Strain-Based Chemical Sensing Using Metal-Organic Framework Nanoparticles

Metal-organic frameworks (MOFs) have received much attention for their potential as chemical sensors, owing to unparalleled tunability of their host-guest response. However, because of the limited compatibility between MOF properties and sensor transduction mechanisms, very few MOFs have successfully been integrated into practical devices. We report the fabrication of a strain-based sensor constructed from MOF nanoparticles deposited directly onto a membrane-type surface stress sensing architecture, which exhibits response times on the order of seconds and ppm-level sensitivity towards volatile organic compounds (VOCs). We show that an array of four types of MOF nanoparticles allows for clear discrimination between VOCs, using principal component analysis of their response profiles. This work opens up the possibility of VOC sensing using a wide range of MOFs, beyond those that are electrically conducting or those that form oriented thin films, with the added advantages of high sensitivity and rapid response. Abstract Metal-organic frameworks (MOFs) have received much attention for their potential as chemical sensors, owing to unparalleled tunability of their host-guest response. However, because of the limited compatibility between MOF properties and sensor transduction mechanisms, very few MOFs have successfully been integrated into practical devices. We report the fabrication of a strain-based sensor constructed from MOF nanoparticles deposited directly onto a membrane-type surface stress sensing architecture, which exhibits response times on the order of seconds and ppm-level sensitivity towards volatile organic compounds (VOCs). We show that an array of four types of MOF nanoparticles allows for clear discrimination between VOCs, using principal component analysis of their response profiles. This work opens up the possibility of VOC sensing using a wide range of MOFs, beyond those that are electrically conducting or those that form oriented thin films, with the added advantages of high sensitivity and rapid response. act as to rapidly the whole layer. prior to this work it to be seen whether receptor layers built from MOF nanoparticles be able to induce sufficient strain in the MEMS sensors, in order to realize effective VOC sensing. We herein report the fabrication and performance of a new MOF–MSS sensor based on nanoparticles of the canonical ZIF family of MOFs. 38 We first demonstrate the concept using ZIF-8 (Zn(2-methylimidazolate) 2 , which exhibits hydrophobic pores with the sod network topology. 38 We observe selectivity in its range of responses to 26 VOCs, with response times of 1-30 s and ppm level sensitivity. Exploiting the versatility of the MSS architecture and the diversity of available MOF nanoparticles, we use spray-coating to fabricate a array of , , and , The chemical diversity these itself wide range sensing profiles, which enable clear discrimination between a range different via component analysis.

owing to unparalleled tunability of their host-guest response. However, because of the limited compatibility between MOF properties and sensor transduction mechanisms, very few MOFs have successfully been integrated into practical devices. We report the fabrication of a strain-based sensor constructed from MOF nanoparticles deposited directly onto a membrane-type surface stress sensing architecture, which exhibits response times on the order of seconds and ppm-level sensitivity towards volatile organic compounds (VOCs). We show that an array of four types of MOF nanoparticles allows for clear discrimination between VOCs, using principal component analysis of their response profiles. This work opens up the possibility of VOC sensing using a wide range of MOFs, beyond those that are electrically conducting or those that form oriented thin films, with the added advantages of high sensitivity and rapid response.

