Anal. Chem. 2004, 76, 5459-5464
Detector for Liquid Chromatography Based on Acoustic Emissions from an Oscillating Flame Kevin B. Thurbide* and Zhongpeng Xia
Department of Chemistry, University of Calgary, 2500 University Drive, N.W., Calgary, Alberta, Canada T2N 1N4
The acoustic flame detector (AFD) is examined as a novel detector for liquid chromatography (LC). It is based upon the acoustic emission frequency of an oscillating hydrogen/ oxygen premixed flame and produces a universal response toward organic molecules. A stable frequency near 1000 Hz, which further depends on mobile-phase composition, is achieved for flow rates in the microliter per minute range. The mass flow sensitivity of the AFD demonstrates a linear response over 3 orders of magnitude and a detection limit (S/σ ) 3) of ∼15 ng of C/s for a series of alcohols. For cyclopentanol, this amounts to an injected mass of ∼77 ng based on a 0.5-µL injection of a 196 ppm solution in methanol (flow rate 20 µL/min methanol; peak width 30 s). Similar sensitivity is observed using a water mobile phase. Low-frequency (1/f ) noise contributions are dominant with or without mobile phase present. The AFD demonstrates a uniform molar sensitivity toward carbon compounds independent of their optical properties or volatility. Results suggest the device might serve as a simple, inexpensive universal LC detector. Liquid chromatography (LC) is an established separation method, and over the years many useful LC detection methods have been developed.1-6 However, few provide a universal response, which is useful when analyzing unknown samples. Currently optical detectors such as the UV-visible absorbance detector, the evaporative light scattering detector (ELSD), and the refractive index detector (RID) are widely used.7 While UVvisible detection is relatively sensitive and linear, it only responds to analytes containing appropriate chromophores.8 ELSD is reasonably sensitive toward analytes with or without chromophores but has limited response linearity and becomes less sensitive as analyte volatility approaches that of the mobile phase.9 Still, it is considered a semiuniversal detector and is often * Corresponding author. Tel.: (403) 220-5370. Fax: (403) 289-9488. E-mail:
[email protected]. (1) LaCourse, W. R. Anal. Chem. 2000, 72, 37R-52R. (2) LaCourse, W. R.; Dasenbrock, C. O. Anal. Chem. 1998, 70, 37R-52R. (3) Bruckner, C. A.; Foster, M. D.; Lima, L. R., III; Synovec, R. E.; Berman, R. J.; Renn, C. N.; Johnson, E. L. Anal. Chem. 1994, 66, 1R-16R. (4) Fielden, P. R. J. Chromatogr. Sci. 1992, 30, 45-52. (5) Tan, Y.; Momplaisir, G.-M.; Wang, J.; Marshall, W. D. J. Anal. At. Spectrom. 1994, 9, 1153-1159. (6) LaCourse, W. R. Pulsed Electrochemical Detection in High Performance Liquid Chromatography, 1st ed.; Wiley & Sons: New York, 1997. (7) LaCourse, W. R. Anal. Chem. 2002, 74, 2813R-2832R. (8) Weston, A.; Brown, P. R. HPLC and CE: Principles and Practice, 1st ed.; Academic Press: London, 1997. (9) Scott, R. P. W. In Analytical Instrumentation Handbook; Ewing, G. W. Ed.; Marcel Dekker: New York, 1997; pp 1123-1203. 10.1021/ac049777r CCC: $27.50 Published on Web 07/28/2004
© 2004 American Chemical Society
employed in applications such as lipid analysis.10 The RID is a universal detector in that it will respond to any analyte with a refractive index different from that of the solvent.9 Accordingly, unlike UV-visible or ELSD, RID is normally incompatible with common LC mobile-phase gradients. While traditionally RID is less sensitive than other methods,9 modifications including laserbased techniques have greatly improved this.11,12 In general, common disadvantages to using optical detectors are the added cost and complexity associated with their operation.13 As well, variable response factors for compounds having different refractive index or molar absorptivity values can be problematic. This can result in UV-visible response not being normalized over a broad range of analytes or the cancellation of RID signals from coeluting solutes that produce positive and negative refractive index changes.14 Thus, although these methods are useful, there remains a need to further develop simple and inexpensive alternative universal LC detectors. The flame ionization detector (FID) is a widely used universal detector in gas chromatography (GC). Its low cost, ease of use, picogram-level sensitivity, and uniform response factor toward carbon compounds over 7 orders of magnitude make it a very useful tool for analyzing unknown samples in GC.15 In LC, the FID has been employed with limited success. For example, when only a pure water mobile phase is used, good LC-FID performance can be obtained.16 However, the FID yields an enormous background response in the presence of organic mobile phases, often obscuring the analytical signal. For instance, in capillary column LC, a modified FID burner was employed to detect analytes introduced at flow rates of less than 1 µL/min.17 When only 5% methanol in water was used as a mobile phase, the resulting detector sensitivity was nearly 2 orders of magnitude smaller than a conventional GC-FID system. Others have used purely organic mobile phases in LC-FID by employing transport systems to evaporate the solvent from the analyte prior to entering the flame.9 However, like the ELSD, this approach is also limited by the inability to detect volatile analytes and remove background signals from nonvolatile solvent residues. (10) McNabb, T. J.; Cremesti, A. E.; Brown, P. R.; Fischl, A. S. Anal. Biochem. 