Identifying and quantifying protein biomarkers is a pressing demand in precision and personalized medicine1. Recent advancements in functional proteomics indicate that there are yet numerous unexplored proteins with potential consequences for the progression of pathological conditions2. For example, the development of various hematological malignancies and solid tumors is associated with changes in specific protein expressions. In general, protein biomarkers are upregulated under oncogenic conditions. Clinical examinations of such biomarkers are instrumental in prognostics, diagnostics, and therapeutics. A persistent challenge in protein detection in biofluids is the occurrence of nonspecific bindings of assay reagents with immobilization surfaces and protein recognition elements, amplifying the background noise. In addition, existing techniques cannot be utilized for a wide dynamic range because of deterioration in the signal-to-noise ratio at low analyte levels. Therefore, there is an increasing necessity to create highly specific and sensitive protein-sensing approaches that employ rapid signal responses to various biochemical stimuli3,4.

Molecular details of protein detection are illuminated using single-molecule methods3,4,5. In particular, single-molecule sensing with nanopores6,7,8 using the resistive-pulse technique9 is adaptable to parallel recording technologies10. In the past decade, substantial progress has been made in peptide and protein analytics using nanopore sensors11,12,13. Nowadays, nanopores are fabricated using a broad range of scaffolds, either in organic or inorganic materials13,14. These sensing elements can identify and quantify peptides at single-amino acid resolution15,16. Moreover, nanopores represent ultrasensitive sensors for revealing critical features of proteins, such as shape and size17, post-translational modifications18,19, enzymatic activity20,21, and mechanical unfolding22,23. In addition, nanopores are utilized to fingerprint proteins using enzymatic degradation24,25. In recent years, these sensors were engineered and formulated for challenging aspects in single-molecule protein sequencing26,27. Despite many benefits, this approach requires the targeted proteins to partition into the nanopore interior. Hence, the detection is conducted under steric restrictions of the nanopore confinement, potentially impairing the strength of specific interactions. If the nanopore diameter is smaller than the hydrodynamic radius of the protein analyte, then this is another obstacle preventing its detection in the natively folded state.

Detecting single proteins outside the nanopore is a practical alternative to sampling the complexity of protein recognition events28,29,30. This task would necessitate an external protein binder (e.g., receptor) covalently attached to a nanopore. However, a transducing mechanism is needed to convert the physical captures and releases of a protein analyte (e.g., its ligand) into a specific electrical signature of the sensor. In addition, changing the system to a different binder-analyte pair requires a lengthy and tedious optimization process that includes amplified difficulties. The heterogeneous architecture, size, charge, and other traits of different binders need extensive protein engineering. This prerequisite is critical for each sensor for a given protein analyte. Earlier studies have suggested that these protein sensors may be limited to established protein fragments of ~100 residues31,32. For example, large protein binders likely induce additional steric constraints, precluding the clearance of the space around the pore opening. Moreover, the interaction interface of the binder must be fully accessible to the protein analyte.

Here, we propose a class of sensing elements for probing proteins at a single-detector precision. These sensors will have an antibody-mimetic protein binder engineered on the tFhuA nanopore31, a monomeric β-barrel scaffold, via a flexible tether. Protein recognition elements, such as proteins31,32, or peptides33, can be covalently tethered at the N terminus of tFhuA through a flexible (GGS)2 peptide tether without the deterioration of its membrane-embedded structure and pore-forming features. Furthermore, this hexapeptide maintains the protein binder in the proximity of the pore opening to acquire a current modulation upon individual binding events. In this study, we apply a similar strategy to develop nanopore sensors as single-polypeptide chain proteins. They are refolded in detergents and reconstituted into a synthetic lipid bilayer at a single-molecule level (Fig. 1). Hence, all data presented in this article are determined at single-molecule precision. We demonstrate that by changing only the binding interface, a different binder-containing nanopore sensor can be obtained and readily implemented into detecting a specific protein. This strategy maintains the sensor’s architecture, high sensitivity, and specificity while featuring its generalization to numerous protein analytes.

Fig. 1: Rational protein design of a generalizable nanopore sensor for protein detection.
figure 1

a A tenth fibronectin type-III domain (FN3)34 or monobody (in green) with the FG, BC, and DE loops highlighted in blue. b The FN3 variants, where cyan, magenta, and yellow were used to represent the binding loops in FN3SUMO45, Mb446, and Adnectin147, respectively. c tFhuA (in red)31, a monomeric β barrel with average internal diameters of ~2.6 and ~3.9 nm, as measured from side chain to side chain. d Functional reconstitution of nanopore sensors into a lipid bilayer. The hSUMO1-specific nanopore is a single-polypeptide unit that comprises a tFhuA, a (GGS)2 tether, and an FN3SUMO monobody (left). WDR5- and EGFR-binding monobodies, Mb4 (center) and Adnectin1 (right), respectively, were also fused to tFhuA in the same way as FN3SUMO. hSUMO1, WDR5, and EGFR are marked in magenta, cyan, and brown, respectively. Δψ indicates the applied transmembrane potential between the cis and trans sides of the membrane. The monobody-analyte complexes are shown as well. The structures of all sensors were predicted by AlphaFold234,35.

In this article, the binder is a monobody, a recombinant protein based on the 94-residue fibronectin type-III (FN3) domain34. Target-specific monobodies are preferable over other antibody-mimetic scaffolds for the following reasons: (i) the monobody has a relatively smaller hydrodynamic radius. This characteristic maintains the tFhuA pore partially open, so it enables a readable output signal through the passage of ions; (ii) the monobody does not partition into the tFhuA pore lumen, and its binding interface is exposed to the aqueous phase for a specific protein target; (iii) monobodies, unlike nanobodies, lack disulfide bridges35. Therefore, they provide an opportunity for straightforward expression, purification, and refolding procedures; and (iv) monobodies can be selected from a broad range of variants against numerous target proteins36. Using monobody-based nanopore sensors with varying binding interfaces, it is possible to detect different proteins that vary substantially in their structural and functional properties. When subjected to a biofluid, this class of sensing elements can report the presence of a protein biomarker at a single-molecule level. This tactic will not only enable overcoming the abovementioned challenges but will also motivate the widespread applications of these sensors.

