Design of the 2D protein crystal “gatekeeper”The construction of the protein gatekeeper-sensor system is illustrated in Fig. 1. The 2D protein crystals to be used as analyte-triggerable membranes were built from l-rhamnulose-1-phosphate aldolase (RhuA)44, a square-shaped C4-symmetric homotetrameric protein (Fig. 1a). In earlier work, we showed that C98RhuA, a variant with Cys residues engineered in its four identical corners (positions 98) assembles via disulfide bond formation into porous, defect-free, and mono-layered 2D crystals with μm-scale dimensions under mildly oxidizing conditions36. Remarkably, the C98-C98 disulfide linkages acted as flexible hinges, enabling C98RhuA crystals to undergo a coherent structural phase change that resulted in the opening and closing of interstitial voids between the protein units; these pores could be induced to open to as large as 7 nm and then close to <1 nm without loss of crystallinity. The fully closed state of C98RhuA was thermodynamically favored due to solvent entropy effects. In subsequent work40, the C98RhuA crystals were engineered with pairs of negatively charged glutamate residues flanking each corner (E57 and E66), such that the fully closed state of the resulting variant (CEERhuA) is disfavored at thermodynamic equilibrium due to electrostatic repulsion between the glutamates, yielding an “ajar” state with a pore size of ~4 nm (Fig. 1a)40.Fig. 1: Design of the chemically triggered protein gatekeeper on a mesoporous silicon sensor.a Renders of the protein gatekeeper CEERhuA, highlighting its square shape, the locations of engineered cysteines at its corners (C98, in black) with proximal glutamate residues (E57/E66, in yellow) and the self-assembled 2D crystal topology, shown in the thermodynamically stable “ajar” state. The coloring indicates the net charge on each face of the CEERhuA monomer; red indicates net negative surface charge density. b Simplified depiction of the dynamic CEERhuA crystal motions that result from the addition of Co2+ ion and then CN–, alongside negative-stain transmission electron microscope (ns-TEM) images of the resulting CEERhuA crystals (unit cell depicted as a yellow square overlaid on the image; 2D Fourier transform is shown in the upper right inset of each image; scale bars: 5 nm). The left image corresponds to the Co2+-free “ajar” state, the central image to the Co2+-bound “closed” state, and the right image to the “ajar” state that results from the interaction of the Co2+-bound “closed” state of the crystal with CN–. Addition of CN– induces formation of the stable Co(CN)64- ion, displacing the Co2+ bound to the E57/E66 residues to allow the 2D structure to open back to its “ajar” state. c Diagram showing deposition of the 2D protein crystals onto the mesoporous silicon (pSi) sensor chip. The net negative charge density on the CEERhuA protein (red circles) enables electrostatic adhesion to the positively charged aminated pSi surface. The 2D crystals stack as multilayers on the chip; the process is assisted by excess Co2+, which coordinate to the negative faces of the CEERhuA crystals during deposition. d Plan-view (left) and cross-sectional (right) scanning electron microscope (SEM) images showing the boundary between a deposited multilayer of the protein and the pSi substrate. The top portion of the left image (labeled CEERhuA) shows the protein overcoat, the lower portion shows an un-coated region of the pSi substrate, revealing the ~15 nm mesopores. The right image shows the edge of a deposited film (80.6 nm thick); corresponding to ~11 individual 2D crystal layers on the pSi surface. Scale bars: 100 nm (left image) and 20 nm (right image).We had previously shown that addition of Ca2+ ions induced these 2D crystals of CEERhuA to collapse to their closed form via specific Ca2+ coordination to the pairs of E57 and E66 glutamate residues near the hinges while also mitigating their electrostatic repulsion40. For the present work, we induced the closed state of CEERhuA using Co2+ as a divalent metal ion (Fig. 1b)40. We chose Co2+ for these experiments because, like Ca2+, Co2+ exhibits a high affinity for carboxylate-based ligands such as the E57 and E66 residues to effect the closing of the CEERhuA crystals. However, in contrast to Ca2+, Co2+ forms a very stable hexacyanide complex [Co(CN)6]4– that we hypothesized would provide a strong and specific driving force to open the gates in the crystal45. To verify that Co2+ could induce the closed state, clarified suspensions of CEERhuA crystals were mixed in a 1:1 v:v ratio with MES-buffered 50 mM CoCl2 solution and incubated for ≥3 d to allow complete metal binding. The negative-stain transmission electron microscope (ns-TEM) images before and