1999, 276, 242-250. (11) Wilson, S. A.; Yeung, E. S. Anal. Chem. 1985, 57, 2611-2614. (12) Abbas, A. A.; Shelly, D. C. Anal. Chim. Acta 1999, 397, 191-205. (13) Dixon, R. W.; Peterson, D. S. Anal. Chem. 2002, 74, 2930-2937. (14) Yeung, E. S.; Synovec, R. E. Anal. Chem. 1986, 58, 1237A-1256A. (15) Sˇ evcˇ´ık, J. Detectors in Gas Chromatography, Journal of Chromatography Library Volume 4; Elsevier: Amsterdam, 1976. (16) Miller, D. J.; Hawthorne, S. B. Anal. Chem. 1997, 69, 623-627. (17) Krejcˇ´ı, M.; Tesarˇ´ık, K.; Rusek, M.; Pajurek, J. J. Chromatogr., A 1981, 218, 167-178.
Analytical Chemistry, Vol. 76, No. 18, September 15, 2004 5459
Previously, we reported a novel passive acoustic sensor for GC, based on frequency changes in the acoustic emissions of an oscillating flame.18 Briefly, a premixed hydrogen-air flame is supported on a glass capillary and the air flow is adjusted until the flame yields audible acoustic bursts of a steady frequency. In this mode, the flame front partially retreats into the capillary, where it terminates against the glass wall. The remaining unburnt gas premixture then travels to the surface of the capillary and is once again ignited by the outer cone of the flame, which remains constant throughout the process. This presents one cycle of flame oscillation. When a hydrocarbon analyte is introduced, the flame’s burning velocity is reduced, causing it to retreat even less into the capillary. This reduces the time required for the remaining gas premixture to travel to the surface for reignition, thereby increasing the frequency of the oscillations as a function of carbon content in the analyte peak. The device, named the “acoustic flame detector” (AFD), yields a universal response toward carbon compounds that is qualitatively very similar to an FID.18 Although the sensitivity and linearity of GC-AFD is lower than GC-FID, it is considerably greater than other passive acoustic sensors.18 Recently, we investigated the effect on response of using different carrier gases in GC-AFD. When methane was employed, it was found that the AFD signal could be nulled to the same values obtained from inert gases, such as helium or nitrogen, by adjusting the premix air flow. More importantly, a similar detector sensitivity was also obtained under these conditions. The ability of the AFD to sense carbon-containing peaks within carbon-based carrier gases suggested that it might also have analytical potential as a universal detector for the organic effluents common in LC. Thus, it would be interesting to know how or if the detector would operate in the presence of organic mobile phases. For example, if this were possible, then the simple, low-cost design and uniform response factor of the AFD could be beneficial for its use as an alternative LC detection method. The objective of this paper is to assemble a functional LC-AFD system and use it to explore the response characteristics of the AFD toward analytes introduced by liquid mobile phases. Factors affecting LC-AFD response will be presented along with the analytical attributes of linearity, sensitivity, noise considerations, and solvent effects. Where possible, comparisons between the developed LC-AFD method and conventional LC detection methods will be presented. EXPERIMENTAL SECTION A detailed description of an AFD construction used for GC has been given earlier.18 Figure 1 presents schematic diagrams of the LC-AFD apparatus (upper figure) and the detector arrangement (lower figure) used currently. This system also resembles the experimental setup of other flame-based detectors examined in LC.16 The AFD was mounted on the original FID detector base of a Varian model 3700 GC, which also served as a temperaturecontrolled oven for the separation column. A cylindrical detector housing (150 mm × 38 mm i.d.) was machined from an aluminum rod and adapted to bolt securely to the detector base. The top 45 mm of the housing was bored out to a 20-mm i.d., while the rest of the housing was bored out to an i.d. of 10 mm. An auxiliary air port was drilled through the side of the housing at a distance of (18) Thurbide, K. B.; Wentzell, P. D.; Aue, W. A. Anal. Chem. 1996, 68, 27582765.