Here, we develop monobody-based sensors for three targeted analytes: (i) human small ubiquitin-related modifier 1 (hSUMO1), a model protein with implications in various cellular processes, such as DNA damage repair, chromosome dynamics, and cell cycle37,38; (ii) WD40 repeat protein 5 (WDR5)39, a chromatin-associated protein hub involved in the epigenetic regulation of histone 3 lysine 4 (H3K4) methylation. WDR5 is a putative biomarker because it is overexpressed under various oncogenic conditions40,41; and (iii) epidermal growth factor receptor (EGFR), a prognosis protein biomarker in lung, colorectal, and breast cancers42,43,44 (Fig. 1; Supplementary Table 1). Therefore, we create three sensors using FN3SUMO45, Mb446, and Adnectin147 monobodies as binders against hSUMO1, WDR5, and the ectodomain of EGFR, respectively. These monobody-based sensors are denoted by FN3SUMO-tFhuA, Mb4-tFhuA, and Adnectin1-tFhuA, respectively (Supplementary Fig. 1).

Results and discussion

Development of monobody-based nanopore sensors

We connected three distinct monobodies to the N-terminus of tFhuA using our standard linker, (GGS)2. We employed AlphaFold2 to predict the overall three-dimensional conformation of the monobody-tFhuA fusion proteins48,49. Hence, our computational prediction of nanopore sensors allowed us to visualize the orientation of monobodies in these structures. We observed that the predicted Local Distance Difference Test (pLDDT), a confidence score for each residue, was between 80 and 100 for most residues of the FN3-tFhuA (Supplementary Fig. 2a, b). This model illustrates that FN3 orients almost perpendicularly on the central axis of tFhuA (Supplementary Fig. 2c). This finding is likely due to long-range electrostatic interactions between clusters of negative charges on tFhuA β turns and positive charges on FN3 loops (Supplementary Fig. 3). Similar results were obtained with FN3SUMO-tFhuA, Mb4-tFhuA, and Adnectin1-tFhuA (Supplementary Fig. 4). Therefore, FN3 monobodies in all sensors potentially block a substantial ionic flow through tFhuA. Inspecting all sensors at a transmembrane potential of +40 mV revealed a relatively quiet single-channel electrical current recorded with FN3SUMO-tFhuA and Mb4-tFhuA, and a slightly noisy signal acquired with Adnectin1-tFhuA (Supplementary Figs. 5 and 6). The unitary conductance of FN3SUMO-tFhuA, Mb4-tFhuA, and Adnectin1-tFhuA were (mean ± s.d.) 0.81 ± 0.03 nS, 0.99 ± 0.04 nS, and 0.90 ± 0.02 nS (Supplementary Table 2), respectively. These are significant conductance reductions compared to the unmodified tFhuA (1.5 ± 0.1 nS) (Supplementary Fig. 7)33. This finding is in accord with the predictions made by AlphaFold2.

Real-time and label-free detection of hSUMO1 using a single FN3SUMO-tFhuA nanopore sensor

A single FN3SUMO-tFhuA sensor was functionally reconstituted into a lipid membrane at an applied transmembrane potential of +40 mV. The presence of hSUMO1 in the cis compartment at nanomolar concentrations produced frequent current blockades (Fig. 2a and Supplementary Fig. 8) between Oon open-substate and Ooff closed-substate. Their normalized current amplitude, A/I0, was (91.5 ± 0.7)%. Here, I0 and A denote the single-channel current of the hSUMO1-released substate and the current amplitude of hSUMO1-produced current blockades, respectively (Fig. 2b, c). In addition, infrequent and brief current spikes were observed when hSUMO1 was added to the cis side of an unmodified tFhuA-containing bilayer (Supplementary Fig. 9). Taken together, these negative-control measurements indicate that hSUMO1 did not produce any significant current blockades due to nonspecific interactions with the cis opening of the nanopore.

Fig. 2: Real-time and label-free detection of hSUMO1.
figure 2

a Representative single-channel electrical traces (in red) of FN3SUMO-tFhuA in the presence of 0, 65, 130, and 260 nM hSUMO1. Oon and Ooff are the hSUMO1-released and hSUMO1-captured substates, respectively. τon and τoff are the durations of the hSUMO1-released and hSUMO1-captured events, respectively. These single-channel electrical signatures were replicated in n = 3 independent experiments. The applied transmembrane potential was +40 mV. Single-channel electrical traces were further low-pass filtered at 3 kHz using an 8-pole Bessel filter. b A current histogram (in orange) of the Oon substate of FN3SUMO-tFhuA. The current amplitude (mean ± s.e.m.) of the Oon substate was 32.1 ± 0.1 pA. c A current histogram (in red) of the Oon and Ooff substates of FN3SUMO-tFhuA at 65 nM hSUMO1. The current amplitude (mean ± s.e.m.) of the Ooff substate was 2.9 ± 0.1 pA. d Histograms of τon at various hSUMO1 concentrations, [hSUMO1]. τon (mean ± s.e.m.) were 125 ± 4 ms (number of events: N = 349), 69 ± 5 ms (N = 623), and 35 ± 1 ms (N = 1168) at [hSUMO1] values of 65 nM, 130 nM, and 260 nM, respectively. e Histograms of τoff at various [hSUMO1] values. τoff (mean ± s.e.m.) were 15 ± 1 ms (N = 354 events), 16 ± 1 ms (N = 633), and 14 ± 1 ms (N = 1180) at [hSUMO1] values of 65 nM, 130 nM, and 260 nM, respectively. f Dependence of the event frequency in the form of 1on on [hSUMO1]. The slope of the linear fit (in red) of 1on versus [hSUMO1] is the association rate constant, kon, of hSUMO1-FN3SUMO interactions because kon = 1/on[hSUMO1]). g Dependence of 1off on [hSUMO1]. The horizontal line is an average fit (in green) of the (1off) data points recorded for various [hSUMO1] values. Data points in panels f and g represent mean ± s.d. obtained from n = 3 different experiments. Source data are provided as a Source Data file.

Moreover, hSUMO1-captured events were noted as concentration-dependent (Fig. 2a). hSUMO1-released and hSUMO1-captured events recorded with FN3SUMO-tFhuA corresponded to the open-substate, Oon, and closed-substate, Ooff, respectively. However, hSUMO1-captured events were not detectable when hSUMO1 was added to the trans compartment (Supplementary Fig. 10), confirming that tFhuA and its derivatives insert into the membrane with a single orientation50.

