All sample collections and studies performed were approved by the Institutional Review Board of The Feinstein Institutes for Medical Research in accordance with the Declaration of Helsinki.

Blood was obtained from patients after obtaining written informed consent. The sex distribution among patients with CLL is ~2:1, male to female, and collection of samples was not influenced by sex. The samples collected and used represent the male to female ratio seen in the disease.

RNA from blood mononuclear cells of CLL patients was extracted and converted into cDNA as described (19). Cloning, expression, and purification of mAbs were performed as reported (62).

Targeted mutagenesis of the IGHV-IGHD-IGHJ and IGKV-IGKJ DNA sequences of CLL IGs was performed by GENEWIZ, Inc. (South Plainfield, NJ).

Purified CLL IGs were sent to the vendor (Invitrogen) to probe ProtoArray^® Human Protein Microarray V 4.0. Details of the assay are available in the Supplementary Material section.

Influenza virus A/TEXAS/1/77/H3N2 was purchased from BiosPacific, CA, USA. Recombinant hemagglutinins were purified as described (63). Viruses or hemagglutinins were coated at 2μg/ml to Polystyrene plates (Nunc, Roskilde, Denmark) at RT for 30min. Plate was saturated with blocking buffer consisting of phosphate-buffered saline (PBS; pH 7.4; Gibco) supplemented with 0.1% Tween 20 (PBS-T), 3% human serum albumin (Calbiochem, CA, USA) and incubated at RT for 2h. IGs were diluted in PBS and incubated on plates at 4°C for overnight. The plates were then washed 3 times with PBS-T and bound IGs were detected with horseradish peroxidase conjugated goat anti-human IgG. After 1h, plates were washed 4 times with PBS-T and developed for 15min with TMP Sure Blue 1-component substrate (KPL, Gaithersburg, MD), stopped with 1M HCl, and absorbance measured at 450nm. HAs from groups 1 and 2 from 6 H3N2 strains (A/Texas/1/77, A/Hong Kong/1/68, A/Alabama/1/8, A/Philippines/2/82/, A/Panama/2007/1999, A/Wyoming/3/03), 4 H1N1 strains (A/Fort Monmouth/1/47, A/Texas/36/91, A/New Caledonia/20/99, A/California/04/09), and one H2N2 strain (A/Japan/305/57) were available and used.

To determine IG binding to double (ds) and single stranded (ss) DNA, the precoated ELISA and reagents of QUANTA Lite™ dsDNA and QUANTA Lite™ ssDNA ELISA kits (INOVA Diagnostics, Inc., San Diego, CA) were used. Calibrators were not used in this ELISA as signal to unit conversion was not required. IGs were tested following the manufacturer’s instructions at indicated concentrations. The signals were used for plotting.

The human Ramos B cell line cells were induced to undergo apoptosis by γ-irradiation (4000-5000R) ~15h before staining. Cells were then incubated with CLL IGs (5 - 50 μg/mL) for 1h at 4°C and binding detected by FITC-conjugated F(ab’)₂ goat anti-human IgG (Southern Biotech, Birmingham, AL). Apoptosis was measured by staining with Annexin V-PE (BD Pharmingen, BD Biosciences, San Jose, CA) as recommended by the vendor. Samples were acquired using a LSRFortessa flow cytometer (Becton Dickinson, San Jose, CA) and analyzed using the FlowJo (LiveTree, San Diego, CA) software.

The ability of SS#4IGs to mediate signal transduction was tested using the TKO cell line (7) that does not express pre-B or mature B-cell IG components due to inactivation of the RAG2 and γ5 genes. The cell is also functionally impaired by making the adaptor molecule, SLP65, which is essential in BCR signal transduction, unavailable until exposed to 4-hydroxytamoxifen (4-OHT). The approaches used for inserting SS#4IGs into TKO cells and the subsequent measurement of Ca⁺⁺ flux are as reported (9, 64). Details of these are available in Supplementary Material .