5460 Analytical Chemistry, Vol. 76, No. 18, September 15, 2004
Figure 1. Schematic illustration of the LC-AFD apparatus (upper diagram) and the detector housing for the acoustic flame (lower diagram).
80 mm from the top, and threaded to fit a 1/16-in. Swagelok tubing union. The AFD burner was made out of a piece of 6-mm-o.d. quartz tubing (typically 100 mm × 1 mm i.d.) and was connected to the stainless steel gas port of the detector base using a 1/4-in. Swagelok nut and a Vespel ferrule. Mobile phase was delivered through the system using an Isco model 260D syringe pump (Isco, Lincoln, NE) operated in the continuous-flow mode. A 70-cm length of stainless steel tubing (1/16 in. o.d. × 0.25 mm i.d.) was used to connect the pump to a Rheodyne model 7520 injector equipped with an internal loop volume of 0.5 µL (Alltech, Deerfield, IL). A 10-cm length of stainless steel tubing (1/16 in. o.d. × 0.10 mm) was led from the outlet of the injection valve and through the oven wall, where it was connected to the separation column. Separations were performed using a LUNA C18 column (150 mm × 1.00 mm i.d.) packed with 3-µm particles (Phenomenex, Torrence, CA). The column was connected to a 37 cm × 0.05 mm i.d. piece of deactivated fused-silica tubing using a 1/16-in. low dead volume union (Alltech). This tubing was extended into the quartz burner, where it was found that the optimal position for it, in terms of signal-to-noise ratio and flame stability, was 40 mm below the surface. At this position within the burner, hydrogen and oxygen flowed concentrically around the tubing and mixed with the mobile phase while moving into the flame. For optimizations and calibrations where a flow injection mode was used, the separation column was removed and the fused-silica tubing was connected directly to the solvent system. All effluents and flame exhausts were directed into a fume hood. ELSD measurements were made using
a Sedex 75 unit (Scientific Products, Concord, ON, Canada), and GC-FID measurements were made on the original Varian GC and a Shimadzu model GC-8A system. The acoustic signals were detected using a simple desktop microphone connected to the soundcard of a Pentium computer. The AFD was monitored in real time using Spectrogram acoustic software (version 6.6, Richard Horne author, www.visualizationsoftware.com). This program routinely allowed chromatographic profiles to be displayed and measured. However, for relatively long chromatograms, archiving the data was more demanding since the program stored it as very large sound (i.e., wav) files. When further data analysis or exportation to a spreadsheet was required, Praat phonetics software was used (version 4.0.26, Paul Boersma and David Weenink authors, www.praat.org). To increase sound wave intensity and avoid damage, the microphone was fitted with a glass tube that was directed to the top of the detector housing to assist signal collection. A screened cap made for the top of the housing was also occasionally used to dampen ambient air currents. While in the absence of such currents the detector could also operate without the housing, the latter did provide control over the auxiliary air flow and facilitated direction of the sound waves toward the microphone. It is important to emphasize that the acoustic flame response is frequency and not amplitude dependent. As such, to monitor the AFD, it is only necessary that the acoustic signal has sufficient intensity to be detected by the microphone. For typical conditions employing a mobile-phase flow rate of 20 µL/min, the AFD was operated using a hydrogen flow of 28 mL/min, an oxygen flow of 15 mL/min, and an auxiliary air flow of 230 mL/min. Normally, the detector temperature was held at 250 °C while the oven was maintained at controlled temperatures between 25 and 40 °C. Tetradecanol (97%), cyclopentanol (99%), dodecanol (98%), cyclohexanol (99%), benzyl alcohol (99%), and all HPLC grade solvents were purchased from Aldrich. High-purity helium, hydrogen, air, and oxygen were obtained from Praxair. All other variations and details are described in the text. RESULTS AND DISCUSSION General Operating