Next, we pursued detailed statistical analyses of the hSUMO1-released, and hSUMO1-captured durations, whose mean values were denoted by τon and τoff, respectively. The maximum likelihood method51 and logarithm likelihood ratio (LLR) tests52 were employed to determine the distribution model of these time constants. The duration of hSUMO1-released and hSUMO1-captured events showed a single-exponential distribution in the form of a single-peak in a semilogarithmic plot (Fig. 2d,e). Although the bin size was identical in these histograms, we represented them differently for clarity. Increasing the hSUMO1 concentration, [hSUMO1], decreased the τon but did not alter τoff (Supplementary Table 3). The association rate constants, kon, were consistent for all [hSUMO1] values (Supplementary Table 4). Here, kon = 1/([hSUMO1] τon). In addition, the frequency of hSUMO1-captured events, f, where f = 1/τon, was proportional to [hSUMO1] in a ratio 1:1 (Fig. 2f), indicating a bimolecular association process of the hSUMO1-FN3SUMO complex. Using the linear fit of f([hSUMO1]), we obtained a kon value (mean ± s.e.m.) of (1.12 ± 0.02) × 108 M−1 s−1. The dissociation rate constant koff was determined as the reciprocal of the mean hSUMO1-captured durations (1/τoff). This value was independent of [hSUMO1] (Fig. 2g), suggesting a unimolecular dissociation mechanism of the hSUMO1-FN3SUMO complex. A linear fit of koff([hSUMO1]) versus [hSUMO1] resulted in its mean ± s.e.m. of 74.5 ± 2.4 s−1, to yield an equilibrium dissociation constant (KD) of 665 ± 24 nM (Supplementary Table 5).

Detection of a chromatin-associated protein hub using a single Mb4-tFhuA nanopore sensor

We employed the same approach and experimental conditions to detect WDR5 using a functionally reconstituted Mb4-tFhuA sensor into a lipid bilayer. When added to the cis compartment at nanomolar concentrations, WDR5 produced frequent current blockades (Fig. 3a; Supplementary Fig. 11) between Oon open-substate and Ooff partly closed-substate with a normalized current amplitude of (14 ± 1)% (Fig. 3b, c). Again, this kind of current blockades was not noted when an unmodified tFhuA was exposed to WDR5 added to the cis side (Supplementary Fig. 12) or when Mb4-tFhuA was subjected to WDR5 added to the trans side (Supplementary Fig. 13). These findings suggest that specific WDR5-Mb4 interactions bring about WDR5-induced current blockades. WDR5-released (Oon) and WDR5-captured (Ooff) events also followed a single-exponential distribution (Fig. 3d, e). In addition, the frequency of WDR5-captured events was proportional to the WDR5 concentration, [WDR5] (Fig. 3f), whereas their duration was independent of [WDR5] (Fig. 3e, g; Supplementary Tables 6 and 7). Using linear fits of the functions f([WDR5]) and koff([WDR5]), we obtained a kon value (mean ± s.e.m.) of (0.83 ± 0.01) × 108 M−1s−1 and a koff value (mean ± s.e.m.) of 72.4 ± 3.7 s−1, resulting a KD of 872 ± 45 nM (Supplementary Table 8). It should be noted the kinetics of WDR5-Mb4 interactions undergo fast association and dissociation rates, which were also detected with hSUMO1-FN3SUMO interactions.

Fig. 3: Single-molecule sensing of WDR5.
figure 3

a Representative single-channel electrical traces (in magenta) of Mb4-tFhuA in the presence of 0, 50, 150, and 300 nM WDR5. Oon and Ooff are the WDR5-released and WDR5-captured substates, respectively. τon and τoff are the durations of the WDR5-released and WDR5-captured events, respectively. These single-channel electrical signatures were replicated in n = 3 independent experiments. The other conditions were the same as those in Fig. 2. b A current histogram (in magenta) of the Oon substate of Mb4-tFhuA. The current (mean ± s.e.m.) corresponding to the Oon substate was 39.5 ± 0.1 pA. c A current histogram (in magenta) of the Oon and Ooff substates of Mb4-tFhuA at 50 nM WDR5. The current (mean ± s.e.m.) corresponding to the Ooff substate was 34.5 ± 0.1 pA. d Histograms of the WDR5-released durations on) at various WDR5 concentrations, [WDR5]. τon (mean ± s.e.m.) were 178 ± 6 ms (number of events: N = 466), 65 ± 7 ms (N = 1175), and 34 ± 4 ms (N = 2235) at [WDR5] values of 50 nM, 150 nM, and 300 nM, respectively. e Histograms of the WDR5-captured durations off) at various [WDR5] values. τoff (mean ± s.e.m.) were 12 ± 2 ms (N = 441 events), 10 ± 3 ms (N = 1127), and 14 ± 2 ms (N = 2034) at [WDR5] values of 50 nM, 150 nM, and 300 nM, respectively. f Plot (in red) illustrating the dependence of the event frequency in the form of 1on on [WDR5]. g Plot (in green) illustrating the dependence of 1off on [WDR5]. The horizontal line is an average fit of the (1off) data points. Data points in panels f and g represent mean ± s.d. obtained from n = 3 different experiments. Source data are provided as a Source Data file.

An orthogonal method proves the rapid association and dissociation kinetics of WDR5-Mb4 interactions

To validate the fast kinetics recorded with the Mb4-tFhuA sensor, we performed additional measurements using biolayer interferometry (BLI)53. Mb4-tFhuA-containing micelles were immobilized onto the BLI sensor surface via a cysteine sulfhydryl engineered on the external L4 loop of tFhuA for biotin-streptavidin chemistry (Methods; Supplementary Fig. 14a). Hence, this experimental design mimics a sensing measurement with an Mb4-tFhuA sensor reconstituted into a lipid bilayer. WDR5 was added to different wells at increased concentrations. The association phases were recorded in real-time by placing the BLI sensors in WDR5-containing wells (Supplementary Fig. 14b). The dissociation phases were then recorded by putting the same BLI sensors in WDR5-free wells. However, the rates of these kinetics are beyond the time resolution of BLI. Nevertheless, BLI sensorgrams acquired at various WDR5 concentrations qualitatively confirm the rapid kinetics of association and dissociation of WDR5-Mb4 interactions noted with the Mb4-tFhuA sensor (Supplementary Table 8).

A single Adnectin1-tFhuA nanopore sensor reveals bimodal protein recognition of EGFR

The ectodomain of EGFR is proteolytically released into the bloodstream, allowing this biomarker to be used for screening, diagnosis, and disease progression54. Hence, we employed Adnectin-1 against the ectodomain of EGFR47. A single Adnectin1-tFhuA sensor exhibited some current noise at +40 mV (Supplementary Fig. 4c). However, its traces showed a relatively quiet signature at a lower transmembrane potential of +20 mV (Fig. 4a and Supplementary Fig. 15). Interestingly, when EGFR was added to the cis side of the bilayer containing the Adnectin1-tFhuA sensor, reversible current blockades were observed in a broad temporal range and with various current amplitudes. In contrast, we noted only low-amplitude and brief current spikes when EGFR was added to the cis side of the bilayer containing tFhuA alone (Supplementary Fig. 16). A two-peak distribution was found for the current amplitudes of individual EGFR-captured events (Fig. 4b). For example, at 40 nM EGFR, the normalized current blockades of the two peaks were (65.0 ± 2.1)% and (86.1 ± 1.6)% with the probabilities of 0.72 ± 0.02 and 0.28 ± 0.02, respectively. Furthermore, the relative position and probability of these peaks were independent of the EGFR concentration, [EGFR] (Supplementary Table 9).