The protein structures of CLL240 asymmetric antibody dimers (PDB ID: 5DRX) and hemagglutinin of the A/Hong Kong/1/1968 (H3N2) influenza virus (PDB ID: 5T6N) were docked using the ClusPro server (57). Due to the symmetrical nature of hemagglutinin’s homo-trimeric biological assembly, repulsive masking was applied to one monomer out of three to eliminate symmetrically identical protein-protein complexes resulting from docking. After masking was applied to the structure, the three step docking process was initiated: rigid body docking, clustering of 1000 lowest energy structures, and relaxation of the resulting structures using energy minimization. Rigid body docking was performed using PIPER, an algorithm based on Fast Fourier Transform (FFT). The 1000 lowest energy poses were then taken, and clusters were formed for neighboring poses with <9 Angstrom Root-Mean-Square Deviation (RMSD). The representative poses from each cluster were then relaxed via energy minimization and ranked based on cluster size and energy. Next, the top-ranking docking solution which involves both antibody monomers interacting with the hemagglutinin (as suggested by experimental data) was selected as the final model. The resulting model was analyzed and amino acids within 5 Angstroms of each other were defined as residues forming the protein-protein interface. Modeling can be viewed at: https://figshare.com/articles/figure/Structure/28239857?file=51805436.

Virus A/Philippines/2/1982/H3N2 was grown in 8- to 10-day-old embryonated chicken eggs (Charles River Laboratories) at 37°C for 48h. MDCK cells were seeded at a concentration of 2×10⁵ cells/ml were seeded in white polystyrene 96-well plates (Costar Corning). The next day, 60µl of mAbs serially diluted 2-fold starting from 400μg/ml in RPMI 1640 medium (Life Technologies) was incubated with 60µl of virus dilution (1,250 PFU/60µl) for 1h at RT on a shaker. MDCK cells were washed once with 220µl of PBS, and then 100µl of the virus-serum mixture was added to MDCK cells. The mAb-virus mixture and the MDCK cells were incubated overnight at 37°C. For phase contrast microscopy, a poly-D-lysine treated coverslip placed in well was used for MDCK cell growth and neutralization. The coverslip was gently washed with PBS and host cell morphology was captured.

For statistical analysis, GraphPad Prism software was used. The statistical significance (P-value) of two group comparison was calculated using unpaired student T-test. Statistical significance was defined by a P-value of less than 0.05 (2-tailed). Data are presented as mean ± SEM, unless otherwise indicated.

Data produced will be shared to ensure transparency and reproducibility and to avoid duplication of research effort and expense. Flow cytometry files will be shared in.fcs format. All processed ELISA data will be made available in.xls format. Modeling data can be accessed at https://figshare.com/articles/figure/Structure/28239857?file=51805436.

Most CLL IGs are polyreactive (39–45), binding a variety of microbes (35, 46–48) and diverse sets of (auto)antigens, including apoptotic cells (35–37, 49). Contrary to this paradigm, IGs from SS#4 clones react with an unidentified surface membrane molecule on viable human lymphocytes (38).

To determine if SS#4IGs bound to other, unappreciated targets, we probed ProtoArray^® Human Protein Microarrays v4.0, a platform containing 8,222 antigenic targets, with three SS#4IGs (183, 240, 342) and 25 other CLL patient derived rIGs ( Supplementary Table S1 ). Among the latter, 16 belonged to other stereotyped BCR subsets and 9 were not assigned to a subset.

Each SS#4IgG bound selectively and significantly, in a dose dependent manner, to influenza A/Texas/1/77 virus (H3N2) on the array ( Figure 1A ). In striking comparison, none of the other 25 CLL-derived mAbs did. The non-binders included two IGHV4–34 IGs that do not use IGHD5-18/IGHJ6 (CLL141, DO8). (We will refer to these IGHV4-34, non-subset #4 IGs hereafter as “4-34-nSS#4IGs”). Although antibodies induced during (influenza) viral infections often use IGHV1-69 (50, 51), a series of CLL-derived IGs from clones expressing IGHV1–69 belonging to subsets 6, 7A, 7B, and 31, and an IGHV1–69 not belonging to a subset (CLL358), did not bind the virus ( Figure 1A ).

To confirm this reactivity, we performed an ELISA using H3N2 A/Texas/1/77 virus as the target. Consistent with the array findings, each SS#4IG showed dose-dependent binding that fit one site-specific binding curves ( Figure 1B ). In contrast, two 4-34-nSS#4IGs and IGs from stereotyped subsets 6 and 31 (both IGHV1-69), 8 (IGHV4-39) and 2 (IGHV3-21) did not react with virus, even at the highest concentration tested.

