The structural basis for the selective antagonism of soluble TNF-alpha by shark variable new antigen receptors Authors: Obinna C Ubah, Eric W Lake, Kendahl L Ott, Stella Priyanka, Sasha Celeda, Jayden L West, Gihan S Gunaratne, Ke Shi, Nicholas H Moeller, Andrew J Porter, Hideki Aihara, David Kosoff, Aaron M LeBeau, Caroline J Barelle Published: 2025 ABSTRACT AbstractThe pro-inflammatory cytokine tumor necrosis factor-alpha (TNF-α) is synthesized as transmembrane TNF-α (tmTNF-α) where proteolytic processing releases soluble TNF-α (sTNF-α). tmTNF-α can act as either a ligand by activating TNF receptors, or a receptor that transmits reverse (outside-to-inside) signalling after binding to native receptors. All TNF-α therapies bind tmTNF-α and induce reverse signalling which can result in immunosuppression leading to infection. We present crystal structures of two anti-TNF-α Variable New Antigen Receptors (VNARs) in complex with sTNF-α via two distinct epitopes. The VNAR-D1 recognizes an epitope that selectively engages sTNF-α while VNAR-C4 binds an epitope that partially overlaps with other biologic therapies. In activated CD4+ T cells, our VNARs do not induce reverse signalling in contrast to currently available therapies. Our findings suggest that neutralization through a unique mechanism may lead to anti-TNF-α agents with an improved safety profile that will benefit high-risk patients.Subject terms: X-ray crystallography, Antibody fragment therapy, Flow cytometry Introduction Tumor necrosis factor-alpha (TNF-α) is a clinically relevant cytokine that regulates activities such as cell growth, carcinogenesis, inflammation, viral replication, immune surveillance, and autoimmunity1. As a proinflammatory cytokine, TNF-α is a principal mediator of autoimmunity and inflammatory diseases, including Crohn’s disease (CD), ulcerative colitis (UC), and rheumatoid arthritis (RA). It exerts its function by binding two structurally distinct membrane receptors, TNFR1 and TNFR2. TNFR1-mediated signaling is predominantly associated with inflammation and cell death, whereas TNFR2 mainly regulates T-cell migration and tissue regeneration2,3. Blocking the interaction of TNF-α with its two receptors has been exploited as a therapeutic mechanism for treating inflammatory and autoimmune diseases3–5. Soluble TNF-α (sTNF-α) is generated through the proteolytic cleavage of its precursor form, transmembrane TNF-α (tmTNF-α), by the enzyme TNF-α converting enzyme (TACE or ADAM-17). tmTNF-α is expressed on most cell types including activated macrophages and lymphocytes, and, contrary to sTNF-α, which acts at sites distal from TNF-α-producing cells, tmTNF-α functions through cell-to-cell contact6–8. tmTNF-α has mechanistic duplexity; it can function as a ligand that binds to TNFR1/TNFR2 and as a receptor that conducts reverse (outside-to-inside) intracellular signaling on tmTNF-α expressing cells. It is therefore believed that tmTNF-α plays a critical role in local inflammation9–13. Anti-TNF-α biologics have been successfully implemented in the clinic for the treatment of chronic inflammatory diseases such as RA, UC, CD, psoriatic arthritis (PsA), psoriasis (PsO), and ankylosing spondylitis (AS). However, these clinical successes have been accompanied by a two-to-four-fold increased risk of granulomatous infectious diseases, including active tuberculosis (TB), reactivation of latent TB, invasive fungal infections, and bacterial and viral infections by a range of opportunistic pathogens11,14,15. There are also differences in the clinical efficacy of anti-TNF-α therapies for certain inflammatory disease indications. For instance, Infliximab and Etanercept are equally effective in RA, but only Infliximab is effective in CD. One explanation for this is that Infliximab forms a highly stable complex with tmTNF-α, resulting in the induction of reverse signaling, while Etanercept does not. Functional characteristics resulting from interaction with tmTNF-α or Fcγ receptor-expressing cells can result in degradation and inflammation of tissue, contributing to the overall clinical and adverse events documented for each anti-TNF-α biologic (Fig. 1). Many adverse events, including immunogenicity, can be explained by the binding of anti-TNF-α biologics to tmTNF-α, and the triggering of reverse signaling that induces cell-cycle arrest, suppression of T-cell proliferation, and apoptosis of tmTNF-α bearing cells16–22. It is likely that the induction of tmTNF-α mediated reverse signaling by itself does not explain all adverse events with anti-TNF-α agents, but manipulation of this characteristic of anti-TNF therapies provides a route to rationally design safer and more broadly effective anti-TNF-α agents with minimal or abrogated tmTNF-α interaction. Such an approach may reduce overall side-effects from TNF-α therapies and provide, for the first time, access to anti-TNF-α drugs for patient groups that are currently unable to benefit from this drug class (e.g. patients with latent bacterial or viral infections, patients at risk of malignancies). Anti-inflammatory mechanisms utilized by anti-TNF-α agents. Broadly grouped into the classical neutralization of soluble TNF-α (sTNF-α) and transmembrane TNF-α (tmTNF-α), Reverse signaling mediated by the neutralization of membrane-bound TNF-α (tmTNF-α), and Fc-mediated apoptosis, which is only triggered by those anti-TNF-α agents capable of binding both Fcγ receptors (FcγRs) and complements. Created in BioRender. Ott, K. (2025) https://BioRender.com/sd9wef3. Here, we present co-crystallization data of anti-TNF-α variable new antigen receptors (VNARs) complexed with TNF-α that provides evidence of VNARs engaging epitopes that are unique among the currently landscape of existing anti-TNF-α biologics. VNARs are small (~11 kDa), naturally occurring, independent binding domains isolated from the adaptive immune repertoire of sharks. One defining characteristic of VNARs is their unusual targeting properties that allow the VNAR paratope to access and bind epitopes inaccessible to conventional biologics23–29. We have previously reported the isolation of high affinity, super-potent, clinically applicable anti-TNF-α VNARs (D1 and C4) that are specific for human TNF-α and possess low or no immunogenicity in murine models25,30,31. The recessed epitope on sTNF-α recognized by VNAR-D1 may be the ultimate discriminator between sTNF-α and tmTNF-α, consequently allowing for the rational design of safer and broadly efficacious anti-TNF-α therapies. Results VNAR D1 