Disentangling the recognition complexity of a protein hub using a nanopore

A nanopore sensor reveals three distant binding events

At a transmembrane potential of −20 mV, a relatively quiet single-channel current was recorded with a single MLL4_WintFhuA nanopore (Fig. 2a). The presence of WDR5 in the cis compartment at nanomolar concentrations produced infrequent current blockades (Supplementary Fig. 2 Table 3), likely because of the entropic penalty of MLL4_Win to partition into the WDR5 cavity. These current blockades occurred as WDR5 was held in the proximity of the pore opening during its reversible captures by MLL4_win, obstructing a fraction of the ionic flux through the nanopore. They were not noted when an unmodified tFhuA nanopore^12,13, without the MLL4_Win ligand, was exposed to WDR5 added to the cis compartment. However, very rare and brief current spikes were observed, likely due to collisions of WDR5 with the opening of the unmodified tFhuA nanopore (Supplementary Figs. 3, 4). Taken together, these findings indicate that WDR5 did not produce significant current blockades due to nonspecific interactions with the cis opening of the nanopore. The lack of current blockades without MLL4_win shows that the ligand must first bind to WDR5 to enable WDR5-produced current blockades.

WDR5-released and WDR5-captured events recorded with MLL4_WintFhuA corresponded to the open-substate, O_on, and closed-substate, O_off, respectively. WDR5-captured events were noted in a concentration-dependent manner when WDR5 was added to the cis compartment (Fig. 2b–d; Supplementary Fig. 5). These single-channel electrical traces were low-pass filtered at a frequency of 100 Hz using an 8-pole Bessel filter. Yet, WDR5-captured events were not detectable when WDR5 was added to the trans compartment (Supplementary Fig. 6). This result agrees with our previous studies^12,13,34, which showed that tFhuA and its derivatives insert into the membrane with a single orientation. Moreover, WDR5-captured events were not detectable at a positive transmembrane potential of +20 mV (Supplementary Fig. 7). This outcome is in accord with the observation of current blockades produced by the positively charged WDR5 at a negative transmembrane potential (e.g., pI_WDR5 = 8.27).

Surprisingly, we noted an extensive spectrum of WDR5-captured durations, between ~3 ms and ~40 s. This result stimulated detailed statistical analyses of both the WDR5-released and WDR5-captured durations, whose mean values were denoted by τ_on and τ_off, respectively. To identify potentially distinct subpopulations of these durations, we employed the maximum likelihood method^35,36 and logarithm likelihood ratio (LLR) tests^37,38,39 to determine the most accurate model of these time constants. Durations of WDR5-released events showed a single-exponential distribution (Fig. 2e). Remarkably, the best model for WDR5-captured durations was a three-exponential distribution (Fig. 2f). Fits to a single-, a two-, or a four-exponential model were not better, as judged by the LLR value. The three WDR5-captured durations were almost one order of magnitude apart from each other. For example, the mean durations of WDR5-captured events were ~12 ms, ~120 ms, and ~1370 ms at 1 μM WDR5 (Supplementary Table 4). We called these short-, medium-, and long-lived events, respectively. Event probabilities of these WDR5-captured events, P₁, P₂, and P₃, respectively, were independent of WDR5 concentration, [WDR5], and followed the inequality: P₁ > P₂ > P₃ (Supplementary Table 5).

Moreover, the mean normalized amplitude of WDR5-produced current blockades, (I/I₀), was independent of [WDR5] (n = 5 independently reconstituted nanopores; Fig. 2g; Supplementary Table 6). Here, I₀ and I denote the single-channel currents of the WDR5-released substate of MLL4_WintFhuA and the amplitude of WDR5-produced current blockades, respectively. Histograms of the normalized amplitude of WDR5-produced current blockades showed two peaks, one at ~60% and the other at ~72%. Plots of the normalized amplitude of WDR5-produced current blockades as a function of WDR5-captured duration demonstrate that the short-lived events spanned the broadest range of I/I₀ (Supplementary Fig. 8). Earlier studies indicated that a 3₁₀-helix is the bound conformation of MLL4_Win to the Win binding site³². Here, we interpret that MLL4_Win may also exhibit conformations that deviate from a 3₁₀-helix. Such distinctive and more frequent conformers of a relatively flexible MLL4_Win likely generate the heterogeneity of its interactions with WDR5. The wide range of I/I₀ of the short-lived events likely correlates to many MLL4_Win conformers being able to bring about brief interactions. This interpretation also agrees with the medium- and long-lived events having closely similar I/I₀ values with a narrower range of normalized current amplitudes. Therefore, subtle alterations in current fluctuations between event types suggest that various conformers are present and the medium- and long-lived events are more selective as to what conformation MLL4_Win must have to lead to a strong binding event.