The surface glycoprotein hemagglutinin (HA) of influenza viruses is a major determinant of antigenicity and pathogenicity (52, 53). Therefore, we compared binding of SS#4IGs and other CLL-derived IGs to the HA from A/Texas/1/77 (H3N2). SS#4IGs bound well to HA; this reactivity was significantly greater than that for 4-34-nSS#4IGs or non-IGHV4-34IGs ( Figure 1C ).

Next, we tested SS#4IGs for binding to HAs from a series of viral strains: 6 H3N2 strains from the 1968 A/Hong Kong/1 pandemic (54); 4 H1N1 strains from the 2009 “Swine Flu” pandemic (55), and 1 H2N2 strain from the 1957-1958 “Asian Flu” pandemic (54) ( Figure 1D ). Each SS#4IG bound significantly and similarly to all 6 H3s and to 3 of the 4 H1s, the latter with varying effectiveness. Notably, none of the SS#4IGs reacted with H1 from A/California/04/2009 (H1N1) nor with the H2 of A/Japan/305/57 (H2N2). So, SS#4IGs bind to HA subtypes in this order: H3 > H1 >> H2.

CLL IGs can interact with themselves (9, 16), a form of autoantigen binding. The crystal structures of SS#4IgGs 183 and 240 in the homodimerized, self-associated state have been solved (17). In the complex formed by self-association of two SS#4IgGs, amino acids in the FR1 and CH1 of one IG molecule form an epitope which acts as an “Antigen”, and this surface embraces the side chains of HCDR3 of the “Receptor” IG. Distinct interactions include a salt bridge between FR1 Glu^17H of the Antigen IG and Arg^116H in the VH CDR3 of the Receptor IG, an interaction between the side chain of Tyr^117H of the Receptor IG and a hydrophobic pocket formed by the Antigen IG, and hydrogen bonds and van der Waals’ interactions between VH CDR3 residues Tyr^118H, Tyr^119H, and Tyr^120H with the IgG CH1 domain (17). (All amino acid positions listed here follow the ImMunoGeneTics (IMGT) numbering system (56). To compare the IMGT numbering used here with that used in the crystallography studies see Supplementary Figure S3 ).

Based on these crystallographic findings, we tested if binding to influenza virus was through a standard interaction between HA3 and the CDRs of the SS#4IGs or if binding required a homo-dimerized complex. To address this, we also made use of the finding that replacing Glu17 with Ala (E^17HA) on VH FR1 leads to loss of self-association (17), presumably because the variant cannot create the SS#4 Antigen IgG that interacts with the VH CDR3 Receptor IgG. When testing the abilities of the non-self-associating variant (FR1 240E^17HA) and the self-associated wild type (wt) SS#4IgG to react with influenza virus by ELISA ( Figure 2A ), we found that the variant exhibited 77% reduced binding compared to the wt SS#4IgG (P<0.005).

Next, we checked if binding to the surface membranes of viable human B lymphocytes also required a dimerized SS#4IgG. Comparing reactivity of the FR1 240E^17HA variant and the wtSS#4IG with viable B cells by flow cytometry revealed 84% less binding of the non-self-associating variant than the wt SS#4IgG (P<0.01) ( Figure 2B ).

Thus, SS#4IgGs interact with a foreign antigen (influenza virus hemagglutinin) and a self-antigen (viable lymphoid cell membranes) only when in the self-associated, homodimerized state.

Since homo-dimerization was required for foreign and self-antigen binding, we reasoned that the antigen-binding site of the self-associated complex might not be a conventional one, since at least parts of the variable domains of the component IGs would not be available because of homo-dimerization. This was tested virtually by docking the CLL240 IgG homodimer to the HA of A/Hong Kong/1/1968 (H3N2) influenza virus, using the ClusPro protein docking server (57) ( Figure 3A ).

Using this model, we first tried to determine the structural reasons for the differences in bindings to HA3, HA1, and HA2 ( Supplementary Figure S1 ). Many residues within the HA3 stem region that participate in the interaction with the dimerized SS#4IGs are conserved in the HA from strain A/Fort Monmouth/1/47 (H1N1). The residues differing between the H3 and H2 subtypes are highlighted in red ( Supplementary Figure S1 ). The differences are most substantial in the interface region, which is consistent with CLL240 binding to HA1 as well as HA3, but not HA2 in ELISA ( Figure 1D ).