recognizes an epitope only accessible on soluble TNF-α The VNARs C4 and D1 were previously identified from the direct immunization of a nurse shark with sTNF-α. Both specifically bound human sTNF-α with nanomolar affinities at two separate epitopes as determined by BLI (Supplementary Fig. 1). To determine their structures, C4 and D1 were separately complexed with sTNF-α (aa residues 77–233) and purified by size-exclusion chromatography. Both complexes were subjected to crystallization and yielded crystals. The structures were determined by molecular replacement phasing and refined to 1.99 and 3.43 Å resolution for C4 (Fig. 2a) and D1 (Fig. 2b), respectively. In the structures, VNAR monomers bind to each protomer of the TNF-α trimer in a 1:1 stoichiometry (Fig. 2a, b). Analysis of the crystal structures revealed that VNAR-C4 binds an epitope located near the C-terminal end of the TNF-α trimer similar to the clinically relevant antibody therapeutics Adalimumab and Infliximab (Fig. 2c and Supplementary Fig. 2). When compared with the contact binding surface of Adalimumab, which spans across multiple protomers and covers an area of 2540 Å2, the binding surface of VNAR-C4 covers only 822 Å2 on the surface of one protomer (Fig. 2c)32. This highlights the ability of VNARs to achieve comparable binding affinities to conventional full-length immunoglobulins, but across a much smaller molecular footprint. Interestingly, VNAR-D1 binds a potentially unique epitope, not represented by any known antibodies, including single-chain camelid variable-heavy-heavy (VHH) domains (Supplementary Fig. 2). This epitope is in the groove formed between two adjacent protomers and is enveloped by the N-terminal tail of soluble TNF-α (Fig. 2b). The binding surface of VNAR-D1 is also relatively small covering only 968 Å2 across the sTNF-α cleft, with high-affinity engagement facilitated by a finger-like binding loop that reaches into grooves on the target proteins. X-ray structures of VNARs expose unique TNF-α epitopes. The a and b show the crystal structure of soluble TNF-α trimer (grayscale) bound by multiple monomers of VNAR-C4 (a, red) and VNAR-D1 (b, green). Insets show a second view to highlight the 1:1 stoichiometry of TNF- α trimers to VNARs. c The figure shows a grayscale surface representation of the TNF- α trimer with the epitopes for VNAR-C4 (red) and Adalimumab (blue)(PDB ID: 3WD5). Epitope overlap is colored light blue. Inset shows cartoon representations of each biologic bound to TNF- α as in Supplementary Fig. 2. The structures of VNARs C4 and D1 each represent a facet of shark-derived therapeutics that apply to multiple disease contexts. Our studies with VNARs have discovered that VNARs either 1) possess similar mechanisms of action to existing biologics but achieved with a much smaller binding footprint, or 2) reveal/expose completely unique mechanisms of action that make VNARs viable therapeutic candidates for patient groups resistant to or are at risk by existing therapeutic strategies24. Here, the epitope of VNAR-C4 and its mechanism of action overlap with that of the TNF receptors 1 and 2 (TNFR1 & TNFR2), directly competing with receptor binding (Fig. 3a). By transposing the C4: TNF-α complex onto the structure of TNFR2 bound to TNF-α, it shows C4 completely obscures the interaction of TNFR2 with TNF-α. This receptor-blocking action of C4 also holds true when the C4: TNF-α complex is transposed onto the structure of TNFR1 bound to TNF-α, with the caveat that the TNFR1: TNF-α structure used in this comparison was taken from a complex that was solved with an asymmetric form of the TNF-α trimer. VNAR mechanism of action against soluble TNF-α. a Overlaid crystal structures of TNFR1 (blue) and TNFR2 (yellow) shown as cartoons bound to a surface representation of the TNF- α trimer (grayscale) with an illustration indicating the receptor position relative to the cell surface. A surface representation of VNAR-C4 (red) has been superimposed onto the structures showing the overlap with the TNFRs. b A surface representation of VNAR-D1 (green) has been superimposed onto the crystal structures showing the overlap with the TNFRs. c The figure shows a model of the tmTNF-α trimer tethered to a representation of the cell membrane with individual domains of TNF-α colored as follows: intracellular N-terminal domains = cyan, transmembrane helices = brown, spanning residues = black, RSSSRTPS-motif = red, receptor binding domain = gray. d The panel shows a schematic of the steps of free sTNF-α (grayscale) being bound by free VNAR-D1 monomers with motion lines indicating the conformational flexibility of the RSSSRTPS-motif (red). A red/green gradient arrow indicates that conformational change of the RSSSRTPS-motif as it envelopes VNAR-D1 in the bound state with an arrow to an inset showing the full transition. VNAR-D1 also overlaps with the TNFR1 and TNFR2 binding site leading to receptor blocking activity that is synonymous with that of VNAR-C4 (Fig. 3b). However, the VNAR-D1 epitope suggests a unique mechanism of action (condition 2 above) that allows selective targeting of the soluble form of TNF-α. To understand how VNAR-D1 binds sTNF-α, it is important to understand the process by which sTNF-α is generated from tmTNF-α33. After translation, full-length monomeric TNF-α is trafficked to the cell surface where it quickly forms a trimer, even at low nanomolar concentrations. The structure of full-length tmTNF-α has never been solved but has been predicted by AlphaFold as monomer34,35. By duplicating this form and rearranging the transmembrane tether, a model of trimeric tmTNF-α was constructed to visualize how tmTNF-α might look on the cell surface. AlphaFold predicts that a helix resides in the span of the residues preceding the sTNF-α receptor binding domain (Fig. 3c), and that this same helix is also predicted by the Chou and Fasman secondary structure prediction (CFSSP) server when the sequence was uploaded. Directly in the middle of the predicted helix lies the cleavage boundary (residues A76 and V77, Fig. 3c, red/black) where ADAM-17 canonically cleaves tmTNF-α into sTNF-α, exposing residues 77-85 of the N-terminal tail (RSSSRTPS-motif). Once sTNF-α is released from its cell surface tether, it can freely move and reorientate, allowing the RSSSRTPS-motif to fold around VNAR-D1 when it binds its epitope (Fig. 3d). In contrast, when sTNF-α is tethered to the transmembrane portion of TNF-α, the residues of the RSSSRTPS-motif are wound into a helix, blocking the conformational