after Co2+ treatment indicated that Co2+ binding indeed prompted closure of the CEERhuA crystal pores en masse (Fig. 1b). The closure is further evidenced by the reduction of the unit cell dimension a from ~11.4 to ~11.0 nm, as determined from the fast Fourier transforms of the ns-TEM images (Fig. 1b).We then established the capability of the CEERhuA crystals to be switched from the closed to the ajar state in a cyanide (CN–)-dependent manner in aqueous solution. A mixture of Co2+-bound CEERhuA crystal suspensions 1:1 (v:v) with a buffered 150 mM KCN solution (final CN–:Co2+:CEERhuA molar ratio was 6:1:0.0002 to fully sequester all Co2+ ions, both solvated and gatekeeper-bound) led to the coherent opening of the crystals and restoration of the original unit cell dimension (Fig. 1b). Vigorous mixing of the Co2+-bound CEERhuA crystal suspensions, or dialysis to remove free Co2+ ions, also resulted in the opening of CEERhuA crystals, consistent with the role of Co2+ as a bridging ion. Taken together, these observations demonstrate that not only is the “closed” conformation of a 2D CEERhuA crystal stabilized by Co2+ ions, but also that Co2+-binding can be reversed upon exposure to CN–, restoring the “ajar” state of the crystal. Thus, the CEERhuA lattice represents a 2D framework possessing molecular-scale gaps that can undergo substantial dimensional changes in specific response to CN–. Importantly, the conformational dynamics of CEERhuA lattices are coherent in that their pore sizes change uniformly and synchronously, which distinguishes them from soft, polymer-based porous materials.Construction of the porous Si photonic crystal sensor with a protein crystal gatekeeperWe next deposited the gatekeeper onto the surface of the solid-state sensor (Fig. 1c). The HCN vapor sensing element was a two-dimensional photonic crystal prepared from porous silicon (pSi) and containing the HCN-specific indicator dye monocyanocobinamide (MCbi), described previously46. In addition to its ability to selectively sense HCN vapors, the pSi sensor was chosen for this study because it is susceptible to interference from ambient environmental chemicals (Supplementary Fig. 1) that degrade its performance—providing a validated platform to test the function of our 2D protein crystal as a “gatekeeper”. In the present system, the pSi substrate was modified with primary amines to impart a net positive surface charge and to enable immobilization of the CEERhuA crystals onto the top of the pSi substrate via noncovalent electrostatic interactions. A cyclic azasilane reagent (2,2-dimethoxy-1,6-diaza-2-silacyclooctane; “diazasilane”) was employed for this purpose, which reacts with Si-OH species at the surface of the pSi layer via a ring-opening reaction47. We also explored other amine-functionalized silanes, including the alkoxysilanes APDMES (3-aminopropyl(dimethyl)ethoxysilane) and APTES ((3-aminopropyl)triethoxysilane) and an heterocyclic azasilane (N-n-butyl-aza-2,2-dimethoxy-silacyclopentane), but found these to either yield weaker binding/coverage of the CEERhuA crystals or lead to the collapse of the surface pores and visible flaking (Supplementary Fig. 2). Ultimately, the diazasilane reagent 2,2-dimethoxy-1,6-diaza-2-silacyclooctane was selected due to its efficient reaction with hydroxy-terminated pSi and the lack of appreciable polymerization of the resulting siloxy species, which can lead to micro/mesopore clogging47. The presence of organic signatures assigned to the surface-grafted amine functionalities was confirmed by attenuated total reflectance Fourier transform infrared (ATR-FTIR) spectroscopy (Supplementary Fig. 2). The CEERhuA crystals possess a p4212 plane group symmetry in which the neighboring CEERhuA proteins are arranged in an alternating up-down fashion with respect to the principal four-fold crystal symmetry axis (Fig. 1a). Each CEERhuA protein monomer has one side (the C-terminal facet) that is rich in aspartic and glutamic acids, imparting a net negative charge density to that side. Because of the alternating up-down distribution of the monomers in the 2D crystal, both 2D crystal faces (which are equivalent) contain these negatively charged domains. We expected that these negatively charged domains would drive electrostatic adsorption of the layer of CEERhuA crystals onto the pSi surface (Fig. 1c).To validate this expectation, both amine-functionalized and untreated (“bare”) pSi chips were drop-cast with Co2+-bound CEERhuA crystal suspensions (~5 μM protein concentration) in parallel and allowed to dry overnight, followed by two rinses with a buffer solution to eliminate weakly or non-adsorbed species. For consistency, both of the fabricated pSi chips contained loaded MCbi. Scanning electron microscopy (SEM) imaging revealed extensive coverage of the amine-functionalized pSi wafers with Co2+-CEERhuA crystals, whereas the untreated pSi substrates were bare, displaying only the characteristic ~15 nm pores of the pSi substrate (Supplementary Fig. 2). Partially coated samples, with a portion of the Co2+-CEERhuA layers removed, are shown in Fig. 1d. Cross-sectional SEM images showed 80-to-120-nm thick Co2+-CEERhuA layers coated on the pSi substrates (Fig. 1d). Given that a single 2D CEERhuA crystal is approx. 