Beyond the Snapshot: How Gator Bio BLI Complements Structural Biology

Beyond the Snapshot: How Gator Bio BLI Complements Structural Biology

Author: Benjamin Osborn, PhD

Introduction

Biolayer interferometry (BLI) is a powerful addition to the structural biologist’s toolkit, complementing snapshot techniques like X-ray crystallography and cryo-EM with real-time binding data. BLI’s real-time measurements also don’t require expensive isotopic or atomic labels that drive techniques like NMR and HDX-MS, making it an ideal tool for difficult-to-produce proteins or for studying many similar proteins in succession. However, BLI is not a single-molecule technique – it requires an interaction partner to characterize these protein-protein dynamics. Ten published studies below used Gator BLI alongside the traditional techniques. The examples below start with BLI’s most established structural use, mapping which residues carry an interaction, and finish with its newest. Section 4 covers two papers in which BLI was the first wet-lab measurement a computationally designed binder was subjected to, including the validation set behind BindCraft.

For a more thorough treatment of the methods used, consult the original publications.

The ten papers at a glance

---

1 – Mapping interaction surfaces and binding residues

Studying binding interfaces is one of the most common applications of Gator BLI. Often, this takes the form of epitope binning, where researchers run two binders against the same target in sequence and see whether the first blocks the second. However, as crystal structures and in silico approaches grow more sophisticated, mutagenic screens are commonly used to build a map of which residues are critical to the interaction. Protein-protein pairs are the most common, but Gator BLI can be used to study interactions with other biomolecules as well, such as DNA/RNA [see 1-3].

[1-1] – HIV-1 Integrase must reach the nucleus to integrate viral DNA into the host genome, and Importin 7 had long been implicated as a carrier, with two arginine/lysine-rich motifs in Integrase’s C-terminal domain identified as the likely binding site. Researchers at the Hebrew University of Jerusalem characterized the interface directly, combining HDX-MS and crosslinking MS to map contact regions, AlphaFold3 to model the complex, and Gator BLI to supply affinity data. Integrase CTD was immobilized on amine-reactive biosensors via EDC/NHS coupling, and Importin 7 was titrated as analyte across a broad concentration range, with affinities determined by steady-state analysis. Wild-type Integrase CTD bound Importin 7 with a K_D of 17.9 nM, while a mutant carrying two substitutions in the second nuclear localization signal (NLS) motif (RRKAK → RAAAK) bound poorly at 299 nM, confirming that the NLS is key to high-affinity binding (See Figure 1).

Figure 1, comprising Figure 5A and 5B from the above publication, showing the steady-state binding of HIV-1 integrase (WT and RAAAK mutant) with Importin 7. Adapted from Bana J, Yariv A, Oppenheim T, et al, FEBS Open Bio. 2026. doi:10.1002/2211-5463.70294, used under CC BY 4.0.

[1-2] – Crimean-Congo hemorrhagic fever (CCHF) is a tick-borne viral disease with a case fatality rate reported up to 40%, and the WHO has flagged CCHFV as a public health risk lacking approved vaccines or therapeutics. An earlier study identified 13G8, a mouse monoclonal antibody targeting the viral glycoprotein GP38 that protects mice from lethal viral challenge. This is notable because GP38 is not a neutralization target, leaving binding data as the primary in vitro readout. These researchers isolated anti-GP38 antibodies from a human CCHF survivor and used a BLI competition assay to determine which antigenic sites they engaged; three of seven competed directly with 13G8, and crystal structures confirmed that the survivor-derived CC5-17 shares that epitope while approaching it offset by 22° (Figs 2D, 3A). To test how much sequence variation the shared site could tolerate, the team used Gator BLI to screen point mutants (Fig 3E), introducing residues from the related Aigai virus into CCHFV GP38 to abolish binding and reciprocally introducing CCHFV residues into Aigai GP38 to restore it. Most single substitutions barely affected 13G8 binding, suggesting several changes would have to occur simultaneously for viral escape, while a single G296K substitution significantly reduced CC5-17 affinity, consistent with the steric clash predicted by its shifted binding angle. In terms of disease treatment, the human-derived CC5-17 bound GP38 far more tightly than 13G8 yet conferred lower protection in mice, which the authors attribute in part to its lower association rate.

Figure 2, from Supplementary Figure 4 in the above publication, showing binding curves for 13G8 mouse-derived antibody (panels a-e) and CC5-17 patient-derived antibody (panels f-i) binding to viral GP38 from various strains. Adapted from Durie IA, Tehrani ZR, Karaaslan E, et al, Nature Communications. 2022;13:7298. doi:10.1038/s41467-022-34923-0, used under CC BY 4.0.

