Ethyl cellulose (49% ethoxy content, MP Biomedicals), tridodecylmethylammonium chloride (TDMACl, 98%, Sigma-Aldrich), methyltriethoxysilane (MTEOS, Aldrich), tetramethoxysilane (TMOS, Aldrich), R706 TiO₂ (DuPont), and dioctyl sebacate (DOS, Sigma-Aldrich) were used as received. Methanol (99.9%), ethanol (99.9%), 2-propanol (99.9%), and other chemicals were purchased from Fisher Scientific. Chromoionophore IX (CIX) was either purchased from Fluka or prepared by the reported procedure [26] .

Standard microscope slides (Corning) were used to cut 1 ´ 1 cm² glass sensor substrates. They were washed in a piranha solution (concentrated H₂SO₄ and 30% H₂O₂ in 3:1 ratio) for 30 min, followed by rinses first with deionized water and then with acetone, methanol, and ethanol. These slides were then dried in an oven before use. N-type [100] silicon wafers were similarly cleaned for deposition of thin film sensors that were then used for characterization by scanning electron microscopy (SEM).

A 0.9900% CO₂ gas tank (Airgas), odorless kerosene (Acros), acetone (Fischer), hexanes (Fischer), 29% aqueous ammonia solution, and food products, specifically bread, were used in the interference tests. To expose CO₂ gas to Sensor A, the CO₂ gas tank was directly connected to one of the flowmeters. After establishing a N₂ baseline, CO₂ gas (0.9900%) was introduced into the gas stream. In the interference tests involving kerosene, acetone, hexanes, and aqueous ammonia, approximately 50 mL of each analyte were placed into a temperature controlled jacketed gas impinger for their respective tests. After a baseline was established for Sensor A, nitrogen gas was introduced to the impinger to bubble the specific interferent vapor into the sensor flowcell, where the response was recorded. To test for food products, a breathing mask was fitted into the inlet of the flowcell and a volunteer would breath into the mask several times to establish a baseline of the breath and then ingest bread products. The volunteer would then breathe into the mask again to see if there were any differences between the two experiments.

A custom-built spin-coater was used to make the thin films.Ethyl cellulose (~0.235 g) was dissolved in a 4:1 mixture of toluene (8 mL) and ethanol (2 mL) to give a solution (2.5% wt EC) which, after several tests, was found to be of the optimal for the deposition of the sensor thin films. This solution was sonicated for 20 min and stirred to ensure that ethyl cellulose was completely dissolved. Then CIX (1 - 2 mg) and TDMACl (0.28 mg) were dissolved into 1.00 g of the solution with stirring. The ethyl cellulose solution containing the dye was pippetted on a cleaned glass slide (~1 ´ 1 cm²), which was on the spin coater, and drawn to the edges of the glass slide with a plastic pipette tip. The slide was then spun at ~2600 rpm (revolutions per minute) for approximately 30 s. Schlenk tubes were used to store the freshly made thin-film sensors (Sensor A) under vacuum prior to use.

Sensors were also prepared by incorporating TiO₂ particles (360 nm diameter) to induce the Mie scattering. In the preparation of these sensors, R706 TiO₂ particles (~3.7 mg) were added to EtOH (5 mL) and sonicated for 20 min to disperse the particles. The EtOH solution containing the TiO₂ particles (100 μL) was added with additional EtOH (1.9 mL) and then mixed with toluene (8 mL) to give an ethyl cellulose solution (2.5% wt EC). Sonication was used to ensure that ethyl cellulose was completely dissolved. Thin film sensors on glass slides were prepared from this EC solution (Sensor B) in a process similar to that described above, and then stored under vacuum in a Schlenk tube before use.

Sensor C was prepared on silicon wafers as Sensor A.

A Perkin-Elmer LS55 luminescence spectrometer with a pulsed Xe source was used for fluorescence measurements. The spectrometer was set at the following parameters: λ_ex = 445 nm, λ_em = 555 nm, 10 nm slits, 810 V PMT detector voltage, and a 515 nm emission cutoff filter. For time-based signal measurements, a signal reading was taken every 0.1 s. To improve the signal-to-noise ratio (S/N), signal averaging was used by the arithmetic mean of a point and the points before and after. Sensors were placed in a brass flow cell constructed to fit with a front surface sample cell holder purchased from Perkin-Elmer.

The vapor pressure of each alcohol at various temperatures was calculated by using the Antoine equation (Equation (1)), where A, B and C are constants over a defined temperature (T in K) range [40] .