7 nm thick, this observation indicates that the drop-casting procedure generated multilayers consisting of between 10 and 15 Co2+-CEERhuA crystalline sheets, likely promoted by the presence of excess metal ions in the buffer solutions that mediated interlayer stacking (Fig. 1c).Response of the gatekeeper-modified porous Si sensor to HCNWith the CEERhuA layers successfully tiled onto the pSi sensor surface, we next sought to verify that the pSi sensor element could still detect HCN. Synthesized by controlled electrochemical etch into a polished silicon wafer46,48,49,50,51, the pSi sensor element consisted of a complex stack of mesoporous silicon layers whose porosity was configured such that the resulting pSi photonic crystal displayed two specific resonances, known as stop bands (Fig. 2). These stop bands appeared in the white light reflection spectrum as peaks centered at 575 and 710 nm, and they acted as “Reference” and “Signal” channels, respectively, for the sensor46,48,49,50. The wavelength of the Signal channel was tuned–based on previously determined preparation conditions–to maximally overlap with the principal absorbance band (~580 nm) of CN-bound MCbi46,52,53. The absorbance maximum of the CN-free MCbi appears at 535 nm (Supplementary Fig. 3), so it did not significantly perturb the Signal or Reference channels of the photonic crystal. The sensor “Response” was quantified as a ratio of the “Signal” channel intensity to the “Reference” channel intensity as defined in Eq. 2 (Methods). Our prior work had demonstrated that with this approach, the pSi-MCbi photonic crystal sensors detect HCN reliably at concentrations as low as 5 ppm within 10 min of vapor exposure46. After impregnation of the pSi photonic crystal sample with the MCbi indicator, the sensor element was coated with the CEERhuA layers to complete the sensor-gatekeeper assembly (Fig. 2).Fig. 2: Construction and operation principle of the gatekeeper-sensor assembly.A porous silicon (pSi) photonic crystal is synthesized to display two resonances in the example optical reflectance spectrum, a signal peak, and a reference peak (“pSi photonic crystal”); an example plot for visualization is shown at left. A monocyanocobinamide (MCbi) indicator dye for HCN is then embedded in the porous layer. The wavelength of the Signal Peak coincides with the absorbance maximum of CN–bound to MCbi (580 nm), while the Reference Peak appears in a wavelength region where no significant optical absorbance occurs, either before or after HCN exposure. The dye-loaded film is thus capable of remote optical detection of gaseous HCN (“HCN Sensor”). The CEERhuA 2D crystals are then layered over the top of the amine-modified pSi HCN sensor as described in Fig. 1C. At this point the gatekeeper is in its “closed” form. Introduction of gaseous HCN (10 ppm, in air of 50% RH) disrupts the Co2+ links in the protein, causing it to open into its “ajar” state. HCN then enters the pSi layer, where it binds to MCbi. The resulting increase in optical absorbance of the MCbi-CN complex is detected as a decrease in the measured intensity of the “Signal Peak” relative to the “Reference Peak” in the optical reflectance spectrum. The sensor readout is quantified as the Response function defined in Eq. 2 of the methods section. Error bars are one standard deviation from the mean.When exposed to 10 ppm of gaseous HCN in air of 50% relative humidity, CN-binding to MCbi manifested as a decrease in the intensity of the “Signal” channel (Isignal) relative to the “Reference” channel (IReference) due to an increase in light absorption by the dye, resulting in a decrease in the magnitude of the Response function (Fig. 2). These data indicated that HCN could trigger a response in the pSi sensor even when the sensor was coated with the protein gatekeeper, although it did not establish if the CEERhuA protein layers were acting as a gatekeeper. To address this latter issue, we next looked at the ability of the CEERhuA protein layers to gate the transport of small molecule interferents into the pSi sensor.Rejection of interferents by the protein gatekeeper: molecular dynamics simulationsA common issue for sensors designed to detect trace toxins