Sculpting conducting nanopore size and shape through de novo protein design

Abstract

Transmembrane β-barrels (TMBs) are widely used for single molecule DNA and RNA sequencing and have considerable potential for a broad range of sensing and sequencing applications. Current engineering approaches for nanopore sensors are limited to naturally occurring channels such as CsgG, which have evolved to carry out functions very different from sensing, and hence provide sub-optimal starting points. In contrast, de novo protein design can in principle create an unlimited number of new nanopores with any desired properties. Here we describe a general approach to the design of transmembrane β-barrel pores with different diameter and pore geometry. NMR and crystallographic characterization shows that the designs are stably folded with structures close to the design models. We report the first examples of de novo designed TMBs with 10, 12 and 14 stranded β-barrels. The designs have distinct conductances that correlate with their pore diameter, ranging from 110 pS (~0.5 nm pore diameter) to 430 pS (~1.1 nm pore diameter), and can be converted into sensitive small-molecule sensors with high signal to noise ratio. The capability to generate on demand β-barrel pores of defined geometry opens up fundamentally new opportunities for custom engineering of sequencing and sensing technologies.

One sentence summary:

De novo design enables the generation of stable and quite transmembrane beta-barrel nanopores with tailored sizes, shapes and properties.

Transmembrane β-barrel (TMBs) nanopores formed by a circularly-closed single β-sheet provide rigid scaffolds for the transport of molecules across cellular (e.g. pore-forming toxins (1)) and organelle membranes (outer membranes of bacteria (2), mitochondria (3) and chloroplast (4)). Engineering of naturally occurring nanopores has enabled single-molecule enzymology (5), protein fingerprinting (6), the detection of small molecules and biomarkers (7), and the sequencing of biological and synthetic polymers (8). Of particular note is nanopore-based DNA sequencing (9), which has enabled widely-accessible large-scale genomics, epigenomics and microbiological analysis (10). Despite this success, the development of nanopore sensors for robust analysis of molecules beyond DNA sequencing has so far been challenging. The sensing properties of a nanopore for an analyte of interest can be modulated by introducing mutations into the pore lumen that alter nanopore/analyte interactions (11). It remains however challenging to identify a channel suitable for each of the many applications of interest, because there is only a limited set of engineerable naturally occuring nanopores, and these have mostly evolved for very different functions. Going beyond nature, a conducting pore based on a β-hairpin peptide has been designed that transports poly-lysine peptides (12). Such self-assembling β-hairpins are however not suitable as a general approach to nanopore design, because it is challenging to control the channel size and to assemble the pore in lipid membranes. Monomeric 8-stranded TMBs have been designed that stably assemble in detergent and in lipid vesicles, but they are too small to contain a central conducting channel (13).

Encouraged by the success designing these narrow TMBs, we reasoned that de novo protein design should provide a general approach to creating robust β-barrel nanopore scaffolds for a next generation of nanopore sensors. A key challenge in designing such structures is that the polar-hydrophobic pattern characteristic of globular protein folds must be inverted: the exterior must be largely nonpolar for membrane insertion, and the interior must be largely polar to support a solvated conducting channel. Furthermore, unlike globular proteins, the structure of TMBs must be specified in the vast majority by short-range interactions between residues located on adjacent strands since there is no close-packed core. Finally, the amphipathic β-strands are highly aggregation prone prior to β-barrel assembly, and hence the design must strongly favor intra-chain rather than inter-chain interactions during folding. We set out to develop general methods to overcome these challenges and design stable monomeric channels with tunable pore shapes, sizes, and single-channel conductance.