Fig. 4: EGFR exhibits a bimodal protein recognition.
figure 4

a Single-channel traces (in black) of Adnectin1-tFhuA in the presence of EGFR, which were filtered at 2 kHz. Oon and Ooff are the EGFR-released and EGFR-captured substates, respectively. These traces were replicated in n = 3 independent experiments. The transmembrane potential was +20 mV. b Event histograms (in magenta) of the normalized current blockades A/I0, where A and I0 are the amplitude of blockades and the amplitude of the Oon substate, respectively. The cumulative fits are marked in green. The blue and black curves indicate fits of low- and large-amplitude blockades, respectively. For 10 nM EGFR, these values (mean ± s.e.m.) were (65.0 ± 0.3)% and (84.5 0.3)%, respectively (number of events, N = 467). For 20 nM EGFR, they were (65.2 ± 0.2)% and (85.1 ± 0.2)%, respectively (N = 924). For 40 nM EGFR, they were (65.3 ± 0.2)% and (85.5 ± 0.2)%, respectively (N = 1711). c Histograms (in blue) of the EGFR-released durations on) at various EGFR concentrations, [EGFR]. τon (mean ± s.e.m.) were 0.78 ± 0.04 s (number of events: N = 491), 0.42 ± 0.03 s (N = 843), and 0.25 ± 0.02 s (N = 1641) at [EGFR] values of 10 nM, 20 nM and 40 nM, respectively. Fits are indicated in red. d Histograms (in green) of EGFR-captured durations off) at different [EGFR]. The cumulative fits are marked in black. The red and cyan curves indicate fits for short- and long-lived EGFR captures. For 10 nM EGFR, they (mean ± s.e.m.) were 0.072 ± 0.011 s and 1.2 ± 0.1 s, respectively (number of events: N = 441). For 20 nM EGFR, they were 0.069 ± 0.007 s and 0.96 ± 0.09 s, respectively (N = 806). For 40 nM EGFR, they were 0.066 ± 0.006 s and 0.81 ± 0.11 s, respectively (N = 1598). e Dependence of the event frequencies in the form of 1on-i on [EGFR], where τon-1 (in red) and τon-2 (in green) are the released durations between the short and long captures, respectively. f Dependence of 1off-i on [EGFR], where i = 1 (in red) and i = 2 (in green) are subscripts corresponding to the short- and long-lived EGFR captures, respectively. Data points in panels e and f represent mean ± s.d. obtained from n = 3 different experiments. Source data are provided as a Source Data file.

EGFR-released (τon) and EGFR-captured (τoff) durations followed single-peak and double-peak event distributions (Fig. 4c, d; Supplementary Tables 1013), respectively, as judged by the maximum likelihood method51 and logarithm likelihood ratio (LLR) tests52. Hence, our statistical analyses revealed two subpopulations of binding events, the short-lived and long-lived EGFR-captured events, whose durations were τoff-1 = ~80 ms and τoff-2 = ~1 s, respectively. Interestingly, the probabilities of short-lived EGFR capture durations, P1, were close to those of low-amplitude current blockades (Supplementary Tables 9 and 10). This outcome suggests two distinct mechanisms of binding of EGFR to Adnectin1, which correlate with the extent of the normalized current amplitude of EGFR-captured events and their duration. The event frequencies of short-lived and long-lived EGFR-captured events, in the form of 1/τon-1 and 1/τon-2, respectively, were linearly dependent on the EGFR concentration, [EGFR] (Fig. 4e). Here, τon-1 and τon-2 are the release (e.g., interevent) durations corresponding to the short-lived and long-lived current blockades, respectively (Supplementary Table 12). The slopes of the linear fits of 1/τon-i (i = 1,2) versus [EGFR] are the association rate constants, kon-i, of Adnectin1-EGFR interactions because kon-i = 1/(τon-i[EGFR]). Again, the dissociation constants of the short-lived (koff-1) and long-lived (koff-2) current blockades were independent of [EGFR] (Fig. 4f; Supplementary Table 14). We interpret that these blockades are produced by specific bindings of EGFR to Adnectin1. We obtained the association rate constants, kon-1 and kon-2 (mean ± s.e.m.), of (6.62 ± 0.21) × 107 M−1 s−1 and (2.89 ± 0.10) × 107 M−1 s−1, respectively. The dissociation rate constants, koff-1 and koff-2 (mean ± s.e.m.), were 12.0 ± 0.4 s−1 and 1.01 ± 0.01 s−1, respectively (Supplementary Table 15). These values yield the equilibrium dissociation constants of the short-lived and long-lived current blockades, KD-1 and KD-2 (mean ± s.e.m.), of 181 ± 8 nM and 34 ± 2 nM, respectively.

The EGFR structure in the EGFR/EGF complex (“1NQL”)55 is similar to that of EGFR in the EGFR-Adnectin1 complex (“3QWQ”)47. It is believed to be an inactive receptor form (Supplementary Fig. 17a, b)47,55. Adnectin1 and EGF bind to the EGFR domain D-I with a highly overlapping binding surface (Supplementary Fig. 17c, d). EGFR is a remarkably adaptable molecule with relatively rigid domains D-I and D-III55,56. In contrast, the domains D-II and D-IV can adopt multiple conformations that place domain D-III differently in relation to domain D-I. We speculate that such distinct conformers of a flexible EGFR may likely be responsible for the bimodal protein recognition of EGFR by Adnectin1. The extended time bandwidth of our measurements facilitated the detection and quantification of conformational binding substates of the EGFR-Adnectin1 complex that are hidden in ensemble or low-resolution single-molecule measurements57. Earlier studies using the resistive-pulse technique have also reported multimodal conformational transitions in the case of the dihydrofolate reductase (DHFR) enzyme21.

Are there interconversion transitions between the capture substates?

Next, we asked whether these reversible current transitions may also involve transitions between the two EGFR-captured substates. Hence, a related question is whether a kinetic model including interconversion transitions between these EGFR-captured substates would more accurately reflect experimentally determined rate constants. An interconversion-dependent kinetic model was developed, encompassing two supplementary rate constants between EGFR-captured substates, k12 and k21 (Supplementary Table 16 and Fig. 18). At a confidence level of C > 0.95, we found that fits to an interconversion-dependent kinetic model were not statistically superior over those corresponding to an interconversion-independent kinetic model, as indicated by the LLR test. Finally, to examine the reactivity crosscheck of our sensors, we recorded electrical traces of Adnectin1-tFhuA in the presence of either hSUMO1 (Supplementary Fig. 19) or WDR5 (Supplementary Fig. 20). In both cases, very short-lived and low-amplitude current blockades were noted. These blockades resemble those typically found in the case of nonspecific interactions of folded proteins with the cis opening of tFhuA (Supplementary Figs. 9, 12, and 16). This finding proves that the Adnectin1-tFhuA sensor is highly specific to EGFR.