is interference from naturally occurring chemicals in the environment, which are usually present at much higher concentrations than the target analyte. Interferents can generate false positives, false negatives, reduction in sensitivity, or zero-point drift (drift of the sensor’s signal from its baseline, or resting state value, in the absence of the target analyte). The purpose of the protein gatekeeper in this work was to reject these molecular contaminants. Thus, with the HCN sensing capability of the gatekeeper-coated sensor assembly established, we next addressed the question of whether the opening and closing of the physical gaps in the 2D CEERhuA crystals could impact molecular transport to act as gatekeepers.To assess the theoretical possibility for such switchable permeability, we performed all-atom molecular dynamics (MD) simulations of infinitely periodic disulfide-linked RhuA lattices in the “closed” (Co2+-bound), and “ajar” (post-HCN-exposure) states (see Fig. 1b) to emulate conditions on the sensor surface and modeled the transport of an ensemble of n-hexane molecules through the lattice. Details of the calculations are provided in the Methods section of the Supplemental Materials and in Supplementary Fig. 4. In brief, solution-equilibrated structures of the RhuA lattices, including their first hydration layer around the proteins (expected to exist in the presence of ambient humidity), were determined and held fixed, while hexane was introduced to one side of the lattice. The simulations indicated that a single layer of the 2D crystal showed a 4–8-fold higher hexane transport rate when in the ajar state (>1 hexane/ns) relative to the closed state (Supplementary Fig. 4). For computational economy, we only simulated a single layer of a RhuA crystal. Because the thickness of the gatekeeper coating on the pSi sensors consisted of 10 to 15 stacked layers of CEERhuA crystals, the aggregate multilayer coating can be expected to be substantially less permeable than what was calculated for the single layer in the closed state. Conversely, opening of the lattice pores in a single layer was expected to be sufficient to permit significant transport of gaseous species through a multilayered structure. This expectation was borne out in the experimental data, where no ingress of hexane through the closed layer was detected in the 10-min timescale of the measurements, whereas substantial infiltration was observed for the ajar state (vide infra).Sensing of HCN in complex matrices using the gatekeeper-modified pSi sensorThe gatekeeper-coated pSi sensor was challenged with a complex mixture of non-specific interferant vapors. A typical data trace is given in Fig. 3a, showing sequential addition of dry N2, water vapor in air at 50% relative humidity (RH), a mixture of 43 different volatile organic compounds (VOCs) in air at 50% RH, and finally this same 43-VOC mixture, spiked with 10 ppm of HCN. The mixture of 43 VOCs used in this work was a composition of commonly occurring atmospheric chemicals that has been defined by the US Environmental Protection Agency (EPA) as “TO-14A” mixture46. The response function from the sensor showed little deviation from the zero point through the course of the experiments in the absence of HCN, and a substantial signal was recorded when the sensor was exposed to HCN. These results illustrate that the above interferents did not impede the sensor’s ability to respond to its target HCN analyte.Fig. 3: Performance of CEERhuA-pSi sensors in the presence of common interferents.a Response of the CEERhuA-pSi sensor to a series of potential interferents. The full sensor, consisting of the MCbi indicator dye embedded in the pSi photonic crystal, and coated with the Co2+-CEERhuA crystal multilayer as shown in Fig. 2 is evaluated. Successive 10-min exposures of the sensor to a flowing stream of: nitrogen, air of 50% relative humidity (RH), an EPA TO-14A mixture of 43 common volatile organic compounds (1 ppm of each component) in air of 50% RH, and 10 ppm of HCN spiked into the same EPA TO-14A mixture in air of 50% RH. The sensor response is quantified as the Rt/R0 response function as defined in Fig. 2 (relative to 0 s). Sensor shows negligible responses to all potential interferents but shows a strong response to the HCN analyte. b. Response values as in (a) but determined after 10 min of exposure to the indicated substance. All substances were diluted in flowing air to final concentrations of: [H2O] = RH at 22 °C as indicated; [N2] = 100%, water vapor added to obtain indicated RH at 22 °C; [NH3] = 100 ppm; [HCl] = 100 ppm; [n-hexane] = 3100 ppm; TO-14A mixture in air. Error bars represent standard deviation from triplicate measurements on independently