Fluid surface coatings for solid-state nanopores: comparison of phospholipid bilayers and archaea-inspired lipid monolayers

Nanopore-based resistive pulse recordings are emerging as a powerful single-molecule characterization approach with applications in nucleic acid sequencing [1–7], biomolecule detection [8–11], characterization of single proteins [12–17], and biomarker-based diagnostics [18]. For instance, we showed recently that solid-state nanopores whose walls are coated with a fluid lipid membrane make it possible to fingerprint proteins with regard to their volume, approximate shape, charge, dipole moment and rotational diffusion coefficient [15].

In nanopore-based single molecule characterization, a high-gain current amplifier applies a constant potential difference between two electrolyte compartments separated by a thin insulating membrane. The amplifier records the ionic current through a single, eletrolyte-filled nanopore in the membrane that connects the two compartments. As non-conducting macromolecules transit the pore, they typically cause transient reductions in the ionic current that contain information about the volume, charge, shape, and dipole moment of the translocating species [15, 19–22].

The earliest and much of the recent work in nanopore sensing employed biological protein pores spanning a planar lipid bilayer [23, 24]. The dimensions of these pores are well defined and typically constant from one protein nanopore to the next, as long as the protein's amino acid sequence is not altered; in addition, protein pores often display minimal non-specific interactions with most analytes of interest [3, 25]. The diameters of the narrowest constriction of wildtype protein pores are, however, limited to a range between 0.4 and 3.3 nm, precluding the translocation of midsize and large proteins as well as other large analytes [2]. Recently, nanopores fabricated in synthetic substrates have enabled the analysis of larger particles [11, 26, 27], but surface interactions between these solid-state nanopores and biomolecules can result in undesired non-specific adsorption and clogging [16, 28, 29].

To reduce non-specific adsorption to the substrates of solid-state nanopores, various groups explored surface treatments such as silanization [30–32], self-assembled monolayers of thiols on gold films [33–35], atomic layer deposition of metal oxides [36, 37], and surfactant coatings [38, 39]. Some of these coatings also functionalized the pore surface [14, 40, 41] to improve detection efficiency [13, 42, 43]. In 2011, our group introduced a fluid lipid membrane coating [16], which eliminated or minimized non-specific interactions with the pore walls and enabled conjugation of proteins to mobile lipid anchors in the membrane [16]. This lipid anchor strategy significantly slows the translocation speeds and rotational dynamics of tethered proteins while circumventing undesired non-specific interactions with nanopore walls [15]. Slow translocation speeds enable current recordings that resolve most translocation events fully in magnitude and time and yield a large amount of data from individual translocation events. Together, these benefits of long-lived translocation events bolster the information content and statistics from individual resistive pulses. The strategy of tethering proteins to lipid anchors therefore enables the characterization of individual proteins beyond the determination of their volume and translocation time [15]. In particular, by eliminating or minimizing non-specific interactions with the pore walls, lipid coatings make it possible to quantify and interpret translational and rotational dynamics of lipid-anchored proteins as they move through the nanopore [15]. This approach makes it possible to extract the protein's dipole moment, and rotational diffusion coefficient in addition to its shape and volume from an individual translocation event and is a first step towards identifying and counting individual proteins in mixtures [15]. While alternative nanopore coatings can also minimize clogging of nanopores and have the advantage that they are more straightforward to apply to the pore walls and chemically more robust than lipid coatings, their disadvantage is that they have not yet been able to minimize non-specific interactions to pore walls sufficiently to allow for the quantification of single molecule protein characteristics that rely on translocation and rotation, such as protein charge or dipole [35, 44]. Notably, the ability to anchor a protein to the bilayer allows for the determination of charge based on the distribution of translocation times for that protein [15, 16].