A custom-made, jacketed gas impinger (Figure 1) containing an alcohol liquid was used. The lid for the bubbler contained a medium frit gas diffuser immersed in the alcohol, and the lid was sealed by an O-ring and clamp. On top of the lid, there was a gas inlet for nitrogen gas and an outlet for the alcohol-saturated gas. The temperature of the jacketed impinger was controlled by a Thermo Haake temperature controller at −15.0, −7.0, and 0.0˚C to give 1.000% MeOH (wt%, 1.549 ´ 10⁴ ppm), 1.000% EtOH (1.587 ´ 10⁴ ppm), and 1.000% PrⁱOH (2.310 ´ 10⁴ ppm), respectively. Two separate mass flow controllers (MKS Instruments) (Figure 1) were connected to the inlet of the flow cell in the fluorescence spectrometer, and they were used to control the ratio of nitrogen _­gas and the gas from the impinger to obtain an accurate alcohol concentration. The outlet of the flowmeter was linked to a sealed vial which was attached to a separate bubbler to prevent the backflow of gases.

SEM images of Sensor C were taken using a Leo 1525 Field Emission Scanning Electron Microscope.

The CIX dye, a trifluoroacetyl, reacts with alcohol to give a hemiacetal (Figure 2) and in the process quenches the fluorescence of the dye. However, formation of the hemiacetal takes approximately 20 h to complete. Mohr and coworkers found that a quaternary ammonium hydroxide, TDMACl, catalyzes the reaction resulting in an immediate conversion of the trifluoroacetyl group to the hemiacetal upon exposure to alcohol [33] . To our knowledge, the dye, CIX, has not been used for gas phase detection of alcohols or encapsulated in an ethyl cellulose matrix.

Initially, PVC sensors (Sensor D) were made following the formulation by Mohr and coworkers [33] . Upon exposure to air for several days, the PVC films lost a majority of their orange color, but were still responsive when

tested under 1.000% PrⁱOH pulses. The sensors, with an excitation peak at 450 nm, produced an emission peak at 540 nm (Figure S1, Appendix). During pulses of nitrogen and 1.000% PrⁱOH vapor, Sensor D showed quick response times by displaying a decrease in emission intensity and fast recovery times suitable for online measurements (Figure S2, Appendix). In order to increase the sensitivity, Sensor E was prepared with twice the dye content. In the 0.036% - 1.000% PrⁱOH vapor range (Figure S3, Appendix), Sensor E showed better sensitivity. How- ever, the signal was noisy and it required signal averaging and baseline correction to adjust for the baseline drift.

Sol-gel sensors (Sensor F) were then explored as another alternative matrix for CIX. Figure S4 shows the response of the sol-gel sensor to 1.000% PrⁱOH pulses, Appendix. Initially, the response to PrⁱOH was quick, but after switching back to nitrogen gas, the response was higher than the initial baseline, indicating that the sol-gel matrix perhaps has an affinity for alcohols and once exposed to alcohols reversibility may be problematic.

After several comparison tests, ethyl cellulose (EC) was found to be the best matrix to encapsulate the dye CIX. A focus on EC as the matrix for CIX also helped to provide consistency with other sensors previously studied by Innosense LLC in a multi-channel sensing platform. Sensor A produced an emission peak at 555 nm when the dye was excited at 445 nm (Figure 3).

Figure 4(a) shows a section of a uniform pitted surface of Sensor C on a silicon substrate at 900× magnification from SEM imaging. Figure 4(b) displays a cross section of Sensor C at 3000× magnification that has a

thickness of 1 - 2 µm, highlighted by the two parallel lines in the image. The surface of the sensor shows features (“pitting”) consistent with phase separation that occurs during the processing of the sensor.

In order to increase the sensitivity of the indicator dye, TiO₂ particles were incorporated into ethyl cellulose thin films. TiO₂ particles scatter wavelengths in the visible range about equally [34] . The scattering of light by the TiO₂ particles within the sensing matrix allows for a more effective interaction of light with the indicator molecules in the sensing layer, thus effectively increasing the path length without using thicker sensing layers [34] . The optimum particle size would be approximately 0.5 times the wavelength of interest which, in this case, would be around 250 - 350 nm for the emission wavelength of 550 nm from the dye. This method was adapted from CO₂ sensors that we developed recently [39] . One possible method using TiO₂ particles would be to prepare size-controlled TiO₂ particles so that it would correspond to the optimum particle size needed for the emission wavelength. Nakata and coworkers have fabricated and characterized self-organized porous TiO₂ particle layers where the particle size can be modified depending on the calcination temperature and time [41] . You and coworkers have also developed a process to prepare submicron-sized porous TiO₂ particles by a single-step swelling process of polystyrene template particles with titanium (IV) isopropoxide (TTIP) [42] . However, commercial TiO₂ particles R706 were used instead because it was inexpensive and readily available. Figure 5 shows the response of TiO₂-doped Sensor B. The addition of TiO₂ R706 particles resulted in an increase in the emission intensity. This is most likely because the median size of R706 particles, at 360 nm [43] , is close to the 250 - 350 nm range indicated above. Moreover, according to the DuPont’s R706 product sheet, small TiO₂ particles scatter

blue light more effectively than those with larger particle sizes [43] . In addition, R706 is ideal for the prototype optoelectronic device using a blue LED as its light source discussed below.