prepared samples. c Evaluation of sensor performance in the presence of a massive quantity of hexane vapor interferent. Trace shows successive 10-min exposures of a single sensor to a flowing stream of hexane-saturated air, followed by 10 ppm of HCN vapor in air, followed by hexane-saturated air. Relative humidity (RH) was maintained at 50% throughout the run. Response values as defined in (a). All traces are representative of triplicate measurements on independent samples. Error bars are one standard deviation from the mean.We next tested a range of more reactive chemicals, which might potentially degrade sensor operation due to specific reactions with the protein or indicator dye components. In separate experiments, sensors were exposed to hydrogen chloride (HCl, 100 ppm) ammonia (NH3, 100 ppm), and n-hexane-saturated (3100 ppm) air (Fig. 3b). All the above potential interferants were delivered in humidified air streams of values 20, 50, and 80% relative humidity (RH). The sensors were exposed to the potential interferants for 10 min and the responses were compared with those obtained on separate, unexposed pSi chips using challenges of 5 and 10 ppm HCN vapor. As a reference, the HCN response to non-coated sensors was also evaluated. The (closed) gatekeeper-coated sensor showed no significant response to the vapor interferants at all values of RH tested, and it showed a similar response to HCN as compared to the non-gatekeeper-coated sensor (Fig. 3b). These results illustrate that the sensor did not give a false-positive signal from the potential interferents, and it performed reliably over the range of humidity values between 20 and 80% RH. Humidity conditions above and below these ranges were not evaluated in this study.Rejection of high concentrations of n-hexane by the gatekeeperBeing primary components of gasoline or diesel fuel fumes, aliphatic hydrocarbons are common interferents encountered in the field. For the pSi sensor used in this study (in the absence of the CEERhuA gatekeeper layer), air saturated with n-hexane generated a substantial zero-point drift in the sensor (Supplementary Fig. 1). Thus, hexane-saturated air provided a good test of the gatekeeper’s rejection capabilities. Nonpolar hydrocarbons also present a chemical environment that is significantly different from the aqueous environment where proteins normally function, so hexane-saturated air also provided a challenging medium to test the function of the protein gatekeeper.As a test of the ability of the closed gatekeeper to reject n-hexane but to admit it upon entering the open “ajar” state, the sensor chip (MCbi-impregnated pSi photonic crystal with an 80 nm-thick Co2+-CEERhuA gatekeeper top layer) was sequentially exposed to flowing n-hexane-saturated air, 10 ppm of HCN in air, then n-hexane-saturated air again (Fig. 3c). In the initial (closed) state, the sensor displayed no significant response to hexane-saturated air, consistent with the computational expectations and the experiments of Fig. 3b, which showed that the closed gatekeeper was not permeable to hexane. Opening of the gatekeeper upon exposure to HCN generated the expected response in the MCbi indicator; a large change in the Rt/R0 response function was recorded during a 10-min exposure to the analyte (Fig. 3c) that could not be reversed by flushing with a non-HCN air stream (Supplementary Fig. 5). Subsequent exposure of the sensor to n-hexane-saturated air generated a significant change in the measured signal that was opposite from the HCN response (Fig. 3c). Analogous sequential exposure experiments using NH3, another hydrophilic metal-binding interferent, showed no shift in response under any conditions (Supplementary Fig. 6).To understand the results of Fig. 3c, it must be stressed that the primary source of chemical specificity in the current sensor system is the strong and selective binding of CN– to the MCbi indicator dye, and the subsequent change in its absorbance spectrum. Hexane, as with most chemicals naturally existing in the atmosphere, does not bind to MCbi and thus does not generate such a colorimetric response, and it does not displace the strongly bound CN– from the indicator dye. Thus, the origin of the positive zero-point drift associated with hexane vapor in this system is not from its interaction with MCbi; rather, it derives from its non-specific adsorption within the pores of the pSi sensor. This produces a change in the average refractive index of the pSi film and in the index contrast in the photonic crystal, which results in a red shift of the stop bands and a decrease in their intensity (Supplementary Fig. 1)46. These effects are convoluted with the non-linearity of the spectral response of the spectrometer