Single-molecule detection of a protein biomarker in a biofluid

We challenged this sensor in the presence of 5% (v/v) fetal bovine serum (FBS) to examine the stability of this system in a harsh environment and the ability to distinguish analyte-captured events from other nonspecific transitions of the solution constituents. Under physiological conditions, the serum threshold for the soluble EGFR ectodomain level is 45 ng/ml (~112 nM)58. The tumor state can be evaluated at EGFR levels significantly exceeding this threshold54. Fig. 5a, b show a representative signature of Adnectin1-tFhuA without and with 20 nM EGFR, respectively. However, the addition of 5% (v/v) FBS decorated the standard signature of EGFR-captured events with brief current spikes in the low-millisecond range (Fig. 5c). The power spectral density (PSD) of current fluctuations revealed a transition from white noise in the absence of FBS to 1/f flicker noise in the presence of FBS (Fig. 5d). This outcome suggests low-frequency equilibrium fluctuations in the local mobility and density of charges at the nanopore tip in the presence of FBS59. The brief FBS-induced current fluctuations had a lower current amplitude around the open Oon substate (Fig. 5e–g), indicating that these may result from trafficking moieties of serum constituents at the cis opening of Adnectin1-tFhuA (Fig. 1). An extensive statistical analysis of the current blockades corresponding to the Ooff-1 and Ooff-2 levels confirmed the presence of two EGFR-captured event types in the presence of FBS (Fig. 5h–k; Supplementary Tables 17 and 18). No statistically significant impact of FBS was noted on the koff-1 and koff-2, but small changes, within the same order of magnitude, on the kon-1 and kon-2 (Supplementary Table 19). These changes may result from the interference of serum constituents with the binding interfaces of EGFR and Adnectin1. The mean duration of long-lived EGFR-induced current blockades was τoff-2 = 0.93 ± 0.14 s, much longer than the brief millisecond-timescale FBS-induced closures. Under these conditions, we determined a corresponding τon-2 of 1.7 ± 0.5 s. Using a kon-2 of (2.9 ± 0.2) × 107 M−1s−1 in the absence of FBS, we can evaluate the EGFR concentration in the serum sample, [EGFR]FBS, using the equation [EGFR]FBS = 1/(τon-2kon-2). Employing these values, we determined an [EGFR]FBS of 22.2 ± 5.9 nM in the FBS-containing sample, which is near the actual concentration of 20 nM.

Fig. 5: Single-molecule detection and quantification of EGFR in a biofluid.
figure 5

a An electrical trace of Adnectin1-tFhuA. b The trace in a with 20 nM EGFR. c The trace in b with 5% (v/v) FBS. The transmembrane potential was +20 mV. This subset of single-channel electrical signatures (in green) was replicated in n = 3 independent experiments. Oon and Ooff are the EGFR-released and EGFR-captured substates, respectively. Traces were filtered at 0.5 kHz. d Power spectral density of current noise (S(f)) of traces from ac. Each spectrum represents an average of three independent traces. e A current histogram (in blue) of the Oon substate of Adnectin1-thuA. The amplitude (mean ± s.e.m.) of the Oon substate was 18.2 ± 0.1 pA. f A current histogram (in green) of the Oon and Ooff substates of Adnectin1-tFhuA at 20 nM EGFR. The currents (mean ± s.e.m.) of the Ooff-1 and Ooff-2 substates were 6.4 ± 0.1 pA and 2.3 ± 0.1 pA, respectively. g A current histogram (in magenta) of the Oon and Ooff substates of Adnectin1-tFhuA at 20 nM EGFR and in the presence of 5% fetal bovine serum (FBS). This plot reveals the residual signal produced by the FBS constituents (IFBS). h A histogram (in green) of the EGFR-released durations (τon) at 20 nM EGFR. τon (mean ± s.e.m.) was 0.40 ± 0.03 s (number of events: N = 844). i A histogram (in orange) of the EGFR-captured durations off) at 20 nM EGFR. τoff-1 and τoff-2 (mean±s.e.m.) were 0.044 ± 0.015 s and 0.982 ± 0.049 s, respectively (number of events: N = 734). j A histogram (in green) of the EGFR-released durations on) at 20 nM EGFR and in the presence of 5% FBS. τon (mean ± s.e.m.) was 0.607 ± 0.051 s (N = 738). k A histogram (in orange) of the EGFR-captured durations off) at 20 nM EGFR and in the presence of 5% FBS. τoff-1 and τoff-2 (mean ± s.e.m.) were 0.036 ± 0.013 s and 0.81 ± 0.08 s, respectively (N = 694). In panels h and j, the fits are indicated in red. In panels i and k, the cumulative fits are marked in black. The blue and green curves indicate fits for the short- and long-lived EGFR captures, respectively. Source data are provided as a Source Data file.

Distinct outcomes with the monobody-based sensors

In this study, we provide a detailed signature analysis of single-molecule protein detection of three analytes using three nanopore sensors that share a modular architecture but differ by their binding surface (Supplementary Figs. 2123). Fortuitously, all monobodies partly block the ionic flow through tFhuA, allowing direct electrical detection of analyte bindings without needing any peptide tag32,33. As the protein analytes and their complexes with the specific monobodies drastically vary in size, charge, and structural complexity, distinct current blockades are noted in each case (Fig. 6; Supplementary Tables 1 and  20). For example, WDR5 interacts with a distal FG loop of Mb4 and away from the tFhuA pore opening (Supplementary Fig. S22), suggesting a modest current blockade made by the WDR5-Mb4 complex. In accord with this expectation, we note low-amplitude current blockades produced by WDR5-captured events (Fig. 3a; Supplementary Table 20). In contrast, the conformational complexity and structural properties of the hSUMO1-FN3SUMO and EGFR-Adnectin1 complexes at the tip of tFhuA indicated a potentially large current blockade, as also found by electrical recordings (Fig. 2a and Fig. 4a). In addition, we probed distinct single-molecule kinetic signatures of each analyte without the steric restrictions of the nanopore confinement (Supplementary Fig. 24). These unique characteristics of protein detections using externally engineered complex binding interfaces culminated with the discovery of bimodal protein recognition of EGFR.