As indicated earlier, reactions of the dye CIX with alcohols forming hemiacetals are selective. Primary alcohols with increasing chain lengths show an increasing ratio of the trifluoroacetyl form rather than the product hemiacetals [33] . Figure 6 shows the sensitivity of the EC sensor towards MeOH, EtOH, and PrⁱOH, respectively. Linear calibrations are obtained over large concentration ranges. The detection limits are 9, 13, 21 mg・kg⁻¹, and the quantification limits are 32, 43, and 70 mg・kg⁻¹, respectively, for MeOH, EtOH, and PrⁱOH vapors. The data show that the sensors are in general more sensitive to less bulky alcohols. Since CIX reacts with different alcohols, it is important to place the sensor in a specific location where the type of alcohol is known. For example, the alcohol sensor could be positioned to detect methanol leaks from fuel cells that are used in microelectronics or be adapted into breathalyzers for ethanol intoxication.

Compared to other previously mentioned studies of alcohol sensors, the EC-CIX sensors developed in this study have a much higher dynamic range and lower limit of detection for the alcohols investigated. Commercial alcohol gas sensors such as the ones used in breathalyzers usually have a sensor accuracy of ±0.01 blood alcohol content (BAC), according to commercial product specifications, which indicates the readings can vary by as much as 100 mg・kg⁻¹ [44] - [46] . While the EC-CIX sensors in this study are not totally unaffected by interferences such as moisture and amines, they are less sensitive to other interferent constituents that are found in rocket fuel and plumes such as CO₂ and kerosene. The interference studies suggest that the alcohol sensors should be used in environments where the analyte gas is known so interferent vapors such as moisture/amines and other specific alcohols do not interfere in the detection. The research here shows that EC-CIX sensors may be considered an option as a sensitive alcohol sensor for detection in rocket plumes as well as being adapted and providing an improvement to general alcohol sensing device.

Unlike the PVC sensors (Sensors D and E), the EC sensors (Sensor A) were able to keep their fluorescent orange color and demonstrated durability under ambient environment. An EC matrix helped to stabilize the baseline response which was observed by the small difference between the original and baseline corrected response (Figure S5) when the sensor was exposed to PrⁱOH vapor, Appendix. Since the EC thin-film sensors do not lose their initial sensitivity in ambient environment, a long term study was performed to test the durability of the sensors. In order to simulate real-world field tests, Sensor As were wrapped in a lint-free cloth and then covered

in aluminum foil to protect the sensors from light. Afterwards, the sensors were placed in sealable bags. Fluorescence measurements were performed every 2 weeks on the same sensor (Sensor A1), which was compared to a freshly opened sensor from the same batch. Over a period of eight months, Sensor A1 began to fluctuate in baseline (Figure S6) but the sensor still showed sensitivity when exposed to alcohol vapor (PrⁱOH). In comparison, freshly opened sensors for each measurement showed a baseline and sensitivity similar to that of Sensor A1 from month 1. This long term study demonstrates that Sensor A is able to optimally detect alcohol for 3 months before it needs to be replaced, and the sensor can remain sensitive for up to a year.

To demonstrate that these alcohol sensors can be used in the field, InnoSense LLC has designed and built a miniaturized multi-channel optoelectronic device that can test up to seven different sensors simultaneously from a blue LED light source. Each channel has a slot for an optical filter to enable absorbance or fluorescence based measurements and longpass filters (500 nm) were used to examine the EC-CIX alcohol sensors. Within the optoelectronic sensor unit, there is a built-in temperature and humidity probe to monitor temperature and relative humidity during the testing period. Therefore, any contribution from humidity can be subtracted from the signal, thus giving the signal arising from the alcohol analyte itself. The optoelectronic sensor unit can be connected to a PC where data acquisition is monitored by LiveGraph software. The sensor unit was evaluated by testing three alcohol sensors (Sensor A) and an EC blank and exposing these sensors to varying concentrations of PrⁱOH ranging from 0% - 1.000%. Data was taken every 5 s. Figure 7 displays the results from the sensor unit. Each alcohol sensor shows a similar response to the other and demonstrates an expected decrease in signal upon exposure to PrⁱOH. The sensors showed minute changes in signal within the optoelectronic device when exposing the alcohol sensors to CO₂ gas. However, these results were expected given the previous results above.