Introduction
Sensing of volatile organic compounds (VOCs) is critical to our perception of the environment around us, 1 monitoring of harmful emissions, 2 and healthcare analytics, 3 and so is required for a wide variety of current and future technologies. 4 Sensors based on metal-organic frameworks (MOFs) offer great potential, in particular towards selectivity, owing to their precisely defined pore structure and chemistry. [5][6][7][8] The modular nature of MOFs allows them to be tailored to absorb small molecules with higher selectivity than conventional materials such as polymers, zeolites and porous carbons. 9,10 Luminescence and other optical transduction modes are most widely reported in the MOF sensing literature; 11 however, for practical usage other modes that integrate more easily with existing electronics are more viable. 6,12 Whilst there have been encouraging reports of electronically-responsive MOFs, [13][14][15][16][17] MOFs can be ideal candidates for strain-induced chemical detection because of the deformations of coordination space within their crystal structures caused by host-guest interactions. 9,[18][19][20][21] Micro-electro-mechanical systems (MEMS) sensors coated with ZIF-8 and HKUST-1 thin films have been shown to successfully respond to water, alcohols and other volatile organics, with sensitivity limits that exceed other mass-sensitive sensors. 22,23 However, several challenges remain, including improving the ease of receptor layer preparation, selectivity, sensitivity and response time. 24,25 Often, parameters act against one another; e.g., a thicker film may lead to increased sensitivity but at the cost of response time, owing to the time taken for analytes to diffuse through the MOF. In addition, the perceived necessity for well-adhered and defect-free, oriented thin films to effectively transmit analyte-induced strain to the sensor surface puts limits on the range of MOFs that can be used and introduces stringent, often laborious requirements to MEMS device fabrication. 5,24 On the other hand, a wide range of MOFs can now be easily synthesized in colloidal or nanoparticle form, 26  We recently reported the Membrane-type Surface stress Sensor (MSS), which consists of a Sibased membrane suspended by four piezoresistive beams, composing a full Wheatstone bridge. 27 This architecture generates potential difference upon changes to the strain state of the membrane, with approximately 100 times greater sensitivity than conventional piezoresistive microcantilevers, and can be extended to multiple sensor arrays. 28 In order to induce strain on the membrane, a receptor layer bound to the membrane must undergo strain in response to an analyte. Modelling has predicted that receptor layer materials with higher Young's moduli give better signals. 27,29 Amongst a wide range of MOF materials available, zeolitic imidazolate frameworks (ZIFs) exhibit mechanical properties somewhat intermediate within the MOF class of materials: they are stiffer than so-called "breathing" MOFs such as MIL-53 and pillaredlayer MOFs, but still exhibit measurable flexibility in their crystal structure upon gas uptake. [30][31][32][33] Encouragingly for their potential sensing performance, it has been shown that particle size reduction to the nanoscale can result in rather linear gas adsorption isotherms, 34 and improved mechanical properties. 35 In addition, gas diffusion coefficients of bulk MOFs are often several orders of magnitude lower than in air; 36,37 therefore, it could be reasonably expected that the meso-and macro-pores within a MOF nanoparticle superstructure will act as channels for analytes to rapidly diffuse through the whole receptor layer. However, prior to this work it remained to be seen whether receptor layers built from MOF nanoparticles would be able to induce sufficient strain in the MEMS sensors, in order to realize effective VOC sensing.
We herein report the fabrication and performance of a new MOF-MSS sensor based on nanoparticles of the canonical ZIF family of MOFs. 38 We first demonstrate the concept using ZIF-8 (Zn(2-methylimidazolate)2, which exhibits hydrophobic pores with the sod network topology. 38 We observe selectivity in its range of responses to 26 VOCs, with response times of 1-30 s and ppm level sensitivity. Exploiting the versatility of the MSS architecture and the diversity of available MOF nanoparticles, we use spray-coating to fabricate a 2×2 array of ZIF-8, ZIF-7 (Zn(benzimidazole)2, sod), ZIF-65-Zn (Zn(2-nitroimidazole)2, sod) and ZIF-71 (Zn(4,5-dichloroimidazole)2, rho). The chemical diversity of these MOFs manifests itself in a wide range of sensing profiles, which enable clear discrimination between a range of different VOCs via principal component analysis.