Here, the association rate constants for short-, medium-, and long-lived events, k_on-1, k_on-2, and k_on-3, respectively, were consistent for all [WDR5] values (Supplementary Table 7). Moreover, the frequency of short, medium-, and long-lived events was proportional to [WDR5] in a ratio 1:1 (Fig. 2h; Supplementary Fig. 9). This outcome indicates a bimolecular association process of the MLL4_Win-WDR5 complex. The slopes of linear fits of the event frequency, f (f = 1/τ_on), versus [WDR5] were the corresponding k_on values. Here, k_on-1, k_on-2, and k_on-3 (mean ± s.e.m.) were (1.4 ± 0.1) × 10⁵ M⁻¹s⁻¹, (6.9 ± 1.8) × 10⁴ M⁻¹s⁻¹, and (3.6 ± 1.0) × 10⁴ M⁻¹s⁻¹, respectively. Dissociation rate constants, k_off-i (i = 1, 2, and 3), which were determined as reciprocal of the mean WDR5-captured durations (1/τ_off-i), were independent of [WRD5] (Fig. 2i; Supplementary Table 8). This result suggests a unimolecular dissociation mechanism of the complex. Subscripts 1, 2, and 3 correspond to short-, medium-, and long-lived events, respectively. Fits of the dissociation rate constants versus [WDR5] resulted in their mean ± s.e.m. values of 86 ± 2 s⁻¹, 9.2 ± 0.5 s⁻¹, and 0.78 ± 0.06 s⁻¹, for short-, medium-, and long-lived WDR5 captures, respectively. The K_D for short-, medium-, and long-lived interactions were 631 ± 49 μM, 138 ± 18 μM, and 20 ± 4 μM, respectively (Supplementary Table 9).

Biolayer interferometry validates the slow association process of the MLL4_Win-WDR5 complex

Then, we employed BLI, a real-time approach for determining the kinetics of MLL4_Win-WDR5 interactions in bulk phase (Methods). Here, we covalently attached MLL4_Win onto the BLI sensor and recorded its binding interactions with WDR5. The association and dissociation phases were optically measured using changes in the interference pattern between reflected light waves at the sensor surface (Fig. 3a). The BLI-measured k_on was (2.6 ± 0.1) × 10⁴ M⁻¹s⁻¹ (Supplementary Table 10). The BLI-measured k_off was (1.2 ± 0.1) × 10⁻² s⁻¹, to generate a K_D of (0.46 ± 0.04) × 10⁻⁶ M. BLI was unable to resolve the three binding subpopulations observed when using MLL4_WintFhuA and only yielded one k_on and one k_off. We also employed BLI to conduct a positive-control experiment, testing whether our peptide adaptor exhibits any interaction with WDR5 and found none (Supplementary Fig. 10).

To better understand the mechanistic distinctions of the nanopore and BLI measurements, we simulated the BLI response by using results from single-channel electrical recordings. The simulation shows that the majority of the contribution to the observed BLI response is generated by the long-lived MLL4_Win-WDR5 interactions (Supplementary Fig. 11). This confirms our tentative interpretation that the BLI sensorgram is extensively biased by the impact of the long-lived binding interactions, despite their lowest event frequency. This simulated curve highlights the superiority of single-molecule measurements using our protein nanopore due to their broad time bandwidth.