Here, we sought an ideal lipid-based surface coating that coats synthetic nanopores in silicon-nitride substrates in a straightforward and reproducible manner, while maintaining a stable and low-noise electrical baseline and slowing down the translational and rotational speed of non-spherical, lipid-anchored proteins such that orientation-dependent modulations in current during a translocation event could be resolved and quantified in time. In order to evaluate the quality and integrity of these bilayer coatings, we took advantage of measurements of protein charge. To do so, we recorded translocation time distributions of three lipid-anchored proteins and fit these distributions to the equation for biased first-passage time [45] with protein charge as a fitting parameter. Close agreement of the protein charge determined from the fit with the theoretically expected charge indicated that proteins translocated ideally and as theoretically predicted by the first-passage time equation; if non-specific interactions within the pore would have slowed down the proteins, then the translocation time could not be fitted well by the biased first-passage time equation and the determined protein charge would be significantly lower than expected.

Applying a fluid lipid coating to a solid-state nanopore can increase the low frequency noise in current recordings, likely due to undulations of the coating [46]. As the ability to identify translocation events in a current recording is limited by the signal-to-noise ratio [47], an ideal nanopore coating would add minimal additional noise to the system or would even reduce the noise [16]. A coated nanopore must also yield a baseline current that varies minimally over various time scales, as fluctuations can lead to artifacts in the analysis.

The lateral diffusion coefficient (D_L) of lipids in a membrane is an important parameter in resistive pulse experiments that employ lipid-tethered analytes [16]. Yusko et al and Plesa et al demonstrated that only a small fraction of untethered proteins translocating through nanopores take sufficiently long to be time-resolved due to limitations of the recording bandwidth and the signal-to-noise ratio of the equipment [16, 48]. Proteins with lipid tethers that are anchored in a viscous lipid membrane coating, however, have more than 100-fold prolonged dwell times (t_d) in the pore compared to untethered proteins because the lateral diffusion coefficient of a tethered protein is dominated by the diffusion of its lipid anchor [16]. Slow protein movement is favorable since it increases the fraction of translocation events with dwell times longer than the time-resolution threshold of the recordings [16].

The work presented here, together with earlier work [15, 16, 49, 50], shows that nanopore fabrication methods and post fabrication processing steps affect the ease of coating a pore and the stability of the coating during nanopore experiments. In this paper, we coated pores that were made by three different fabrication techniques: ion-beam sculpting [7, 51], ion-beam drilling [52–54], and dielectric breakdown [50, 55]. A final goal of the work presented here was to identify the attributes of lipid coatings and pore fabrication strategies that facilitate reproducible formation of membrane coatings.

Figure 1 shows the chemical structure of the different classes of lipids tested in this study. In an attempt to decrease noise levels and increase the overall stability of the coating, we added various proportions of cholesterol into bilayers of 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC). Cholesterol increases stability in lipid membranes [56] because of its effect on membrane structure and packing [57–59]. Additionally, membranes with increasing proportions of cholesterol are more viscous, with correspondingly reduced D_L values [16, 60]. Here, we hypothesized that increasing the cholesterol concentration in the coating would prolong protein dwell times and lead to coatings with good stability.

Zoom In Zoom Out Reset image size

We also tested fluid lipid coatings inspired by extremophile organisms of the kingdom archaea, which often have cell membranes composed of bolaamphiphilic lipids [61, 62]. These lipid structures have two terminal polar head groups connected covalently by a hydrophobic alkyl chain and form lipid membranes composed of fluid monolayers instead of bilayers. These lipids are thought to stabilize cell membranes, allowing extremophiles to survive in environments with extreme temperature, pH and salinity [61, 62]. To test if coatings from these lipids would be beneficial, we coated nanopores with a class of chemically synthesized archaea-inspired bolaamphiphilic lipids containing two glycerophosphocholine heads, which were attached to glycerol by an ether-bond to one pendant phytanyl chain and to one end of a membrane-spanning hydrocarbon chain (figures 1(A)–(C)). We have recently examined the ion permeation kinetics of membranes composed of these lipids [63–67] and since we had a selection of these lipids available, we tested eight of them.

For comparison, we additionally tested coatings composed of two commercially available and widely used non-bolaform archaea-inspired lipids: 1,2-diphytanoyl-sn-glycero-3-phosphocholine (DiPhyPC) and 1,2-di-O-phytanyl-sn-glycero-3-phosphocholine (Di-O-PhyPC).