Next we focused on understanding the specific modeled interactions between the self-associated SS#4IgG and HA3 ( Figure 3A ). Notably, the homo-dimerized IgG (green and orange) interacts preferentially with the stem region of the HA (slate blue), not the globular head region (dark blue), the latter being more diverse among strains than the former. Three distinct portions of the homodimer are involved ( Figure 3B ): an extensive area of the H chain of the Antigen IG (AG IG HC); a targeted area of the variable domain of the IG H chain of the Receptor IG (REC IG HC), and a single interaction of constant region of H chain of the Receptor IG.

Concentrating on each portion individually ( Figure 3C ) indicates that, for the first portion, there are 23 contact residues in the constant region of the Antigen IG that likely interact with the hemagglutinin ( Figure 3C , box 1; Table 1 , columns 1 and 2). Since there are multiple amino acids in the HA with which these could partner, specific interactions are not listed in the Table. (Crystal study numbering is used here and in rest of this section. For corresponding IMGT numbering see Supplementary Figure S3 ). For the second portion, there are 18 amino acids of the Receptor IG that can be assigned specific complimentary residues in the HA; 14 of these are in the variable domain of the H chain and 4 in the variable domain of the L chain. Notably, of the interacting residues in these variable regions, for the H chain 6 are in FR1, 3 in FR3, and 5 are in CDR1; for the L chain, 2 interacting residues are in FR2 and 2 in CDR2 ( Figure 3C , box 2; Table 1 , column 3). Thus, 11 interactive residues are in FR and 7 in CDRs. Finally, for the third portion of the homodimerized SS#4IgG that interacts with HA3, there is only one interacting amino acid, which is the Lys at position 214 of the Receptor IG ( Figure 3C , box 3; Table 1 , column 3). Note that for homo-dimerization to occur, the Lys at position 214 is required, albeit in the Antigen (not Receptor) IG.

Thus, the interaction between influenza virus HA3 and the non-conventional binding site is through amino acids in the constant region of the Antigen IG. For the Receptor IG, residues in the constant region and the FR contribute more than the CDRs.

The functional capacity of such a non-conventional, conformational binding site depends on its stability and hence the duration that the Antigen-Receptor complex is intact. To test this, we made use of the finding that homo-dimerization of CLL IGs induces autonomous BCR signaling in an experimental system in which IGs are inserted into the membrane of a reporter B cell line (TKO) that only permits BCR signaling when exposed to 4-OHT (8, 9).

Using this system, we determined the ability of soluble, non-associable components of SS#4IgGs or of the IGs in normal human serum to disrupt the homo-dimerized complex and thereby abrogate autonomous signaling. Specifically, we co-transfected TKO cells with constructs containing membrane versions of wt SS#4IgG and Igκ, and then exposed these cells to the soluble, “Antigen-only” variant (R^116HA/Y^117HA) or to the “Receptor-only” variant (E^17HA) of the 3 SS#4IgGs, 183, 240, 342.

As expected (9, 17), cells expressing wt SS#4IgGs on the TKO surface membrane initiated significant Ca⁺⁺ flux upon 4-OHT exposure ( Figure 4A ). In contrast, SS#4 expressed as IgM could not, confirming that isotype switching is necessary for self-association ( Figure 4A ). However, neither the 240 variants with intact “Antigen” nor intact “Receptor” sites could disrupt the Antigen-Receptor complex nor could they interrupt Ca⁺⁺ flux ( Figure 4B ). Similarly, pooled, affinity-purified human IgG at concentrations of 25– 50 μg/mL ( Figure 4B ) and pooled human IGs up to 10 mg/mL (not shown) did not do so. In contrast, surface SS#4IgM, which contains intact “Receptor” but lacks “Antigen” due to absence of K241 in the human µ H chain, bound to soluble “Antigen-only” (R^116HA/Y^117HA) but not to “Receptor-only” variant (E^17HA) ( Figure 4C ).