Considerations. Methanol was first applied as the liquid phase to investigate LC-AFD sensitivity since it is similar in structure to methane, which was successfully used earlier in the GC-AFD mode. The AFD can operate with either air or oxygen premixed into the hydrogen flow.18 However, the latter produces a greater burning velocity and requires a narrower burner to prevent flash back. While no significant differences in sensitivity are observed between them, oxygen/hydrogen acoustic flames produce a much louder signal that operates over a larger range of frequencies (∼500-2000 Hz). In the LC-AFD system, oxygen/hydrogen acoustic flames were much more favorable for achieving an audible, stable signal and were used throughout these experiments. The effect of mobile-phase flow rate on the detector operating range was initially investigated. This is considered the range of flows that will maintain a stable oscillating flame. To better understand this, LC-AFD experiments were conducted to reveal the hydrogen, oxygen, and methanol flows required to maintain a constant frequency. Figure 2 demonstrates the typical results for combined flow rates that produced an acoustic flame frequency of 894 Hz. As seen for methanol flow rates tested up to 30 µL/ min, the detector produced a constant signal for inputs as low as
Figure 2. Example of the influence of flow rates on the operating range of the LC-AFD system. The diagram demonstrates the methanol and oxygen flow rates required to produce a stable acoustic flame frequency of 894 Hz when using hydrogen flows of 20 (2), 28 (b), 50 (9), and 65 ([) mL/min.
∼10 µL/min and as high as ∼25 µL/min, depending on the hydrogen and oxygen flow rates. Hydrogen flow rates between 20 and 65 mL/min and oxygen flow rates between 8 and 20 mL/ min were examined. Gas and solvent flows outside of these ranges produced an unstable AFD signal. For a given flow of hydrogen, increasing oxygen flows were required to maintain the same frequency as the methanol flow increased. Conversely, for a given flow of oxygen, the hydrogen flow had to be decreased as methanol flows increased. These findings reflect the delicate balance achieved between fuel and oxidant supply, which respectively decrease and increase the acoustic flame burning velocity. Detector Response Characteristics. Overall, the best AFD performance was achieved using a methanol flow rate of 20 µL/ min, an oxygen flow of 15 mL/min, and a hydrogen flow of 28 mL/min. This normally produced a baseline frequency near 1000 Hz. The auxiliary air flow had less impact on the AFD sensitivity but was normally not kept at values below 230 mL/min in order to maintain a smooth signal. While an oxygen/methanol (or oxygen/methane in the GC-AFD mode) premixture will also create a stable acoustic signal, the introduction of hydrogen is necessary for analytes to alter the burning velocity of the flame and elicit a response. LC-AFD sensitivity was examined by introducing a series of alcohols under the same conditions over several days, and it was found that the detector response was quite stable and reproducible. During a given analysis, four replicate analyte injections normally produced peak heights with a relative standard deviation of 4-5%. The linear regression between the peak height in hertz (y) and the micrograms of compound injected (x) for each of the individual alcohol calibrations performed over the course of one week produced the following equations:
for cyclopentanol,
y ) 0.831x + 4.83
(R2 ) 0.9982)
for cyclohexanol,
y ) 1.19x + 12.42
(R2 ) 0.9999)
for dodecanol,
y ) 1.97x + 5.06
(R2 ) 0.9999)
for tetradecanol,
y ) 1.30x + 7.51
(R2 ) 0.9977)
Since it has been demonstrated previously in the GC mode that Analytical Chemistry, Vol. 76, No. 18, September 15, 2004
5461
Figure 3. Illustration of the LC-AFD response toward various amounts of different alcohols. The test analytes were cyclopentanol (9), cyclohexanol (2), dodecanol ([), and tetradecanol (b). Conditions used were 28 mL/min hydrogen, 15 mL/min oxygen, and a mobile-phase flow rate of 20 µL/min methanol. The figure inset shows a sample peak from 5 µg of cyclopentanol.