Computational design

We sought to build from scratch TMB backbones accommodating water-accessible pores starting from the principles elucidated during the design of 8-stranded TMBs lacking pores (13). To modulate the size of the pore, we increased the number of β-strands (10, 12 and 14 strands) while keeping the transmembrane span and the connectivity between β-strands (the shear number (14, 15)) constant. This resulted in an increase of the average β-barrel diameter from 16.4 Å for the previously designed 8 strand β-barrels (13) to 19.4 Å (10 strands), 22.8 Å (12 strands) and 26.4 Å (14 strands) (Figure 1A, Figure S1). By comparison to 8-stranded TMBs, the diameters of the larger β-barrels do not allow long-range sidechain contacts across the pores and the structural properties of the pores (β-strand pairing, β-barrel shape) must be locally encoded. Naturally-occurring TMBs typically feature long, disordered, loops on one side of the barrel (16), which results in noisy electrophysiology recording and challenging data interpretation when the pores are used for sensing applications (17, 18). To design quiet pores and reduce the noise, we connected the β-strands on both sides of the barrels with 2- and 3-residues (Figure S2) β-hairpins, the shortest loops we have previously found to support TMB folding (13). The first-generation backbones corresponding to these designs were assembled with the Rosetta BlueprintBDR (19) application and had similar cylindrical shapes. Such cylindrical β-sheet configurations are strained (20, 21) due to repulsion between side-chains packing the barrel lumen (Figure 1C). Glycine kinks (glycine residues in extended positive-phi conformation (15)) were introduced into the blueprint to relieve the strain and to bend the β-strands to form corners in the β-barrel cross-section. We generated four blueprints with the same topology but different glycine kink distributions to design 12-stranded β-barrel backbones with square, triangle, rectangle or oval-shaped cross-sections (Figure S1). A single glycine kink was used in corners of an angle of >= 90° and several adjacent and/or stacked kinks were placed to form corners of < 90° (Figure 1B). Sequence-agnostic TMB backbones incorporating these constraints were assembled in silico and had the shapes expected based on the placement of the glycine kinks (Figure 1D).

Fig. 1:

Sculpting β-barrel geometry. A. Pore diameter can be controlled through the number of β-strands in the β-barrel blueprint. B. β-barrel 2D interaction map. Strong bends in the β-strands (< 90° bend, right) is achieved by stacking several glycine kink residues (yellow spheres) along the β-barrel axis, as opposed to placing one kink (>90° bend, left). C-D. Cross-sections of explicitly assembled β-barrel backbones without (cylinder, C) and with (D) glycine kinks. The Cβ atoms of the residues facing the pore are shown as sphere’s and colored based on their respective repulsion energy. Glycine kinks positions are shown with arrows; placement at the corners of the embedded rectangular, oval and triangular shapes (dashed lines in D) generates the desired backbone geometries. E. Polar threonine residues are tolerated on the membrane-exposed surface of TMBs (right) as they can form a hydrogen bond to the backbone, mimicking the interactions with water molecules observed in similarly curved areas of water-exposed β-strands (left).

A challenge for TMB design is to balance the optimization of the folded β-barrel state in the membrane with delayed folding in water to reduce misfolding and aggregation that would prevent successful integration into a membrane bilayer (13, 22, 23). For the 8-stranded TMBs, this was achieved by incorporating local secondary-structure frustration (24) to reduce premature formation of aggregation-prone β-strands prior to full barrel assembly: hydrophobic amino acids were designed into the water-accessible pore to disrupt the hydrophobic-polar amino acid alternation pattern characteristic of amphipathic β-sheets. To test whether such balancing is necessary for larger β-barrel designs that need to have water-accessible (and hence more polar) channels, we first designed “optimal” 10 and 12-strand TMBs with only polar and charged amino acids facing the pore. All 16 such designs failed to express in E. coli, as was previously observed for similarly optimal 8-stranded TMB designs. We therefore set out to design larger TMB nanopores incorporating local secondary structure frustration. In the water-accessible pore, networks of polar residues were designed around the canonical TMB folding motif Tyr-Gly-Asp/Glu (13, 25, 26) to optimize strong local β-register defining interactions while alternating with patches of hydrophobic and small, disorder promoting, residues (Gly, Ala, Ser). To compensate for the more hydrophilic pores in the larger TMBs, we further reduced the β-sheet propensity by replacing a small number of β-branched residues with Ser and Thr amino acids on the lipid-exposed surface. Although it is perhaps counterintuitive to expose hydroxyl groups to the lipid environment, model building indicated that Ser and Thr could form a hydrogen bond to the β-strand backbone when placed in close proximity with a glycine kink, effectively mimicking the backbone-water hydrogen bonds observed in strongly bent β-strands of water-soluble β-barrels (Figure 1E).

