Accepted for publication in Analytical Chemistry, August, 2011 SUPPORTING INFORMATION FOR. address:

Transcription

1 Accepted for publication in Analytical Chemistry, August, 0 SUPPORTING INFORMATION FOR Microfabricated Gas Chromatograph for the Selective Determination of Trichloroethylene Vapor at Sub-Parts-Per-Billion Concentrations in Complex Mixtures Sun Kyu Kim,, Hungwei Chang,, Edward T. Zellers,,3* Department of Environmental Health Sciences, University of Michigan, Ann Arbor, MI Center for Wireless Integrated Microsystems, University of Michigan, Ann Arbor, MI Department of Chemistry, University of Michigan, Ann Arbor, MI address: ezellers@umich.edu The information provided below includes the rationale for using 6 m of total microcolumn length, the table of 46 test VOCs, a representative F heating profile, Golay plots and a representative chromatogram from the dual 3-m microcolumn ensemble, descriptions of the printed circuit boards and DAQ boards used for instrument control, and methods used for data analysis and chemometrics. Rationale for Two 3-m Microcolumns. The use of two microcolumns in series, each 3 m in length, can be rationalized on the basis of the required number of theoretical plates, N req, which is defined as follows for the peaks of a critical pair of compounds, and : S N k req 6Rs (eq. S) k where R s is the resolution, α is the separation factor, k is the retention factor of the later eluting compound (TCE in this case), and each of these variables is defined as follows: ( tr t ) R Rs.8 (eq. S) ( fwhm fwhm ) where t Ri is the retention time of compound i, fwhm i is the full width at half maximum the peak; and k k (eq. S3)

2 Accepted for publication in Analytical Chemistry, August, 0 k t t ' R M (eq. S4) where t R is the adjusted retention time and t M is the hold-up time (evaluated as the retention time of methane). For TCE, a k value of 3.3 and a fwhm value of.0 s were assumed on the basis of experimental data, S and values of =. and R s =.5 (i.e., baseline separation) were imposed. This yields an N req value of 7,400 plates. For operation at. ml/min, it is estimated that each microcolumn generates ~,400 plates per meter (derived by extrapolation of the Golay plot presented in Figure S; see below). Therefore, a total microcolumn length 5.3 meters would be required. Table S. List of 46 test compounds and their vapor pressures (p v ). ID Compound p v (torr) a ID Compound p v (torr) a Pentane 54 4 n-octane 4.,-Dichloroethane n-butyl Acetate 5 3 Methylene Chloride Chlorobenzene.8 4 Acetonitrile 73 7 Ethylbenzene Propanol (Isopropyl Alcohol) 44 8 m,p-xylenes Acrylonitrile 97 9 Bromoform 5 7 -Butanone (MEK) o-xylene n-hexane 50 3 Styrene Ethanol 60 3 n-nonane Tetrahydrofuran (THF) 6 33 Cumene 3.5 Chloroform alpha-pinene 4.75 Acetone 3 35 n-propylbenzene 3.4 3,,-Trichloroethane 00 36,,4-Trimethylbenzene. 4 Ethyl Acetate Ethyltoluene 3 5 Carbon Tetrachloride 3 38,3,5-Trimethylbenzene.48 6 Benzene d-limonene.98 7 Cyclohexane 98 40,,4-Trichlorobenzene Trichloroethylene (TCE) 69 4,4-dichlorobenzene Methyl--pentanone 0 4 Naphthalene ,,-Trichloroethane 7 43 n-decane.4 Toluene n-undecane Hexanone 45 n-dodecane 0. 3 Tetrachloroethylene (PCE) n-tridecane 0.08 a from reference S3.