detector. While these phenomena exert relatively minor effects on the reflection spectrum of the pSi photonic crystal, at high concentrations of interferent they can overwhelm the sensor response to its target analyte49. In the present system, hexane-saturated air typically generated a +5% change in the zero-point of the sensor response function in the absence of the gatekeeper (Supplementary Fig. 1). As the absorbance peak of CN-bound Cbi is fixed (Supplementary Fig. 3), sufficiently large shifts in the centers of the reflectance peaks due to infiltration of fouling agents can result in artificially high Rt/R0 values even in the presence of HCN. The consequence can be observed in Fig. 3c after HCN exposure, where subsequent hexane infiltration created a perceived change in the opposite direction of the signal from the HCN analyte; thus, it generated a false negative response in the pSi sensor when the gatekeeper was open, and no response when the gatekeeper was closed (Fig. 3c).In addition to the information given by the response function Rt/R0, which derived from changes in optical absorbance of the MCbi dye upon its binding of CN–, the stop bands also displayed a small red shift, whose magnitude scaled with concentration of hexane in the pSi layer. This wavelength shift is related to the average refractive index of the pSi film, which is more dependent on the total mass of material within the pores than on the optical absorbance of the MCbi indicator dye included in the film, especially in the wavelength range of the “Reference” stop band, where the optical absorbance of MCbi is negligible50. Thus the red shift of the stop bands was used to confirm infiltration of the n-hexane vapor through the gatekeeper structures and into the mesoporous pSi chip (Supplementary Figs. 1, 6–9). The wavelength response to hexane vapors provided a useful in situ diagnostic of whether the gatekeeper was open or closed that was independent of the HCN sensing modality and confirmed that the gatekeeper was effective at blocking ingress of the organic vapors until it was triggered to open.Validation of the gatekeeper functionThe basic gatekeeper function was validated with a series of control experiments, summarized in Fig. 4. The Figure compares the response of the complete CEERhuA gatekeeper-sensor assembly with three control samples, in which the composition of the protein “gatekeeper” coating was varied. The first control used a 2D protein crystal constructed from the C98RhuA variant of the protein (“non-responsive C98RhuA”, Fig. 4a), which thermodynamically favors the closed state due to solvent entropy effects40. For the second control, the 2D CEERhuA crystals were prepared in the absence of added Co2+ (“Co2+-free Gatekeeper”, Fig. 4a), which generates the 2D lattice in its permanently “ajar” state as indicated in Fig. 1b. The third control sample was a “disassembled” form of the protein consisting of CEERhuA monomers that were lacking the disulfide bonds needed to bind their corners together, resulting in an amorphous protein layer when deposited onto the pSi chip (“disassembled gatekeeper”, Fig. 4a). The addition of tris(2-carboxyethyl)phosphate (TCEP), a non-sulfurous reducing agent, to RhuA crystals results in complete dissolution into individual proteins (Supplementary Fig. 10). This latter structure thus was expected to contain no voids either before or after exposure to HCN and was considered as an example of a permanently closed gatekeeper.Fig. 4: Comparison of the CEERhuA gatekeeper with inactive protein control samples: response to n-hexane and to HCN.a Diagram depicting the functional Co2+-CEERhuA-coated sensor in this study (upper left, “CEERhuA Gatekeeper”) and control samples containing the same HCN-active pSi sensor element as in the functional sensor, but with different protein layers on the pSi sensor surface (clockwise, from upper right): “Disassembled Gatekeeper”, an amorphous layer of CEERhuA protein monomers lacking disulfide bonds, co-deposited with Co2+ (expected to result in clogged pSi pores); “Co2+-free Gatekeeper”, a layer of 2D CEERhuA crystals prepared with disulfide links but without Co2+ (an “always ajar” gatekeeper); “Non-Responsive C98RhuA Gatekeeper”, a layer of 2D protein crystals using the C98RhuA variant of the protein (present in an “always closed” state). The small green spheres represent Co2+ ions; yellow bars represent glutamate residues E57/E66. All experiments were carried out on individually prepared chips. b (top) Response of the indicated samples to a flowing stream of n-hexane-saturated air (50% RH). Rt/R0 response function is as defined in Eq. 2; the zero-point of the sensor is Rt/R0 = 1.00. Sample response to n-hexane showed that the