During combinatorial design of sequences for β-barrels of different size, we found that the frequency of incorporation of each amino acid type strongly depended on the curvature of the β-sheet. For each of the generated blueprints, we adjusted the Rosetta solvation and reference energies (27) to achieve the desired balance of frustrated and energetically-favorable contacts (Figure S3). Following several iterations of combinatorial sequence design and structure relaxation, designs were selected based on hydrogen bond network descriptors, secondary structure (28) and aggregation propensities (29) (Figure S4). We previously found that AlphaFold2 (30) could accurately predict the structures of designed TMBs even in the absence of evolution information (from a single sequence input and without a multiple sequence alignment (31)) when weak 3D contacts contained in the sequence were amplified by 48 rounds of molecular model recycling through the prediction network, and that the confidence assigned to the model (plDDT) was a good discriminator of the sequences with higher probability of experimentally folding (32). Therefore, we selected 4-10 designs per blueprint for which AlphaFold2 predicted high-confidence structures closely matching the design models (Figure S5).

Experimental characterization of TMB folding

We first tested two sets of TMBs with 10 (four designs) or 12 β-strands with a square cross-section (nine designs). Genes were synthesized and the proteins were expressed as inclusion bodies in E. coli to avoid the complexity of targeting the outer membrane (33) (Figure 2A). Unlike the 16 “optimal” designs which all failed to express, most sequences incorporating secondary structure frustration were expressed at high-levels (12/13, Figure S6). Since most naturally-occurring TMBs can fold in vitro (34), the purified designs were solubilized in guanidine hydrochloride and refolded by slow dilution into a buffer containing either detergent (fos-choline 12 (DPC) at a concentration double the critical micellar concentration (CMC)) or synthetic lipid vesicles (Material & Methods). As previously observed for the 8-stranded TMB designs, the standard band-shift assay on cold SDS-PAGE used to assess folding of natural TMBs (35) was not informative to identify properly folded synthetic TMBs (Figure S7). Instead, the designs were characterized by size exclusion chromatography (SEC), far UV circular dichroism (CD) in the presence of DPC detergent, and tryptophan fluorescence in DUPC (C_11:0PC) large unilamellar vesicles (LUVs). One 10-strand design (TMB10_163) and one 12-strand design (TMB12_3) with predominantly monomeric SEC profiles (Figure 2A), thermostable CD spectra characteristic of β-sheet (Figure 2B,C) and clear shift of tryptophan fluorescence maximum from ~350 nm (unfolded proteins in 8 M urea or in the absence of lipid) to ~330 nm (folded in LUVs) (Figure S8, S9) were selected for further characterization by urea titration. Both designs showed sharp and reversible folding/unfolding transitions in the presence of DUPC LUVs (Figure 2D) (mid-point urea concentrations for folding (Cm^F): 4.5 ± 0.2 M and 5.5 ± 0.2 M, respectively). The equilibrium unfolding curves were fitted to a two-states transition, with the calculated unfolding free energies (ΔG⁰_UF) of −35.6 ± 2.7 and −63.1 ± 8.0 kJ/mol (for TMB10_165 and TMB12_3, respectively) in the range of natural (ΔG⁰_UF −10 to −140 kJ/mol (36-39)) and previously designed 8-stranded TMBs (−38 and −56 kJ/mol (13)). To confirm that the designs folded by integration into the bilayer rather than partial folding on its surface, the kinetics of folding were recorded in DUPC (C_11:0PC) membranes, as well as in thicker DMPC (C_14:0PC) membranes. Dramatically decreased folding rates were observed with DMPC compared with DUPC LUVs (Figure S10), consistent with integral membrane folding. Encouraged by these results, we assessed the nanopore activity of these two designs, based on their capacity to integrate spontaneously into planar dipalmitoylphosphatidylcholine (DPhPC) membranes after dilution out of DPC micelles, and to conduct Na⁺ and Cl⁻ ions. The 12-strand TMB12_3 inserted successfully into the membrane, producing distinct jumps of current of reproducible intensities (Figure S11), and had stable nanopore conductance. While the design TMB10_163 did not have detectable nanopore activity, one variant (TMB10_165) featuring seven mutations on the lipid-exposed surface (V72I, V102T, V114I, A124L, I126V, I138V and I144V, designed using Rosetta (40)) inserted successfully into DPhPC membranes and was able to conduct ions (Figure S11). TMB10_165 had increased stability by comparison to TMB10_163, as shown by higher stability to protease digestion and more dispersed NMR ¹H-¹⁵N HSQC chemical shift in DPC micelles (Figure S12). Unlike some native pores which can fall off from the membrane stochastically, these two pores remain stably inserted over long periods of time with the longest recording acquired being 2 hours for the TMB12_3 design. Recording of the current-to-voltage response showed an incremental increase in observed conductance under positive and negative changes of voltage, indicative of stable transmembrane channels (I/V curves in Figure S11). Our results on TMB10_163, TMB10_165, TMB12_3 and other TMB12 designs with less, to no, detectable nanopore activity (Figure S14) suggest that membrane integration and nanopore conductance require stable TMB folding in vitro.