the AFD is a mass flow-sensitive detector with a uniform response factor toward carbon,18 it is interesting to also examine these features for the LC mode. Figure 3 therefore displays the same calibration data as the peak height frequency (Hz) obtained for different analyte flows (grams of carbon per second). A log-log format is used for direct comparison with the earlier GC-AFD calibration data.18 Also included (Figure 3 inset) is a sample peak obtained from the injection of 5 µg of cyclopentanol. As seen in the figure, the AFD sensitivity is quite consistent among the compounds, changing very little for alcohols of different molecular weight and structure. The AFD produces a linear response toward these compounds over nearly 3 orders of magnitude. The upper limit of response in Figure 3 is usually obtained for ∼200 µg of injected analyte, where the peak begins to overload the flame and stop oscillations, creating a negative dip in the chromatogram. The lower range of response in Figure 3 corresponds to the minimum detectable limit, which was typically ∼15 ng of C/s. This was determined at a signal-to-noise ratio of 3, where noise was measured as the standard deviation of the peak-to-peak baseline fluctuations over at least 10 analyte peak base widths. In terms of injected mass, this corresponds to ∼77 ng of cyclopentanol based on a 0.5-µL injection of a 196 ppm solution in methanol (flow rate 20 µL/min methanol; peak width 30 s). Overall, these values agree within a factor of 3 with our own comparative GC-AFD experiments, and also within a factor of 10 to those presented earlier for a fully optimized GC-AFD system.18 Using a lower oxygen flow rate of 2 mL/min and otherwise similar conditions, the precision, linear range, detection limit, and sensitivity of AFD carbon response (as methanol and ethanol) did not change significantly for a water mobile phase. Thus, similar to the GC mode,18 the detector apparently also yields similar response characteristics under different LC conditions. Since the latter often produced a baseline around 1100 Hz, it suggests that the different optimal parameters involved may be a function of achieving a common AFD frequency (i.e., capillary depth). Detector noise was found to be quite stable and followed a Gaussian distribution producing minimal drift. A 6-min sample of 5462
Analytical Chemistry, Vol. 76, No. 18, September 15, 2004
baseline data centered at 1250 Hz generated a standard deviation of 4 Hz with a negligible slope of 0.003. Since data were normally collected at 10-ms intervals, one pass of a 20 point moving average digital filter removed much of the noise without causing significant distortion to the peak shape.19 Fourier transformations of the noise revealed lower frequencies near 40 Hz to be dominant, similar to the 1/f character found for GC-AFD.18 This and other noise characteristics did not change with and without mobile phase present, indicating that LC introduction in AFD does not deteriorate detector performance. In general, noise increased with the baseline frequency but did not alter in character. Mobile-Phase Effects. Since the AFD responds universally to carbon compounds, mobile phases other than methanol were also investigated for their relative effect on the detector operating frequency. In addition to water, several single-carbon solvents, acetonitrile, ethanol, tetrahydrofuran (THF), and octanol (OcOH) were also examined. In general, these solvents increased AFD frequency in the following order:
H2O < H2CO, HCO2H < CHCl3 < CH2Cl2, CS2 < CH3OH < CH3CN < CH3CH2OH < THF < OcOH
Water does not respond in the AFD, resulting in a frequency similar to the pure hydrogen/oxygen acoustic flame. Hydrocarbons respond in the AFD by reducing the flame’s burning velocity through capturing key propagating species such as hydrogen radicals.18 This is thought to occur by their reaction with saturated hydrocarbon decomposition products such as methane.18,20,21
H• + CH4 f H2 + •CH3
Accordingly, it is generally found that solvent molecules possessing more saturated hydrocarbon fragments generate a greater AFD frequency. This is also true of the FID, which responds greatly toward saturated hydrocarbons but very little toward chlorinated molecules, formaldehyde, carbon disulfide, or formic acid.15 The FID is also believed to obtain its universal response through first reducing all hydrocarbons to methane prior to flame ionization.22 Thus, the role of methane in the FID and the AFD may possibly be related. Since the AFD frequency is mass flow sensitive, different flow rates were also examined. Figure 4 displays this for water, methanol, and 50% water/methanol. The observed frequency does not change significantly for water, but as methanol flows increase, it greatly rises and eventually levels off, receding the oscillations. As for all of the solvents examined, the effect of a 50% solution is intermediate to the pure mobile phases and the frequency is quantitative with respect to their carbon content and composition. Therefore, under isocratic conditions, the AFD frequency is constant; however, for changing flow rates or mobile-phase composition, it shifts accordingly. Occasionally, the latter changes (19) Sun, X.-Y.; Singh, H.; Millier, B.; Warren, C. H.; Aue, W. A. J. Chromatogr., A 1994, 687, 259-281. (20) Gaydon, A. G.; Wolfhard, H. G. Flames, 4th ed.; Wiley & Sons: New York, 1979. (21) Miller, D. R.; Evers, R. L.; Skinner, G. B. Combust. Flame 1963, 7, 137142. (22) Holm, T.; Madsen, J. O. Anal. Chem. 1996, 68, 3607-3611.
