AI-validated fusion proteins for local inhibition of interleukin-17A

Abstract

IL-17A is a potent immunomodulatory cytokine that is upregulated in many disorders characterized by chronic inflammation, including autoimmune-mediated dry eye observed in patients with Sjögren’s Disease. Existing IL-17A inhibiting therapies are delivered systemically and have been associated with increased infection rates in patients. Inhibiting IL-17A signaling locally, only in affected tissues, may provide therapeutic benefits without compromising systemic immune functions. To address this unmet need, we have developed a local IL-17A inhibiting therapy by fusing HAP, a peptide which binds and inhibits IL-17A, with elastin-like polypeptides (ELPs), which are intrinsically disordered thermosensitive protein-polymers that can be engineered to undergo liquid-liquid phase separation to form depots after injection. We show tuning of ELP composition allows us to manipulate the pharmacokinetics, and specifically mean absorption time of our local immunosuppressive depots, demonstrating the potential for extended, local therapy. Furthermore, the fusions retain the binding and inhibitory activity of the HAP peptide, aligning with our in silico predictions using AlphaFold3. Finally, we confirmed that our formulations, which are bacterially derived, do not measurably stimulate the innate or adaptive immune system, which is critical for therapies for autoimmune diseases. Together, these findings establish the utility of our novel peptide-ELP fusions for extended local IL-17A inhibition to address chronic inflammation with fewer off-target effects.

1. Introduction

Interleukin-17A (IL-17A) is an evolutionarily ancient cytokine that is highly conserved across species and is critical to host defense. It mediates inflammation, immune cell recruitment, and tissue remodeling in response to bacterial or fungal infections. [1,2] IL-17A is the most potent member of the IL-17 cytokine family, which also includes other isoforms IL-17B, IL-17C, IL-17D, IL-17E, and IL-17F, which dimerize to form cytokines with different cognate receptor complexes and downstream effects. In this work, IL-17A refers to the IL-17A/IL-17A homodimer, of which the heterodimeric receptor complex is composed of IL-17 receptors A and C (IL-17RA and IL-17RC). IL-17A is secreted primarily by Th17 cells, but also by other immune cell subsets, including γδ T cells, Type 3 innate lymphoid cells, dendritic cells, and macrophages. [3] Following secretion, IL-17A is understood to first bind IL-17RA, resulting in the recruitment of IL-17RC to the complex. The formation of the heterodimer receptor complex allows binding of the Act1 adaptor protein, leading to signal transduction and activation of the canonical NF-κB cascade, which results in the secretion of pro-inflammatory factors by the target cell. [3–5] These include antimicrobial peptides, chemokines, and other cytokines, which, in normal function, work to build a response to an invading pathogen. [6,7]

However, dysregulated IL-17A expression is observed in many autoimmune disorders, where unnecessary and prolonged pro-inflammatory signaling leads to chronic inflammation and tissue damage. These disorders include psoriasis, rheumatoid arthritis, and Sjögren’s Disease (SjD), where elevated IL-17A levels are detected in affected tissues. [8–11] SjD is a relatively common but medically underserved autoimmune disease, affecting an estimated 1–4 million people in the United States and 0.01–0.05% of the global population. [12,13] Patients display numerous systemic symptoms, such as joint pain, fatigue, and rashes, in addition to the canonical symptoms of dry eye and dry mouth. These latter effects are caused by inflammation and gradual loss of function of the salivary gland and tear-producing lacrimal gland (LG), events in which IL-17A has been implicated. Specifically, IL-17A is thought to contribute to immune cell infiltration into the affected tissues, leading to the development of ectopic lymphoid structures, which further promote inflammation and cause damage to parenchymal tissue. Evidence of IL-17A involvement in the disease includes the fact that it is present at significantly higher concentrations in the saliva and tears of SjD patients, compared to people suffering from oral or ocular dryness without SjD, as well as healthy controls. [14–16] Furthermore, strong positive staining for IL-17A in lymphocytic infiltrates in salivary gland biopsies support this finding. [17]

Biologics inhibiting IL-17A activity are available on the market, including Secukinumab and Ixekizumab, which target IL-17A, and Brodalumab, which targets IL-17RA. These have been used to successfully treat plaque psoriasis [18,19], psoriatic arthritis [20], and anky-losing spondylitis [21], validating IL-17A’s importance in these autoimmune disorders. However, these antibody therapies are administered subcutaneously to inhibit IL-17A systemically, which has led to unwanted side-effects. Due to IL-17A’s protective role against pathogens in a healthy immune system, its systemic inhibition has been shown to increase rates of fungal infections of Candida in patients receiving these treatments. [22–25]

To reduce the risk of negative side effects in off-target tissues, local therapy to limit IL-17A inhibition to only affected organs may be beneficial. This can be accomplished through engineered delivery technologies, including the use of in situ forming materials to retain drugs at the delivery site. For example, Atrigel is an in situ forming PLGA hydrogel used to deliver drugs locally to treat prostate cancer [26] and periodontitis [27]. Similarly, products with temperature-sensitive vehicles that spontaneously gel after injection have been approved for use in local drug delivery. For example, OTIPRIO is an antibacterial fluoroquinone formulation that contains poloxamer 407, which undergoes a sol-gel transition upon exposure to body temperature. [28] Other temperature sensitive delivery systems, including those based on elastin-like polypeptides (ELPs) are also in development. [29–34]

ELPs are a class of synthetic protein-polymers derived from the human tropoelastin sequence which can be engineered to undergo liquid-liquid phase separation (LLPS) upon heating above characteristic transition temperature (Tt). They are comprised of the repeat motif (Val-Pro-Gly-Xaa-Gly)_n, where the guest residue (Xaa) and chain length (n) can be used to tune the lower critical solution temperature or Tt. Specifically, more hydrophobic guest residues and longer lengths bring down the Tt, which can range from below room temperature to above 80 °C. As their Tt is exceeded, ELPs undergo LLPS to form a highly viscous secondary liquid phase called coacervate. This thermoresponsive property allows for injectable formulations that coacervate in situ, forming depots that prolong local retention. ELPs also enable column-free purification via inverse transition cycling (ITC), and their tropoelastin-like composition minimizes immunogenicity. Unlike most other temperature-sensitive drug delivery materials, ELPs are protein-polymers, allowing them to be fused to therapeutic proteins or peptides, resulting in monodisperse molecules that contain both the bioactive component and excipient. This provides key advantages over existing mixtures of drug and synthetic temperature-sensitive excipient with drug. One such advantage is the monodisperse nature of proteins, which allows for more predictable properties and less batch-to-batch variability. Also, drugs are covalently bound to the depot-forming excipient, which may lead to longer retention and more predictable behavior, without the need for chemical conjugation reactions. Furthermore, their intrinsically disordered nature allows for easier handling, as there is no secondary structure that can be disrupted by temperature fluctuations or freeze-thaw cycles. Peptide-ELP fusions further capitalize on this, as neither domain has secondary structure to preserve.

