is important for antibody-dependent complement activation on bacteria32. Intriguingly, SpA-B did not effectively reduce C3b deposition induced by the mixture of two anti-capsular antibodies (Fig. 6g). This indicates that the process of Fc:Fc interactions between neighboring antibodies and ultimately hexamerization does not completely drive the synergistic activity of these anti-capsule antibodies.
Altogether, these results show that anti-capsule antibodies can jointly activate complement because they strengthen each other’s binding to the surface via a mechanism that is mostly independent of hexamer formation.