[1-3] – Not every useful binding pair is protein-protein. J2 is the gold-standard antibody for detecting double-stranded RNA in cells, spotting viral replication intermediates, and screening mRNA therapeutics for inflammatory contaminants. Despite three decades of use, nobody had defined what it binds. Researchers at NIDDK used BLI to perform the bulk of that characterization effort. Running J2 IgG as ligand on Gator Bio Protein A biosensors with nucleic acids as analytes, they showed J2 binds dsRNA robustly while dsDNA, ssRNA, ssDNA, and RNA-DNA hybrids gave no appreciable signal, which means that J2 doesn’t detect cellular R-loops (Fig 1C, 1D). Before turning to mutagenesis they ran a series of dsRNA lengths, which identified the practical binding threshold at roughly 14 bp, well below the ~40 bp previously estimated by atomic force microscopy (see Figure 3, below). Based on a dsRNA/J2 co-crystal structure, the team mutated several CDR residues and used BLI to measure the effect. Two tyrosine residues proved essential, with Y50A and Y52A sharply reducing binding and Y50F and Y52F barely registering a change, pointing to van der Waals contacts from the aromatic rings in the minor groove of the dsRNA, rather than the phenolic hydroxyls. They also tried swapping in basic residues at N101 and S33 on the theory that more positive charge should help bind a phosphate backbone. Both compromised binding rather than improving it. Finally, the team checked whether AlphaFold3 could reproduce the complex. It predicted the J2 Fab and its CDR loops accurately, landing within about 1 Å of the crystal structure, but it placed the dsRNA nowhere near the true epitope, which BLI had already established. The authors attribute the miss to how few antibody-nucleic acid interfaces exist in the PDB for AlphaFold3 to train on.

Figure 3, from Figure 1I in the above publication, showing binding curves for different lengths of dsRNA, and 1J, bottom right, showing steady-state binding curves for each. Adapted from Bou-Nader C, Juma KM, Bothra A, et al, Nature Communications. 2026;17:635. doi:10.1038/s41467-025-67414-z, used under CC BY 4.0.

---

[BACK TO TOP]

2 – Tracking binding across variants and species

Beyond locating a binding site, Gator BLI can track how that site changes over time as a species mutates, or how binding affinity changes between different species. BLI’s plate-based format allows you to run many binders against many targets in rapid succession without having to prep the samples multiple times. The papers below cover viruses evading host antibodies and viruses acquiring the ability to infect new hosts.

The same set of experiments is run in industrial drug development. Before a therapeutic antibody goes into toxicology studies, you have to show that it binds the target in the model species with an affinity close to human. Otherwise you cannot read much into the tox results. You can do all of that on one plate. You run the antibody against the human, cynomolgus, mouse and rat orthologs, and you use the same biosensor chemistry for all four.

[2-1] – SARS-CoV-2 spilling over into animal hosts had been a concern since the start of the pandemic, and mice are a particular worry as a possible host. Mouse ACE2 (mACE2) doesn’t bind ancestral Spike well at all. Researchers at EPFL wanted to know whether the mutations that were driving immune escape in humans were simultaneously expanding host range. They captured dimeric ACE2-Fc on Gator Mouse Fc biosensors and titrated trimeric Spike as the analyte. An initial screen across cat, dog, and mink showed that each of these other species bound wild-type Spike protein. In contrast, mouse ACE2 showed almost no binding in the original Wuhan strain but later variants of concern bound far better (Fig S1). Full titrations showed that wild-type Spike barely bound to mACE2 even at 300 nM, while Alpha, Beta, and Gamma came in between 25 and 45 nM, and Omicron was around 2-3 nM (See Figure 4). Fig S10 features a mutation screen that introduced point mutants N501Y, K417N, or E484K on their own into the WT Spike background and found that single mutants weren’t enough to confer mACE2 binding, but a dual-mutant N501Y/E484K did.

Figure 4, from Figure 1 in the above publication, showing binding curves for various SARS-CoV-2 Spike proteins for ACE2 from mouse and human, alongside the differences in binding residues among several species. Adapted from Ni D, Turelli P, Beckert B, et al, PLOS Pathogens. 2023;19(4):e1011206. doi:10.1371/journal.ppat.1011206, used under CC BY 4.0.