permanently ajar “Co2+-free Gatekeeper” is permeable, generating a positive zero-point drift in response to hexane infiltration in the pSi nanostructure. The “CEERhuA Gatekeeper”, the “Disassembled Gatekeeper” and the “Non-Responsive C98RhuA Gatekeeper” lattices were all impermeable and did not respond to n-hexane. Relative to “Co2+-free Gatekeeper”, the closed forms exhibited statistically distinct responses (CEERhuA Gatekeeper, p = 0.023, Disassembled Gatekeeper, p = 0.0085, C98RhuA Gatekeeper, p = 0.045). (bottom) Response of the sensors to HCN. In contrast, exposure of the sensors to HCN triggered a strong response in the functional sensor (“CEERhuA Gatekeeper”) and in the always-open “Co2+-free Gatekeeper” (p = 0.067); no significant response was detected in the “Disassembled Gatekeeper” (p = 7.8 × 10−6) and the “Non-Responsive C98RhuA Gatekeeper” (p = 9.2 × 10−6). The reported p values are relative to the “CEERhuA Gatekeeper”. Whereas the “Co2+-free Gatekeeper” permitted entry of both gaseous species, only the fully functional “CEERhuA Gatekeeper” sensor exhibited both HCN responsiveness and minimal zero-point drift (prior to HCN exposure). Data points represent averages of three independent measurements on independent samples. Error bars are one standard deviation from the mean. Statistical significance was obtained through Tukey’s multiple comparison test (summarized in Supplementary Table 3).The response of each of these controls, relative to the full gatekeeper-sensor assembly, to HCN and hexanes is shown in Fig. 4b. Response of the samples to a flowing stream of n-hexane-saturated air (50% RH) showed that the permanently ajar “Co2+-free Gatekeeper” control was permeable to n-hexane, generating a positive zero-point drift in response to hexane infiltration in the pSi nanostructure that differed significantly from the other samples, reaching a positive drift of >0.05 from the starting point. As noted above, the positive drift in the Rt/R0 value following interferent infiltration can produce false negative signals in this sensor, highlighting the need for selective permeability. In contrast, the “CEERhuA Gatekeeper”, the “Disassembled Gatekeeper” and the “Non-Responsive C98RhuA Gatekeeper” lattices were all impermeable and did not respond to n-hexane, with triplicate measurements within error of each other and remaining close to the initial zero-point of 1.00 (Fig. 4b). Consistent with the hexane response of each sample, exposure of the sensors to HCN triggered a strong response in the functional sensor (“CEERhuA Gatekeeper”) and in the always-open “Co2+-free Gatekeeper”, and no response in the “Disassembled Gatekeeper” and the “Non-Responsive C98RhuA Gatekeeper” (Fig. 4b). The HCN response was tightly distributed around the mean for all except the “Co-Free Gatekeeper”, which exhibits large standard deviations in the Response curve, though it is distinct from the impermeable controls. It is possible that the hypothesized role of Co2+ in mediating interlayer stacking (Fig. 1c) is important for a coherent gatekeeper response on the surface, and its absence leads to less consistent deposition that could account for the wider variance of this specific control.Of the above variants of the sensor, only the fully functional “CEERhuA Gatekeeper” sensor (as described in Fig. 1 and characterized in Fig. 3) exhibited both HCN responsiveness and minimal zero-point drift (prior to HCN exposure). Crucially, the lack of response for the “C98RhuA gatekeeper” demonstrates both that the presence of Co2+ alone is insufficient to produce a false-positive signal, and that conformational responsiveness of the crystal pores to metal binding is a requirement for selective permeability. Relatedly, the fact that the individual proteins can only occlude the surface but not selectively open in response to HCN (with all other aspects of the system remains the same; Fig. 4 and Supplementary Fig. 8) indicates that crystallinity is essential for the gatekeeper’s function.Additional control experiments were performed to verify that all components of the gatekeeper-integrated sensor were necessary for HCN-gated sensing via sequential exposure of hexane, HCN, and hexane. We also confirmed that NH3, another metal-binding vapor, could not substitute directly for HCN. The findings are summarized in Supplementary Table 2 and Supplementary Figs. 1, 6–9). These data confirm that the CEERhuA gatekeeper in its closed form effectively rejects small molecule interferents such as hexane, and when triggered into its open form, it allows the ingress of all molecules, including the HCN analyte. We discovered a singular exception to this behavior for ethanol (an amphiphilic molecule found in exhaust fumes) at 80% RH