Finally, we hypothesized that the discrepancy in ease of coating between different pores may be due to heterogeneities in the 'sharpness' of the radii of curvature of the corners at the brims of the pores. As we have reported previously, the success rate of forming a lipid coating on a nanopore chip can vary from 10% to 50% depending on the chip and we found that nanopores with diameters smaller than 17 nm [50] and pores with sharp edges of extreme curvature could, so far, not be coated with lipid membranes. Similar limitations of forming lipid membranes with extreme curvature have been reported for phospholipid vesicles [68] and for lipid coatings of nanoparticles [69–71]. We therefore tested if lipid membrane compositions that form more flexible coatings with regard to membrane curvature, when compared to those composed of cylindrically shaped lipids, would be able to conform more readily to sharp corners and thereby coat nanopores with extremely small radii of curvature. In order to enhance the coating's ability to conform to highly curved surfaces, we chose a 'curvature-tolerant' lipid composition containing lipids with a cone-shaped geometry (1-oleoyl-2-hydroxy-sn-glycero-3-phosphocholine or LysoPC), with a complementary inverted cone shape (1,2-dioleoyl-sn-glycero-3-phosphoethanolamine or DOPE), and with 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC), a cylindrically shaped lipid [72].

Figure 2 shows that coating a nanopore with a lipid membrane had only a relatively small effect on the noise in the most important frequency range, above 5 kHz for most lipid compositions, as reflected in the power spectra (distributions of noise power versus frequency, figures 2(B)–(D)). In contrast, lipid coating increased mostly the noise present in current recordings in the low frequency range (1/f noise) [47]. Figure 2(A) shows plots of the root mean square (rms) current noise after coating a nanopore chip divided by the rms noise before coating a chip with lipid membranes of various compositions. Bilayer coatings of the phytanyl-bearing lipids DiPhyPC and Di-O-PhyPC produced the largest increase in noise, likely due to unruptured liposomes near the pore entrance, shown in figure 3. Curvature-tolerant lipid bilayers also significantly increased the noise in current recordings. These elevated noise levels putatively stem from the diversity of lipid shapes present in the bilayer. Given that membranes containing conical lipids are prone to forming curved regions [73], membrane undulations along the pore walls are likely more pronounced when the membrane is a mixture of curvature-tolerant lipids compared to membranes from non-conical lipids.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Increasing the proportion of cholesterol in POPC membranes had no significant effect on the noise; in general lipid coatings of pure POPC with 10 to 40 mol% cholesterol showed the smallest increase in noise upon coating with a tendancy towards the lowest noise values with 20 mol% cholesterol. These favorable noise properties (figures 2(A) and (C)) are likely due to the structural changes cholesterol imparts to lipid membranes [57, 74]: since the addition of cholesterol improves the packing of membranes, cholesterol-doped membranes may be more stable and fluctuate less than in the absence of cholesterol. Overall, we observed the lowest noise recordings with coatings from POPC-cholesterol mixtures and from archaea-inspired lipid membranes that contained 80 to 100 area% of archaea-inspired lipids. Comparing the various types of archaea-inspired lipids shown in figure 1(A), we observed no differences with respect to the noise and stability of nanopore coatings.

With most of the lipid compositions tested here, the current baseline measured from lipid-coated nanopores was stable for the average duration of a protein translocation experiment (around one hour). Experiment-ending instability was rare, and when it did occur, it generally began after the addition of protein. Lipid coatings made from DiPhyPC and Di-O-PhyPC bilayers were more prone to unstable baseline currents than coatings of other lipid compositions. Epifluorescence microscopy revealed that during the coating procedure with these two lipids, many small unilamellar vesicles (SUVs) did not burst despite the osmotic pressure gradient. The resulting surface-attached residual SUVs likely contributed to baseline fluctuations due to the proximety of these liposomes to the nanopore entrance. Figure 3(A) displays a fluorescent image of a well-coated nanopore made with liposomes from POPC lipids, with the corresponding stable current trace in figure 3(B). In contrast, figure 3(C) shows a fluorescent image of liposomes from DiPhyPC lipids on the chip window with the corresponding unstable current trace in figure 3(D).