HAP is a peptide inhibitor of IL-17A and a potential alternative to the available IL-17A-inhibiting biologics. It was developed through optimization of a sequence identified in phage display and has been shown to bind IL-17A with nanomolar affinity (2.6 ± 0.5 nM). [35] Importantly, this binding occurs in a pocket critical for engagement of IL-17RA, resulting in neutralization of the cytokine’s activity. Like other peptides, its short length of just 15 amino acids results in a lack of higher order structure to preserve, unlike proteins which can be denatured and lose function. However, its low molecular weight also makes it unsuitable for local administration, due to the likelihood of rapid clearance. This shortcoming can be addressed by fusion to an ELP, which would result in a structurally robust, depot forming inhibitor of IL-17A that could be for tissue-specific IL-17A inhibition. Fusion of ELP to a therapeutic protein or peptide, however, cannot be assumed to have no impact on its bioactivity. In fact, studies show that ELP fusion can impact activity based on the fusion order, analogous to site-dependent changes in activity of PEGylation. Recent advances in protein folding and interaction tools may be useful in predicting these effects.

AlphaFold3 is an openly available AI-based protein folding and interaction simulation tool that predicts protein structures from sequence alone, and may be useful in predicting loss of therapeutic activity of a peptide or protein following carrier protein fusion. [36] The first iteration of AlphaFold was released by DeepMind in 2021 and was the first model to utilize deep learning techniques to train a protein folding simulation tool, dramatically outperformed existing models in prediction accuracy. Unlike traditional molecular dynamics simulations, which are sensitive to simulation parameters, such as the force field, water model, and temperature, AlphaFold does not directly model protein folding, but rather trains on structures in the Protein Data Bank, combined with sequence alignments of homologous proteins from various species, in order to determine weights in its neural network that implicitly capture protein folding dynamics and result in accurate protein structure predictions. Importantly, AlphaFold and other AI-based protein folding tools do not require specialized hardware or experience in selecting appropriate simulation parameters, as is the case for many traditional molecular dynamics approaches, democratizing the ability to perform protein folding simulations. Furthermore, the latest iterations of AlphaFold, AlphaFold3 and AlphaFold2-Multimer predict not only structure, but also protein-protein interactions. [37]

In this work, we developed an easy-to-use pipeline to assess whether AlphaFold3 predicts changes in binding interactions upon fusion of carrier proteins, like ELPs, to bioactive peptides. We used this process to validate novel HAP-ELP fusions for local immunosuppression in silico. These favorable predictions were confirmed by our in vitro investigation of HAP-ELP fusions, which show that they retain the binding and inhibitory activity of the HAP peptide. We also demonstrate control over the thermosensitive properties of HAP-ELPs through engineering of the ELP sequence and show that this can be used to drive depot formation in vivo following local administration. Finally, the immunogenicity of HAP-ELP was evaluated in vivo, showing that our purification protocol results in a formulation that is appropriate for use in anti-inflammatory applications. Together, these findings show that HAP-ELPs are a good candidate for local IL-17A inhibition, which may address the side-effects caused by existing treatments.

2. Materials and methods

2.1. In silico validation of HAP-ELP activity

To evaluate the structural feasibility of HAP-ELP fusions, we used AlphaFold3 Software (https://alphafoldserver.com/, accessed 2025) to predict the structure of the HAP peptide fused to various ELPs in complex with IL-17A. The sequence of the simulated structures can be found in Table 1. The sequence of the simulated HAP peptide is MIHVTI-PADLWDWINKG, which includes an initiating methionine and a terminal glycine, to match the expressed HAP-ELPs. The simulations of the scrambled HAP peptide, scHAP, used the sequence MIWNIIKLPH-DADTWVG which is a randomized order of the HAP sequence, also including an N-terminal methionine and C-terminal glycine to match the simulated HAP peptide. The deviation of the predicted structures compared to the solved structure of the interaction (PDB: 5HHV) [35], called the reference structure, was analyzed using PyMOL (https://www.pymol.org/, OpenGL version 2.1, accessed December 2024) using a Python script developed with ChatGPT 4.0. The script has been made available in the Supporting Information (Script S1), along with a protocol for its use (Supplementary Method 1). Each interaction was assessed using at least 15 AlphaFold3 predictions. The simulated IL-17A was aligned with the reference IL-17A and the distance of each amino acid in the HAP structure was compared to the corresponding amino acid in the reference HAP. The simulated IL-17A was comprised of two copies of residues 24–155 from the human interleukin-17A sequence (Uniprot Q16552), which represents the mature biologically active protein after cleavage of the signal peptide. To account for the two possible binding sites on the IL-17A homodimer, the script aligns a simulated IL-17A monomer with both reference IL-17A monomer and selects the orientation with better agreement. The final residue of HAP in the reference structure is not resolved, so only the first 14 amino acids were compared. The coordinates of these amino acids were then used to determine root mean square deviation (RMSD) between each residue in a simulated pose with that of the same residue in the reference structure.

Table 1.

Sequence and biophysical characterization.

Construct	Sequence*	Yield (mg/L)**	Expected MW*** (kDa)	Observed MW (kDa)	Tt at 50 μM (°C)	Slope (°C/Log(μM))	Intercept (°C)
HAP-A96	M-HAP-(VPGAG)96-Y	53.4	38.79	38.67	72.7 (3.5)	−16.26 (0.81)	87.85 (1.26)
HAP-V96	M-HAP-(VPGVG)96-Y	90.8	41.48	41.38	26.1 (1.6)	−2.45 (0.56)	29.07 (0.88)
A96	G-(VPGAG)96-Y	51.7	36.86	36.88	82.1 (1.0)	−15.19 (0.76)	96.99 (1.36)
V96	G-(VPGVG)96-Y	101.8	39.55	39.62	31.6 (0.1)	−2.95 (0.10)	34.71 (0.15)

^*HAP Sequence: IHVTIPADLWDWINK

^**Units of yield are mg purified protein per liter bacterial culture; a typical batch was 6 L yielding ~300–600 mg purified protein

^***Expected MW calculated by ExPasy ProtParam based on open reading frame.

2.2. Plasmid construction

pUC57 plasmid containing HAP, a ribosome binding site, a 5′ un-translated region, and flanking restriction sites was purchased from Azenta Life Sciences. Cloning was performed as previously described. [38] Briefly, the HAP insert was excised using NdeI and BamHI and ligated into the T7 expression region of a pET-25b(+) plasmid. This intermediate construct was then digested with BseRI and BssHII and ligated into a pET25b(+) vector encoding ELP with a Class IIS restriction site to facilitate cloning of fusion partners at its N-terminus. All constructs were verified by Sanger sequencing (Azenta Life Sciences). Plasmids will be deposited to Addgene for public access.