This functional estimation of “Antigen-Receptor” affinity suggests that the interaction between the two components is relatively strong and prevents disruption of the homo-dimerized IG by microenvironmental antigens. Analyses of the affinity of the two components in a soluble setting (17) corroborate our membrane estimates of the strength of the IG-IG interaction and support our conclusion. Moreover, the finding that routine screening of SS#4IgG binding reveals interactions with influenza virus HA which requires homo-dimerization supports that the self-associated state is relatively stable and long lived.

Influence of IG isotype. Next, we set out to track the impact of “conventional/classical” (non-dimerized) and the “non-conventional” (homo-dimerized) antigen-binding on the development of a normal B cell bearing the SS#4IG IGHV-D-J rearrangement. Since SS#4 B cells always express IgG, we first determined if SS#4IgMs reacted with the same foreign (influenza virus HA) and self (viable B-cell membranes) antigens that the SS#4IgGs did. After creating recombinant variants of two CLL SS#4IGs (183 and 240) bearing the wt IGHV-D-J rearrangement linked to either Cµ or to Cγ and also linked with the corresponding wt IGKV-IGKJ rearrangement, binding to influenza virus and to viable and apoptotic human B cells was analyzed.

Whereas the wt Cγ-linked SS#4IGs exhibited substantial binding to influenza virus A/Texas 1/77 (H3N2) in ELISA, binding of the Cµ-linked versions of 183 and 240 IGs was significantly reduced (reduction percentage IgM vs. IgG- IG183: 62.9%, P < 0.001; IG240: 68.4%, P < 0.001; Figure 4D ). Similar significant differences were found when we tested reactivity of the wt Cγ-linked and the Cµ-linked versions with viable human B cells by flow cytometry ( Figure 4E, left : reduction percentage IgM vs. IgG - IG240: 91.5%, P<0.01; IG342: 86.5%, P<0.01; IG183: 45.5%, P<0.01; Supplementary Figure S2A ).

Remarkably, the SS#4IgM interacted significantly with single stranded DNA (ssDNA; Figure 4F, left ) and double stranded DNA (dsDNA; Figure 4F, right ), whereas as expected the IgG-linked SS#4IGs did not (P<0.0001). Similarly, we detected binding of the SS#4IgMs, but not the SS#4IgGs, to apoptotic/dead cells ( Figure 4E, right : Figure 4E, right : IG240: P<0.01; IG342: P<0.05; IG183: P=0.256; ( Figure 4E, right ; Supplementary Figure S2B ).

Thus, the SS#4IgGs and their Cµ-linked variants bind distinct types of antigens, using discrete antigen binding domains. So, the emerging naïve SS#4 IgM-expressing B cell was not driven to survive and expand by antigens that drove the mature SS#4 IgG-expressing B cell. Also, these interactions were mediated by the classical/conventional/non-dimerized binding site. The antigens involved here were at least those typical autoantigens with which most CLL clones bindInfluences of somatic mutations shared by multiple SS#4IgGs on influenza virus binding, autoreactivity, and B-cell maturation. To discriminate the effects of somatic mutations on the development and function of normal B cells bearing a SS#4 IGHV-DJ, we focused on the mutations in the IGHV-D-J rearrangement. First, we reverted all the IGHV-D-J mutations in IG240 to the germline sequence (GL-HC) ( Figure 5A ) and paired this with the wt IGKV-J. This significantly reduced reactivity with influenza virus compared to wt SS#4IgG 240 (P<0.001) ( Figure 5B ). Since GL-HC IgG corresponds to the IGHV-D-J germline precursor of mAb240 that was present at the time of isotype switching from an IgM⁺ to an IgG⁺ B cell, this finding indicates that somatic mutations occurring somewhere in the germline H chain variable domain enhanced binding to the virus.

Because alterations of the amino acids within the IGHV are not affected by the assembly of the IGHV-D-J and hence result from antigenic stimulation and potential selection, we next focused on somatic mutations in IGHV4-34. Specifically, all mutations in IGHV4-34 (the SS#4 shared mutation as well as the mutations unique to patient 240) were reverted to the germline sequences (GL-IGHV4-34), while retaining the remainder of the gene in the wt configuration ( Figure 5A ). Surprisingly, reverting all the somatic mutations present in IGHV4–34 led to a 28-fold increase in binding affinity compared to that of wt IG240 (P<0.05) ( Figure 5B ). Thus, the development of these mutations had a negative effect on HA binding and mutations developed outside of IGHV4–34 significantly enhanced binding to the virus.