change necessary to envelop the VNAR-D1 binding site. The RSSSRTPS-motif is essential for the high affinity interaction of VNAR-D1 with sTNF-α, and this requirement is also evident in the crystal structures. The B-factor values show that the electron density of the RSSSTRPS-motif is well ordered when stabilized by VNAR-D1, and even though an identical sTNF-α construct (residues 77-233) was used for the VNAR-C4 complex, the RSSSRTPS-motif is unresolved in that structure. In fact, the RSSSRTPS-motif is not visible in any published structures of TNF-α, demonstrating that it can only be visualized when fully stabilized. To confirm that this motif was required for recognition, we screened VNAR-D1 against a pro-TNF-α (residues 57-233) and did not detect any binding by BLI. Interestingly, VNAR-C4 also did not bind pro-TNF-α, suggesting that some degree of a conformational change must occur for its epitope to be fully formed (Supplementary Fig. 3). This suggests a completely new mechanism of action of TNF-α neutralization and one that may offer new therapeutic possibilities. Analysis of the VNAR CDR3 engagement with the “unique” epitopes on the surface of TNF-α The presentation of binding loops on a VNAR allows them to pack tightly into their epitope, with their protruding CDRs reaching into structural features that would be difficult to access by larger antibodies with planar binding faces formed from two separate binding domains. Covering a total area of 970 Å2 on the TNF-α trimer, the VNAR-D1 epitope is unique among existing anti-TNF-α biologics (Supplementary Fig. 2). This is because the VNAR-D1 epitope is in the cleft of two protomers, with one protomer interface accounting for 634 Å2 and another accounting for the remaining 336 Å2. This interface is repeated in triplicate on the whole TNF-α trimer, allowing up to three VNAR-D1 molecules to engage. The “fingertip” of the VNAR-D1 CDR3, which consists of residues Gly90, Leu91, and Ala92, reaches deep into the cleft formed between two protomers of the TNF-α trimer and is enveloped by the residues of the RSSSRTPS-motif (Fig. 4a). The hydrophobic fingertip engages with hydrophobic residues in the cleft, primarily with Val89, Ala114, and Leu112 of the first protomer of TNF-α (Fig. 4a, white) as well as Leu131 of the second protomer (Fig. 4b, yellow). The sidechain carbon segments of Thr83 and Ser85 of the first protomer and Gln203 of the second protomer are also oriented towards the hydrophobic finger, helping to stabilize the packing of VNAR-D1 into the cleft (Fig. 4a, b, yellow and white). Of note, Leu91 of VNAR-D1 contributes 1.57 kcal/mol of solvation free energy to this interaction, which accounts for approximately one order of magnitude of the overall affinity. The N-terminal Arg78 of TNF-α at the end of the RSSSRTPS-motif donates two hydrogen bonds (h-bonds) to Thr31 and Ser32 of VNAR-D1 (Fig. 4a), and its carbonyl receives an h-bond from Lys64 of VNAR-D1 (Fig. 4b). A substantial interaction accounting for a total of −2.03 kcal/mol of solvation energy is the result of Arg107 of the first TNF-α protomer coordinating a h-bond network with the sidechains of Glu86, Glu93, and Asp95 of VNAR-D1. Also, on the CDR3 finger, Tyr89 receives a h-bond at its carbonyl group from Ser81 of the RSSRTPS-motif of protomer 1, and it also forms an h-bond with Glu129 of the second protomer. In addition, and to reinforce the strength of this interaction, the VNAR-D1 CDR3 backbone amine of Gly90 donates an h-bond to Gln201 of the second TNF-α protomer. Outside of CDR3, Arg25 of VNAR-D1 can donate an h-bond to Thr165 (Fig. 4b). The carbonyl of Ser162 of the second protomer receives multiple h-bonds from both the backbone amine and the sidechain of VNAR-D1 Ser27 (Fig. 4b). VNAR-D1 His26, depending on its sidechain orientation, can donate a h-bond to Glu203 or receive an h-bond from Ser162 of the second TNF-α protomer (Fig. 4b). In total, the majority of the VNAR-D1 interaction occurs with CDR3, but its small profile allows multiple interactions across its surface. VNAR:TNF-α interactions are mediated primarily through the CDR3. The center of the figure shows an interaction map as a surface representation of the TNF-α trimer with protomer 1 colored white, protomer 2 colored yellow, and protomer 3 colored pale cyan. VNARs are represented as cartoon structures with VNAR-D1 colored green and VNAR-C4 colored red. Square boxes indicate the binding interfaces with letter labels corresponding to the approximate location of each of the zoomed in figures. a Cartoon representation of a view of the VNAR-D1:TNF-α interface with sticks showing interacting residues with dashed lines representing hydrogen bonds and spheres representing hydrophobic interactions. Residue labels are colored to match their structures which are colored as in the central interaction map. b A view of the VNAR-D1interaction surface rotated 180 degrees from panel A. Residue labels are colored to match their structure, which is colored as in the central interaction map. c, d A view of the VNAR-C4:TNF-α interaction with sticks representing interacting residues and dashed lines representing hydrogen bonds. Residue labels are colored to match their structures, which are colored as in the central interaction map. e A view of the VNAR-C4:TNF-α interaction shown as in c and d with added transparent spheres over residues with hydrophobic interactions. Water molecules are shown as solid red spheres with dashed lines indicating hydrogen bonds with residues. Residue labels are colored to match their structures, which are colored as in the central interaction map. In contrast, VNAR-C4 binds an epitope that partially overlaps with most clinically approved biologics targeting TNF-α (Fig. S1). The epitope covers a single protomer that is accessible on all three protomers, allowing VNAR-C4 to bind in triplicate. The surface area of the interaction covers 766 Å2. Here again, the interaction interface of VNAR-C4 highlights the unique finger-like nature of the CDR3 loop, with residues 88–93 engaging with a single protomer of TNF-α (Fig. 4c). The carbonyl of Trp88 receives an h-bond from the sidechain of TNF-α Gln67; TNF-α Gln67 receives another h-bond from VNAR-C4 Arg93, whose sidechain is stabilized by an internal h-bond with the carbonyl of Gln90. Gln67 of TNF-α also donates an h-bond to the carbonyl of VNAR-C4 Thr89, forming a backbone-backbone interaction common among VNARs24. This same backbone-backbone interaction also takes place between the amine of VNAR Asp91 and the