2.3. Protein expression and purification

All HAP-ELP fusion proteins and ELP-only controls were expressed in ClearColi BL21(DE3) cells and purified using a previously described ITC protocol. [39,40] Cultures were grown in Terrific Broth supplemented with 10 g/L NaCl and 100 μg/mL carbenicillin, induced with 0.5 mM IPTG when OD600 exceeded 0.6, and incubated overnight at 27 °C. Bacterial pellets were resuspended in PBS and lysed by probe sonication, followed by removal of genetic contaminants through precipitation using polyethyleneimine and centrifugation at 4 °C. To trigger ELP coacervation, clarified lysates were incubated at 37 °C, with NaCl added up to 2 M as a kosmotropic agent if needed. Coacervates were pelleted (“warm spin”), resolubilized in cold PBS and centrifuged again at 2 °C (“cold spin”), followed by collection of the supernatant. Together, a warm spin and cold spin constitute an ITC cycle. Each batch was purified with at least three ITC cycles. Protein concentration was determined by the Beer-Lambert Law using theoretical molar extinction coefficients calculated using ExPASy ProtParam and absorbances at 280 nm measured using a NanoDrop 200 (ThermoFisher).

2.4. Fluorescent labelling

HAP-ELPs used in pharmacokinetic studies were fluorescently labelled using NHS-Rhodamine (ThermoFisher). A 10 mg/mL stock solution of NHS-Rhodamine was prepared in DMSO and added to protein solutions at a 6:1 molar ratio of fluorophore to HAP-ELP. The reaction was incubated on a rotator at 4 °C for 2 h, then dialyzed against PBS to remove unreacted dye. Rhodamine concentration was quantified by absorbance at 570 nm using the Beer-Lambert Law, assuming a molar extinction coefficient of 80,000 M⁻¹ cm⁻¹. The rhodamine-labelled HAP-ELPs were mixed with non-labelled HAP-ELP to normalize both stocks to a 75 mol% rhodamine labelling efficiency.

2.5. Protein identity and purity confirmation

For SDS-PAGE, 5 μg of each sample was denatured with β-mercaptoethanol and heated at 95 °C for 10 min. Denatured samples were loaded into precast gels (Bio-Rad Mini-PROTEAN TGX) alongside a protein ladder (Bio-Rad Precision Plus Protein Kaleidoscope). For MALDI-TOF, samples were reduced with β-mercaptoethanol, diluted in water, and heated to 95 °C for 10 min to reduce disulfide bonds. Samples were then centrifuged briefly and mixed with a 2,5-dihydroxyacetophenone (DHAP)-based matrix solution (30 mg/mL DHAP in 50:50 aceto-nitirle/water with 0.2% formic acid) at 1:5 or 1:10 dilution ratios. 0.5 μL aliquots were spotted onto a MALDI target plate, air dried, and analyzed using a Bruker rapifleX MALDI Tissuetyper in intact protein mode. Spectra were processed with baseline subtraction and peak identification to confirm expected molecular weights.

2.6. Thermoresponsive behavior characterization

Phase transition behavior of ELP-containing proteins was characterized by temperature-dependent UV–Vis spectrophotometry and dynamic light scattering (DLS). Turbidity measurements were performed on a Beckman Coulter DU800 spectrophotometer at 350 nm with a 1 °C/min temperature ramp with data collected every 0.3 °C in a 1 cm pathlength quartz cuvette. The transition temperature (Tt) was defined as the point of maximum first derivative of absorbance with increasing temperature. DLS measurements were conducted on a Malvern Zetasizer system. Samples were filtered through a 0.2 μm syringe filter and loaded into 1 cm polystyrene cuvettes at a concentration of 15 μM. Measurements were performed using a 90° side scatter with a 1 min equilibration at each temperature.

2.7. Binding affinity quantification

Binding affinity of IL-17A to HAP-ELP constructs was measured using a Biacore T200 surface plasmon resonance (SPR) instrument. Recombinant human IL-17A (Peprotech) or HAP-A96 were immobilized onto CM5 sensor chips via carbodiimide chemistry in 10 mM sodium acetate buffer (pH 4.5). For immobilized IL-17A, HAP-ELPs were injected as analytes, along with ELP-only controls. In the opposite orientation, with HAP-A96 immobilized, recombinant human IL-17A was injected along with BSA as a negative control. All solutions were sterile filtered prior to use. Experiments were performed in pH 7.4 HBS-EP running buffer prepared from a 10× stock solution. The chip surface was regenerated between cycles using 20 mM HCl.

2.8. IL-17A depletion from supernatant

10 nM, 2 nM, 0.4 nM, and 0.08 nM solutions of IL-17A were mixed in a 1:1 volume ratio with 20 μM HAP-A96, HAP-V96, V96, or PBS to a final volume of 500 μL per tube. Samples were incubated for 10 min on ice before being transferred to a 37 °C water bath to induce coacervation. Samples were then spun at 4000 ×g for 10 min at 37 °C to pellet coacervates. 200 μL of the supernatant was sampled immediately after centrifugation and assayed for IL-17A concentration using an enzyme-linked immunosorbent assay (ELISA) kit (Biolegend).

2.9. In vitro inhibition of IL-17A signaling

IL-17A signaling inhibition was evaluated using the HEK-Blue IL-17 reporter assay cell line (InvivoGen), which produces secreted alkaline phosphatase (SEAP) in response to IL-17A stimulation, in accordance with the manufacturer’s instructions. Briefly, cells were seeded in 96-well plates, switched to starvation medium, and treated with mixtures of IL-17A and HAP-ELP or ELP overnight. After incubation, supernatants were collected and SEAP activity was quantified using QUANTI-Blue reagent (InvivoGen), which is a colorimetric substrate of SEAP that is read using absorbance at 620 nm.

2.10. Endotoxin removal

Endotoxin was removed using ThermoFisher Detoxi-Gel resin, degassed and packed into a gravity-flow column, then regenerated with 0.05% sodium deoxycholate, which was flushed out with PBS. Protein samples were loaded and incubated with pre-cooled resin for ≥1 h at 2 °C, followed by elution and sterile filtration through a Mustang E membrane syringe filter (Pall Corporation). Enxotoxin quantification was performed using HEK-Blue LPS cells (InvivoGen) according to manufacturer instructions using QUANTI-Blue SEAP substrate.

2.11. Animals

BALB/c mice (JAX strain #: 000651) were purchased from The Jackson Laboratory and bred in vivarium facilities at the University of Southern California (USC). All experiments were approved by the Institutional Animal Care and Use Committee (IACUC) at USC. Local doses were administered subcutaneously above the lacrimal gland (“supra-LG” injection) in male mice using a 31G insulin syringe. [38] Mice were anesthetized using isoflurane (2–3% in 100% O₂) using an induction chamber and nose cone for local injections and IVIS imaging. For longitudinal blood sampling, a restraint device was used to limit movement. Mice were euthanized by intraperitoneal administration of a 80–100 mg/kg ketamine and 5–10 mg/kg xylazine solution prior to blood collection by cardiac puncture and cervical dislocation.