The three binding events involve Win binding site interactions

Next, we employed D92N_WDR5, a WDR5 variant, to find out which event types are related to the partitioning of MLL4_Win into the cavity. Also, the detection of D92N_WDR5 provided proof that our MLL4_wintFhuA can recognize a somatic cancer mutant of WDR5⁴⁰. Asp-92 is a residue deeply located at the inner tip of the WDR5 cavity (Fig. 1b; Supplementary Fig. 12). First, we used BLI to evaluate MLL4_Win-D92N_WDR5 interactions. Interestingly, MLL4_Win binds very weakly to D92N_WDR5, and we were unable to quantify these interactions using BLI (Fig. 3b). D92N_WDR5 produced current blockades with an amplitude comparable to that of WDR5-captured events (Fig. 4a, b). Moreover, D92N_WDR5-captured events appeared in a concentration-dependent manner (Fig. 4b–d). As in the WDR5 case, we noted a broad span of D92N_WDR5-captured event durations. Remarkably, D92N_WDR5 also showed three distinct binding events, as judged by the LLR test (Fig. 4e, f; Supplementary Table 11). However, the frequency of D92N_WDR5-captured events was drastically reduced with respect to that value of the WDR5-captured events. This result indicates the influential role of Asp-92 in MLL4_Win-WDR5 interactions. In addition, the three events had D92N_WDR5-captured durations similar to those recorded with WDR5. Therefore, the conservation of the three-component binding events illuminates that these subpopulations resulted from interactions deeply located within the WDR5 cavity. Event probabilities of short-, medium-, and long-lived D92N_WDR5-captured events also followed the inequality: P₁ > P₂ > P₃ (Supplementary Table 12). Again, it is likely that these three interactions correspond to three different modes of D92N_WDR5 recognition by the flexible MLL4_Win ligand⁴¹. The similarity between D92N_WDR5 and WDR5 shows that these different MLL4_Win conformers were present in the interactions of MLL4_Win with each protein. This outcome confirms closely related multimodal binding mechanisms of the two proteins. The D92N mutation influences the k_on, but not the k_off of short-, medium-, and long-lived D92N_WDR5-captured events (Supplementary Table 13). Furthermore, the frequency of each of the three event types depended on D92N_WDR5 concentration, [D92N_WDR5], in a 1:1 ratio (Supplementary Figs. 13, 14 Table 14). Hence, our findings suggest that the three events were specific binding events. K_D values corresponded to about one order of magnitude weaker interactions than those obtained with WDR5, regardless of event type. For short-, medium-, and long-lived D92N_WDR5-produced interactions, K_D values were 5.4 ± 1.3 mM, 721 ± 18 μM, and 218 ± 22 μM, respectively (Supplementary Table 15). This result reveals that our protein nanopore is able to probe extremely weak interactions in the low millimolar range. Our approach also provides quantitative distinctions between the binding interactions of D92N_WDR5 and WDR5 (Supplementary Tables 16,17).

Fits of individual rate constants, which are presented above, were conducted with the assumption that no transitions occur among the three capture substates. However, it is not clear whether a kinetic model encompassing interconversion transitions among these capture substates would be better suited for experimentally determined rate constants. An interconversion-dependent kinetic model was created, including six additional rate constants among the capture substates (Supplementary Tables 18, 19 Figs. 15–16). At a confidence level C > 0.95, we found that fits to an interconversion-dependent kinetic model were not statistically better than those corresponding to an interconversion-independent kinetic model, as judged by the LLR test.

A negative charge removal on the WDR5 local surface does not impact the WDR5-MLL4_Win interaction

The weak interaction between D92N_WDR5 and MLL4_Win prompted the examination of D172A_WDR5 (Supplementary Fig. 17), a second WDR5 mutant, which served for an additional positive-control experiment. Here, the goal was to observe how a small alteration in the local surface charge of WDR5 influences the observed kinetic fingerprint of its interactions with MLL4_Win. The removal of an Asp residue on the protein surface was critical to have a reliable comparison with D92N_WDR5. D172A_WDR5 was selected, because it maintains the integrity of the WDR5 cavity while altering the local surface charge of the protein in the proximity of the Win binding site. We found that D172A_WDR5 produced current blockades with release and capture durations similar to those noted with WDR5 (Supplementary Figs. 18, 19; Supplementary Tables 20–25). These results show that the decrease in the k_on observed with D92N_WDR5 occurs only because this mutation changes the electrostatic environment in the binding cavity.