Figure 4 shows D_L values obtained via fluorescence recovery after photobleaching (FRAP) experiments that were carried out on the SiN_x membrane of nanopore chips after coating with various lipid compositions. The inset shows that D_L values of coatings from POPC lipids could be varied by a factor of ∼3 by including 10 to 50 mol% of cholesterol in the lipid composition. Overall, the results in figure 4 show that D_L values measured in nanopore coatings composed of membrane-spanning archaea-inspired lipids were approximately 10-fold smaller than for the D_L of bilayers of pure POPC or DiPhyPC, DOPE and LysoPC. Because of their strongly increased viscosity, these monolayer-forming coatings from archaea-inspired lipids doubled the fraction of protein translocation events with dwell times longer than 400 μs compared to experiments with lipid membranes of pure POPC when we used the same relative amount of lipid anchor and protein. This high viscosity of coatings from archaea-inspired lipids also reduced the event frequency by a factor of approximately 10 compared to bilayer coatings at the same concentration of tethered protein and resulted in detectable event frequencies smaller than 1 Hz. Such low event frequencies pose two challenges: first, they lengthen the required duration of each experiment to collect a representative number of translocation events, thereby necessitating long-term stability of membrane coatings. Second, resistive pulses stemming from noise artifacts (which have a very low but non-negligible statistical likelihood [75]) during long recording times gain more influence over the population of collected events. To avoid impractically long durations of the experiments, the frequency of resistive pulses should range between 1 and 10 Hz, and hence the slow diffusion of lipid-anchored analytes should be compensated for by increasing the density of lipid anchors in the coating. For the experiments reported here, we used a fraction of 1 mol% of potential lipid anchors but a fraction of 10 mol% can be achieved with most anchors if desired.

Zoom In Zoom Out Reset image size

To investigate whether the lipid tether or the overall lipid composition of the coating dominated the observed diffusion of lipid-anchored proteins, we used an archaea-inspired membrane-spanning lipid to anchor translocating proteins in both bilayer-forming and monolayer-forming lipid compositions. To our surprise, lipid anchors from monolayer- or bilayer-forming lipids did not produce a significant difference in the event frequency or distribution of t_d values when compared to respective experiments that used a shorter, bilayer-forming lipid to anchor the proteins. In addition to t_d values, we also used FRAP experiments to compare the diffusion constants of fluorescently-labeled bilayer-forming and fluorescently-labeled monolayer-forming lipids both in a supported lipid membrane made from archaea-inspired monolayer-forming lipids and in a coating of bilayer-forming lipids. The resulting D_L values likewise were not significantly different as a function of the lipid to which the fluorophore was attached but rather depended on the viscosity of the respective overall lipid composition of the membrane. These results indicate that the background lipid composition defined the lateral diffusion of the tethered species. Figure 1 shows that we conducted experiments with six different archaea-inspired lipids. Results with regard to membrane viscosity were similar across this set of lipids [65], especially when compared to non-bolaamphiphilic lipid membranes, which is why we did not differentiate between different archaea-inspired lipids.

During this investigation, we found that nanopores fabricated by ion beam sculpting produced stable coatings more frequently than nanopores fabricated by ion beam drilling or dielectric breakdown [50, 52]. None of the lipid compositions that we tested, including the curvature-tolerant lipid mixture, had a significant effect on the success rate or ease of coating. We found, however, that pores made by ion beam drilling or dielectric breakdown, which initially resisted coating, often could be coated more readily by all lipid compositions after ten or more cleaning cycles (see Materials and Methods). Since these cleaning cycles were frequently accompanied by an increase in the pore diameter as quantified from current measurements, we hypothesized that pores fabricated by these two methods initially contained rougher surfaces or sharper edges than those formed by ion beam sculpting and that these edges were smoothened over time by the slow surface etching induced by the cleaning solution or by the aqueous electrolyte during experiments [76].

