Anal. Chem. 2010, 82, 8034–8038
Elemental Composition of Nanoparticles with the Nano Aerosol Mass Spectrometer† Christopher A. Zordan, M. Ross Pennington, and Murray V. Johnston* Department of Chemistry and Biochemistry University of Delaware, Newark, Delaware 19716 The nano aerosol mass spectrometer (NAMS) irradiates individual, size selected nanoparticles with a high energy laser pulse to generate a mass spectrum consisting of multiply charged atomic ions. The elemental composition of the particle is determined from the ion signal intensities of each element, which requires deconvoluting isobaric ion signals at 4 m/z (O4+ and C3+) and at 8 m/z (O2+ and S4+). A method to deconvolute these ion signals using sucrose and ammonium sulfate as calibrants is presented. The approach is based on the assumption that the charge state distribution of a given element is independent of the chemical form of that element in the particle. Relative to previously reported methodology, the new approach permits accurate and precise determination of sulfur, which is crucial for interpretation of ambient nanoparticle data sets. With this approach, the differences between expected and measured elemental ratios of C, O, N, and S for a variety of test particles were generally much less than 10%, although a difference as high as 16% was observed. Nanoparticles in ambient air are of great interest owing to their potential to affect global climate by growing into the size range of cloud condensation nuclei1-3 and to negatively impact human health by inhalation.4-6 Understanding the sources of ambient nanoparticles and the risk they may pose requires the development of real time analytical methods. Because nanoparticles contain so little mass, airborne nanoparticle analysis is difficult and only a few methods have been reported.7 One such method, the nano aerosol mass spectrometer (NAMS), irradiates individual, size-selected nanoparticles with a high energy laser pulse to reach † Part of the special issue “Atmospheric Analysis as Related to Climate Change”. * Corresponding author. Phone: 302-831-8014. Fax: 302-831-6335. E-mail:
[email protected]. (1) Kerminen, V.-M.; Lihavainen, H.; Komppula, M.; Viisanen, Y.; Kulmala, M. Geophys. Res. Lett. 2005, 32, L14803. (2) Laaksonen, A.; Hamed, A.; Joutsensaari, J.; Hiltunen, L.; Cavalli, F.; Junkermann, W.; Asmi, A.; Fuzzi, S.; Facchini, M. C. Geophys. Res. Lett. 2005, 32, L06812. (3) Merikanto, J.; Spracklen, D. V.; Mann, G. W.; Pickering, S. J.; Carslaw, K. S. Atmos. Chem. Phys. 2009, 9, 8601–8616. (4) Hoek, G.; Boogaard, H.; Knol, A.; Hartog, J. d.; Slottje, P.; Ayres, J. G.; Borm, P.; Brunekreef, B.; Donaldson, K.; Forastiere, F.; Holgate, S.; Kreyling, W. G.; Nemery, B.; Pekkanen, J.; Stone, V.; Wichmann, H.-E.; Sluijs, J. v. d. Environ. Sci. Technol. 2010, 41, 476–482. (5) Ibald-Mulli, A.; Wichmann, H.-E.; Kreyling, W.; Peters, A. J. Aerosol Med. 2002, 15, 189–201. (6) Pope, C. A., III; Dockery, D. W. J. Air Waste Manage. Assoc. 2006, 56, 709–742. (7) Bzdek, B. R.; Johnston, M. V. Anal. Chem. 2010, DOI: 10.1021/ac100856j.
8034
Analytical Chemistry, Vol. 82, No. 19, October 1, 2010
Figure 1. NAMS spectrum of 25 nm diameter HEPES particles. The spectrum is an average of 20 single particle spectra.