Materials and methods
Synthesis. All chemicals are purchased from Tokyo Chemical Industry Co. Ltd., Sigma-Aldrich Co., Wako Pure Chemical Industries, Ltd., Kanto Chemical Co. Ltd. and Nacalai Tesque, Inc. ZIF-8 nanoparticles were synthesized following a literature protocol, 39 washed thrice in methanol to remove residual byproducts and resuspended in alcohol. ZIF-7, ZIF-65-Zn and ZIF-71 nanoparticles were synthesized following similar literature routes (see ESI Section S1 for full details). 40,41 Chemical composition, particle size and phase purity of bulk samples were confirmed by powder-XRD, FTIR, SEM, DLS, zeta potential, TGA and BET before and after activation under similar conditions used for sensor preparation (see ESI Section S2-7). Nitrogen gas sorption isotherms (ESI Section S6) indicate significant mesoporosity and/or macroporosity in addition to the expected MOF microporosity, in line with previous reports. [39][40][41] Sensor preparation. Materials were deposited directly onto the MSS membrane without any adhesive layer. For selectivity and VOC discrimination experiments, ZIF nanoparticle suspensions were deposited onto a MSS membrane array by spray coating 30 layers followed by washing with methanol; the membrane temperature was held at 100 °C, in order to rapidly evaporate the carrier solvent, and promote sintering and adhesion of the nanoparticles to the surface. For investigation of sensitivity and response time, a ZIF-8 nanoparticle suspension was deposited on a single MSS membrane by inkjet printing; an inkjet spotter (LaboJet-500SP) and a nozzle (IJHBS-300), which were purchased from the MICROJET Corporation, were utilized. The ZIF-8 nanoparticle suspension was loaded into the inkjet printer module, and up to 2500 sequential droplet depositions were performed. The inkjet stage was heated at 80 °C to control evaporation.
Sensing. Selectivity tests using the spray-coated ZIF-8-MSS were performed under ambient temperature using the saturated vapours of 26 VOCs, including those a range of alcohols, carbonyls, arenes and alkanes. Gases were introduced to the sensor for 30 s via a custom-built setup (see ESI Section S8) and purged with nitrogen gas for 30 s. Four injection-purge cycles were performed for each gas and data were recorded at a sampling rate of 20 Hz by applying a voltage of -0.5 V to the Wheatstone bridge. VOC discrimination tests using the 2x2 MOF-MSS array were performed in an identical manner. Sensitivity and response time were investigated using the inkjet-printed ZIF-8-MSS for 12 VOCs under conditions of constant temperature and humidity using a separate setup (see ESI Section S9). Gases were diluted to 2 %, 5 % and 10 % of their saturated vapour concentrations and humidified at 0 %, 10%, 40 %, 70 % and 90 % RH prior to injection. 10 injection (10 s) -purge (10 s) cycles were performed for each measurement. Limits of detection were determined from the mean reversible response of cycles 2-4, divided by the electrical noise inherent in the MSS device (approximately 1 μV 27 ) to give the signal-to-noise ratio. The effective limit of detection was then estimated by dividing the concentration of analyte present by the signal-to-noise ratio.

Principal component analysis (PCA).
The responses of the 2x2 MOF-MSS membrane array were analysed following the methodology of Shiba et al (see ESI Section S10). 42 Briefly, the features of each response profile were decomposed into four parameters, defined as the rise rate, plateau rate, recovery rate and response magnitude. Parameters from the latter three of four cycles were used as inputs for PCA using Origin software, which determined the projection weights for a set of orthogonal principal components in order to maximise the total response variance.
Results and discussion MOF-MSS sensor concept. We fabricated a MOF-MSS sensor by spray-coating ZIF-8 nanoparticles directly onto an MSS membrane. Its response to the saturated vapour of 26 VOCs is shown in Fig. 1a (for details see ESI section S11). All gases were found to elicit a measurable response within seconds, including a range of alcohols, carbonyls, arenes, and alkanes. The irreversible signal-which we attribute to residual molecules that remain in the MOF pores on the timescale of these experiments-apparent in the first cycle for most gases is largely absent from the second cycle onwards. Whilst these irreversible signals in the first cycle can be also caused by the enhanced concentration of VOC vapours in the head space of the vial prior to the measurements, they are not included in the following analyses. Different classes of VOCs give rise to quite distinct response profiles. Water and alcohols, such as methanol (Fig. 1b) and ethanol ( Fig. 1c) give amongst the highest output voltages and profiles that typically reach saturation within 30 s. Molecules with carbonyl functionality, such as ketones, esters and amides, including acetone ( Fig. 1d), as well as chloroform and tetrahydrofuran also give large magnitude responses. The response of acetic acid (Fig. 1e) is unusual amongst the VOCs studied in that it is almost entirely irreversible on these timescales. Aromatic compounds, such as toluene (Fig. 1f), elicit intermediate responses that typically do not reach saturation within 30 s, whilst linear alkanes, including hexane ( Fig. 1g), give rise to small responses that rapidly reach saturation and then decrease. We attribute the different saturation rates to the different diffusivities of these class of VOCs in ZIF-8; larger molecules will generally pass through the structure more slowly. Clearly, the response is a complex function of many factors, including VOC concentration, host-guest interactions, diffusion through the receptor layer and analyte-induced framework distortion.