Table 1. Typical Properties of the AFD and Other Common LC Detectors
mass LODb linear rangec universal analyte response approximate costd
ELSD
UV-visible
RID
AFDa
500 ng 102-103 semi mass/volatility 11000
1 pg 103-104 no optical properties 6000
10 ng 102-103 yes optical properties 6000
80-120 ng 103 yes carbon content 400
a AFD values from alcohols in this work. b Mass limit of detection values obtained from ref 14. c UV-visible ref 9; ELSD ref 29; RID refs 9 and 12. d Prices in $US; ELSD and UV-visible from the 2004 Alltech Chromatography Sourcebook, Deerfield IL; RID from Thomson Instrument Co., Oceanside, CA.
Table 2. Comparison of LC-AFD, LC-ELSD, and GC-FID Response LC-AFD
cyclopentanol cyclohexanol dodecanol tetradecanol
LC-ELSD
GC-FID
mass LOD (ng)
linear range
mass LOD (µg)
linear range
mass LOD (pg)
linear range
80 120 120 110
2.5 × 103 1.7 × 103 1.7 × 103 1.8 × 103
20 20 1.8 2.4
25 10 39 83
380 210 260 200
>1.0 × 106 >9.5 × 105 >9.0 × 105 >1.0 × 106
may be minimized by choosing solvents similar in carbon number or adjusting the oxygen flow to offset the effect. However, due to its hydrocarbon sensitivity, the AFD does not appear practically compatible with LC gradient elution. Conversely, since AFD response is not influenced by mobile-phase optical properties, it is compatible with a wide range of solvents. Further, while aqueous solutions of nonvolatile salts and buffers generally disrupt the response or plug the transfer line near the flame, more volatile acids, such as formic acid, do not produce any adverse detector effects. Comparative Properties. Table 1 compares typical properties of the AFD and some common LC detectors. Generally, linearity in the AFD is intermediate to the ranges normally observed for the ELSD, RID, and UV-visible detector.9,23,24 For example, nearly 4 orders of linearity has been reported for a laser-based UVvisible method,25 whereas the ELSD often produces 1-3 orders.10,26-28 While detection limits in the AFD are also intermediate to those of ELSD or RID methods,10,14,26-28 they are still several
orders larger than UV-visible detection.9,23,24 Since the current LC apparatus was primarily assembled to examine liquid-phase introduction in the AFD, it is likely that a more optimized chromatography system providing sharper peaks should improve the detection limits. By further comparison, the AFD and the RID elicit universal response, whereas UV-visible and ELSD methods do not. Additionally, AFD response is proportional to molecular carbon content. Therefore, it is uniform toward a broad range of organic analytes regardless of their optical properties. For example, similar GC-AFD response was generated for aromatic and n-alkane analytes with the same carbon number.18 In the LC mode, the respective AFD sensitivity (Hz‚s/nmol of compound) for cyclohexanol and benzyl alcohol was about 2.9 and 2.7. Therefore, consistent response is obtained for compounds similar in structure but diverse in their molar absorptivity. Compared to ELSD,29 AFD response is independent of analyte volatility. For instance, large changes in LC-ELSD response were reported for various poly(ethylene glycol) formulations.30 In extreme cases, the smallest analytes detectable by ELSD were relatively nonvolatile (e.g., C24) even though volatile supercritical CO2 was the mobile phase31,32. In contrast to this, a similar AFD sensitivity is produced, for example, toward methanol and tetradecanol despite their difference in volatility. Table 2 compares calibration data for the alcohol standards used earlier as determined by LC-AFD and LC-ELSD using a methanol mobile phase. As seen, similar detection limits and linear ranges are achieved for the standards analyzed by AFD. By comparison, ELSD response is greater for the larger alcohols and more varied in linearity. As such, the AFD values are about 10-100 times larger than those of ELSD.