2.12. In vivo pharmacokinetic studies

To evaluate depot pharmacokinetics following local delivery, 150 μL of 250 μM HAP-ELP (75 mol% Rhodamine-labelled) was administered via supra-LG injection in male BALB/C mice (14–22 weeks). Syringes and injection materials were kept on ice prior to injection. Systemic absorption was assessed via longitudinal tail vein blood sampling, and depot persistence was monitored using an IVIS Spectrum imaging system (PerkinElmer). Systemic clearance was characterized following IV bolus administration of 100 μL of 50 μM HAP-ELP and subsequent longitudinal tail vein sampling. Pharmacokinetic parameters were obtained by non-compartmental analysis using equations previously described. [38]

2.13. In vivo safety studies

The acute inflammatory response to HAP-ELP was evaluated in BALB/c mice (11–13 weeks, n = 4) using 150 μL of 250 μM HAP-V96, V96, PBS as a negative control, or 2 μg/mL lipopolysaccharide (LPS; E. coli O55:B5, Invivogen) as a positive control. Blood was collected by cardiac puncture and plasma C-reactive protein (CRP) levels were measured using a Mouse CRP Quantikine ELISA kit (R&D Systems). To assess immunogenicity, BALB/C mice (17–18 weeks, n = 3) received supra-LG injections of 150 μL of 250 μM HAP-V96, or PBS on days 0 and 14. On day 28, blood and LGs were collected. Serum was tested for antidrug antibodies (ADAs) using high protein binding Nunc Maxisorp ELISA plates (Biolegend) coated with HAP-V96, V96, and BSA as a negative control. Detection was performed using HRP-conjugated antimouse IgG/IgA/IgM secondary antibodies (ThermoFisher) with Assay Diluent A (Biolegend) used for blocking and dilutions. A previously developed anti-ELP antibody, AK1, was used as a reference standard. [41] LGs were fixed, sectioned, and stained with hematoxylin and eosin (H&E) to assess tissue architecture and inflammation.

2.14. Statistical analysis

All statistical analysis was performed in Graphpad prism. Linear regression of the log of concentration vs temperature was used to determine the slope and intercept of the LLPS phase boundaries. Curve fitting for EC50 of IL-17A and IC50 of HAP-ELPs in HEK-Blue IL-17 cells was performed using a 4PL sigmoidal fit. Significance of IL-17A pull-down results was assessed by a global of ANOVA of the log of the concentration, followed by Tukey’s multiple comparison test. Significance of AlphaFold RMSD was assessed using one-way ANOVA followed by Dunnett’s test compared to the HAP predicted structure. Similarly, CRP and ADA significance was assessed by one-way ANOVA followed by Dunnett’s test compared to serum from PBS-injected or PBS-inoculated animals. The lower limit of quantification (LLOQ) was defined as the lowest concentration on the standard curve that had less that 20% deviation from the predicted fluorescence based on the linear region of the standard curve.

3. Results

3.1. In silico simulation of fusions of HAP and ELP binding with IL-17A

The ability of HAP to bind and inhibit IL-17A was previously validated. [35] This analysis included a crystal structure of the HAP and IL-17A interaction which was solved by X-ray diffraction (PDB 5HHV). To assess the ability of AlphaFold3 to correctly predict the binding pose of HAP, we simulated the interaction of HAP, as well as a scrambled sequence with the same amino acid composition, scHAP, with IL-17A. After alignment of the simulated IL-17A with reference IL-17A, the alignment RMSD never exceeded 0.53 Å. It was found that AlphaFold was able to recapitulate the binding pose and secondary structure of HAP, with a mean RMSD of 2.33 ± 1.49 Å. However, scrambling of the sequence resulted in significantly increased deviation from the reference with an RMSD of 21.46 ± 4.24 Å, as well as loss of secondary structure (Fig. 1 A, D). The impact of fusion order was also assessed by AlphaFold3 using simulations of V4-HAP and HAP-V4, which revealed that neither fusion orientation significantly changed the accuracy of the binding pose prediction compared to the simulation of the peptide alone (Fig. 1B, D). Finally, we predicted the binding poses of HAP-ELPs using the longer ELPs, up to the intended full length A96 and V96 ELPs that would endow the thermosensitive properties required for extended local pharmacokinetics that would make the fusions useful for the treatment of local IL-17A overexpression. Secondary structure was observed in the ELP domain for both HAP-A96 and HAP-V96 (Fig. 1C). Quantification of the simulated structures revealed that longer ELPs resulted in greater deviation from the known binding pose. In addition, HAP fused to Alanine ELPs had greater deviation from equivalent length fusions with Valine ELPs (Fig. 1E).

Fig. 1.

In silico analysis of HAP-ELP binding to IL-17A. (A) AlphaFold3 predicted binding poses of HAP (green) and scHAP (orange) with an IL-17A homodimer (beige and white) compared to that of the solved ground truth crystal structure of this interaction (purple) (PDB: 5HHV). (B) Representative AlphaFold3 predicted binding poses HAP-ELP (blue) and ELP-HAP (yellow) with IL-17A compared to the ground truth interaction (purple). (C) AlphaFold3 predicted interactions of HAP-A96 (red) and HAP-V96 (blue) with IL-17A showing binding as well as likely-hallucinated secondary structure of ELP. (D) RMSD of AlphaFold3 outputs compared to ground truth showing that scrambling of the HAP-ELP sequence is predicted to disrupt binding, while N- or C-terminal ELP fusion is not. Each point represents an AlphaFold3 output structure, shown with the mean and SD. (E) Mean distance of AlphaFold3 simulated HAP amino acids in various HAP-ELP fusions to the corresponding amino acids in the solved structure, showing longer ELPs leading to increased divergence, but not to the extent of scrambling the HAP sequence. Error bars represent the 95% CI. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

3.2. Ligation, expression, and purification of HAP-ELPs

Successful ligation of the HAP-ELP constructs was confirmed by Sanger sequencing (Fig. 2A, B). The chromatograms reveal the in-frame insertion of the HAP-encoding sequence into the open reading frame of both the A96 and V96 ELP. The insertion depicted in the figure is illustrative and does not reflect the true steps taken to construct the HAP-ELP plasmids, as can be observed in the differences in the 5’ UTRs of the HAP-ELP sequence and their respective ELP controls. All four fusion peptides, were purified using ELP-mediated phase separation, giving yields from 51.7 mg/L to 101.8 mg/L (Table 1). The expressed and purified constructs, as well as the rhodamine-labelled constructs used for the pharmacokinetics experiments, were run on an SDS-PAGE gel, which showed a single band for all unlabelled constructs. The fluorescently-labelled constructs were observed to have a minor band at roughly double the target molecular weight, but gel intensity analysis nevertheless indicated over 90% purity for both constructs Supplementary Fig. S1C,D). Furthermore, fluorescent signal was only from the labelled constructs and the 75 kDa and 25 kDa ladder markers, as expected (Supplementary Fig. S1B). The distance travelled on the gel corresponded to the approximate expected molecular weight. However, A96 and HAP-A96 ran higher on the gel than their corresponding valine-ELP constructs. The individual and overlaid gel images can be seen in Supplementary Fig. S1-A,B. Finally, these bands remained stable for at least 3 days when incubated 1:1 with human serum (Supplemental Fig. S2).

Fig. 2.

Plasmid ligation and expression of HAP-A96 and HAP-V96 along with their respective ELP-only controls. (A-B) Sanger sequencing chromatograms of the ligated HAP-A96 (A) and HAP-V96 (B) constructs showing insertion of HAP peptide sequence into the open reading frame of the respective ELPs. (C) Coomassie-stained SDS-PAGE gel containing 10 μg protein per well demonstrating purity and approximate MW. (D) MALDI-TOF spectra showing MW shift caused by ligation of HAP to ELPs.