carbonyl of TNF-α Asp140. The sidechain of VNAR-C4 Asp91 receives an h-bond from TNF-α Lys65, and, together with VNAR-C4 Trp92, donates an h-bond to the backbone carbonyl of TNF-α Leu142 (Fig. 4c). And lastly, the sidechain of VNAR-C4 Gln90 interacts with the sidechain of TNF-α Asp140, which in turn is stabilized by an intra-residue h-bond with its backbone amine. While CDR3 residues are the main drivers of binding, additional residues outside this loop region support the final interaction. VNAR-C4 residues Tyr37 and Lys84 each form an h-bond with the sidechain of TNF-α Glu107, and the sidechain of VNAR-C4 Glu61 accepts an h-bond from the backbone amine of TNF-α His73 (Fig. 4d). VNAR-C4 Thr58 is the only residue in the paratope that interacts with a different protomer in the epitope: Glu110 of the third protomer of TNF-α. Lys51 of VNAR-C4 Lys51 donates two intermolecular h-bonds to the sidechains of TNF-α Ser71 and Glu10, respectively, but it also donates a stabilizing intramolecular h-bond to VNAR-C4 Glu57, forming a nearly perfect tetrahedral bonding geometry. A second but similar h-bonding geometry occurs from the side chain of TNF-α Ser71, which accepts two h-bonds from VNAR-C4 Thr34 and Lys51, while donating its hydrogen to the carbonyl of VNAR-C4 Thr34. Two additional interactions further stabilize this dual-tetrahedral h-bond network with the backbone amine of TNF-α Ser71 donating an h-bond to the carbonyl of VNAR-C4 Ser32, and VNAR-C4 Glu57 forms an intramolecular interaction with Thr34 (Fig. 4d). Though the VNAR-C4:TNF-α interaction is primarily h-bond mediated, there is a subtle hydrophobic attraction between VNAR-C4 residues Leu31 and Ile59 with the methyl group of TNF-alpha Thr72 (Fig. 4e). This epitope:paratope space also contains a water-mediated hydrogen bond network between VNAR-C4 residues Ser32, Ser33, Tyr35, Trp88, and Thr89 with TNF-α Cys69, Thr72, Arg138, and Tyr141 (Fig. 4e). This water network is quite stable due to the favorable h-bond angles and the fact that the spatial location of the water molecules is conserved across all three interfaces in the crystal structure. Functional characterization of the tetravalent human Fc fusion VNAR constructs The in vitro neutralizing potency and TNF-α binding affinities of the monomeric VNAR D1 and C4 and a corresponding biparatopic tetravalent human Fc fusion proteins (referred to as ELN22-108) consisting of 2x D1 and 2x C4, have been previously reported (Fig. 5a)25,31. As the epitopes of VNAR D1 and C4 were confirmed through crystallography as being distinct, two monoparatopic tetravalent Fc fusion protein, ELN22-129 (4x D1) and ELN22-130 (4x C4), were assessed for binding to sTNF-α (Fig. 5a). Given the loss of biparatopic binding, the monoparatopic anti-TNF-α VNARs showed a predicted but acceptable loss in affinity and neutralizing potency for sTNF-α compared to the biparatopic construct. We predicted this loss in affinity from previous work with the monoparatopic and biparatopic dimers and trimers of the anti-TNF-α VNARs25. Biparatopic ELN22-108 recorded affinities ranging between 9 and 74 pM depending on whether TNF-α is captured on the biosensor via biotin-streptavidin or presented in solution in the assay wells in 96-well plates (Table 1). The monoparatopic ELN22-129 and ELN22-130 constructs had affinities in the range of 10–230 pM, also depending on the antigen presentation approach. Affinities independently reported for Etanercept, Infliximab, and Adalimumab vary from group to group and are dependent on measuring instrument and assay method. However, affinities of these anti-TNF-α mAbs range between 23 pM and 1.15 nM19,32. Functional characterization of the tetravalent human Fc fusion VNAR constructs. a Depiction of the mono- and biparatopic tetravalent VNAR constructs composed of VNARs D1 (green) and C4 (red). b ELISA data comparing the ability of the mono- and biparatopic tetravalent constructs to inhibit the binding of sTNF-α to TNFR1 and TNFR2. The data shown are the mean ± SD. c Efficacy assessment in standard L929 cell-based assay. Neutralization of 0.3 ng/mL hTNF-α-induced cytotoxicity in an actinomycin-D primed fibrosarcoma cell line (L929). 2V-huFc is an isotype VNAR negative control. The reported results are the mean ± SEM (n = 3, with 6 replicates. d Flow cytometry experiment documenting the ability of the tetravalent constructs and Adalimumab to bind tmTNF-αon the surface of LPS-activated THP-1 treated with and without the weak ADAM-17 inhibitor Thymoquinone (TQ). Table 1Summary of Octet R8 binding kinetics of anti-TNF-α mono- and biparatopic tetravalent VNAR constructsAnti-TNF-α IDELN22-108 (AHC2)ELN22-108 (SAX)ELN22-129 (AHC2)ELN22-129 (SAX)ELN22-130 (AHC2)ELN22-130 (SAX)KD (pM)74.59.21981392309.8Ka (1/Ms)2.4 × 1051.09 × 1062.67 × 1054.87 × 1051.77 × 1053.54 × 105Kdis (1/s)1.7 × 10−71 × 10−55.3 × 10−56.75 × 10−54.07 × 10−53.48 × 10−6Full R20.98840.99610.99890.98920.99930.9586 Summary of Octet R8 binding kinetics of anti-TNF-α mono- and biparatopic tetravalent VNAR constructs We next compared the ability of the mono- and biparatopic tetravalent constructs to inhibit the binding of sTNF-α to TNFR1 and TNFR2 by ELISA (Fig. 5b). Biparatopic ELN22-108 was vastly more effective at inhibiting sTNF-α binding to both receptors compared to the monoparatopic constructs. Additionally, ELN22-108 had superior neutralizing potency (EC50) of ~3.9 pM compared to the 184 pM and 8.9 pM for monoparatopic ELN22-129 and ELN22-130, respectively, in the gold standard L929 cytotoxicity assay (Fig. 5c). There is a clear affinity and in vitro potency improvement achieved by targeting two distinct epitopes on TNF-α. The affinity obtained for the monoparatopic tetravalent VNARs is comparable with the values observed for the clinically approved anti-TNF-α biologics, Adalimumab, and Infliximab32,36. We then determined the ability of the tetravalent constructs to bind tmTNF-α on the surface of activated monocytes. The expression of tmTNF-α was stimulated in THP-1 cells by LPS followed by treatment with the ADAM-17 inhibitor Thymoquinone (TQ) to enrich for populations of uncleaved tmTNF-α (Fig. 5d)(Supplementary Figs. 4 and 5). Both monoparatopic constructs showed minimal binding to the THP-1 cells compared to Adalimumab. Binding to THP-1 cells was observed for ELN22-108. One explanation for the increased binding of ELN22-108 compared to the other constructs is that TQ is a poor inhibitor of ADAM-17 and some cleaved TNF-α protomer exposing the D1 epitope possibly exists within the TNF-α