where limited permeability could be inferred via peak shift (Supplementary Fig. 11), though this effect was not observed at lower %RH, and no false-positive signal was observed at any conditions due to the selectivity of the MCbi dye. We characterized the effect of increasing ethanol concentrations on CEERhuA lattices in solution and confirmed that >10–20% (v/v) ethanol can induce lattice opening (Supplementary Fig. 12). These experiments indicated that despite the sensor robustness to a wide category of interferents, combined interactions between ethanol and water may disrupt the protein interfaces, which are also partially solvent-mediated40. Despite this narrow limitation, the key in achieving very high specificity in this system was the combination of specific interactions between HCN and the CEERhuA gatekeeper, and between HCN and the MCbi indicator dye.In conclusion, this work demonstrates the successful incorporation of a genetically engineered and structurally well-defined protein-based assembly into a solid-state device and the exploitation of its unique structural and dynamic attributes. The use case in this study focused on a passive indicator for HCN gas that can be assessed remotely through optical interrogation while remaining impervious to ambient clutter. The rejection of non-HCN chemical species present in the environment was achieved through the flexibility of the CEERhuA crystals, which only became permeable upon loss of specific chemical interactions (metal coordination) that were directly severed by HCN molecules. As a result, the sensor demonstrated resistance to erroneous readouts due to a variety of non-specific interferents: changing relative humidity, HCl or NH3 vapors in air, air saturated with hexane, and a complex mixture of VOCs (EPA TO-14A) in air, while displaying a response to low-ppm levels of HCN. The gatekeeping ability of the CEERhuA crystals was enabled by the specific interaction between the Co2+ metal ions and the carboxylate sidechains on the CEERhuA protein, which were engineered to maintain the crystals in their closed form upon binding to Co2+. The high affinity of Co2+ for HCN was key in displacing these protein-Co2+ bonds and triggering the protein into its open state, providing a closed/opened opening. The 2D structure of CEERhuA crystals was also important in that it enabled effective coverage of the pores in the sensor, blocking the ingress of interferent molecules that would normally cause an unacceptable zero-point drift in the sensor. Because the sensor transducer employed an optical readout, the transparency of the protein crystals in the visible region of the spectrum was also essential. Finally, the crystallinity of the protein mesostructure allowed it to exhibit its gatekeeper function in a non-liquid environment and immobilized on the surface of a Si chip, conditions that are generally not favorable for protein structure or function. Given that the structure and dynamics of the CEERhuA crystals were maintained through disulfide bonds with minimal footprints between the protein monomers, it should be possible to further engineer the protein surfaces such that the lattice opening/closing dynamics, and therefore, the gatekeeping modality, can be rendered responsive to different stimuli. In addition, future investigations could further explore gatekeeper performance, stability, and long-term reliability through evaluating gatekeeper responses at lower limit concentrations of HCN, shelf-life studies as well as more challenging conditions such as high-humidity or low-humidity environments.Proteins have featured prominently in functional devices and used in practical applications outside their native biological contexts, owing to their advantageous structural and physical properties, their functional specificities, and their ready accessibility to genetic and chemical manipulation54,55,56,57,58. Some notable examples of such applications include blood sugar monitors (glucose oxidase)59,60, antigen detection kits (antibodies and derivatives)61,62, and biosensors/DNA sequencing devices (membrane channels)63,64,65. In all these cases, this particular utility of a protein depends on its inherent structure and function as an individual unit. A major distinction of the present system is that it takes advantage of an emergent physical property (i.e., coherently dynamic 2D crystallinity) that is not present in the original protein building blocks but obtained through their self-assembly into a well-ordered material through properly chosen and placed chemical linkages. This assembly-based approach establishes a paradigm for the incorporation of designed protein assemblies into solid-state devices as functional and dynamic components.