3.3. Thermoresponsive behavior of HAP-ELPs

behavior of the fusion proteins was characterized by temperature-dependent turbidity and dynamic light scattering (DLS). UV–vis absorbance temperature ramps show rapid increase in absorbance at 350 nm, characteristic of liquid-liquid phase separation that causes the formation of coacervate droplets the dramatically increase light scattering (Fig. 3A). Temperature ramp experiments were run at multiple concentrations and the results are summarized a phase diagram (Fig. 3B). The phase diagram shows that HAP fusion to the ELP slightly reduced the (Tt) compared to the parent ELP, as shown in the constructed phase diagrams. We performed linear regression analysis with Tt as a function of log₁₀ of concentration in nM to determine equations to predict concentration dependent phase behavior of each construct, the slope and intercept of which can be found in (Table 1). DLS measurements at 17 °C, 27 °C, and 37 °C show that temperature-dependent droplet formation occurs for HAP-V96, but not HAP-A96. Soluble HAP-ELP molecules have sub-10 nm hydrodynamic radius, while coacervate droplets are between 100 and 1000 μm, with higher temperatures resulting in larger coacervate droplets (Fig. 3C). Hydrodynamic radius determined by DLS was plotted against temperature, showing that HAP-V96 undergoes LLPS at a lower temperature than V96, in line with UV–Vis results, suggesting a modest influence of the HAP domain on ELP solubility (Fig. 3D). Coacervation behavior was retained for at least 3 days when incubated 1:1 with human serum (Supplementary Fig. S3).

Fig. 3.

Characterization of thermosensitive liquid-liquid phase separation behavior of HAP-ELPs. (A) Temperature-ramp absorbance curves of 250 μM HAP-ELPs and ELP controls showing rapid phase separation behavior upon heating. (B) Phase diagram showing the mean and standard deviation of concentration-dependent transition temperatures for HAP-ELPs and ELP controls across three independent temperature ramps. Above the phase boundary, liquid-liquid phase separation results in coacervate formation, while below the phase boundary the constructs remain water soluble. Ligation of HAP to ELPs results in a small decrease in transition temperature. (C) Distribution of radius of particles by volume % as measured by DLS for HAP-ELPs (15 μM) at various temperatures. (D) Temperature-dependent hydrodynamic radius of the major population of HAP-ELPs and ELP controls, showing temperature-dependent coacervate droplet formation of HAP-V96 and V96. Each point represents the mean and standard deviation of three DLS measurements using 15 μM samples. All error bars represent SD.

3.4. HAP-ELP binding and sequestration of IL-17A

To assess binding between HAP-ELP constructs and IL-17A, SPR was performed. IL-17A was covalently bound to a sensor chip, followed by injection of various fusion proteins. HAP-A96 and HAP-V96 both bound IL-17A in a dose-dependent manner, which revealed that, despite C-terminal ELP fusion, the peptide still bound IL-17A in a dose-dependent manner, in line with our in silico predictions (Fig. 4A and S4A,B). Furthermore, A96 and V96 did not result in any measurable binding to IL-17A, indicating that binding occurs through the HAP peptide domain (Fig. 4B and S4C,D). In the opposite configuration, HAP-A96 was bound to a sensor chip, followed by injection of IL-17A solutions. This orientation resulted in much higher signal to noise and was used to quantify the binding affinity of IL-17A and HAP-A96. The association rate constant (ka) was determined to be 2.11E5 1/Ms. and the dissociation rate constant (kd) was 1.55E-4 1/s, resulting in an equilibrium constant of 734.2 pM (Fig. 4C). BSA, however, showed no measurable interaction (Fig. 4D). These data confirm that fusion to ELP does not prevent HAP’s ability to bind IL-17A, in agreement with our AlphaFold3 predictions. To confirm binding in an orthogonal assay, the capacity of these constructs to deplete IL-17A from solution was examined using coacervation-based enrichment. Among the tested constructs, only HAP-V96—capable of both LLPS and IL-17A binding—effectively removed IL-17A from the supernatant (Fig. 4E). V96 underwent LLPS but lacked affinity for IL-17A, while HAP-A96 bound IL-17A but did not phase separate under the assay conditions. Notably, HAP-A96 did not interfere with detection antibody binding in a commercial IL-17A ELISA, indicating that its epitope does not overlap with the detection region. Furthermore, HAP-V96 mixed with human serum either prepared fresh or incubated at 37 °C for 3 days both retained their ability to deplete IL-17A from supernatant and enrich it in the pellet (Supplementary Fig. S5). An IL-17A/ IL-17RA TR-FRET inhibition assay kit (ACROBiosystems) was also used to quantify inhibition of IL-17A/IL-17RA binding by HAP-ELPs under various conditions (Supplementary Fig. S6). Finally, Rho-HAP-V96, but not Rho-V96 was shown to enrich IL-17A on the surface of coacervate droplets when imaged by confocal microscopy using a FITC-labelled anti-IL-17A antibody for detection (Supplementary Fig. S7).

Fig. 4.

Binding to and depletion of IL-17A by HAP-ELP. (A) Representative surface plasmon resonance (SPR) response curves of IL-17A ligand with HAP-A96 and HAP-V96 analytes showing concentration dependent binding. (B) SPR response curves of IL-17A ligand with ELP control analyte showing a lack of a binding response, indicating HAP is critical for interaction with IL-17A. (C) SPR response curve of HAP-A96 ligand against various analyte concentrations of IL-17A, indicating a KD of 734 pM. (D) SPR response curve of HAP-A96 ligand against BSA analyte, showing a lack of binding as a control for IL-17A. (E) Depletion of IL-17A from supernatant by HAP-V96, but not by HAP-A96, V96, or PBS after heating to 37 °C and centrifugation as measured by enzyme-linked immunosorbent assay (ELISA). Only HAP-V96 both binds IL-17A and undergoes phase separation at 37 °C, resulting in sequestering of IL-17A in the pellet. A global ANOVA of the log of the measured concentration followed by Tukey’s multiple comparisons test showed significance (p < 0.0001) in all comparisons between HAP-V96 and other groups with the same initial IL-17A concentration.

3.5. In vitro inhibition of IL-17A

Inhibition of IL-17A signaling was further assessed using an engineered HEK293 cell line that overexpresses IL-17RA and IL-17RC and secretes SEAP under the transcriptional control of an NF-κB promoter. Treatment with recombinant IL-17A induced a robust, dose-dependent SEAP response, with an EC50 of 267 pM (Fig. 5A). A comparable concentration of 250 pM was used in subsequent inhibition assays to determine the half-maximal inhibitory concentration (IC50) of each construct. HAP-A96 and HAP-V96 inhibited IL-17A signaling with IC50 values of 26.36 nM and 33.42 nM respectively, whereas control ELPs (A96 and V96) lacking the HAP peptide showed no inhibitory activity (Fig. 5B).

Fig. 5.