homotrimer on the cell surface. Membrane TNF-α-mediated expression of E-selectin (CD62E) and IFN-γ in PHA activated CD4+ T cells tmTNF-α mediated induction of the adhesion molecule, E-selectin (CD62E) in PHA activated CD4+ T cells was measured as a marker of reverse signaling. Unstimulated CD4+ T cells from healthy individuals did not express tmTNF-α, whereas treatment with 0.1 µg/mL PHA for 12 h resulted in the expression of tmTNF-α. The treatment with acidified buffer to dissociate the cell-surface sTNF-α did not alter the intensity of the staining following binding of anti-TNF-α biologics to tmTNF. MAB210 (R&D systems), a commercially available anti-human TNF-α binding and neutralizing antibody, and Infliximab showed consistent and significantly superior surface tmTNF-α binding compared to the anti-TNF-α VNARs at 16, 24, and 48 h timepoints (Fig. 6a–c). The binding intensity of MAB210 and Infliximab peaked at 24 h (Fig. 6b), with significantly higher binding at this timepoint compared to earlier and later timepoints. This peak likely reflects a dynamic balance of processes such as tmTNF-α internalization and recycling, antibody-induced upregulation of tmTNF-α expression, and/or avidity effects from the multivalent binding of IgG domains. By 48 h, surface-bound antibody levels significantly decreased (Fig. 6c), consistent with previously published data confirming internalization of tmTNF-α bound anti-TNF-α antibodies37–39. The observed surface binding kinetics also correlates with high levels of E-selectin induced at 24 h with both MAB210 and Infliximab (Fig. 6d). Similar anti-TNF-α: tmTNF-α binding and E-selectin expression profiles were observed when the samples were analyzed by flowcytometry (Supplementary Figs. 6 and 7). E-selectin was not expressed when anti-TNF-α test articles were immobilized on the assay plates prior to the addition of 0.1 µg/mL PHA stimulated CD4+ T cells (data not shown). The effect of these anti-TNF-α biologics on the induction of IFN-γ via tmTNF-α in PHA activated CD4+ T cells was then assessed. Previous reports have shown that IFN-γ is produced in response to an increase in monocyte apoptosis following tmTNF-α stimulation by anti-TNF-α antibodies33,40. At the 24 h timepoint, MAB210 and Infliximab at 100 nM induced ~ 10–15 ng/mL IFN-γ (Fig. 6e, f). The anti-TNF-α VNARs were unable to induce any detectable levels of IFN-γ in these PHA stimulated CD4+ T cells (Fig. 6f). VNAR-2V-huFc is an isotype control isolated from a naïve library and has no inherent binding property against any known target antigen41,42. tmTNF-α dependent binding and downstream induction by anti-TNF-α agents on PHA activated normal human CD4+ T cells. Fluorescence quantified surface binding of anti-TNF-α agents to tmTNF-α on PHA stimulated human CD4+ T cells at a 16 h, b 24 h, and c 48 h timepoints, respectively. At 24 and 48 h and 100 nM: Infliximab vs ELN22-108, ELN22-129 and ELN22-130 (***p < 0.0001). MAB210 and Infliximab showed significant surface binding differences at 16 and 48 h with 100 nM dosing. d Fluorescence quantified E-selection expression at 24 h timepoint. At 24 h, Infliximab vs ELN22-108, ELN22-129, and ELN22-130 (** p < 0.0035). e, F IFN-γ secretion induced by the binding of anti-TNF-α agents to tmTNF-α on PHA stimulated CD4 + T cells was quantified using human IFN-γ DuoSet Sandwich ELISA kit. The results are the mean ± SD (n = 4) with 2 replicates per experiment. Statistics were performed using a two-way ANOVA with Dunnett’s multiple comparisons post hoc test. Discussion All five anti-TNF-α biologics (Adalimumab, Infliximab, Golimumab, Etanercept, and Certolizumab) currently in clinical use effectively neutralize TNF-α; however, increasing evidence demonstrates that all five drugs have differences in their therapeutic efficacy and the occurrence of unwanted side effects. One possible explanation for at least some of these observed and well-documented differences is how well they recognize tmTNF-α or, more importantly, the degree to which they can or cannot trigger full tmTNF-α reverse-signaling in tmTNF-α expressing cells8,10,11,43. There is growing evidence implicating binding and induction of tmTNF-α reverse signaling and associated tmTNF-α enabling mechanisms (i.e., CDC, ADCC) in the cytotoxic lysing and clearing of tmTNF-α-bearing immune-surveillance T cells, antigen-presenting cells, germinal centres, and memory B cells43–45. These therapeutically unhelpful tmTNF-α mediated mechanisms can result in patients receiving treatment experiencing severe and life-threatening opportunistic infections and malignancies. For other patients, their clinical history combined with the risk of these side effects excludes them from ever receiving systemic anti-TNF-α therapies. In this study, we presented two anti-TNF-α VNARs that recognized distinctly different epitopes on TNF-α. While VNAR-D1 bound a unique epitope not previously recognized by any of the existing anti-TNF-α agents, VNAR C4 recognizes an epitope that partially overlaps with existing agents. VNAR-D1 and C4 have calculated contact areas of only 970 Å2 and 822 Å2, respectively (on the surface of each TNF-α protomer), which would be considered a small molecular footprint compared to the surface contact footprint of conventional antibodies on target proteins32,46,47. Generally, larger surface area footprints result in increased complex stability, affinity, and neutralizing potency. However, our monoparatopic and biparatopic VNARs demonstrated equal or increased affinity and potency (in vitro) when compared to clinically approved anti-TNF-α biologics. This was achieved by packing multiple residue interactions with the CDR3 binding loops that protrude deep into their epitope on TNF-α48,49. Despite their small molecular footprint, VNAR-D1 and C4 effectively achieved stabilization of the sTNF-α homotrimer. With the epitope of VNAR-D1 in the cleft of two protomers, three VNAR-D1 domains were able to bind simultaneously to the homotrimer (Fig. 2b), Similarly, the VNAR-C4 epitope covered a single protomer that was accessible on all three protomers also allowing for binding in triplicate (Fig. 2a). VNAR-D1 and C4 engagement with TNF-α stabilized the homotrimers by promoting interactions between the protomers and preventing a TNF-α monomer exchange reaction. This is like the protein:protein stabilization achieved with Adalimumab and Infliximab32,46,49,50. The low nanomolar binding affinities of VNAR-D1 and C4 monomers, respectively, correlate with the TNF-α homotrimer stabilization. By