Fig. 2:

Biophysical characterization of designed nanopores. Top row: 10-stranded design (TMB10_163); Bottom row: 12-stranded design with a square cross-section (TMB12_3). Both designs elute as one major species with retention time consistent with a monomeric protein in complex with DPC detergent (A), show distinct negative maxima in far UV CD spectra at 215 nm (B) that remain stable up to >70°C (D), and cooperative and reversible folding/unfolding transitions in DUPC LUVs (obtained at 25°C) (D),

We next sought to solve the structures of the designs to assess the accuracy of the computational design methods. While the design TMB10_165 did not form crystals in the conditions screened, TMB10_163 formed crystals which diffracted to 2.5 Å resolution (Table S1). The seven surface-exposed mutations between TMB10_165 and TMB10_163 are shown in Figure 3A. The four copies of the TMB10_163 in the asymmetric units had a structure very similar to the original Rosetta design, with an average RMSD of 1.4 Å over all backbone heavy atoms (Figure 3B) and featured the expected β-strand connectivity (shear number of 12). Although TMB10_163 nanopore activity was not observed, analysis of its structure using PoreWalker (41) and MOLE 2.5 (42) indicated the presence of a water-accessible cylindrical pore with an average diameter ranging from 4.2 to 5.3 Å in the four subunits (Figure 3C, Figure S15), matching the diameter of the pore calculated from TMB10_163 design model (4.6 Å). Most of the side-chains lining the pore had similar rotameric states in the crystal structure and the design model, with remarkable similarity at the level of the designed Tyr-Gly-Asp/Glu folding motifs (Figure 3D). We further determined the structure of TMB12_3 by NMR spectroscopy. Optimization of the in vitro folding conditions showed that the protein was structured in aqueous solution in LDAO detergent micelles, as indicated by well-dispersed amide and side chain methyl spectra (Figure 3E, Figure S16). Secondary chemical shifts indicated the presence of twelve β-strands arranged into polypeptide segments expected from the design (Figure S17). Amide and side chain methyl NOEs spanned a dense network of experimental connectivities that reached around the barrel circumference and thus confirmed the correct arrangement of the strands into the predicted barrel structure (Figure 3F). The TMB12_3 features the designed β-strands connectivity (shear number of 14) with the barrel closed by the canonical antiparallel β1-β12 seam (Figure 3G, Figure S18, Table S2). These experimental structures demonstrate that our computational design method can design TMB nanopores with precisely controlled shear, channel width and shape.