In contrast, most of the nanopores that coated reliably well in this work were fabricated using an ion beam sculpting technique, in which He ions, which are directed towards a focused ion beam-drilled pore (∼100 nm diameter), cause the pore to shrink to a desired diameter through mass flow of silicon nitride [7]. Once the sculpting process was complete, each ion beam sculpted pore was temperature treated during an annealing process in an inert atmosphere [76]. Most of the pores produced by ion beam drilling or dielectric breakdown that were difficult to coat initially did not undergo annealing during production [52, 77]. Rollings et al [76] and we [50] have shown that annealing smoothens the shape at the rim of silicon nitride nanopores and that this effect likely renders nanopores amenable to coating with lipids. Rollings et al also demonstrated an increase in the stability towards slow etching in electrolyte solution of uncoated ion-beam sculpted pores after annealing, possibly due to a decrease in the number of dangling bonds left over from the fabrication process [76]. AFM images taken by Rollings and TEM images taken by us indeed indicated a smoother surface after annealing [50, 78]. Asghar et al showed that annealing also alleviates residual stress in silicon oxide nanopores [79]. We hypothesized that these effects together may have increased the success rates for generating lipid coatings of high quality. This hypthesis is supported by the observation that pores formed by controlled dielectric breakdown [55, 80, 81] initially were difficult to coat, while an annealing step doubled the success rate of coating and improved the noise characteristics in both uncoated and coated states [50]. Altogether, these results indicate that pore morphology is a critical parameter for the success of coating nanopore chips with lipid membranes.

One major advantage of coating nanopore walls with fluid lipid coatings is that this approach makes it possible to slow down the translocation of proteins by means of attaching them to a lipid anchor. The viscous drag of this anchor slows down the translocation of tethered proteins by two orders of magnitude, making it possible to resolve complete distributions of translocation times, t_d. These t_d distributions reveal the net charge of the protein by fitting the distribution to the equation for a biased first-passage time process[16, 45]:

$$\begin{eqnarray}&&P\left({t}_{d}\right)=\,\displaystyle \frac{{l}_{p}}{\sqrt{4\pi {D}_{L}{t}_{d}^{3}}}{e}^{-{\left({l}_{p}-v{t}_{d}\right)}^{2}/4{D}_{L}{t}_{d}},\end{eqnarray} \tag{ 1 }$$

where v = ∣z∣eV_pD_L/l_pk_BT. The parameter ∣z∣ (unitless) represents the net number of unit charges carried by the protein, V_p (V) is the drop in electrical potential across the nanopore, k_B (J K⁻¹) is the Boltzmann constant and T (K) is the temperature of the system. This approach assumes that the lateral diffusion coefficient of the lipid anchor with the attached protein is the same as the average D_L value of the lipids in the bilayer coating [16].

For charge determination in previous work [15, 16], we have typically fit the experimentally-determined probability distribution of t_d values with equation (1). However, when we applied this approach to data from experiments with nanopore coatings from archaea-inspired lipids, the determined charge values were unrealistically high and the model could not fit the data (with adjusted R² values <0.5 in all cases). Close inspection of the recorded t_d data revealed a large population of short dwell times (<40 μs) that dominated the distribution and were too short to originate from proteins that were lipid-anchored to a highly viscous coating from archaea-inspired lipids. We hypothesized that this first population of short translocation times primarily represented false events due to fluctuations in the baseline noise [75] and proteins that have lost their lipid tether during the course of the experiment.

To account for these artifacts, we first integrated equation (1) to obtain the cumulative density function (CDF). This change allowed us to perform fitting on the data without the influence of binning. Next, we expanded the CDF to allow for the fitting of three separate populations within the cumulative distribution of measured t_d values. This expansion of the fitting equation made it possible to account for (1) the first population of t_d values shorter than 40 μs that we attributed to artifacts, (2) an intermediate population with the range of expected t_d values, and (3) a small, third population with dwell times longer than 1 ms, representing events of a few proteins that may have interacted non-specifically with the pore. We then determined the charge of three different proteins from the fit to the 2nd population in the middle of the distribution, resulting in mean charge values within ±30% of reference values (see figure S1, available online at stacks.iop.org/NANO/30/325504/mmedia) [15]. Further details on the integration of equation (1) and the function for three-population fits can be found in the Supplementary Information.