the rational design of the biparatopic VNAR-D1 and C4, we improved this stabilization, resulting in several orders of magnitude enhancement in neutralizing potency and binding affinity25. The structural elucidation of VNAR-D1:TNF-α epitope implied a binding preference for sTNF-α over tmTNF-α. This binding preference resulted in the absence of any induction of reverse signaling by activated CD4+ T cells. As markers of reverse signaling induction, we measured E-selectin and IFN-γ levels33,40,51. There was no significant binding by any of the three anti-TNF-α VNAR formats tested (ELN22-108, ELN22-129, and ELN22-130); however, in stark contrast, there was a clear, dose- and time-dependent increase in tmTNFα binding that peaked at 24 h (Fig. 6a–c) time- and dose-dependent binding response observed for MAB210 and Infliximab with maximum binding occurring at 24 h (Fig. 6a–c). We propose that this transient maximum reflects a dynamic interplay of tmTNFα internalization and recycling, antibody-induced reverse signaling, and avidity effects inherent to bivalent IgGs. After binding, tmTNFα-antibody complexes are internalized via tmTNFα-mediated endocytosis previously described in dendritic cells, macrophages, and CD4+ T cells, but a portion of these complexes is recycled back to the surface rather than degraded. At the same time, engagement by high-affinity, multivalent antibodies triggers reverse signaling through tmTNFα, boosting its transcription and promoting renewed surface expression. Moreover, the bivalent structure of IgG cross-links tmTNFα molecules, enhancing their retention on the cell surface or within recycling compartments and amplifying net accumulation by 24 h. Bio-layer interferometry confirms that both antibodies possess rapid association and slow dissociation rates (Supplementary Fig. 8), ruling out a slow on-rate as a driver of the 24 h binding peak and supporting the roles of recycling, reverse signaling, and avidity in shaping this profile37–39. While the crystallography data did not suggest that VNAR-C4 may preferentially bind to sTNF-α, there was no measured binding to surface tmTNF-α on PHA activated CD4+ T cells at all timepoints or to pro-TNF-α by BLI. Possible explanations could be either rapid internalization of the anti-TNF-α VNAR-C4 upon binding to tmTNF-α or because the smaller contact surface (epitope) on TNF-α recognized by this VNAR-C4 is conformationally different in both the transmembrane and soluble versions of TNF-α. The latter is more plausible since we do observe this level of flexibility in the structure of TNF-α. For instance, the E-F loop of TNF-α plays a crucial role in the interaction with adalimumab and infliximab Fab fragments; however, this region is completely unobserved in the complex structures of TNF-α with TNFR2 or certolizumab, indicating that the E-F loop is flexible and not involved in these two interactions5,32,46,47. Supporting this notion of subtle epitope variation, our BLI binning experiments demonstrate that, despite overlapping binding regions, VNAR-C4, infliximab, and adalimumab occupy distinct bins (Supplementary Fig. 9). An E-selectin expression time course has been reported to be different between CD4+ T cells induced by tmTNF-α, with peak expression captured between 12 and 24 h51. In this batch of CD4+ T cells, we were able to measure dose-dependent induction of E-selectin and IFN-γ expression above baseline at the 24 h timepoint for only MAB210 and Infliximab. Again, the mono-paratopic and bi-paratopic VNARs did not induce the expression of these molecules. The biparatopic anti-TNF-α VNAR, ELN22-108 is capable of binding and neutralizing sTNF-α via unique epitopes, with at least one of two epitopes believed to be an important differentiator between tmTNF-α and sTNF-α binding. In a previous study, we demonstrated that the bi-paratopic ELN22-108 is superior to Adalimumab in controlling sTNF-α-driven polyarthritis in Tg197 transgenic mouse model of polyarthritis, while in another study, we showed evidence that VNARs targeting an unrelated antigen are amenable to binding unique epitopes that consequently resulted in previously undocumented mechanism(s) of antigen neutralization24,31. A combination of a unique mechanism of action with enhanced safety profile and superior TNF-α neutralizing potency positions ELN22-108 as an ideal first-line therapeutic for certain subsets of high-risk, autoimmune inflammation patients. Although the primary focus of this manuscript has been structural; defining the binding of VNAR-D1 and VNAR-C4 on TNF-α and assessing the functional implications of those interactions within a defined biological context, we recognize that a deeper mechanistic and functional dissection of TNFR1 and TNFR2 signaling, particularly in response to tmTNF-α, will be of great interest. While TNFR1 signaling results in the undesirable pro-inflammatory effects of TNF-α, TNFR2 signaling is known to induce favorable responses that include immune cell homeostasis and tissue remodeling52–54. Several studies have documented the development of TNFR1-specific antagonists that preserve TNFR2-mediated signaling54–56. Based on our BLI data with pro-TNF-α and our flow cytometry data using LPS-activated THP-1 cells, we do not believe that our VNARs would inhibit tmTNF-α-induced TNFR1 and TNFR2 signaling. This is postulation, and additional experiments are needed for confirmation and also to interrogate what the effects of uninhibited tmTNFα-induced TNFR1 and TNFR2 signaling are on target cells and the immune system. In summary, there is a strong case to be made for a class of anti-TNF-α targeting therapy with preference for sTNF-α or complete abrogation of tmTNF-α binding and induction. Also, in the absence of surface antigen binding, CDC and ADCC activation are impractical. It is understandable that such proposed mechanisms may have limited clinical efficacy benefit for patients with granulomatous diseases such as Crohn’s disease, although we believe there are additional approaches that could lead to optimal clinical outcomes for CD patients via sTNF-α-only targeting. This proposed mechanism will ensure enhanced safety profile for anti-TNF-α-treated patients. Methods Expression and purification of soluble TNF-α and VNARs. Synthetic genes encoding the soluble fragment of human TNF-α (residues 77–233 of the membrane-bound form, uniprot ID PO1375), and the VNARs D1 and C4 with N-terminal, 3C protease-cleavable hexa-histidine tag were synthesized at GeneArt Gene Synthesis Thermo Fisher Scientific. E. coli