Optimisation of response magnitude and time.
In order to determine the detection limits of the ZIF-8-MSS sensor, the response magnitude and time were first optimized by investigating the effect of receptor layer volume (see ESI Section S12). Devices were fabricated by inkjet printing between 100 and 2500 droplets of a ZIF-8 nanoparticle suspension onto the MSS membrane. A device fabricated from 2100 inkjet droplets exhibited the best compromise between high output voltages and fast response times for selected gases from the five VOC classes investigated previously (methanol-alcohol, acetone-carbonyl, toluene-arene, heptanealkane and water-other). Notably, for most receptor layer volumes, the response times for methanol, acetone and heptane were found to be less than 5 s, whilst for toluene and water they were consistently below 30 s. The 2100 droplet device was therefore chosen for the subsequent investigation of detection limits.
Detection limits. The sensitivity of the 2100 droplet ZIF-8-MSS sensor were determined for 12 VOCs under variable humidity at 298 K (Fig. 2). Gases were diluted to 2 % saturated vapour. To calculate the sensitivity, the corresponding VOC concentration was multiplied by the ratio of 1 µV (the noise inherent to the MSS architecture) 27,28 and the average output voltage of three ON-OFF cycles. The sensor exhibited sensitivities to most gases well below 10 ppm and sub-ppm sensitivity to some, including 1-hexanol and heptane. Interestingly, sensitivity appears to improve with increasing humidity in some cases, suggesting that cooperative analyte-water interactions may be beneficial to the sensing mechanism. Multichannel array sensing. Despite reasonable selectivity that differentiates somewhat between VOCs, cross-sensitivity means that ZIF-8 is unable to unambiguously discriminate between multiple analytes. We adopted an multichannel array approach previously demonstrated for chemoresistive carbon nanotubes 43 and 2-D MOFS, 14 and exploited the versatility of the MOF-MSS concept to spray-coat a 2×2 array of MSS membranes 27,28 with nanoparticles of four ZIFs, ZIF-8, 44 ZIF-7, 40 ZIF-65-Zn and ZIF-71 41 (Fig. 3). These particular MOFs were chosen because of their variety in composition, network topology, pore aperture, and diameter (Table 1); which may be expected to affect the adsorption of gases according to size, shape and/or functionality in different ways, thus leading to a diversity of responses and thus improved VOC discrimination.

MOF
Composition Zn(4,5-dichloroimidazole)2 rho 4.2 16.5 Table 1. Compositions, nets and pore aperture diameters (da), and pore diameters (dp ) of four ZIFs used in the 2x2 MOF-MSS array. Simultaneous sensing experiments using the MOF-MSS array reveal a wide variety of responses of the four ZIFs to 26 VOCs (Fig. 1h, Fig. 4; see also ESI Section S13). The relative responses of the sod structures ZIF-7, ZIF-8 and ZIF-65 are qualitatively similar, reflecting the similarity in their network topologies and pore apertures. However, certain differences are apparent in, for example, response magnitudes (e.g., acetic acid) or the relative responses of related compounds (e.g., methanol vs. ethanol; acetic acid vs. acetone). Like in ZIF-8, arenes and linear alkanes elicit medium and low responses, respectively, in ZIF-7 and ZIF-65-Zn. It is particularly interesting that the responses of ZIF-65-Zn to both alkanes and arenes tend to decrease as molecular size increases. Its pore aperture is slightly bigger than the other sod analogues; perhaps this allows for more linear discrimination between compounds. The relative responses of ZIF-71 are qualitatively different to its sod analogues. Arenes and alkanes elicit higher responses, whilst methanol and ethanol give rise to lower responses. In this case, the response to alcohols appears to increase with increasing size of the molecules. ZIF-71 also gives a more uniform response across all VOCs. We tentatively attribute this to the openness of the rho net, which increases the reversibility of gas sorption compared to the other ZIFs, all of which exhibit the denser sod net. Relative mean response magnitudes are calculated from three measurements for each gas; error bars represent two standard deviations. Colours correspond to those used in Fig. 1.