In vitro inhibition of IL-17A in engineered HEK293 cells. (A) IL-17A dose response curve of HEK-Blue IL-17A cells (Invitrogen) which secrete alkaline phosphatase in response to IL-17A treatment, allowing a colorimetric substrate to measure IL-17A-dependent activation. EC50 for this response was found to be 0.267 nM IL-17A. Error bars represent SD. A sigmoidal (B) HAP-ELP dependent inhibition of IL-17A response following 0.250 nM treatment of HEK-Blue IL-17A cells. IC50 was determined to be 26.36 nM and 33.42 nM for HAP-A96 and HAP-V96 respectively, while no inhibition was observed from ELP controls. Sigmoidal 4PL fits were used to determine EC50 and IC50 values and are shown with 95% confidence bands. All error bars represent SD. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

3.6. Pharmacokinetics of HAP-ELPs

Pharmacokinetic behavior was assessed by administering rhodamine-labelled HAP-ELPs via supra-LG injection in male BALB/c mice. All injected doses were below 0.05 EU/kg bodyweight, as confirmed by HEK-Blue LPS analysis (Invivogen). IVIS imaging of Rho-HAP-V96 injected animals showed stable, persistent depots at the injection site, as visualized by longitudinal IVIS imaging (Fig. 6A). Quantification of this data revealed that an average of 52% of the fluorescent signal from Day 0 was observed on Day 12 for HAP-V96, while HAP-A96 became undetectable after two days (Fig. 6B, Supplementary Fig. S8). Analysis of non-compartmental analysis of IVIS data resulted in an estimated mean residence time (MRT) of 12.3 h for HAP-A96 compared to 81 h for HAP-V96 following supra-LG administration (Table 2). The rapid relative absorption of HAP-A96 was also reflected in the plasma concentrations following supra-LG administration, with HAP-A96 having a detectable Cmax at 0.35 μM at 4.3 h, while HAP-V96 remained below the LLOQ at all timepoints (Fig. 6C). Intravenous (IV) bolus administration of Rho-HAP-A96 and subsequent tail nick sampling was used to determine a clearance rate of 1.89 ± 0.30 mL/h, AUC of 2.69 ± 0.47 μLxh, and a t_1/2 of 16.55 ± 5.45 h (Fig. 6D). By comparison with the IV administration, it was determined that HAP-A96 has a bioavailability of 27.7% and a mean absorption time (MAT) over 2 days following supra-LG administration (Table 2).

Fig. 6.

Pharmacokinetics of HAP-ELPs. (A) Representative timecourse IVIS images showing ELP-dependent absorption of HAP-ELPs from the supra-lacrimal gland (supra-LG) injection site. HAP-A96, which remains soluble at body temperature, is depleted from the gland more rapidly than the HAP-V96, which undergoes phase separation. (B) Quantification of photon flux at the injection site from timecourse IVIS images. At Day 12, an average of 52% of the initial signal is observed for HAP-V96 injected mice, while HAP-A96 was undetectable. (C) Absorption of HAP-ELPs into circulation following supra-LG administration. HAP-A96 is quickly absorbed with a high average Cmax of 0.71 μM and early average Tmax of 4.25 h, while HAP-V96 is absorbed more slowly with an average Cmax of 0.001 μM and Tmax of 20 h. (D) Plasma concentration of HAP-A96 following IV bolus administration with an inset showing early timepoints. All error bars represent SD.

Table 2.

Pharmacokinetic parameters of HAP-ELPs determined through various routes of administration.

	HAP-A96	HAP-V96
Supra-lacrimal gland Administration
Dose (nanomoles)	37.5	37.5
Mouse weight (g)	29.3 (2.2)	30.3 (1.3)
IVIS parameters
AUC (signal x h)	1510 (30)	16,000 (8400)
AUMC (signal x h2)	18,500 (800)	1,300,000 (800,000)
MRT (h)	12.3 (0.6)	81 (49)
Plasma parameters
MRTLG (h)	67.8 (17.9)	*
AUCLG (μM x h)	4.7 (0.4)	*
CL/F (mL/h)	8.1 (0.8)	*
Cmax (μM)	0.35 (0.03)	*
Tmax (h)	4.3 (2.5)	*
t1/2 (h)	22.0 (3.6)	*
Intravenous Bolus Administration
Dose (nanomoles)	5	*
Mouse weight (g)	29.3 (1.5)	*
Plasma parameters
Vd (mL)	14.9 (8.3)	*
Vss (mL)	23.9 (13.4)	*
MRTIV (h)	10.4 (5.3)	*
AUCIV (μM x h)	2.3 (0.2)	*
CL (mL/h)	2.2 (0.2)	*
t1/2 (h)	19.2 (6.2)	*
Combined Parameters
MAT (h)	57.4 (18.7)	*
Bioavailability (%)	27.7 (2.5)	*

Values are indicated as mean (SD).

^*Measured plasma concentrations were below the lower limit of quantification

3.7. Cytotoxicity and in vivo innate and adaptive immunogenicity of HAP-ELPs

Safety and immunogenicity were evaluated in BALB/c mice by measuring plasma levels of CRP following supra-LG injection of HAP-ELPs. All injected doses were below 0.05 EU/kg bodyweight, as confirmed by HEK-Blue LPS analysis (Invivogen). As a positive control, mice treated with lipopolysaccharide (LPS), a potent bacterially-derived immunostimulant, exhibited marked increases in plasma CRP. Conversely, no CRP elevation was observed in animals treated with HAP-ELP at therapeutically relevant doses (Fig. 7A). Additionally, mice dosed twice over 28 days with HAP-V96 or V96 did not generate antibodies (IgG, IgA, or IgM) specific to the inoculation material (Fig. 7B). Detection of a spiked-in anti-ELP IgM antibody in mouse plasma (10 μg/ mL) as well as the ability to generate a standard curve using ELP-coated plates and AK1 antibody confirmed the sensitivity of the ELISA (Supplementary Fig. S9). Histological analysis of LGs by H&E staining revealed no inflammation or tissue damage in the target tissue adjacent to the injection site (Fig. 7C). In vitro, none of the constructs demonstrated cytotoxicity in either Jurat or HEK-Blue IL-17A cells up to at least 10 μM in an LDH assay (Supplementary Fig. S10).

Fig. 7.

In vivo evaluation of innate and adaptive immune response following supra-lacrimal gland (supra-LG) administration of HAP-V96 and V96. (A) Enzyme-linked immunosorbent assay (ELISA) determination of CRP measured in serum obtained 16 h after administration of 150 μL PBS, 2 μg/mL LPS, or 250 μM HAP-V96 or V96. (B) Anti-drug antibodies (ADAs) measured in serum after two supra-LG doses by ELISA, using the inoculation material as the capture reagent. Serum from PBS injected mice spiked with 10 μg/mL AK1, an anti-ELP antibody, was used as a positive control. For both experiments, statistical analysis was performed using a one-way ANOVA followed by Dunnett’s test compared to serum from PBS-injected mice. (C) Representative H&E sections collected from mice from the multi-dose ADA experiment. Blood vessels (arrowheads) and ducts (arrows) are surrounded by acinar structures, with no discernable changes in typical markers of an immune response, such as vasodilation, fibrosis, or cellular infiltrates. Scale bars: top = 200 μm, bottom = 50 μm.