Fig. 3:

Experimentally determined nanopore structures closely align with the computational design models. A. Seven surface mutations differentiate TMB10_163 from TMB10_165. B-D. Crystal structure of TMB10_163 B. backbone superposition C. superposition of side-chains involved in key folding motifs in the lumen D. cross-sections superposition with the water-accessible pore shown as a gray surface. Water-molecules crystallized in the pore are shown as red spheres. E-G. TMB2_13 structure in LDAO micelles. E. 2D [¹⁵N,¹H]-TROSY NMR spectrum of [U-²H,¹⁵N]-TMB12_3 with sequence-specific resonance assignments. F. Long-range NMR NOE contacts mapped to the expected TMB12_3 hydrogen bonds (dashed black lines). Residues with amide assignment are shown in white and green, unassigned residues are shown in ash gray. Residues with β-sheet secondary structure are shown as squares, all others as circles. Bold outlines indicate available methyl assignments. NOE contacts are shown as red lines (long-range amide-amide, dashes indicating diagonal overlap) and blue lines (contacts involving side chain methyl groups). G. Ensemble of the 20 lowest energy solution NMR structures (β-sheets shown in brown).

Electrophysiology

Encouraged by the success in designing 10- and 12-stranded β-barrels, we set out to design TMBs with different numbers of β-strands and different shapes. We designed 12-stranded β-barrels with a triangular cross-section (eight designs), an oval cross-section (seven designs), or a rectangular cross-section (nine designs), as well as 14 β-stranded β-barrels (nine designs), incorporating the design features described above for the 10- and 12-stranded TMBs. The designs were obtained as synthetic genes and the proteins were again expressed in inclusion bodies. A relatively lower fraction of 12-stranded TMB designs with a rectangular (4/9 designs) and oval (4/7 designs) cross-section showed a prominent expression band SDS-PAGE gel in comparison to the square-shaped designs (8/9). This could be the result of less homogeneous distribution of β-sheet destabilizing amino acids (which are easier to introduce in bent than in flat β-sheet regions) in these designs, as suggested by a higher density of strong β-sheet islands co-localizing with predicted early folding regions (43) (Figure S19). The difficulty of de novo β-barrel design thus depends not only on the size of the TMB pore but also on the shape encoded into the blueprint. Because of the small number of designs pre-selected in silico thanks to the new AlphaFold2 filter, we proceeded to screening designs for nanopore activity directly after validating their folding into monodispersed β-sheet structures in DPC micelles (Figure S20). We evaluated the designs for their capacity to insert into planar membranes from dilute detergent solution and form conducting pores (Figure 4). We obtained both 12 (three triangular-shaped, three oval-shaped and two rectangle-shaped) and 14 stranded (two) TMBs that exhibited consistent and stable conductances at positive and negative voltage (Fig 4, 3rd and 5th columns), with multiple sequential insertions corresponding to current jumps of small integral multiples of the base pore conductance (Figure 4, 4th column). Based on the intensities of the current jumps, we estimated the conductances of single-channel events, which increased with pore size, as expected: the 10 stranded TMB design described above had a conductance of 108 ± 1.4 pS, which based on the cylindrical pore access resistance model (44) corresponds to a nanopore diameter of approximately 3.5 Å. The 12-stranded designs had similar conductances to each other (210-230 pS) despite their different shapes, consistent with a cylindrical nanopore of around 5 Å. The 14-stranded design had a conductance of 427 ± 2.7 pS consistent with a calculated pore diameter of 7 Å. The predicted diameters are close to the average expected diameters of 4.6 ±0.7 Å, 9.4 ± 0.8 Å and 10.6 ± 1.4 Å (calculated along the pore of TMB10_165, TMB12_3 and TMB14_8 design models, respectively, using MOLE 2.5 (42) (Figure S15)). In comparison to naturally-occurring pores used for sensing, such as OmpG which undergoes both transient and complete occlusion events by its solvent-exposed loops over a timescale of 100 ms (18, 45), our TMB designs show remarkably quiet conductances, with no occlusion events detected over 10 sec measurements (Figure S11). Varying the pore shape of the pore while keeping the size constant (Figure 4, first column) did not have a large effect on monovalent ion conductance, likely because of the relatively large pore constrictions and the flexibility of the long polar side-chains lining the channels (Figure S21). Nevertheless, we anticipate that modulation of the nanopore shape and chemical lining should allow control over the permeability of the pores to larger and more complex solutes in the future.