(23) Moslemi, P.; Najafabadi, A. R.; Tajerzadeh, H. J. Pharm. Biomed. Anal. 2003, 33, 45-51. (24) Wang, H.; Yi, E. C.; Ibarra, C. A.; Hackett, M. Analyst 2000, 125, 10611064. (25) Rosenzweig, Z.; Yeung, E. S. J. Chromatogr. 1993, 645, 201-207. (26) Toussaint, B.; Duchateau, A. L. L.; van der Wal, Sj.; Albert, A.; Hubert, Ph.; Crommen, J. J. Chromatogr., A 2000, 890, 239-249.
(27) Nair, L. M.; Stephens, N. V.; Vincent, S.; Raghavan, N.; Sand, P. J. J. Chromatogr., A 2003, 1012, 81-86. (28) Lo´pez- Lo´pez, A.; Castellote-Bargallo´, A. I.; Lo´pez-Sabater, M. C. J. Chromatogr., B 2001, 760, 97-105. (29) McNabb, T. J.; Cremesti, A. E.; Brown, P. J.; Fischl, A. A. Sem. Food Anal. 1999, 4, 53-70. (30) Alexander, J. N., IV J. Microcolumn Sep. 1998, 10, 491-502.
Figure 4. LC-AFD frequency obtained for various flow rates of a water ([), methanol (9), and 50% water/methanol mobile phase (2).
Analytical Chemistry, Vol. 76, No. 18, September 15, 2004
5463
∼4 orders of magnitude and no signal could be detected for even 200 µg of injected analyte. Thus, by comparison, LC-AFD has the advantage of screening both volatile and nonvolatile analytes. Figure 5 demonstrates a substantial AFD response for compounds that widely differ in size and volatility. Figure 5A presents an LC-AFD chromatogram for 50 µg each of cyclohexanol (100 g/mol) and tetradecanol (214 g/mol). Figure 5B illustrates a 250ng injection of gramicidin D, a mixture of polypeptides (1850 g/mol) each composed of 15 amino acids.33 Note the similar signal generated in each figure, despite a 200-fold difference in injected mass. This reflects AFD carbon sensitivity since the much larger polypeptide contains ∼100 carbon atoms/mol. Finally, in terms of relative cost, AFD is minimally expensive. With minor adaptations and a simple computer required for operation, a functional AFD could be produced for about 15-30 times less than the current commercial cost of the other detectors in Table 1. This is further reduced for laboratories already equipped with a working FID data system. Additionally, since the AFD is readily assembled and operated, it should be a widely accessible method.
Figure 5. (A) LC-AFD chromatogram illustrating the peaks obtained for 50 µg each of cyclohexanol and tetradecanol (in order of elution). (B) LC-AFD response toward a 250-ng injection of the polypeptide gramicidin D. Conditions are as in Figure 3.
For comparison, Table 2 also includes GC-FID calibration data since it is an excellent universal detector for volatile compounds. The values are within about a factor of 10 to a fully optimized commercial GC-FID15 and are clearly superior to those of AFD by almost 3 orders of magnitude. However, for nonvolatile analytes requiring LC separation, these detectors greatly differ when operated with organic mobile phases. Indeed, in our attempts to include LC-FID data in Table 2, the flame chemistry was saturated and analyte response was obliterated. Of note, methanol flow rates examined up to 50 µL/min increased the background level by
5464
Analytical Chemistry, Vol. 76, No. 18, September 15, 2004
CONCLUSIONS LC-AFD demonstrates a universal response toward carbon compounds carried by organic mobile phases. The current limits of detection are considerably larger than typical UV-visible values but compare well with ELSD and RID. Response linearity is within a factor of 10 of these methods. Similar to RID, AFD is not compatible with mobile-phase gradients. Conversely, the AFD responds to organic analytes regardless of their optical properties or volatility. Results suggest that this relatively simple and inexpensive device may be beneficial as an alternative universal LC detection method. ACKNOWLEDGMENT The authors are grateful to the Natural Sciences and Engineering Research Council of Canada for an operating grant in support of this research. Mike Fogwill is acknowledged for his helpful assistance. Received for review February 9, 2004. Accepted June 24, 2004. AC049777R (31) Berry, A.; Ramsey, E.; Newby, M.; Games, D. E. J. Chromatogr. Sci. 1996, 34, 245-253. (32) Strode, J. T. B.; Taylor, L. T. J. Chromatogr. Sci. 1996, 34, 261-271. (33) Stankovic, C. J.; Delfino, J. M.; Schreiber, S. L. Anal. Biochem. 1990, 184, 100-103.