the so-called “complete ionization limit”.8-11 The laser pulse induces formation of a plasma that completely disintegrates the particle into multiply charged atomic ions. In this limit, the sum of ion signal intensities from a single element is proportional to the amount of that element in the particle, and the ratio of summed ion intensities of different elements gives the elemental composition of the particle. An example spectrum is shown in Figure 1 for 25 nm massnormalized diameter HEPES (N-2-hydroxyethylpiperazine-N′-2ethanesulfonic acid; C8H18O4N2S) particles. Ions detected include the 2+ to 5+ charge states of C, N, and O, the 3+ to 5+ charge states of S, and H+. Determining the elemental composition from the mass spectrum is complicated by isobaric interference, specifically overlap of C3+ and O4+ at 4 m/z and O2+ and S4+ at 8 m/z. Although theoretically possible, deconvoluting isobaric signals based on isotopic signal intensities and natural abundances does not work in practice owing to the inherently low abundances of 13C, 18O, and 34S. The low signal-to-noise ratios of the corresponding ions inhibit accurate and precise determination of the elemental ratios. The process described herein for deconvoluting these signals requires an assumption about the relative populations of ions as a function of charge state. (8) Reents, W. D.; Ge, Z. Aerosol Sci. Technol. 2000, 33, 122–134. (9) Reents, W. D., Jr.; Schabel, M. J. Anal. Chem. 2001, 73, 5403–5414. (10) Mahadevan, R.; Lee, D.; Sakurai, H.; Zachariah, M. R. J. Phys. Chem. A 2002, 106, 11083–11092. (11) Lee, D.; Park, K.; Zachariah, M. R. Aerosol Sci. Technol. 2005, 39, 162– 169. 10.1021/ac101700q 2010 American Chemical Society Published on Web 08/30/2010
Inspection of the N2+ to N5+ charge states in Figure 1, for which there is no isobaric overlap, shows that the signal intensity and hence the ion population reaches a maximum for the +3 charge state. This intensity distribution reflects two competing processes: heating inside the plasma as the particle disintegrates and cooling outside the plasma as ions are drawn toward the mass analyzer. In previous work, a simple deconvolution model was constructed based on the similarity of the second, third, and fourth ionization energies of C and O.12 If it is assumed that the energy distribution imparted to these atoms during heating and cooling is independent of the specific element, then the fraction of atoms reaching each charge state should be the same: NC(n+) NO(n+) ) NC(total) NO(total)
(1)
where NC(n+) is the number of carbon atoms having a n+ charge, NC(total) is the total number of carbon atoms, etc. Thus, each charge state gives an independent measure of atomic composition: NC(2+) NC(3+) NC(4+) ) ) NO(2+) NO(3+) NO(4+)
(2)
In practice, using eq 2 along with the signal intensities at 5.33 m/z (O3+), 4 m/z (C3+ and O4+), and 3 m/z (C4+) to separate the 4 m/z signal intensity into contributions from O4+ and C3+ requires solving a quadratic equation. For most oxygenated organic compounds including laboratory generated secondary organic aerosol, this approach has proven to work well. However, the inherent variation of signal intensity from one laser shot to another leads to situations where solutions to the quadratic equation are negative or exceed the total signal intensity. In these cases, the offending spectra must be removed from the data set being analyzed. Furthermore, this approach suffers from poor precision and accuracy when deconvoluting O2+ and S4+ signals at 8 m/z since the ionization energies of O and S do not match in the same way as O and C. For these reasons, an alternative approach for signal intensity deconvolution is needed, especially for analysis of ambient nanoparticles that contain substantial amounts of sulfate. In the work reported here, a new method for signal intensity deconvolution is described. Unlike the previous approach which assumed a relationship among the same charge states of different elements, the new approach assumes a relationship among different charge states of the same element. The method is applied first to deconvolution of the O and C signals at 4 m/z and then to the O and S signals at 8 m/z. Performance is assessed through the analysis of standard particles having known compositions. EXPERIMENTAL SECTION The nano aerosol mass spectrometer (NAMS) has been described previously.12,13 Briefly, charged particles are sampled through an inlet consisting of an aerodynamic lens optimized for nanoparticle transmission and a quadrupole ion guide optimized (12) Wang, S.; Zordan, C. A.; Johnston, M. V. Anal. Chem. 2006, 78, 1750– 1754. (13) Wang, S.; Johnston, M. V. Int. J. Mass Spectrom. 2006, 258, 50–57.
for “high” m/z ion transmission. Particles emerging from the inlet are captured and focused into the center of a digital ion trap.13 The frequency applied to the ring electrode of the trap determines the size range of particles that are captured. Captured particles are then ablated with a high energy laser pulse, and the resulting ions are extracted into a time-of-flight mass analyzer. In this work, particles were passed through a unipolar charger before entering the mass spectrometer. For the size range of interest, particles entering the mass spectrometer had either zero or one positive charge. Of these, only the positively charged particles were efficiently captured and analyzed. The frequencies applied to the ring electrode of the digital ion trap selected singly charged particles having a mass-normalized diameter between 21 and 25 nm. The mass normalized diameter is defined as the diameter of a unit density particle that has the same mass (and mass-to-charge ratio) as the particle being analyzed. Since the densities were different for the various standards studied, the physical diameters of the analyzed particles also were different although the masses were the same. Single component aerosols of sucrose, adipic acid, sebacic acid, suberic acid, succinic acid, phthalic acid, phenylsuccinic acid, ammonium sulfate, HEPES (N-2-hydroxyethylpiperazine-N′-2ethanesulfonic acid), MOPS (3-(N-morpholine)propanesulfonic acid), and MES (2-(N-morpholino)ethanesulfonic acid) were generated from aqueous solutions of individual solutes using an electrospray aerosol generator (model 3480, TSI, Inc., Shoreview, MN). The generator produced a monodisperse aerosol of charged droplets that were neutralized with a radioactive source and then dried to produce the final particle size distribution. Sucrose solutions were prepared in aqueous ammonium acetate buffer to facilitate the electrospray process. Chemicals were obtained from the following sources: adipic acid (Fluka, St. Louis, MO); ammonium sulfate (Flinn Scientific, Batavia, IL); sucrose, MES, MOPS, phtahlic acid, phenylsuccinic acid, sebacic acid, suberic acid, and succinic acid (Sigma-Aldrich, St. Louis,MO); glycine and HEPES (Fisher, Fair Lawn, NJ). A scanning mobility particle sizer consisting of a nano differential mobility analyzer and condensation particle counter (nano-SMPS, model 3936, TSI, Inc., Shoreview, MN) was used to monitor the particle size distribution and number concentration during data collection. In most cases, several thousand single particle spectra were obtained for each standard. To reduce the impact of shot-to-shot variations in the mass spectra, 20 single particle spectra were averaged to generate one averaged particle spectrum. In the discussion below, each “particle spectrum” represents a 20-particle average. Custom programs for signal averaging, flight time-to-m/z conversion, peak area integration, and deconvolution of isobaric signals were written in LabVIEW v6 (National Instruments, Austin, Texas). RESULTS AND DISCUSSION The basic premise underlying this work, that the charge state distribution of an individual element is independent of the chemical form of that element, is illustrated in Figure 2. Shown in this figure are the charge state distributions for C, N, O, and S from the mass spectra of different composition particles. The C, O, and S distributions were obtained from the deconvolution procedure described below, while the N distribution was taken directly from the mass spectra. Within the inherent variation of Analytical Chemistry, Vol. 82, No. 19, October 1, 2010
8035
Figure 2. Charge distributions of C, O, N, and S in the NAMS spectra of several test particles. Error bars represent 1 standard deviation.