Statistical analysis.
Principal component analysis (PCA) is an unsupervised method of classifying multi-sensor data, which reduces the dimensionality of the dataset by representing the sensor contributions as linear combinations of the original variables in typically, two or three principal components (PCs). 45 Noting that the MOF-MSS response profiles contain a wealth of information beyond the simple magnitudes of response, we extracted parameters to describe the reversible response, uptake rate, plateau gradient and recovery rate of each ZIF in the 2x2 array as input data for PCA (see ESI Section S10). 28,42 It was found that the four classes of VOCs could successfully be discriminated using just two PCs, with only small ambiguities in the case of overlaps between alcohols and carbonyls (e.g., acetone), and arenes and alkanes (e.g., shorter chain alkanes and dichlorobenzenes) (Fig. 5). Interestingly, alcohols appear to be subdivided into two groups, one with the smallest molecules (methanol and ethanol) and the other with larger molecules. The weightings of PC1 and PC2 were 55.9 % and 20.1 %, respectively, and together they describe 75.0 % of the total variance. A third principal component (PC3 = 8.2 %) was found to improve the description to 83.2 % and, when viewed with PC1, suggests a much closer grouping of the alcohols (see ESI Section S14). Within each class, the repeatability of our measurements is apparent in the ability to clearly discriminate between different VOCs. This is an advantage for VOC identification applications, for which there is prior knowledge of a given analyte's response. For example, methanol and ethanol are clearly distinguishable from each other, as are methylethylketone and acetone, both pairs of which differ by just one CH2 group. Alkanes and aromatic molecules follow clear trends with molecular size, which could be useful in monitoring separation processes.

Conclusions
We have demonstrated that MOF nanoparticles as a receptor layer on the MSS platform can be highly effective for strain-based chemical sensing. Response times of 1-30 s represent an order of magnitude improvement over existing MOF strain-based sensors (see ESI Section S15). Sub-ppm sensitivity towards a range of VOCs again represents an improvement in strain-based sensing using MOFs, bringing it on par with hard-to-fabricate photonic crystal thin films 46 and colorimetric sensors, 47 the latter of which was limited in analyte scope to water. Importantly, we have demonstrated that, in principle, any MOF that can be made in nanoparticle form 26 can now be employed in strain-based sensing, which allows for an enormous diversity of chemistries and VOC selectivities. We constructed a 2x2 MSS array using four ZIFs, which have different responses to a range of VOCs and, when their combined responses are subjected to PCA, are able to successfully discriminate between them. Whilst the performance of the MOF-MSS sensor is promising, the exact mechanism of sensing remains to be proven. Fundamental studies in areas such as MOF-guest interaction energetics, diffusion and flexibility will undoubtedly aid progress in this regard.  Contents S1. Synthesis S2. Powder X-ray diffraction S3. Infrared spectra S4. Scanning electron microscopy S5. Dynamic light scattering and zeta potential S6. Nitrogen sorption S7. Thermogravimetry S8. Selectivity and discrimination experiments S9. Sensing experiments: sensitivity and response time. S10. Principal component analysis for VOC discrimination S11. ZIF-8 response to 26 VOCs -detailed plots S12. Optimisation of receptor layer volume. S13. Relative responses of 2x2 array of ZIFs to 26 VOCs. S14. Principal component analysis plots using three PCs. S15. Performance comparison with other MOF-based sensors. S16. References S1. Synthesis ZIF-7 nanoparticles. Nanoparticles of ZIF-7, Zn(bIm)2 in sod topology, were synthesized by adapting the method of Li et al. 1 In detail, zinc nitrate hexahydrate (302 mg, 1 mmol) was dissolved in N,N-dimethylformamide (10 mL) and poured rapidly into a solution of benzimidazole (769 mg, 6.4 mmol) in N,N-dimethylformamide (10 mL) under stirring at room temperature. Stirring continued for 12 hours after which the milky suspension was centrifuged at 15000 xg for 30 minutes. The supernatant was decanted and replaced by methanol and the mixture sonicated for one minute to redisperse the particulate matter. The centrifuge / washing process was repeated thrice more. Half the resulting suspension was kept for further use, and half was dried (60 °C in air, then 180 °C under active vacuum overnight), yielding an off-white solid, 104 mg (71 % yield based on Zn).