BL21 Gold pLysS cells were transformed with expression vectors encoding human sTNF-α, VNAR D1, and VNAR C4 sequences with an N-terminal 3C protease-cleavable hexa-histidine tag. Positive transformants were selected on LB-agar containing appropriate antibiotics, and expression was induced in LB medium using 0.5 mM isopropyl β-D-1-thiogalactopyranoside (IPTG). Harvested cells were resuspended in PBS supplemented with Complete protease inhibitors and 5 mM β-mercaptoethanol and lysed by three passes through an EmulsiFlex C5 homogenizer. Cleared lysates were mixed with Ni-NTA agarose beads equilibrated in PBS, and bound protein was eluted with 500 mM imidazole (prepared in 20 mM Tris HCl, pH 8.0, 150 mM NaCl, 5 mM β-mercaptoethanol (Buffer A). Eluted protein was cleaved with 3C protease-GST tag and dialyzed overnight in 20 mM Tris HCl, pH 8.0, 150 mM NaCl, 5 mM β-mercaptoethanol buffer. Uncleaved protein, cleaved hexa-histidine tag, and 3C protease-GST tag were removed by passing the dialysed sample through a Glutathione agarose and a Ni-NTA column. The untagged protein (sTNF-α, VNAR D1, and VNAR C4) flowthrough was further purified using cation exchange and/or size exclusion chromatography and dialysed into PBS. The tetravalent VNAR constructs were designed, gene synthesized, expressed, and purified as previously described25,31. The tetravalent constructs were approximately 105 kDa fusion molecules designed with two varying lengths of glycine-serine (GlySer) linkers, a shorter 2–3 (Gly4Ser) fusing the first anti-TNF-α VNAR-D1 or C4 to the hinge of a wild-type human IgG1 Fc region, and a 4–5 (Gly4Ser) linker fused to the C-terminal of the human IgG1 Fc CH-3 domain linking the second anti-TNF-α VNAR-D1 or C4. NB: Co-crystallization of the anti-TNF-α VNARs with soluble TNF-α was conducted using monomeric VNAR-D1 and VNAR-C4. sTNF-α – anti-TNF VNARs complex assembly and size exclusion chromatography (SEC) of complexes of TNF-α with VNARs Complex assembly was first tested at microgram scale. Successful assembly protocols were scaled up to milligram amounts. At small scale tests, 100 µg sTNF-α (5.7 nmol) was mixed with a 2-fold molar excess (11.4 nmol) of VNAR in 100 µL Buffer A. sTNF-α-VNAR mixtures were incubated for 30 min on ice and centrifuged for 10 min, 21,000 × g, at 4 °C. Supernatants were analyzed by SEC using Superdex 200 10/300 column, and SEC fractions were analyzed by SDS PAGE. Fractions containing target complexes were pooled, concentrated, and used for crystallization screens. Crystallization and X-ray structures determination for VNAR-D1 and sTNF-α complex Crystallization conditions for sTNF-α-VNAR-D1 were screened following the sitting drop vapor diffusion protocol57. Drops were monitored regularly using an optical microscope. Initial hits were refined in 48-well plates using hanging drop setups with 500 µL reservoir solution and 1–6 µL drops with streak seeding. The best crystals grew after several rounds of seeding in a condition containing 19% PEG 8000, 0.2 M ammonium sulfate, and 0.1 M MES pH 6.0 at 18 °C. Crystals were harvested with nylon loops, transferred to a cryoprotectant (reservoir solution supplemented with 10–20% glycerol, ethylene glycol, propylene glycol, or PEG 400), and flash-frozen in liquid nitrogen. X-ray diffraction data were collected at beamline P14, German Electron Synchrotron DESY, Hamburg, and a complete dataset was collected at 3.43 Å resolution (Table S1). Diffraction images were processed using XDS58. Phases were obtained by molecular replacement with PHASER using the TNF trimer from Protein Data Bank (PDB) entry 1TNF59. The model was completed with COOT (Emsley, Cowtan, 2004) and refined using REFMAC560,61. Protein-protein interfaces were analyzed using PISA. PyMOL and COOT were used for refinement of the crystal structure and comparison of the sTNF-α-VNAR-D1 structure to other known soluble huTNF-α/anti-TNF-α complexes. The VNAR-D1: sTNF-α crystallization and X-ray structural determination were carried out at moloX GmbH, Berlin, Germany. The coordinates and structure factors have been deposited in the Protein Data Bank (PDB) with the accession code 9DJW. Crystallization and X-ray structures determination for VNAR-C4 and sTNF-α complex Purified sTNF-α and VNAR-C4 were mixed at ~1:1 molar ratio, and the complex formed was isolated using a Superdex200 size-exclusion chromatography (SEC) column operating with 10 mM Tris-HCl (pH 7.4) and 150 mM NaCl. The isolated complex was concentrated by ultrafiltration to ~10 mg/ml and subjected to crystallization. The sTNF-α-VNAR-C4 complex was crystallized by the sitting drop vapor diffusion method, using a reservoir solution containing 0.1 M MES, pH 6, 0.1 M magnesium chloride, and 8% w/v polyethylene glycol 6000. The crystals grew to full size in three days. The crystals were cryo-protected by briefly soaking in the reservoir solution supplemented with 25% ethylene glycol and flash-cooled by plunging into liquid nitrogen. X-ray diffraction data were collected at the Advanced Photon Source NE-CAT beamline 24-ID-C. All X-ray diffraction data were processed using XDS version 20210205. Molecular replacement calculations were performed using PHASER version 2.859. Iterative manual model building and refinement were done using COOT version 7766 and PHENIX version v.1.19.2-4158, respectively62,63. The summary of data collection and model refinement statistics is shown in Table S2. The VNAR-C4:sol huTNF-α crystallization and X-ray structural determination were carried out at the Department of Biochemistry, Molecular Biology, and Biophysics, University of Minnesota, Minneapolis. The coordinates and structure factors have been deposited in the PDB with the accession code 9BN7. Inhibition of TNF-α:TNFR1/2 binding by anti-TNF-α VNARs via ELISA The TNFR1/2 Inhibitor Screening ELISA kits (Acro Biosystems Cat No. EP-143, EP-147) were used following the manufacturer’s protocol. Anti-TNF-α VNARs were serially diluted and tested across a concentration range of 10.0 - 0.020 μg/mL. Transmembrane TNF-α binding on activated human monocytes Human monocytic leukemia cell line, THP-1, was cultured in supplemented RPMI at a density of 0.5 × 106 cells/mL in a 24-well flat bottom cell culture plates, 37 °C, 5% CO2. THP-1 cells were activated with 1 μg/mL lipopolysaccharide (LPS) and treated with 10 μM Thymoquinone (TQ) or 0.1% DSMO for 4 h. LPS activated cells were washed with PBS three times and stained with Human TruStain FcX (Biolegend, Fisher Scientific, Cat no. NC9067567) per