Fig. 4:

Conductance of designed nanopores. A. TMB10_165, B. TMB12_3, C. TMB12_oval_4, D. TMB12_rect_8, E. TMB12_tri_12, F. TMB14_8. i) Top view cartoon representation. ii) Vertical cross sections of the pore. iii) single channel conductance (smallest observed conductance jump). iv) sequential insertions of designed pore in planar lipid bilayer membrane from detergent solubilised sample at low concentrations. v) histogram of smallest measured current jumps for each design up to 50 pA. The applied voltage across the bilayer was 100mV and experiments were performed in a buffer containing 500 mM NaCl. A gaussian fit was carried out for the single channel current histograms for each design. For TMB10_165, 38 independent single channel jumps were identified from 3 recordings to plot the histogram shown. Similarly, 44 single channel insertions were identified for TMB12_3 (4 recordings), 29 insertions for TMB12_oval_4 (3 recordings), 30 insertions for TMB12_rect_8 (3 recordings), 45 insertions for TMB12_tri_12 (5 recordings) and 32 insertions for TMB14_8 (3 recordings) to plot the above depicted histograms.

Conclusion

Our results demonstrate that it is now possible to systematically design transmembrane β-barrels with conducting pores spanning a range of sizes and shapes. Despite the inversion of the hydrophobic-polar core compared to globular proteins, and the almost entirely local nature of the sidechain interactions, our approach enables TMB design with atomic level precision, as highlighted by the close agreement between the experimentally determined crystal and NMR structures and the corresponding design models. Whereas the shapes of globular proteins are largely determined by the packing of hydrophobic residues in a central core, the TMB shapes can be specified by strategic placement of glycine residues at which bending takes place to reduce strain. As previously observed for 8-stranded TMBs, a delicate balance between the optimization of tertiary structure energy and negative design (introduction of locally frustrated residues) to disfavor premature β-strand formation before membrane insertion was critical for the expression of the larger TMB nanopores in E coli inclusion bodies. In comparison with previously designed oligomeric protein nanopores - built from self-assembling α-helical (46, 47) or β-hairpin peptides (12) - the nanopores presented here have the advantage of being built from a single chain which enables controlled assembly of monodisperse nanopores without alternative oligomeric states and much greater control over the shape and specific surface properties of the transmembrane channel. While nanopore designs from oligomeric β-hairpins require lipid nanoparticles for solubility and assemble under electric current at lipid-lipid interfaces (12), the monomeric TMB designs fold efficiently into detergent micelles and lipid vesicles. The stability of the monomeric designs enables their spontaneous insertion into planar lipid membranes following dilution out of detergent micelles, opening the door to the use of synthetic transmembrane nanopores in commercial flowcells developed for single-molecule sensing and sequencing.

Monomeric integral TMBs such as OmpG have been turned into sensors by incorporating analyte-recognition motifs (48, 49) or biotin-bound (50, 51) antibodies in the solvent-exposed loops (7). However, the long disordered loops of naturally-occurring TMBs result in noisy reading, and the engineering of quiet pores with shorter or mutated loops has been a long-standing problem (17, 18, 45, 52). The TMBs described here were designed with the shortest loops compatible with folding, and allow up to 2 hours of quiet recording which would enable the detection and quantification of molecules at low concentration. As illustrated in the accompanying manuscript, the stability and robustness of the designed nanopore structures and conductances enables ready conversion into ligand gated channels with considerably lower noise and more comprehensible signal analysis than previously engineered channels.

Further advances in understanding the relation between the chemical properties of a nanopore and the directed detection of an analyte in the pore lumen (11, 53, 54), coupled with our new ability to design custom pores, opens up exciting new directions for sequencing and sensing since the pore size and the side chains lining the pore can be designed specifically for the desired application. Unlike native pores, which are finite in number, there is no limit on the number of distinct designed pores that can be generated. De novo design now enables pore geometry and chemistry to be custom built to be optimal for applications ranging from detection and selective transport of a wide range of molecules of interest to biopolymer sequencing.