ZIF-8 nanoparticles.
Nanoparticles of ZIF-8, Zn(mIm)2 in sod topology, were synthesized by adapting the method of Cravillon et al. 2 In detail, zinc nitrate hexahydrate (297 mg, 1 mmol) was dissolved in methanol (20 mL) and poured rapidly into a solution of 2-methylimidazole (649 mg, 7.9 mmol) in methanol (20 mL) under stirring at room temperature. Stirring continued for six hours 30 minutes, after which the milky suspension was centrifuged at 15000 xg for one hour. The supernatant was decanted and replaced by fresh methanol and the mixture sonicated for 5 minutes to redisperse the particulate matter. The centrifuge / washing process was repeated twice more. Half the resulting suspension was kept for further use, and half was dried (90 °C in air, then 180 °C under active vacuum overnight), yielding a pale yellow solid, 38 mg (35 % yield based on Zn).

ZIF-65-Zn nanoparticles.
Nanoparticles of ZIF-65-Zn, Zn(nIm)2 in sod topology, were synthesized by adapting the method of Tu et al. 3 In detail, zinc acetate dihydrate (110 mg, 0.5 mmol) was dissolved in N,N-dimethylformamide (5 mL) and poured rapidly into a solution of 2-nitroimidazole (113 mg, 1.0 mmol) in N,Ndimethylformamide (5 mL) under stirring at room temperature. Stirring continued overnight, after which the yellow milky suspension was centrifuged at 15000 xg for 30 minutes. The supernatant was decanted and replaced by N,N-dimethylformamide and the mixture sonicated for one minute to redisperse the particulate matter. The centrifuge / washing process was repeated thrice more, using methanol as the fresh solvent. For the final centrifugation, the suspension was split into two equal fractions. One half was dried under vacuum at 55 C overnight, yielding a yellow solid, 22.3 mg (39 % based on Zn). The other half was redispersed in IPA for sensing experiments.
ZIF-71 nanoparticles. Nanoparticles of ZIF-71, Zn(dcIm)2 in rho topology, were synthesized by adapting the method of of Tu et al. 3 In detail, zinc acetate dihydrate (220 mg, 1 mmol) was dissolved in N,N-dimethylformamide (10 mL) and poured rapidly into a solution of 4,5-dichloroimidazole (960 mg, 6 mmol) in N,N-dimethylformamide (10 mL) under stirring at room temperature. Stirring continued for four hours after which the milky suspension was centrifuged at 15000 xg for 30 minutes. The supernatant was decanted and replaced by methanol and the mixture sonicated for one minute to redisperse the particulate matter. The centrifuge / washing process was repeated thrice more. Three quarters of the resulting suspension was kept for further use, and one quarter was dried (60 °C in air, then 180 °C under active vacuum overnight), yielding a brown powder, 75 mg (92 % yield based on Zn).