4. Discussion

This work was motivated by the need for local IL-17A inhibition to treat local, tissue-specific inflammation seen in a number of chronic autoimmune conditions, including LG dysfunction in Sjögren’s Disease. Local therapy is preferable to the existing systemically delivered IL-17A neutralizing biologics available on the market, as a local treatment could mitigate side-effects, including the increased rate of infection seen in patients receiving the currently available, systemic IL-17A inhibiting therapies. Moreover, local therapy may improve drug bioavailability in tissues affected by IL-17A-overexpression, such as the exocrine glands in Sjögren’s Disease, where biologics are unlikely to accumulate when administered systemically. To this end, we hypothesized that fusion of an IL-17A inhibiting peptide, HAP, to ELPs, which can be engineered to undergo LLPS and thereby influence absorption kinetics, could be the basis for a novel therapy to treat IL-17A driven chronic inflammation in a tissue-specific manner.

Previous findings have shown that the fusion of ELP and other carrier proteins can impact the activity of a therapeutic peptide or protein, which was considered a potential risk to our proposed therapeutic strategy. Given the recent development of accessible and accurate protein folding and interaction prediction tools, we set out to develop a user-friendly in silico analysis pipeline to screen for carrier protein fusion strategies that are predicted to negatively impact bioactivity. Alpha-Fold3 was the obvious choice in running our simulations, because it can predict not only structure, but also protein-protein interactions, based on sequence alone. While workarounds can be developed, such as using a single sequence with a long, unstructured linker between the sequence of two proteins whose interaction is to be tested, only AlphaFold3 and AlphaFold2-Multimer do this natively. [36,42,43]

Our basic approach was to simulate the interactions of HAP and HAP-ELP fusions with a homodimer of IL-17A, and to compare the position of HAP peptide domain in the simulations with those of the solved crystal structure of the interaction obtained from the Protein Data Bank (PDB: 5HHV). This comparison was performed by aligning the simulated IL-17A homodimer with the cytokine in the reference structure, then measuring the distances between corresponding amino acids in the HAP sequence. Importantly, the close agreement of the IL-17A in the simulated interaction with the reference IL-17A, which always had an RMSD below 0.53 Å, allowed for the binding poses to be compared. AlphaFold3 was able to successfully recapitulate the binding pose of HAP and the IL-17A homodimer. This is not surprising as the solved crystal structure of the interaction (PDB: 5HHV), being deposited to the Protein Data Bank in 2016, was likely a part of the AlphaFold3 training set. Nevertheless, inclusion in the training set does not guarantee accurate prediction, so we were pleased to find that AlphaFold3 was able to predict the binding pose of HAP (Fig. 1A, D). Furthermore, the simulated HAP peptide included the addition of an N-terminal methionine and a C-terminal glycine, to match the expressed HAP-ELP constructs, suggesting that the inclusion of these residues does not impact binding. In contrast, scrambling of the sequence did disrupt the binding, as reflected in the significant change in RMSD compared to the reference. We also explored differences in simulated binding of N-terminal ELP fusion to HAP compared to C-terminal fusion. This was motivated by previous work showing fusion order can play a significant role in preserving activity. However, the simulation results show that AlphaFold3 does not predict the orientation of fusion to impact binding affinity significantly.

Since our ELP-encoding plasmids were designed for convenient ligation at their N-terminus, we constructed plasmids for two HAP-ELP fusions: HAP-A96 and HAP-V96. V96 has previously been characterized as undergoing LLPS below physiological temperatures, while A96 remains soluble. These constructs are appropriate for comparison, given their differential thermosensitive properties and similar molecular weight. Furthermore, both valine and alanine amino acids are present in the disordered domains of the human tropoelastin sequence, which may make these synthetic ELPs less immunogenic. [44–46] The HAP-ELP constructs were successfully ligated and purified through inverse transition cycling, the standard method of ELP purification, resulting in single bands when run on SDS-PAGE and the expected molecular weight measured by MALDI-TOF (Fig. 2). As expected, HAP-V96 transitioned below physiological temperature, while HAP-A96 did not phase transition below 37 ° C at any of the concentrations tested (Fig. 3A–B). Interestingly, fusion of HAP to ELP shifted transition temperature down at all concentrations, but did not significantly affect the slope of the phase boundary. While we did not explore the cause of this behavior, similar shifts have been observed in other ELP fusions. [47,48] A paper by Luo et al. describes how ELP fusion with collagen-like peptide dramatically reduced Tt from 80 °C to <4 °C, due to increased local concentration of ELP caused by collagen-like-peptide triple-helix formation. [49] We hypothesize that transient interactions of the HAP domain with other HAP peptide domains or with ELP on other molecules may increase the propensity for coacervation to occur. The transience or weakness of these interactions can be inferred by the modest decrease in T_t observed in our fusions and more explicitly seen in the similar size of HAP-ELP and ELP monomers as measured by DLS at 17 °C (Fig. 3D).

In line with our AlphaFold3 predictions, both HAP-A96 and HAP-V96 were able to bind and inhibit IL-17A. Using IL-17A as the ligand, we observed concentration-dependent binding with both HAP-ELPs, but not with ELP controls, showing that the HAP peptide domain of HAP-ELPs is responsible for binding with IL-17A, and that both constructs have activity. Similarly, using HAP-A96 as the ligand, we observed binding to IL-17A, but not BSA. While both ligand-analyte orientations of the HAP-ELP and IL-17A interaction showed binding, we report the KD from the experiment with immobilized HAP-ELP and IL-17A as the analyte, due to higher quality sensorgrams (Fig. 4A–D). The observed KD of 734 pM is within an order of magnitude of the K_D between HAP peptide and IL-17A reported by Liu et al. (2.6 ± 0.5 nM) when the peptide was initially developed. The opposite orientation, where IL-17A was immobilized, resulted in higher signal-to-noise ratio and kinetic constants that approached the limits of what could be measured by the instrument. This may be due to movement of the intrinsically disordered ELP domains causing fluctuations in the refractive index. The sensorgrams from all concentrations tested in this orientation, as well as relevant calculated kinetic constants are shown in Supplementary Fig. 2. To confirm binding in an orthogonal assay, we demonstrated that HAP-ELP can be used to sequester IL-17A from solution in a temperature dependent manner. Specifically, solutions of IL-17A were prepared at various concentrations, and mixed with PBS, HAP-A96, HAP-V96, or V96. These mixtures were then heated to 37 °C and centrifuged. This resulted in HAP-V96 and V96 coacervating and forming a pellet. As expected, only HAP-V96 depleted IL-17A from the supernatant (Fig. 4E). While coacervation was observed for V96, it was not able to bind IL-17A and therefore did not impact IL-17A concentrations in the supernatant. Conversely, HAP-A96 theoretically bound IL-17A but did not undergo phase separation, which would allow it to be pelleted, which also led to no changes in supernatant concentration compared to PBS. This elegantly demonstrates the dual IL-17A binding and thermosensitive nature of our fusion proteins, which was further confirmed through confocal imaging, which showed IL-17A was enriched on coacervate droplets of HAP-V96, but not V96 (Supplementary Fig. S7).