[2-2] – In addition to increasing the affinity of Spike for ACE2 in mice and humans, Omicron also dodged N3-1, a potent antibody isolated from a first-wave COVID-19 patient that had neutralized every variant of concern until then. Only one of Omicron variant BA.1’s fifteen receptor binding domain mutations even contacts the antibody, and substituting that residue back didn’t restore binding (measured using a human Fc Gen II biosensor). Despite carrying fewer mutations than later variants, BA.1 showed a much faster off-rate. This is because a serine-to-proline substitution at position 373 inverts a nearby loop, covering the hydrophobic pocket the antibody reaches into without altering the pocket itself. As a result, the Omicron spikes hold their receptor binding domains down far more often than the ancestral strain, so the arrangement N3-1 needs is rarely exposed. With Strep-Tactin XT biosensors, the team screened every possible substitution at the clashing residue, running the mutants straight from clarified expression supernatants (with no purification step), and several improved affinity, but no amount of engineering of the site of interest overcame the larger conformational rearrangement. Strep-Tactin XT is exclusive to Gator Bio. Direct capture from cell-free and crude expression removes the purification step, and it may explain why three 2026 AI binder-design preprints used Gator Bio platforms to run their binding assays: BoltzGen, BoltzProt-1, and Candido et al. on protein language models.

Figure 5, from Figure 4A in the above publication, showing patient-derived N3-1 binding to wild-type (WHU1) versus Omicron SARS-CoV-2 strains, illustrating how Omicron escapes antibody binding. Adapted from Goike J, Hsieh CL, Horton AP, et al, Communications Biology. 2023;6:1250. doi:10.1038/s42003-023-05649-6, used under CC BY 4.0.

---

[BACK TO TOP]

3 – Membrane proteins and lipid environments

Where previous sections dealt with soluble proteins, membrane proteins are notorious for being difficult to characterize, and they’re conformationally coupled to the system they’re in, such as nanodiscs [3-1, 3-2] or detergent micelles [3-3]. Membrane proteins are also expensive to prepare and problematic to purify, meaning low yields and a product that is often too fragile to push through a fluidic system. Gator BLI needs 40 µL per well, so a difficult prep goes further. XT Optics gives 3x signal to noise on top of that, which is what turns a low-shift nanodisc measurement into useful results. Together they let you hold a membrane protein in a pseudo-native conformation and still resolve analyte binding against it.

[3-1] – The team at the Weis lab at Stanford developed a way to ensure completed complexes are present in the same nanodisc with a split GFP tether. Quantitative Western blotting confirmed about two Frizzled4 per nanodisc for the homodimer and one Frizzled4 plus one LRP6 for the heterodimer. To show functionality, they used BLI. Biotinylated nanodiscs were immobilized on streptavidin biosensors, and Norrin bound the Frizzled4 homodimer at 0.46 nM while DKK1 bound the Frizzled4–LRP6 heterodimer at 5.6 nM. Then, to measure the kinetics of Dishevelled-2 DEP domain against nanodisc-immobilized Frizzled4-LRP6 heterodimers, they used Gator Bio XT Optics-enabled SMAP biosensors, capped at a 10 nm shift to avoid crowding artifacts from packing nanodiscs too densely. They ran repeated association-dissociation cycles on a single sensor, with a ligand-free first cycle followed by walking DEP from 10 nM to 32 µM. To account for any drift or non-specific DEP binding, they ran a matched empty-disc reference sensor against the same samples concurrently on a separate tip. DEP bound at roughly 200 nM with or without Norrin present, showing that heterodimerization doesn’t change how Frizzled4 recruits Dishevelled. This work predates SA XT. For biotinylated nanodisc capture today we would recommend SA XT, which pairs the same streptavidin chemistry with the XT Optics sensitivity gain that makes a low-shift titration like the DEP series easier to resolve.

Figure 6, from Supplementary Figure 4a and 4b in the above publication, depicting Norrin binding to biosensor-immobilized lipid nanodiscs. Adapted from Bruguera ES, Mahoney JP, Weis WI. Journal of Biological Chemistry. 2022;298(4):101628. doi:10.1016/j.jbc.2022.101628, used under CC BY 4.0.

[3-2] – Norrin activates the β-catenin pathway like Wnt does, but it needs a co-receptor, the tetraspanin Tspan12. As the complex forms, do Norrin and Tspan12 interact with each other? Researchers in the Weis lab at Stanford purified Tspan12, reconstituting it into biotinylated lipid nanodiscs, and immobilized those on streptavidin biosensors, and titrated Norrin as the analyte. Norrin bound at 10.4 nM, and the interaction was stronger in the nanodisc than in a GDN detergent micelle. They then modeled the complex with AlphaFold and confirmed the putative interaction site with mutagenesis. Charge reversals on Norrin alone reduced affinity by up to two orders of magnitude, but compensation on the paired binding site (Norrin K102E/R121E with Tspan12 E196K/S199K) restored binding, confirming the interaction site precisely.