S2. Powder X-ray diffraction
Samples were analysed on Rigaku Ultima3 or RINT2000 instruments in a θ-2θ flat plate geometry using Cu K-α radiation. Data were collected from 2 ° to 50 ° 2θ. ZIF-7 nanoparticles exhibited reasonable crystallinity when as-synthesized and washed, with peak positions consistent with the theoretical pattern simulated using data from Zhao et al. 4 Upon heating under vacuum, crystallinity was largely lost, as evident from the reduction of definition and intensity of the diffraction peaks. Figure S1. XRD patterns of ZIF-7 nanoparticles: as-synthesized, washed and dried in vacuo at room temperature, and activated at 180 °C in vacuo. The theoretical, simulated pattern is shown below for comparison. ZIF-8 nanoparticles exhibited a good match to the pattern simulated using data from Park et al. 5 and very little difference between XRD patterns as-synthesized and washed and dried. The crystal structure was maintained even when activated at 180 °C under vacuum. Figure S2. XRD patterns of ZIF-8 nanoparticles: as-synthesized, washed and dried in vacuo at room temperature, and activated at 180 °C in vacuo. The theoretical, simulated pattern is shown below for comparison.
ZIF-65-Zn nanoparticles also exhibited a close match to the pattern simulated using data from Banerjee et al. 6 Very little difference was observed between the patterns of as-synthesized, washed and dried, and activated material. Figure S3. XRD patterns of ZIF-65-Zn nanoparticles: : as-synthesized, washed and dried in vacuo at room temperature, and activated at 180 °C in vacuo. The theoretical, simulated pattern is shown below for comparison. ZIF-71 nanoparticles exhibited lower crystallinity when as-synthesized than the other materials investigated, although the positions of the XRD peaks match the pattern simulated using data from Banerjee et al. 6 well. A similar degree of crystallinity is retained after washing and drying, and after activation. Figure S4. XRD patterns of ZIF-71 nanoparticles: as-synthesized, washed and dried in vacuo at room temperature, and activated at 180 °C in vacuo. The theoretical, simulated pattern is shown below for comparison.

S3. Infrared spectra
Fourier transform infrared (FTIR) spectra were recorded on a Thermo Nicolet spectrometer with ATR attachment, under a flow of nitrogen.

S4. Scanning electron microscopy
Scanning electron microscopy (SEM) images were acquired using a Hitachi SU8000 FE-SEM in backscattered electron imaging mode, with 2.5 kV and 1.5 kV acceleration and deceleration voltages, respectively.

S5. Dynamic light scattering and zeta potential
Dynamic light scattering (DLS) and zeta potential measurements were performed on diluted nanoparticle suspensions using an Otsuka Electronics ELSZ-2000 instrument.  Table S1. Dynamic light scattering data for the ZIF nanoparticle suspensions.

S6. Nitrogen sorption
Nitrogen sorption isotherms were collected at 77 K on samples activated at 180 C for 48 hours in the pressure range 10 -6 < P/P0 < 1, using a Quantachrome Autosorb iQ instrument. Surface areas were calculated from Brunauer-Emmett-Teller (BET) theory using data in the region 0.05 < P/P0 < 0.3.

S7. Thermogravimetry
Thermogravimetric analysis was performed using a Perkin Elmer Diamond combined TG/DTA instrument. Activated samples were heated beside an Al2O3 reference at a rate of 10 °C/minute under a constant air flow. Figure S16. Thermogravimetry of ZIF nanoparticle samples post-activation.

S8. Sensing experiments: selectivity and VOC discrimination.
Saturated vapour sensing was performed with the following experimental setup. The MSS coated with various ZIFs was mounted on a chamber and the chamber was carefully sealed with an O-ring. Two mass flow controllers (MFCs; FCST1005C-4F2-F100-N2, purchased from Fujikin Inc.) were utilized to introduce nitrogen into the chamber at the flow rate of 100 mL/min. One MFC was for purging (i.e. accelerating desorption of adsorbents), and the other one was for introducing sample vapor together with nitrogen as a carrier. In the present case, 1 mL of 23 sample liquids (pure water, formaldehyde solution (35-38%), acetic acid, methanol, ethanol, acetone, methyl ethyl ketone, n-hexane, nheptane, n-octane, n-nonane, n-decane, n-undecane, n-dodecane, benzene, toluene, xylene, 1,2-dichlorobenzene, 1,3-dichlorobenzene, chloroform, ethyl acetate, tetrahydrofuran and N,N-dimethylformamide) was added into a small vial capped with a rubber lid. Two needles connected to a PTFE tube was stuck into the headspace of the vial through the rubber lid. One end of the PTFE was connected to MFC and the other end of the PTFE tube was connected to a vacant vial, so-called 'mixing vial' to make the mixed gas sample homogeneous. Another PTFE tube stuck into the mixing vial was connected to the chamber. Another MFC and vacant vial were set in the same way and connected to the mixing vial. The two MFCs were switched every 30 seconds to perform a sample introduction-purging cycle. This cycle was repeated four times, and the data were recorded at the bridge voltage of -0.5 V and sampling rate of 20 Hz. The data collection program was designed by LabVIEW (National Instruments Corporation). All the above experiments were conducted under an ambient condition without any temperature/humidity control.