Fig. 6: Schematic of the relative current blockades acquired with FN3SUMO-tFhuA, Mb4-tFhuA, and Adnectin1-tFhuA.
figure 6

Oon and Ooff are the protein analyte-released and protein analyte-captured substates, respectively. τon and τoff are the durations of the protein analyte-released and protein analyte-captured events, respectively. a Graphic representation of stochastic sensing of hSUMO1 using an FN3SUMO-tFhuA sensor, which maintains an open-state current (left panel). When hSUMO1 is added to the cis side, the analyte produces large-amplitude (A), medium-lived current transitions between two current substates (right panel). b Mb4-tFhuA sensor maintains an open-state current (left panel). When added to the cis side, WDR5 produces low-amplitude, medium-lived current transitions (right panel). c Adnectin1-tFhuA protein maintains an open-state current (left panel). When added to the cis side, EGFR produces short- and long-lived current transitions (right panel). Note that there are two EGFR-captured durations, τoff-1 and τoff-2. The short-lived transitions feature a relatively lower current amplitude than the long-lived transitions.

Validation of the monobody-based sensors

Next, we examined the binding affinity of detergent-refolded sensors with their cognate analytes using steady-state fluorescence polarization (FP) anisotropy. If the labeled protein analyte interacts with the corresponding monobody-containing sensor, its tumbling rate (e.g., the coefficient of rotational diffusion) decreases, increasing the FP anisotropy. In accordance with our prediction, the FP anisotropy substantially increased at elevated sensor concentrations (Supplementary Fig. 25). Conversely, tFhuA at increased concentrations did not alter the FP anisotropy, confirming no interaction between labeled proteins and tFhuA-containing detergent micelles. The calculated KD values of hSUMO1 and WDR5 with their respective nanopore sensors were 186 ± 16 nM and 455 ± 59 nM, respectively. These data agree well with those derived from single-channel electrical recordings (Supplementary Tables 5 and 8). EGFR is unsuitable for this assay because of its large molecular weight, producing a longer tumbling rate than the fluorescence lifetime of most fluorophores. However, the KD determined for the long-lived EGFR-captured events using Adnectin1-tFhuA sensor is in accord with a previously reported study (Supplementary Table 15)47. It should be mentioned that restraining one binding partner to a surface can decrease the affinity by one order of magnitude53. Hence, this explains a significantly weaker binding interaction with the monobody-containing nanopore sensor immobilized on a lipid bilayer than that value measured in solution by steady-state FP spectroscopy.

Advantages of these nanopore sensors and their implications in nanobiotechnology

In this study, we engineered nanopore sensors made of a single-polypeptide unit that feature a selective protein binder adaptable with atomic precision. The monomeric nature of these sensors circumvents the necessity of tedious purification steps of the assembly reaction, otherwise required for multimeric nanopores. The overall architecture of the sensors can be maintained while changing the interaction interface of the antibody-mimetic binder. This way, such an approach substantially extends the applications of these sensing elements for numerous protein biomarkers. This critical benefit is facilitated by the genetically encoded nature of these sensors so that they can create combinatorial libraries of tethered binders. For instance, the loops of monobodies are analogous to the complementarity-determining regions (CDRs) of antibodies. One significant advantage of monobodies is their ability to interact with challenging binding surfaces that are not generally exposed to the CDRs of antibodies34. In addition, there is no fundamental limitation in replacing the monobody with another synthetic protein binder60,61,62. Furthermore, the main benefits of using antibody-mimetic proteins include strong binding affinities with different epitopes, straightforward expression and purification procedures, and high thermodynamic stability34.

A shortcoming of most detection techniques is the nonspecific binding of assay reagents and immobilization surfaces. This problem determines an increase in background noise, interfering with the signal produced by protein binding events. Enzyme-linked immunosorbent assay (ELISA) is the most common protein detection technique in basic research and molecular diagnostics. This technology is also formulated as single-molecule arrays that rely on bead-based, femtoliter-size wells1,63. The significant advantage of ELISA is its high detection performance in the pg/ml range. However, its dynamic range is limited to four orders of magnitude.

Notably, our method has the potential to detect and characterize functionally distinct subpopulations of specific binding events in a challenging biofluid. This is a significant advantage that existing technologies in the ensemble lack, such as BLI, surface plasmon resonance (SPR), isothermal titration calorimetry (ITC), and ELISA. BLI, SPR, and nanopore sensors are amenable to parallel recording technologies, but ITC cannot be used at high throughput and necessitates large amounts of proteins. Our modular sensors rely on a single-molecule detector that enables direct measurements of time-resolved protein binding events at adjustable protein concentrations. These nanopore sensors feature high-affinity protein binders that do not cross-react. In addition, our sensors may be further developed for concurrently identifying two or more target proteins of varying binding affinity and specificity because each protein would provide a specific signature. These competitive binding interactions against the same protein binder cannot be analyzed using bulk-phase techniques. We probe the complexity and heterogeneity of protein recognition events without requiring any additional exogenous tag or peptide tail. In addition, this sensor formulation includes a system that precludes the occurrence of nonspecific binding events or protein inactivation at the liquid-surface interface, as in the case of surface immobilization-based sensors. Our proposed approach shows prospects for discovering rare and short-lived binding events, which are unlikely to be detectable by prevailing technologies. In extreme conditions of unusually high kon, such as those in the range of 107−109 M−1s−129, we show that our method can be utilized to measure such values (e.g., for hSUMO1 and WDR5). In nanopore-based sensing, the koff can be recorded up to a value of ~105 s−18. In contrast, SPR and BLI have challenges detecting this value greater than 1 s−1. Nanopore sensors with external protein recognition elements may detect protein levels at least two orders of magnitude below the KD in solution31. If these nanopores encompass very high-affinity binders, such as affibodies, with a KD in a low-picomolar range64, then their detection sensitivity may attain pg/ml levels. Hence, these sensors can operate at clinically relevant concentration ranges of proteins and with an extended time bandwidth of at least eight orders of magnitude. In this process, the analyte-induced events are unambiguously distinguished from other nonspecific current blockades of biofluid constituents. With further developments, these sensors can be integrated with high-throughput technologies for biomarker profiling in biomedical diagnostics.


Computational grafting of monobodies onto tFhuA

For the structural prediction of nanopore sensors, the amino acid sequence of each monobody (FN3, FN3SUMO, Mb4, and Adnectin1) was inserted at the N-terminus of tFhuA via a (GGS)2 peptide tether31,33. 3D structural models of the nanopore sensors were generated in silico using AlphaFold248,49. All parameters were kept the same for all nanopore sensors. The predicted structures of sensors were confirmed by comparisons with individual structures of FhuA and monobodies.