Figure 7, from Figure 1 in the above publication, depicting that the Norrin-Tspan12 interaction, and how the LEL (the Large Extracellular Loop) of Tspan12 interacts with Norrin, and that this loop is specific to Tspan12 (Tspan11’s LEL loop did not bind Norrin at concentrations tested). Adapted from Bruguera ES, Mahoney JP, Weis WI. eLife. 2025;13:RP96743. doi:10.7554/eLife.96743, used under CC BY 4.0.

[3-3] – Detergent choice for a GPCR is usually a stability question, and lauryl maltose neopentyl glycol (LMNG) is the most common choice by far, appearing in over half of all solved GPCR structures. Stability for a crystal structure is not the same thing as being in an active conformation. They purified a thermostabilized neurotensin receptor 1 variant into three detergents and measured agonist affinity on a Gator Prime. Instead of immobilizing the receptor, they loaded the small peptide agonist NT (8–13) onto Ni-NTA biosensors through a 6-His tag and titrated the receptor as analyte, which keeps the variable in solution. Affinity came in at 13.7 nM in decyl maltoside, 35.9 nM in dodecyl maltoside, and 94.9 nM in LMNG, meaning that ligand binding and receptor stabilization were, in this case, inversely proportional.

Figure 8, from Figure 2 in the above publication, showing how binding affinity changes depending upon membrane composition from least restrictive detergent structure (top) to most restrictive (bottom) for this GPCR. Adapted from Bower JB, van der Velden WJC, Gomez KP, et al, Protein Science. 2026;35(2):e70475. doi:10.1002/pro.70475, used under CC BY 4.0.

Running membrane proteins in nanodiscs or detergent? Our applications team can work through the immobilization strategy with you before you commit material.

Get in touch with us

---

[BACK TO TOP]

4 – Validating designed structures

Every study so far validated something that already existed in the lab. Structural prediction has advanced to the point that AlphaFold will model almost any complex you ask for, and then design tools like Boltz will hand you a binder for it. But until wet-lab data shows the interaction is real, it’s just a hypothesis.

AlphaFold has shown up several times earlier in this review: [1-1] paired AlphaFold3 with NLS mutagenesis, and [3-2] confirmed an AlphaFold Multimer interface through reciprocal charge swaps. Both aligned with the prediction. [1-3] shows the other outcome, where AlphaFold3 nailed the antibody fold to within 1 Å but placed the RNA nowhere near the real binding site.

De novo binder design is hit-or-miss by nature, which is why many AI-driven discovery programs generate thousands of binders in silico before they even touch a wet lab. The ones that survive often still number in the hundreds, and each one represents a potential data point for retraining and refining the AI algorithm. That’s why BLI, a high-throughput method that can measure kinetics and affinity without a purification step, is often the first wet-lab measurement a new binder sees. We mentioned a few AI-guided binder preprints in our previous blog post; the examples below add two more published works that used Gator BLI to validate de novo designs.

[4-1] – BindCraft is an open-source de novo binder development approach built on an iterative loop with AlphaFold2. The authors report that the method works without high-throughput wet-lab screening, citing experimental success rates from 10% to 100% (averaging 46%) across twelve targets. BLI was the first-pass method for validating these binder designs, with representative traces in Fig 2b, and it also carried the off-target analysis that checks whether a binder is merely sticky or selective. The team captured Fc-tagged PD-1, PD-L1, and IFNAR2, three immunoglobulin-like fold receptors that look alike structurally, on Gator Bio Protein A biosensors and dipped them into each binder at 1 µM. None of the binders cross-reacted.

[4-2] – Pathogenic E. coli steals iron by ripping the heme cofactor out of host hemoglobin. The transporter responsible, ChuA, grabs hemoglobin just long enough to strip the heme and let go, an interaction so transient that the team couldn’t capture the complex by crystallography or cryo-EM. BLI could measure it. With hemoglobin immobilized on Ni-NTA probes and ChuA titrated as analyte, they got a K_D of 71.5 nM with fast on and off rates, which aligns with the bind-extract-release transporter hypothesis. Using AlphaFold, they identified extracellular loops 7 and 8 as possible binding sites for the interaction, and they used RFdiffusion and ProteinMPNN to design binders that would cover those loops. In total, they generated about 20,000 in silico and took their 96 top candidates to the lab for bacterial growth assays. Eight inhibited bacterial growth on hemoglobin agar. BLI put three of these inhibiting proteins at 127, 84.9, and 71.4 nM, while those same inhibitors showed no binding against three unrelated E. coli proteins, confirming the designs hit only their target.