S9. Sensing experiments: sensitivity and response time.
For sensing experiments in much lower concentration ranges, we designed another measurement system. The MSS chip coated with various ZIFs was mounted in a Teflon chamber, and the chamber was carefully sealed with Orings. The chamber was placed in a constant temperature bath kept at 25 o C. The sample gases-the vapors of the 12 solvents (pure water, ethanol, 1-hexanol, hexanal, n-heptane, methylcyclohexane, toluene, ethyl acetate, acetone, chloroform, aniline and propionic acid) generated via bubbling-were injected into the chamber with a gas flow system equipped with three MFCs. Nitrogen was used as a carrier gas. The concentrations of the sample gases were calibrated by measuring the decrease in the weight of the solvents before and after a gas flow. The relative humidity (RH) was controlled by providing a saturated water vapor to the gas flow line. The concentration of the sample gases and the humidity of the carrier gas were adjusted by controlling the flow rates of the three MFCs. The total gas flow rate of the three MFCs was set at 100 mL/min. The surface stress caused by the gas adsorption/desorption in the ZIF layer was electrically read by a Wheatstone bridge circuit consisting of the piezoresistors embedded on the bridges [1]. In the present study, a voltage of -0.5 V was applied to the circuit, and the relative resistance changes of piezoresistors were detected as output signals. Each measurement was performed through 10 cycles of 10 seconds sample injection and 10 seconds nitrogen purge. The sample gases were diluted to 2, 5 and 10% of their saturated vapor concentration. The carrier gas was humidified at 0%, 10%, 40%, 70% and 90% RH.

S10. Principal component analysis for VOC discrimination
Principal component analysis (PCA) was performed for discrimination of the 23 samples, following the methodology of Shiba et al. 7 To perform PCA, features of a signal measured by the MSS were expressed by four parameters which are defined as follows: where a, b, c, d, e, ta, tb, tc, and td are denoted in Fig. S17. In this case, tb = ta + 5 [s], tc = ta + 30 [s], and td = ta + 35 [s] were used. Three sets of the parameters were extracted from the latter three signals where ta = 90, 150 and 210 out of the four repeated curves in the response signals, since the latter cycles could provide reproducible signals without initial fluctuations such as mixing of sample gases and pre-adsorbed gases. Then, Origin software (ver. 2017) was used to perform PCA. PCA finds projection weights for sensor response data that maximize the total response variance in principal components (PCs), where the dimension capturing the greatest variance is given by PC1, and the second greatest variance (subject to being orthogonal to PC1) is given by PC2.

S12. Optimisation of receptor layer volume.
The receptor layer volume was optimized by measuring the sensor response to selected VOCs as a function of the number of inkjet droplets deposited. The magnitude of responses (Fig. S23) to water, methanol, acetone, heptane and toluene appear to increase linearly with respect to mass up to 900 droplets. Thereafter, maxima are found around 1700-2100 droplets in each case. Response time, defined here as the time taken to reach 80 % of the maximum output voltage, was found to be less than 10 s in most cases without a linear trend with respect to layer volume (Fig.  S24). With the exception of the layer of 2500 droplets, both acetone and heptane elicit response times less than 5 s. Figure S23. Output voltage as a function of receptor layer volume, measured as number of inkjet printed droplets, for selected VOCs.