Synthetic gene construction

Three derivatives of wild-type fibronectin type-III (FN3) were used to develop these sensors. The cDNA sequences of these fn3 genes, namely fn3sumo, mb4, and adnectin1, were fused to the 5′ end of the tfhua gene via a (GGS)2-encoding linker by a restriction-free cloning method65. The cDNA sequences of Mb4 and Adnectin1 were synthesized by Eurofins Genomics (Louisville, KY) and Integrated DNA Technologies (IDT, Coralville, Iowa), respectively. The construction of the fn3sumo gene was made based on ySMB945. The cDNA sequence of all three fibronectin derivatives was first amplified using Q5 high-fidelity DNA polymerase (New England BioLabs, Ipswich, MA) from their respective template DNA. PCR products were separated on 1% agarose gel and purified using a Gel extraction kit (Promega, CA). Sequences of forward and reverse primers are listed in Supplementary Table 21. Amplified products of fn3sumo and mb4 genes were then fused to the 5′ end of tfhua cloned in pPR-IBA1 plasmid (IBA, Goettingen, Germany). adnectin1 was joined at the 5′ end of tfhua in pET28a (EMD Millipore, Burlington, MA). The pET28-tFhuA plasmid was constructed by inserting the gene between BamHI and XhoI restriction sites after amplification with forward and reverse primers of tFhuA (Supplementary Table 21). All the gene sequences were verified by sequencing (MCLab, San Francisco, CA). The pET11a-hSUMO1 was kindly provided by Fauke Mechior (Addgene plasmid #53138).

Protein expression and purification

For the expression of FN3SUMO-tFhuA, Mb4-tFhuA, and Adnectin1-tFhuA, the plasmids mentioned above were transformed into E. coli BL21(DE3) cells. These monobody-containing protein nanopores were purified from inclusion bodies. Cells containing these overexpressed protein nanopores were lysed using a microfluidizer (Model 110L; Microfluidics, Newton, MA). Then, cells underwent a series of washes in 1% (v/v) Triton. Cell lysates were centrifuged at 108,500 × g for 30 min at 4 °C. This way, the insoluble pellet and supernatant were separated. The insoluble pellet was solubilized in 8 M urea. This was followed by a first purification step achieved through anion-exchange chromatography (Q12-Sepharose; Bio-Rad, Hercules, CA). Then, protein samples were purified further using size-exclusion chromatography (HiLoad 16/600 Superdex-75; GE Healthcare Life Sciences, Pittsburg, PA). The protein purity was validated by sodium dodecyl-sulfate–polyacrylamide gel electrophoresis (SDS–PAGE) analysis (Supplementary Fig. 1). No noteworthy differences were detected among the productions of the three sensors. In the case of hSUMO1, BL21(DE3) cells were transformed with pET11a-hSUMO1 and grown in Luria–Bertani (LB) medium at 37 °C until OD600 attained a value of ~0.5. Then, the temperature was changed to 20 °C. Expression was initiated by inducing the cells with 250 μM IPTG. After induction, the cells were cultured for ~18 h at 20 °C. Cells were centrifuged at 3700 × g for 30 min at 4 °C, then resuspended in 50 mM Tris-HCl, 50 mM NaCl, and pH 8.0. The lysozyme was added to the suspended cells and incubated on ice for 15 min. Cell lysis was accomplished using sonication (30 s on, 60 s off × 4 times). The cell lysate was centrifuged at 108,500 × g for 30 min at 4 °C. Then, the supernatant was filtered using a 0.22 µm filter. The supernatant was loaded onto a Q-Sepharose column (Cytiva, Marlborough, MA), which was washed with 50 mM Tris-HCl, 50 mM NaCl, pH 8.0, and eluted with 50 mM Tris-HCl, 1 M NaCl, pH 8.0 in a gradient manner. The desired fractions were collected, dialyzed, and concentrated. Furthermore, the protein sample was loaded on an S75 gel-filtration column (GE Healthcare, Chicago, IL). Pure fractions were collected and dialyzed against 20 mM Tris-HCl, 150 mM NaCl, pH 8.0, and 0.5 mM TCEP overnight at 4 °C.

WDR5 was expressed using ROSETTA cells (New England Biolabs, Ipswich, MA). Cells were induced utilizing 100 μM IPTG. The cell lysis was conducted using a microfluidizer (Model 110L; Microfluidics, Newton, MA). The lysates were centrifuged at 3700 × g for 30 min at 4 °C. The WDR5-containing supernatant underwent an initial purification process via a metal-affinity chromatography (5 ml, Bio-Scale Mini Profinity IMAC cartridge; Bio-Rad, Hercules, CA). Then, a Tobacco Etch Virus (TEV) protease (New England Biolabs) enzymatic assay was used to remove the hexahistidine tag, which was followed by the addition of the benzonase nuclease (Sigma-Aldrich, St. Louis, MO) to digest DNA contaminants. Finally, the sample was again passed through a metal-affinity column to remove protein aggregates. Finally, a 10 kDa-molecular weight concentrator (Millipore Sigma, St. Louis, MO) was utilized to prep the final protein samples.

For the purification of the ectodomain of epidermal growth factor receptor (EGFR), Expi293F cells (Thermo Fischer Scientific; Cat# A14527) were seeded at 106 cells/ml density in 1 l of Dynamis growth medium (Gibco) 24 h before the transfection and supplemented with Tryptone/Glucose. For the sake of simplicity, we name this EGFR throughout this article. The culture was transfected with 2 µg/ml of the pCMV_EGFR plasmid containing the signal peptide with 3.75 × polyethylenimine (PEI). Transfected cells were cultured for five days, and the protein was allowed to excrete from the cells. Five days post-transfection, the culture was pelleted, and the supernatant was filtered. The sample was loaded onto an immobilized metal-affinity column (1 ml, HIStrap HP column, GE Healthcare), which was washed with 50 mM sodium phosphate (NaPi) (pH 8.0), 300 mM NaCl, 20 mM imidazole. The protein was eluted using 50 mM NaPi (pH 8.0), 300 mM NaCl, and 500 mM imidazole. Peak fractions were collected and confirmed by SDS–PAGE (Supplementary Fig. 26). Finally, the protein sample was concentrated and exchanged with phosphate buffer saline (PBS, pH 7.5) using a PD10 column (GE Healthcare) and stored at −80 °C. The purity of all protein analytes was tested by SDS–PAGE analysis.

Protein refolding

The purified FN3SUMO-tFhuA, Mb4-tFhuA, and Adnectin1-tFhuA were adjusted to a final concentration of ~10 µM. Next, n-dodecyl-β-d-maltopyranoside (DDM) was added to denatured samples to a final concentration of 1% (w/v). The protein samples were immediately dialyzed against the buffer containing 200 mM KCl, 20 mM Tris-HCl, pH 8, at 4 °C for 96 h. The dialysis solution was replaced at 24-h intervals. These refolded protein samples were centrifuged to eliminate any protein precipitations, and the supernatant was used as the running sample for single-channel electrical recordings. Protein concentrations were determined by their molar absorptivity at a wavelength of 280 nm.