Figure 9, from Figure 5f in the above publication, showing the K_D of the growth-inhibiting proteins for ChuA. Adapted from Fox DR, Asadollahi K, Samuels I, et al, Nature Communications. 2025;16:6066. https://doi.org/10.1038/s41467-025-60612-9, used under CC BY 4.0.

If you are triaging designed binders and want the wet-lab step to keep up with the design step, tell us what you are expressing and at what scale.

Get in touch with us

---

[BACK TO TOP]

Final thoughts

These ten studies used Gator BLI to measure what the snapshot methods could not. ChuA grabs hemoglobin and lets go too fast for crystallography or cryo-EM to capture the complex. BLI measured it at 71.5 nM with fast on and off rates, which is what you would expect if the transporter binds, strips the heme, and releases.

No single Spike substitution was enough to let SARS-CoV-2 bind mouse ACE2. The virus needed both N501Y and E484K. To establish that, the team ran full titrations across five variants on a single plate.

Membrane proteins are harder. In [3-3] the team purified a neurotensin receptor into three detergents and measured a sevenfold range in agonist affinity, and the detergent that stabilised the receptor best gave the weakest binding. Measurements like that produce small shifts, and XT Optics gives you 3x signal to noise to read them.

Designed binders are the newest. In BindCraft and the ChuA work, BLI was the first wet-lab measurement anyone made on a new design. The antibody work in [2-2] is not de novo design, but it ran the same way. The team screened every substitution at one residue directly from clarified expression supernatant, with no purification step, on Strep-Tactin XT biosensors. Strep-Tactin XT is exclusive to Gator Bio.

These ten papers used seven capture chemistries on the same instrument, with no setup change between them. Those were Protein A, SA, Human Fc Gen II, Ni-NTA, Strep-Tactin XT, Amine Reactive and SMAP. That is what the biosensor menu gives you.

If you are working on a structural biology problem, our applications team can help you design the assay around it.

Get in touch with us

---

[BACK TO TOP]

References

[1-1] Bana J, Yariv A, Oppenheim T, et al. Importin 7 mediates the nuclear import of HIV-1 integrase via a specific interacting interface. FEBS Open Bio. 2026. doi:10.1002/2211-5463.70294

[1-2] Durie IA, Tehrani ZR, Karaaslan E, et al. Structural characterization of protective non-neutralizing antibodies targeting Crimean-Congo hemorrhagic fever virus. Nature Communications. 2022;13:7298. doi:10.1038/s41467-022-34923-0

[1-3] Bou-Nader C, Juma KM, Bothra A, et al. Structural basis of double-stranded RNA recognition by the J2 monoclonal antibody. Nature Communications. 2026;17:635. doi:10.1038/s41467-025-67414-z

[2-1] Ni D, Turelli P, Beckert B, et al. Cryo-EM structures and binding of mouse and human ACE2 to SARS-CoV-2 variants of concern indicate that mutations enabling immune escape could expand host range. PLOS Pathogens. 2023;19(4):e1011206. doi:10.1371/journal.ppat.1011206

[2-2] Goike J, Hsieh CL, Horton AP, et al. SARS-CoV-2 Omicron variants conformationally escape a rare quaternary antibody binding mode. Communications Biology. 2023;6:1250. doi:10.1038/s42003-023-05649-6

[3-1] Bruguera ES, Mahoney JP, Weis WI. Reconstitution of purified membrane protein dimers in lipid nanodiscs with defined stoichiometry and orientation using a split GFP tether. Journal of Biological Chemistry. 2022;298(4):101628. doi:10.1016/j.jbc.2022.101628

[3-2] Bruguera ES, Mahoney JP, Weis WI. The co-receptor Tetraspanin12 directly captures Norrin to promote ligand-specific β-catenin signaling. eLife. 2025;13:RP96743. doi:10.7554/eLife.96743

[3-3] Bower JB, van der Velden WJC, Gomez KP, Pan M, Bumbak F, Vaidehi N, Ziarek JJ. Stabilization versus flexibility: detergent-dependent trade-offs in neurotensin receptor 1 GPCR ensembles. Protein Science. 2026;35(2):e70475. doi:10.1002/pro.70475

[4-1] Pacesa M, Nickel L, Schellhaas C, et al. One-shot design of functional protein binders with BindCraft. Nature. 2025;646:483–492. doi:10.1038/s41586-025-09429-6

[4-2] Fox DR, Asadollahi K, Samuels I, et al. Inhibiting heme piracy by pathogenic Escherichia coli using de novo-designed proteins. Nature Communications. 2025;16:6066. https://doi.org/10.1038/s41467-025-60612-9