The three protein recognition modes are conserved at a higher transmembrane potential

We then explored whether this kinetic fingerprint of MLL4_Win^_WDR5 interactions exists under different experimental conditions. Therefore, we first conducted single-molecule experiments at a transmembrane potential of −40 mV. The addition of WDR5 to the cis compartment produced closely similar binding events to those described above (Supplementary Figs. 20, 21). The frequencies of all three WDR5-captured events depended on [WDR5], in a 1:1 ratio, whereas their durations were independent of [WDR5] (Supplementary Fig. 22 and Tables 26–30). This finding suggests that the three observed events were specific binding modes between one tethered MLL4_win ligand and one WDR5 protein. The unaltered k_off-i under this condition with respect to values obtained at a transmembrane potential of −20 mV shows that the integrity of the deep binding interface of WDR5 is not influenced by the transmembrane potential.

Voltage dependence of MLL4_Win-WDR5 interactions

WDR5 has a slightly positive charge under our experimental conditions and binding interactions occur near the nanopore opening. Therefore, we hypothesized that the k_on-i is voltage dependent, so we compared single-channel electrical traces at various transmembrane potentials when 8 μM [WDR5] was kept unchanged in the cis compartment. In accord with our expectation, the three WDR5 capture rate constants, k_on-i (i = 1, 2, and 3), increased at elevated negative transmembrane potentials (Fig. 5a–d and Supplementary Figs. 23, 24, Table 31). Our detailed event analyses also confirmed that the k_off-i (i = 1, 2, and 3) and distribution probabilities of the three protein recognition events were independent of the transmembrane potential (Fig. 5e and Supplementary Tables 32, 33). In addition, a semilogarithmic representation of the voltage dependence of all k_on-i values revealed an identical slope for the three binding events, suggesting a similar mechanism of these WDR5 recognition modes. Hence, this enabled the calculation of the association rate constants at a zero transmembrane potential for the three binding events, k_on-i (0) (Supplementary Table 34). For example, the lower-limit association constant of the long-lived events at a zero transmembrane potential, k_on-3 (0), was (1.9 ± 0.2) × 10⁴ M⁻¹s⁻¹, which compares well with the BLI-determined k_on under similar buffer conditions (Supplementary Table 10). The reduction in the activation free energy of all WDR5-released events, ΔΔG_on, was ~0.7 kcal/mol at a transmembrane potential of −40 mV with respect to that value determined at a zero transmembrane potential (Supplementary Table 35). This data was used to determine a small relative net charge, z, of ~0.8 for WDR5 (Supplementary Table 36).

Quantitative and conceptual comparisons of nanopore sensing with BLI

In a recent study⁴², we have shown that the k_on values of Win motif peptides with WDR5 are influenced by their tethering on a sensor surface. In addition, the partitioning of MLL4_Win into the WDR5 cavity significantly reduces the k_on due to the entropic penalty of this process³². The nanopore-determined k_on ranged between 10⁴ and low 10⁵ M⁻¹s⁻¹. These values and the BLI-determined k_on value are at least two orders of magnitude lower than one would expect for the k_on of peptide-protein complexes (10⁷–10⁸ M⁻¹s⁻¹)^43,44. Interestingly, the BLI-determined k_on was similar to k_on-3 probed by nanopore recordings. This finding suggests that the similar tethering and entropic penalty in both techniques influence the k_on. On the other hand, the BLI-determined k_off was much lower than all nanopore-inferred k_off-i values. Yet, it must be noted that the inability of BLI to distinguish the three binding events and its relatively limited time resolution make it challenging to compare dissociation rate constants. Instead, our approach can be employed to detect a significantly weaker interaction than that probed by BLI. This protein nanopore is the only tool sensitive enough to resolve the three modes of protein recognition in real time.