3 Temperature ( C) Accepted for publication in Analytical Chemistry, August, s Time (s) Figure S. Representative (experimental) heating profile for the F during desorption/ injection. Application of 36 V leads to an increase 5 to 5 C in 0.45 s (440 C/s). Subsequent application of 6 V maintains the F between 5 and 50 C. Microcolumn Characterization. In order to compare the chromatographic efficiency of the current microcolumn design, with chamfered corners, to that of the previous design, with right-angle corners, the relationship between the linear velocity and the plate height was determined using n-octane as the test compound. For these experiments two series coupled 3-m microcolumns of a given design were connected between the injector and FID of a conventional GC (7890, Agilent Technologies, Pal Alto, CA) and mounted in the GC oven held at 30 C. Serial injections (000: split) of methane and n-octane (headspace above liquid, k = 3.7) were performed over a range of N carrier gas inlet pressures that produced a range of average linear velocities, u, estimated from the methane hold-up times. Retention times and peak widths were used to calculate the number of theoretical plates, N, from which the theoretical plate height, H, was determined (i.e., H = N/L, where L is the microcolumn length; N = 5.54(t R /fwhm), where t R is the adjusted retention time and fwhm is the full-width at half maximum of the peak). These data were used to create the Golay plots for n-octane shown in Figure S. The value of the optimal velocity, u opt, was ~ 0 cm/s (0. ml min - ) for both types of microcolumns, but those with the chamfered corners yielded a minimum value of H = 0.0 cm, which was 0% smaller than those with sharp corners (minimum H = 0.07 cm). This results in a commensurate increase in N produced by the dual 3-m microcolumn ensemble from,000 to 7,300 plates (i.e., 4,550 plates m - ). 3

4 H (cm) Accepted for publication in Analytical Chemistry, August, 0 a) b) c) u (cm/s) Figure S. SEM images of sub-sections of the etched-si channels used in the 3-m-long microcolumns of the GC prototype prior to sealing and coating with PDMS stationary phase: a) previous design with right-angle corners; b) current design with chamfered corners. The Golay plots in c) were generated with n-octane (000: split, k = 3.7) using N as the carrier gas by connecting the dual 3-m microcolumn ensemble between the injector and FID of a bench-scale GC: previous design (filled symbols) and current design (unfilled symbols). The chromatogram in Figure S3 shows the.-min separation of a 0-component mixture, including TCE, using the dual 3-m microcolumns of the current design, configured as described above for generating the data in Figure S. The separation conditions, including the temperature program used with the integrated microcolumn heaters, are given in the caption of Figure S3. All compounds were baseline separated, TCE eluted in about 45 s, and the entire mixture eluted in <. min. The fwhm values ranged from 0.6 to.9 s. On the basis of 6 replicate injections, the retention time variations ranged from 0.35 to 0.89% of the average values. 4

5 Accepted for publication in Analytical Chemistry, August, 0 TCE Time (min) Figure S3. TCE separation from 0 VOC interferences using a conventional (bench scale) GC inlet/injection port and FID, and the dual 3-m microcolumns of the current design. (0. L injection of the neat mixture; inlet pressure: 4 psi; inlet temperature: 50 o C; split ratio: 00:; Temperature program of st microcolumn: hold at 5 C for 60 s, heat to 60 C at 70 C/min, heat to 00 C at 80 C/min, hold at 00 C for 30 s. Temperature program of nd microcolumn: hold at 5 C for 60 s, heat to 60 C at 70 C/min, heat to 0 C at 0 C/min, hold at 0 C for 30 s. Compounds:, n-hexane;, benzene; 3, TCE; 4, toluene; 5. -hexanone; 6, PCE; 7, ethylbenzene; 8, o-xylene; 9, nonane; 0, cumene;, n-propylbenzene. Microcolumn Temperature Programming. For applied dc biases of 5-5 V (0.6-5 W), the corresponding steady-state microcolumn temperature ranged from C, respectively. The maximum heating rate, taken from the initial stages of the 5-W curve is 7.6 C s -. Higher heating rates may be possible but were not explored. Actuation and Control Circuitry. A custom pneumatic control circuit board and associated digital I/O card (USB-650, National Instruments, Austin, TX) for actuating the valves, pumps, and the heaters on the pre-trap and sampler, were located beneath the manifold on the chassis floor. A second printed circuit board and associated 6-bit multi-functional DAQ card (USB-68, National Instrument, Austin, TX) for monitoring and controlling the devices in the analytical subsystem (i.e., the µf and microcolumn heaters and temperature sensors, and the sensors in the CR array ) were located beneath the analytical subsystem components. A USB hub permitted connections to a laptop computer running a control program written in LabView (Ver. 8.5, National Instruments, Austin, TX). 5