To directly test the potential influence of the recurrent, stereotyped change Gly to Glu/Asp at position 28, we next solely reverted this residue to germline (E^28HG), keeping the remainder of the molecule in the wt IgG state. Remarkably, this single reversion significantly reduced viral binding compared to wt IG240 (P<0.001) ( Figure 5B ). So, the acquisition of the stereotyped G28E mutation, which contributes to some extent to self-association, contributed positively to viral reactivity during B-cell development ( Figure 5B ).

Finally, we asked if the mutations tested above were involved in the lack of autoreactivity of SS#4IgGs. SS#4IgG 240 bound negligibly to DNA ( Figure 5C ). Unexpectedly, reverting the entire H chain variable domain to the germline sequence (GL-HC/wt-LC) did not lead to DNA binding. Notably, when the same germline sequence was linked to Cµ (Cμ-GL-HC), the resulting mutant showed significant binding to DNA. When only the mutations in IGHV4–34 were reverted to their germline amino acids (GL-IGHV4-34), DNA reactivity was significantly increased compared to wt SS#4IgG240 (P<0.001). Thus, the mutations in IGHV4-34, which reduced foreign antigen binding, also suppressed the autoreactivity of SS#4IgG. Moreover, reverting only the stereotyped, negatively charged amino acid E, which could repel DNA, back to G (E^28HG), led to substantial increased binding to ssDNA (P<0.001) ( Figure 5C ).

Thus, the stereotyped G28E mutation and the mutations unique to IG240 appears to have been selected to inhibit autoantigen binding. In addition, the G28E mutation, which is shared by many SS#4 patients, increased influenza virus HA interactions. So, collectively these mutations seem to have allowed the SS#4 B cell to avoid negative selection and permit clonal survival and growth, and the stereotyped mutation could have promoted grow induced via the unconventional antigen-binding site.

In CLL, (auto)antigen binding to the BCR is critical for the survival and expansion of the leukemic clone. In the healthy state, secretion of the SS#4 IgG by a normal B cell could be of physiologic importance. Therefore, we tested if the distinct structural features of SS#4IGs might be advantageous to the host as an effector molecule. To do so, we analyzed the capacity of SS#4IgGs and its progenitors (Cµ-linked and somatic mutations reverted) to neutralize influenza virus ( Figure 6A ).

Specifically, we tested if the various IGs could protect Madin-Darby canine kidney (MDCK) cells from infection by A/Philippines/2/82 (H3N2) virus. MDCK cells, incubated with wtSS#4IGs (IgG183 and IgG240) and then exposed to the virus, were protected from influenza virus infection in a dose-dependent manner (~80% inhibition at the highest tested concentration) ( Figure 6A ). Cells incubated with the Cμ-linked SS#4 (IgM183) did not inhibit infection, even at highest concentration; this lack of reactivity was equivalent to that of the negative control. The Cγ-linked, IGHV-D-J germline precursor of mAb240 (GL-HC) showed minimal inhibition, as expected from the binding curves ( Figure 5B ). In contrast, the progenitor Cγ-linked GL-IGHV4–34 showed enhanced viral inhibition compared to wt SS#4 IG240 and was the best inhibitor among all tested variants, consistent with the ELISA binding data ( Figure 5B ).

As illustrated by phase-contrast light microscopy ( Figure 6B ), MDCK cells exposed to virus after incubation with SS#4IgG183 retained a healthy morphology and an intact, attached monolayer, resembling cells grown in the absence of virus. This differs for cells exposed to influenza virus after incubation with SS#4IgM183 in that cell numbers were remarkably reduced and the monodispersed and spherical shapes of healthy cells was lost. The latter was very similar to those incubated with the control, virus non-reactive IgG282 ( Figure 6B ).

Thus, these in vivo actions corroborate the in vitro studies above. In addition, they suggest that soluble homo-dimerized SS#4IgGs can protect against infection of cells by influenza virus. However, since the efficiency of this protection is considerably lower than that of a therapeutic mAb, the SS#4IgGs most likely function as B-cell receptors for foreign and self-antigens rather than as soluble effector molecules.