Distinctive outcomes with WDR5 and D92N_WDR5

In this work, we developed a modular protein nanostructure capable of discriminating between three individual kinetic components existing in a complex binding distribution (Fig. 6a). Furthermore, a weakly binding mutant, D92N_WDR5, confirmed the three-binding event distribution. D92N_WDR5 does not show an altered k_off, because Asp-92 does not form any stabilizing bonds with MLL4_Win^32,33. Only Cys-261, Phe-133, Ser-91, and Asp-107 are responsible for the stability of this interaction (Supplementary Fig. 1 Table 2). Therefore, both proteins show a similar activation free energy of dissociation, ΔG_off (Fig. 6b). However, Asp-92 of WDR5 does provide a negative charge to assist with the electrostatic pulling of the critical Arg residue of MLL4_Win into the WDR5 cavity (Supplementary Table 1 Fig. 12)⁴⁵. In contrast, Asn-92 of D92N_WDR5 does not exert this role, so the activation free energy barrier of association, ΔG_on, is significantly amplified, reducing the frequency of reversible binding events.

Advantages of this protein nanopore

Our nanopore experiments with WDR5, D92N_WDR5, and D172A_WDR5, along with prior crystallographic studies^32,33, suggest that all three binding events are specific interactions and occur deeply within the WDR5 cavity. In addition, both D92N_WDR5 and D172A_WDR5 showed normalized current blockades and capture durations similar to those noted with WDR5 (Fig. 2g; Supplementary Figs. 8 and 25, Tables 4, 6, 11, 20, 22, and 37). Through this cascade of binding scenarios, we show that our protein nanopore is capable of sampling complex binding interfaces. It could also be used to study other WDRs and groove-containing binding systems. These include unfolded protein chains entering chaperone tunnels, such as those of the GroEL⁴⁶ and Clp⁴⁷ families. Furthermore, we show that our protein nanopore can also probe very weak interactions with affinities up to low micromolar. Again, this substantially extends the application spectrum of our nanopore and highlights its significant sensitivity. For example, the interaction details observed in our study could only be observed at single-molecule precision. Other bulk-phase techniques, such as isothermal titration calorimetry (ITC)⁴⁸ and surface plasmon resonance (SPR)⁴⁹, can be used to study these peptide-protein interactions, but they have a narrower time bandwidth. These limitations of ensemble studies prevent us from identifying binding details like those present in a realistic biological system.

Implications and prospects in biotechnology

Recently, it has been discovered that the high-affinity Win binding site of WDR5 is a key player in transient interactions with dozens of proteins⁵⁰. These include 3-phosphoinositide-dependent protein kinase 1 (PDPK1) and proteins involved in phosphatidylinositol 3-kinase (PI3K) signaling. Here, we speculate that WDR5 selectively utilizes the Win binding site. For example, it is likely that the long-lived interactions of WDR5 have physiological relevance for the functional integrity of large multi-subunit complexes of methyltransferases⁵¹, which is required for optimal enzymatic activity^52,53,54. On the other hand, weaker interactions of WDR5 with other proteins may assist in cell signaling pathways⁵⁰. Hence, the wide range of binding affinities of these specific interactions is in accord with the multitasking feature of the Win binding site in subnuclear PPIs^55,56. Furthermore, there is significant interest in exploring the Win binding site, because WDR5 is implicated in numerous cancers^57,58. Therefore, our protein nanopore can serve as a tool to acquire real-time and comprehensive pharmacokinetics for these clinically important interactions.

In summary, we probed the complexity of MLL4_Win-WDR5 interactions with no steric hindrance of nanopore confinement. The heterogeneous event distribution unambiguously revealed three distant modes of protein recognition with diverse affinities. The interactions of MLL4_Win with a weakly binding WDR5 derivative were not quantifiable using BLI, but our nanopore sensor provided a complete characterization of the binding kinetics. This unusual kinetic fingerprint of MLL4_Win-WDR5 interactions was further confirmed and fully characterized at varying voltage biases. Therefore, this study demonstrates that our approach can illuminate a wide span of multimodal protein recognition events and strengthen quantitative protein interaction studies.