6 Accepted for publication in Analytical Chemistry, August, 0 Table S. Timetable for SPIRON µgc operation. Component Sampling (5-6 min) Focusing (3 min) Stabilization (3 min) Separation & Detection (3.5 min) Manifold valve On On Off Off Off Off Off Manifold valve Off Off Off Off Off Off Off Manifold valve 3 Off On Off Off Off Off Off Manifold valve 4 Off Off Off Off Off Off Off Manifold valve 5 Off Off Off Off Off Off Off Manifold valve 6 Off Off On On On On On Analysis pump Off Off On On On On On Sampling pump On On Off Off Off Off Off Pretrap Off Off Off Off Off Off Off Sampler Off On Off Off Off Off Off Focuser Off Off Off On Off Off Off Column Off Off On On On On On Sensor Off Off On On On On On Data Analysis and Chemometrics. Peak heights and peak areas were determined after importing the raw response data into GRAMS AI/3 (Ver. 6.00, Thermo Scientific Inc., West Palm Beach, FL), and linear regressions of calibration data were performed using Excel. The performance of the CR array in differentiating among TCE and several potential interfering VOCs eluting nearby was assessed using Monte Carlo simulations coupled with extended disjoint principal components regression (EDPCR) classification models. Details of this approach to array assessment have been published elsewhere (see, for example, refs 4d, 4g, and 4h in the main article) and are summarized in the following paragraph. Using the experimental sensitivity values, synthetic MPN-CR responses to each vapor were generated by randomly selecting a vapor concentration within the range of 5-0 LOD, where the LOD was dictated by the least sensitive sensor in the array to ensure that all sensors contributed to the response patterns. The response was calculated from the calibration-curve regression equation for each sensor. Then, error was introduced by adding to the response a value obtained by multiplying that response value by a factor derived from randomly sampling a Gaussian distribution with a mean of zero and a standard deviation corresponding to the random sensitivity errors derived from the calibration data (Figure 3, main body) for each sensor for TCE (i.e., C8, 8.%; DPA,.7%; OPH,.%; HME, 9.5%). The error enhanced responses from all sensors were combined and the location of the resulting response vector was projected onto the principal component corresponding to the original calibrations for each vapor via EDPCR. The identity of the vapor assigned to this synthetic response vector was determined by the shortest Euclidean distance. This procedure was performed iteratively (i.e., 500 samples) to yield a statistical estimate of recognition rate (RR) for each vapor. 6

7 Accepted for publication in Analytical Chemistry, August, 0 Table S3. Confusion matrix for single-vapor discrimination. a Compound HEX BEN TCE MIBK TOL HEX BEN TCE MIBK TOL Recognition rate (%) a Based on Monte Carlo simulations and EDPCR classification models (see text); HEX = n-hexane, BEN = benzene, TCE = trichloroethylene, MIBK = 4-methyl--pentanone, TOL = toluene; actual identities are listed in the top row and assigned identities are listed in the first column; n = 500 iterations for each vapor. References S. Jennings, W.; Mittlefehldt, E.; Stremple, P. Analytical Gas Chromatography, nd Ed., Academic Press, 997. S. Chang, H.; Kim, S. K.; Sukaew, T.; Bohrer, F.; Zellers, E. T. Procedia Engineering 00, 5, S3. Department of Chemistry at the University of Akron, The Chemical Database, available at 7