In addition to confirmation of binding to IL-17A, we performed in vitro assays to demonstrate that binding of IL-17A led to its inhibition. To accomplish this, we used the HEK-Blue IL-17 cell line (Invivogen), a HEK 293-derived cell line that is engineered to constitutively express the IL-17A receptor complex, comprised of IL-17RA and IL-17RC, and the Act1 adaptor protein, critical for intracellular signal transduction, which has been used to quantify inhibition of IL-17A signaling in prior studies. [50–52] The cells are also modified to express secreted embryonic alkaline phosphatase (SEAP) under the transcriptional control of AP-1 and NF-κB, which are activated by IL-17A. SEAP levels are then quantified by mixing the cell culture supernatant with a colorimetric substrate. We established an IL-17A response curve and determined an EC50 value, which guided our treatment conditions for HAP-ELP IC50 determination. Importantly, this has greater physiological relevance than the binding and depletion assays, as it demonstrates that HAP-ELP binding to IL-17A results in inhibition of signaling through cell surface-expressed receptor. In line with simulated AlphaFold3 experiments and SPR experiments, no meaningful differences in inhibitory activity were found between HAP-A96 and HAP-V96.

As expected, the transition temperature of HAP-ELPs strongly influence absorption kinetics following local administration. Rhodamine-labelled HAP-ELPs were administered via a supra-LG injection, in the subcutaneous space between the ear and the eye. Timecourse imaging of the supra-LG depots showed faster decrease of fluorescent signal in the injection site in HAP-A96 injected mice, leading to undetectable signal at 5 days, indicating rapid absorption. In contrast, HAP-V96 injected mice showed an average of 52% of the initial signal at day 12 (Fig. 6A–B). This confirms that fusion of HAP to an ELP with a T_t below physiological temperature leads to longer depot residence time. IVIS signal was normalized to signal on day 0 to account for differences in apparent initial concentration, which is likely an artifact of spreading between supra-LG administration and acquisition of the first IVIS image which occurred at 4 h, in order to allow for frequent tail vein sampling immediately after injection. As can be seen in Fig. 6A, the initial signal of HAP-A96 is significantly higher than that of HAP-V96 on day 0; we believe this to be a result of rapid spreading from the injection site, including toward the skin surface, which would increase signal. The non-normalized data, which shows high initial signal and rapid clearance for HAP-A96 and low initial signal, but strong retention of HAP-V96, can be found in Supplementary Fig. S8. Mechanistically, ELP-driven LLPS results in the formation of coacervate droplets, which are harder to clear than HAP-A96, which remains soluble even after injection. Tail nick blood sampling following local injection also reflected this effect, with rapid clearance in HAP-A96 animals resulting in an earlier Tmax and higher Cmax compared to HAP-V96 (Fig. 6C). Systemic clearance of HAP-ELP was assessed using HAP-A96 as a proxy for both constructs, as HAP-V96 is expected form large coacervate droplets following IV bolus administration at a dose appropriate for reliable concentration measurements in the plasma, which is not representative of HAP-V96 being absorbed from the injection site. These data were used to calculate the mean absorption time and bioavailability of both HAP-ELP constructs (Table 2), which further reflect the impact of LLPS on absorption. The low bioavailability of HAP-V96 is beneficial to treatment, as it indicates lower propensity for off-target tissue effects.

Finally, the immunogenicity of our lead candidate therapeutic, HAP-V96 was assessed in vivo. This is critical as activation of the immune system would be directly counterproductive to the desired anti-inflammatory effect. The HAP peptide was developed in phage display, and therefore carries a non-endogenous sequence to both humans and mice. Furthermore, the HAP-ELP fusions were expressed in bacteria and insufficient purification may elicit an inflammatory response. Our established endotoxin removal protocol of incubation in Detoxi-gel resin (ThermoFisher), followed by filtration through Mustang E endotoxin removal syringe filter consistently resulted in formulations that allowed us to inject below the 5 EU/kg limit mandated by the FDA for injectable products. This purification resulted in a lack of innate immune response, as measured by serum CRP 16 h after injection. [53] To test whether our formulations stimulated the adaptive immune system, which would lead to the generation of ADAs, we developed an ELISA to measure serum IgG, IgA, and IgM specific to our inoculation material. HAP-V96, V96, or BSA was coated onto high binding ELISA plates and incubated against serial dilutions of a mouse anti-ELP antibody, AK1. Bound anti-ELP antibody was quantified using an HRP-linked antibody that detects mouse IgG, IgA, and IgM, and incubated with HRP substrate to develop a standard curve (Supplementary Fig. 3). As expected, BSA coated plates captured significantly less AK1. The secondary antibody was selected to detect a wider range of potential immunoglobulins developed in response to the administration of our formulations. Serum from mice inoculated with a multi-dose regimen of HAP-V96, V96, or PBS were assayed using this ELISA, which revealed that no ADAs were developed, while AK1 was detectable when spiked into serum using this assay. These data show that purified HAP-ELPs prepared by our standard protocol are safe for injection and do not engage the innate or adaptive immune system.

5. Conclusion

We leveraged the openly available protein-folding AI from Deep-Mind, AlphaFold3, to confirm that our proposed HAP-ELP fusions are not predicted to interfere with the binding and inhibitory activity of HAP. This workflow is simple and accessible even for those without experience in molecular dynamics simulations and may be useful for those hoping to design and test novel fusion proteins. Using recombinant DNA techniques, HAP was fused to two elastin-like polypeptides, A96 and V96, one with a Tt above and the other with a Tt below physiological temperature. The resulting in fusion proteins, HAP-A96 and HAP-V96, were shown to have distinct depot dissolution kinetics following a local supra-lacrimal injection in mice as a result of their differing transition temperatures, indicating the critical role of ELP coacervation in modulating depot residence time. Our results demonstrate the possibility for extended local therapy using HAP-V96.

It was further confirmed that both fusion proteins retain the ability to bind and inhibit IL-17A in vitro, in line with our in silico results. As such, these fusion proteins demonstrate the utility of peptide-ELP fusions in endowing thermosensitive depot-forming properties to peptides without compromising their biological activity. Importantly, despite being expressed in bacteria, our fusion proteins do not activate the innate or adaptive immune system significantly, which is critical for an anti-inflammatory therapeutic. This novel class of ELP-mediated local cytokine blockade is therefore an attractive potential alternative to the existing systemically administered IL-17A targeting biologics in treating tissue-specific symptoms of autoimmunity, such as chronic inflammation of the LG in SjD.

Abbreviations:

SjD

Sjögren’s Disease

HAP

high affinity peptide

ELP

elastin-like polypeptide

MD

molecular dynamics

Tt

transition temperature

LG

lacrimal gland

ITC

inverse transition cycling

supra-LG

supra-lacrimal gland

DLS

dynamic light scattering

LLPS

liquid-liquid phase separation

CRP

C-reactive protein

LPS

lipopolysaccharide

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.jconrel.2026.114890.

Data availability

Data will be made available on request.