Single-channel electrical recordings

Electrical detection of protein ligands at single-molecule precision was conducted using planar lipid bilayers66. The two halves of the chamber were divided by a 25 µm-thick Teflon septum (Goodfellow Corporation, Malvern, PA). A planar lipid bilayer was made of 1,2-diphytanoyl-sn-glycero-phosphatidylcholine (Avanti Polar Lipids, Alabaster, AL) across an ~100 μm-diameter aperture of the Teflon septum. For all experiments, the buffer solution contained 300 mM KCl, 10 mM Tris-HCl, and pH 8.0. In addition, this buffer included 0, 0.5, and 1 mM TCEP in experiments with EGFR, hSUMO1, and WDR5, respectively. The nanopore protein samples (final concentration, 0.5–1.5 ng/μl) and analytes were added to the cis compartment, which was grounded. Single-channel electrical currents were acquired using an Axopatch 200B patch-clamp amplifier (Axon Instruments, Foster City, CA) controlled by Clampex 10.7 (Axon Instruments). The applied transmembrane potential was +40 mV, unless otherwise stated. The electrical signal was sampled at 50 kHz using a low-noise acquisition system (Model Digidata 1440 A; Axon Instruments). A low-pass Bessel filter (Model 900; Frequency Devices, Ottawa, IL) was further employed for signal filtering at 10 kHz. For the data processing and analysis, the electrical traces were digitally filtered with a low-pass 8-pole Bessel filter at 3 kHz, unless otherwise stated. All single-channel electrical recordings were acquired at a temperature of 24 ± 1 °C.

EGFR detection in a heterogeneous solution

For the detection and quantification of EGFR in heterogeneous solutions, fetal bovine serum (FBS, GibcoTM, Thermo Fisher Scientific, Pittsburgh, PA) was used. FBS was sterilized through a syringe filter before being stored at −80 °C. For single-channel recording, an aliquot was defrosted on ice and kept at room temperature before adding to the chamber. Single-channel electrical traces were recorded in the presence of FBS at a final concentration of 5% (v/v). At this FBS concentration, the lipid bilayers were stable, and no clogging of nanopores was noted for long recording periods32. Under these conditions, the nontarget FBS protein concentration for analyzing single-molecule EGFR detection was 1.5–2.5 mg/ml (provided by GibcoTM). These traces were filtered with a low-pass 8-pole Bessel filter at 500 Hz.

Biolayer interferometry (BLI) assay using immobilized proteomicelles

These experiments were conducted using an Octet Red384 instrument (FortéBio, Fremont, CA) at 24 °C67. For BLI experiments, a site-specific insertion of cysteine at position 287 was achieved in the long L4 loop of Mb4-tFhuA by site-directed mutagenesis (Q5 mutagenesis kit, New England Biolabs). This cysteine-containing Mb4-tFhuA was expressed and purified as described above, except for the presence of a reducing agent. Cys287 was biotinylated using maleimide chemistry. A flexible (PEG)11 linker was used between the biotin and maleimide groups. The BLI running buffer contained 300 mM KCl, 20 mM Tris-HCl, 1 mM TCEP, 1% DDM, 1 mg/ml bovine serum albumin (BSA), pH 8.0. It was used to soak streptavidin (SA) sensors for 30 min. The 50 nM Mb4-tFhuA_Cys287-(PEG)11-Biotinyl was loaded onto the sensors for 2.5 min via biotin-streptavidin chemistry. By dipping the sensors in a protein-free solution for 6 min, the unattached Mb4-tFhuA_Cys287 was washed away. The association process was examined using various concentrations of WDR5, ranging from 1.5 µM to 6 µM. The BLI sensors were dipped in a WDR5-free running buffer to inspect the dissociation phase. For all WDR5 concentrations, the Mb4-tFhuA_Cys287-free BLI sensors were run in parallel as controls. The baseline and drift in the sensorgrams were subtracted using these controls. FortéBio Octet HT acquisition 11 and FortéBio Octet HT kinetic analysis 11 (FortéBio) were used for data acquisition and analysis, respectively.

Steady-state fluorescence polarization (FP) measurements

hSUMO1 and WDR5 were labeled with fluorescein and rhodamine, respectively, at pH9.0 by primary amine chemistry. These labeled proteins were added to the well at a final concentration of 50 nM. Steady-state fluorescence polarization (FP) anisotropy assays were conducted in triplicate with an 18-point serial dilution of FN3SUMO-tFhuA, Mb4-tFhuA, or unmodified tFhuA, against a fixed concentration of labeled proteins on black 96-well plates. All steady-state FP measurements were recorded using a SpectraMax i3x plate reader (Molecular Devices, San Jose, CA) controlled by SoftMax Pro 6.4 (Molecular Devices). All measurements were conducted at 0 min and after a one-hour incubation at room temperature in the dark. The resulting dose-response data were averaged and fitted using logistic regression to obtain each interaction’s dissociation constant (KD).

Statistics and reproducibility

pClamp 10.7 (Axon Instruments) was used for the data acquisition and analysis. Capture and release events were collected using single-channel event searches by ClampFit 10.7 (Axon Instruments), and figures were prepared by Origin 9.7 (OriginLab, Northampton, MA). The probability distribution function (PDF) was generated using a kinetic rate matrix, and the kinetic rate constants were determined by fitting the data using the maximum likelihood method51. To evaluate the results of multiple models and select the number of statistically significant peaks that are best matched to the data, a logarithm likelihood ratio (LLR) test was performed52. At a confidence number of C = 0.95, a single-exponential fit was the best model for the release and capture durations of hSUMO1 and WDR5. For EGFR, a two-exponential fit was the best model for the capture durations. Single-channel traces were recorded for long periods, so a single statistical analysis encompassed several hundred events. This way, satisfactory reproducibility was achieved from one experiment to the next, as indicated by values of standard deviations. These statistical analyses were not subjected to randomizing and blinding.

Molecular graphics

All cartoons showing molecular graphics were prepared using PyMOL (Version 2.4.0; Schrödinger, LLC) and Chimera X (Version 1.4; The University of California at San Francisco). Entries “3RZW”, “6BYN”, “1NQL”, “3QWQ”, “1BY3”, and “1FNF” from Protein Data Bank were used in this article for visualizations and molecular graphics of hSUMO1-FN3SUMO1, WDR5-Mb4, EGFR ectodomain, EGFR-Adnectin1, tFhuA, and FN3, respectively.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.