manufacturer’s recommendations. Activated cells were stained with Alexa Flour™ 647 (Invitrogen, Thermo Fisher Scientific Cat No. A20006) labeled anti-TNF-α VNARs at a concentration of 125 nM in Flow Cytometry Staining buffer (eBioscience, Thermo Fisher Scientific Cat no. 00-4222-26) and incubated for 1 h on ice. Cells were washed three times, and fluorescence intensity was measured using an Attune™ NxT flow cytometer (Thermo Fisher Scientific). Data analysis was performed using FlowJo software (Tree Star, Inc.). In vitro functional characterization of the monoparatopic and biparatopic constructs The TNF-α sensitive mouse fibrosarcoma cell line (L929 cells) was grown to 90% confluence, seeded onto 96-well flat-bottom microtiter cell culture plate at 5000 cells per well, and incubated for 48 h. Anti-TNF-α VNAR constructs were added to actinomycin-D, and sTNF-α-treated L929 cells. Cytotoxicity or cell survival was determined by adding tetrazolium salt (WST-1) cell proliferation reagent, and absorbance was read at 450 nm. The reported results are the mean ± SEM (n = 3, with 6 replicates). Octet R8 (Sartorius) biolayer interferometry (BLI) was used to determine binding kinetics of the monoparatopic tetravalent VNAR construct to sTNF-α. Anti-TNF-α VNAR constructs prepared in 1% BSA in PBS + 0.01% Tween 20, pH 7.4 (assay buffer) were loaded on assay buffer-equilibrated anti-human IgG Fc Capture (AHC2) biosensors at a final concentration of 10 nM for 10 min, followed by 30 min association step with serial dilutions of sTNF-α prepared in assay buffer and a 30 min dissociation step. Data were analyzed using the Octet® BLI Analysis Studio software 12.2., and global 1:1 fit model. Binding kinetics were also measured using biotin labeled sTNF-α prepared in assay buffer (1% BSA in PBS + 0.01% Tween 20, pH 7.4) at a final concentration of 10 nM. Biotinylated sTNF-α was loaded on assay buffer-equilibrated High Precision Streptavidin (SAX) biosensors for 10 min followed by 30 min binding association of serially diluted anti-TNF-α VNARs and a final 30 min binding dissociation step. Transmembrane TNF-α binding and reverse signaling in phytohemagglutinin (PHA) activated CD4+ primary human T cells Healthy human CD4+ T cells (BioIVT, lot no. HumanCD4-0111394) were cultured in supplemented RPMI at density of 0.2 × 106 cells/mL in 96-well flat bottom cell culture plates, 37 °C, 5% CO2. CD4+ T cells were stimulated with 0.1 µg/mL PHA for 12 h. PHA activated cells were washed with acidified buffer (50 mM glycine-HCl containing 150 mM NaCl, pH 3.0) for 3 min at 4 °C to remove traces of sTNF-α. Acidified buffer was removed by washing the cells three times with PBS containing 1 % BSA. PHA activated cell were treated with anti-TNF-α VNAR constructs and commercial antibodies at different concentrations (100 nM, 10 nM, and 1 nM) and incubation timepoints (16 h, 24 h, and 48 h). Cells were then stained with a mouse anti-human Fc-FITC conjugated antibody and were fixed with 0.5% paraformaldehyde (PFA) for 15 min, 4 °C, and fluorescence intensity measured using CLARIOstar® plus microplate reader and Attune™ NxT flow cytometer. Cells were stained for E-selectin using anti-human CD62E-FITC conjugated antibody and fixed with 0.5% paraformaldehyde (PFA) for 15 min, 4 °C, and fluorescence intensity measured using CLARIOstar® plus microplate reader and Attune™ NxT flow cytometer. IFN-γ levels were measured in anti-TNF-α-treated activated CD4+ T cells supernatant using the human IFN-γ DuoSet Sandwich ELISA kit (R&D Systems, Cat no. DY285B-05). The results are the mean ± SD (n = 4) with 2 replicates per experiment. Results were statistically analyzed using a two-way ANOVA with Dunnett’s multiple comparisons post hoc test. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Reporting Summary Transparent Peer Review file Source data Source data Acknowledgements This work was supported by the UKRI FLF/MRC Award MR/V026283/1 (O.U.), NIH/NIGMS R35 GM118047 (H.A.), NIH/NCI R01 CA237272 (A.M.L.), NIH/NCI R01 CA233562 (A.M.L.), a 2018 Prostate Cancer Foundation Challenge Award (A.M.L.), a Prostate Cancer Foundation Young Investigator Award (A.M.L.), and Andy North and Friends (A.M.L.). This work used an Advanced Photon Source (APS) NE-CAT beamline, which is supported by NIH P30 GM124165. Author contributions K.S., N.H.M., and H.A. performed crystallographic experiments and structural determination. E.W.L. analyzed structural data. O.C.U., S.P., S.C., and J.L.W. performed affinity and binning measurements. O.C.U., S.P., and K.L.O. performed flow cytometry and fluorescence measurements. K.L.O. and S.P. performed ELISA assays. O.C.U., E.W.L., K.L.O., S.P., G.S.G., A.J.P., D.K., A.M.L., and C.J.B. contributed to the conceptualization and design of the study. O.C.U., E.W.L., K.L.O., A.J.P., A.M.L., and C.J.B. wrote the manuscript, with all authors contributing to its final version. Peer review Peer review information Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available. Data availability The authors declare that the data supporting the conclusions of this article are included in the main text and supplementary materials. The structural data generated in this study have been deposited in the RCSB Protein Data Bank. Structures of TNF-α in complex with VNAR-D1 and VNAR-C4 are original to this work and can be found with accession codes 9DJW and 9BN7. Source data is available with this paper as a Source data file. Source data are provided with this paper. Competing interests O.C.U., S.P., A.J.P., and C.J.B. are employees of Elasmogen, Limited. The remaining authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Obinna C. Ubah, Eric W. Lake. Contributor Information Obinna C. Ubah, Email: obinna.ubah@elasmogen.com Aaron M. LeBeau, Email: aaron.lebeau@wisc.edu Supplementary information The online version contains supplementary material available at 10.1038/s41467-025-66967-3. References Associated Data Supplementary Materials Supplementary Information Reporting Summary Transparent Peer Review file Source data Data Availability Statement The authors declare that the data supporting the conclusions of this article are included in the main text and supplementary materials. The structural data generated in this study have been deposited in the RCSB Protein Data Bank. Structures of TNF-α in complex with VNAR-D1 and VNAR-C4 are original to this work and can be found with accession codes 9DJW and 9BN7. Source data is available with this paper as a Source data file. Source data are provided with this paper.