Another challenge for detection in optical sensors is interferences. To evaluate the response of the alcohol sensor

to interferent vapors, carbon dioxide (CO₂), kerosene, ammonia (NH₃), moisture (H₂O), hexanes, acetone, and food were directly exposed to Sensor A in triplicate and recorded. Kerosene is a rocket fuel, and CO₂ is a major component of rocket plumes. NH₃ is a primary product of hydrazine (H₂NNH₂) decomposition, and hydrazine has been used as a liquid rocket propellant [47] [48] . Moisture is a byproduct generated through the combustion of 2-propanol with liquid oxygen in order to simulate a vacuum for rocket testing [16] . Acetone, hexanes, and food products, specifically bread, are some of the interferences that have attributed to false positives in general commercial alcohol sensors [49] .

As expected, Figure S7 shows that CO₂ gas has a minute effect on Sensor A, and the observation is consistent with the fact that CO₂ does not react with the CIX dye. Exposing Sensor A to kerosene vapors caused the fluorescence of the sensor to slightly quench in intensity, suggesting that the introduction of the hydrophobic kerosene to the sensor film may have changed the environment, shifting to a small degree the equilibrium of the reaction between the CIX dye and alcohol (Figure 2). It is expected that vapors of other organic hydrocarbons may lead to a similar, small effect.

However, exposing Sensor A to ammonia vapor and moisture caused a large decrease in fluorescence of the sensor. Other amines, especially primary amines, may react with the CIX dye and quench its fluorescence [50] . Moreover, moisture can react with the CIX dye, a ketone, forming geminal diols and thus affect the reaction of the dye with alcohol [50] . Since Sensor A is being used in an optoelectronic sensor unit with a built in humidity probe, the relative humidity can be measured during the testing period. Thus, the relative humidity can be measured and be subtracted from the signal. Studies are also underway to develop a dual optical sensing approach to address the impact of moisture, and the results will be reported in the future.

General alcohol sensors such as commercial breathalyzers may also be susceptible to outside interferences such as food and other chemicals in the environment, leading to readings higher than normal and false positives [50] . The US National Highway Traffic Safety Administration (NHTSA) establishes guidelines and screenings for commercial breathalyzers in order to be approved as an alcohol sensing device [51] . These guidelines do not require a certain limit of detection to be met but require devices to conform to a set of tests to evaluate the precision and accuracy of the alcohol sensing device [51] .

One of the major flaws of IR breathalyzers is the lack of specificity, since it is often the methyl group in ethanol that is being detected [49] . Thus the sensor will treat any compound with a methyl group as ethanol which includes many chemical compounds whose molecular structures are compatible with IR filters. Two IR wavelengths and an algorithm for the detection of interferents have been used in new breathanalyzers to calculate the ratio of absorbance values at the two measured wavelengths [50] . Also, the instrumentation for IR breathanalyzers can be complex with some requiring five filters to address volatile interferences [52] . Although such an approach reduces interference, acetone, e.g., was found to still have an effect on the new, IR-based breathanalyzers [53] . Semiconductor breathalyzers have been shown to absorb many substances which can give positive alcohol readings even when no ethanol is present [49] . Studies have confirmed the existence of a wide variety of compounds on the human breath [54] . Acetone and hexanes are among chemicals that are commonly found on the breath of normal, healthy individuals but are of insignificant levels to affect breathalyzers. However, dieters and diabetics not in control of their blood sugar can have acetone levels hundreds or even thousands of times higher than normal which may create falsely high results in a semiconductor breathalyzer [49] . Breathalyzers based on fuel cell technology give concerns due to sensitivity loss and sensor degradation after long term exposure in dry conditions [55] . The primary reason for sensitivity loss and sensor degradation is from the loss of electrochemically active surface area of the platinum electrode, which is seemingly irreversible [55] . Another concern for fuel cell sensors, albeit to a lesser extent, is the loss of proton conductivity as a result of membrane dehydration but can be alleviated by rehydrating the membrane in humid conditions [55] . There has been a report of a false-positive breath-alcohol test using a fuel-cell based analyzer after a ketogenic diet [56] .

In comparison to commercial breathalyzers, Sensor A appears to be less prone to common interferences of current commercial alcohol sensors. Sensor A was directly exposed to interferents that are known for commercial breathalyzers such as acetone, hexanes, and food but gave a minute response to these interferences (Figure S7). The small responses are perhaps a result of solvent effects, as acetone [57] and hexanes do not react with the CIX dye. It is unclear why food interferes with common breathalyzers since it can cause one of the highest false positives in commercial breathalyzers [58] . However, it does not affect the response of Sensor A. The interference study here demonstrates that interferences without active N-H or O-H groups are unlikely to interfere with the response of Sensor A. The current study looks into a new method centered on a fluorescence based system for the detection of alcohol and provides an improvement over existing alcohol sensors by limiting the possible interferences that may affect a reading such as chemicals or dry conditions.