signal intensity from spectrum to spectrum, the N charge state distributions do not change with composition, which confirms the basic premise behind deconvolution. The C, O and S distributions are also the same within experimental error, although the similarity is not a rigorous test of the procedure since charge state distributions were assumed during deconvolution. The deconvolution procedure is described below, first for O and C at 4 m/z and then for O and S at 8 m/z. Deconvolution of O4+ and C3+ at 4 m/z. Deconvoluting the O and C signals at 4 m/z is a two step process. First, a calibration standard is used to determine the charge state distribution for O that best fits the known composition. Second, the charge state distribution of the standard is applied to the spectrum of an unknown particle to determine its O/C elemental ratio. In the procedure described below, sucrose was used as the standard and its charge state distribution was applied to the analysis of several organic acids. The charge state distribution for O in the standard is represented by eqs 3 and 4:
( )
(3)
( )
(4)
5+
(OR4+) ) (O5+)
4+
IPO
3+
(Oβ4+) ) (O3+)
N
IPO
IPO
IP
M
O4+
In these equations, the charge state intensities (Ox+, x ) 3-5) are expressed as ratios of the corresponding ionization potentials raised to a power (N, M). For a given (N, M) pair, 4+ values for O4+ R and O β are determined and then averaged 8036
Analytical Chemistry, Vol. 82, No. 19, October 1, 2010
4+ which is then used to determine the together to give Oave O/C elemental ratio of the particle as shown in eq 5:
O C
( )
) particle
4+ Onon-isobaric + Oave 4+ Cnon-isobaric + (I4 - Oave )
(5)
where (O/C)Particle is the elemental ratio in the particle, Ononisobaric is the O signal intensity at nonisobaric m/z values (8, 5.33, 3.2, 2.67 for sucrose), Cnonisobaric is the C signal intensity at nonisobaric m/z values (6, 3 for sucrose), and the C intensity at 4 m/z is given by the total intensity at 4 m/z (I4) minus O4+ ave. Since (O/C)Particle is known, the exponents (N, M) are iterated until the ratio calculated by eq 5 matches the known ratio. Iteration is performed for a set of standard particle spectra (465 in the present example) by a least-squares of error method where for each combination of N and M, a total error QN,M is calculated across all the standard particles. The minimum value of QN,M is found and the corresponding values of N and M are used for deconvoluting the total signal at 4 m/z into O4+ and C3+ signal intensities. If the calculated area of O4+ ave for an individual spectrum is larger than the total signal at 4 m/z, then the total area is assigned to O4+ and the C3+ area is set to zero. In principle, the above deconvolution procedure could have been performed in alternative, but mathematically equivalent, ways. For example, the charge state distributions in eqs 3 and 4 could be represented by simple ratios that are iterated rather than as exponents. In practice, we have found that directly iterating ratios gives less satisfactory results (poorer accuracy and precision when predicting unknown compositions) than iterating exponents. This difference may arise from the ability
Table 1. Prediction of O/C Elemental Ratio in Test Particles O/C sucrose “calibration” set (n ) 465)
phthalic acid (n ) 150)
adipic acid (n ) 150)
phenylsuccinic acid (n ) 150)
expected calcd value % RSD % diff expected calcd value % RSD % diff expected calcd value % RSD % diff expected calcd value % RSD % diff
0.917 0.91 7.1 -0.87 0.50 0.55 15 9.6 0.667 0.61 4.9 -8.5 0.40 0.43 2.5 7.3
of the exponent method to interrogate a greater portion of the QN,M surface with a step size appropriate for the shape of the surface. Another possibility is to iterate the C charge state distribution rather than O. Again, we find in practice that less accurate and precise results are obtained. This difference most likely arises from the shapes of the C and O distributions in Figure 2. The C distribution peaks at 4 m/z and a relatively small uncertainty in its contribution to the total signal intensity at 4 m/z causes a relatively large error in the calculated O/C ratio. The second step of the process is to calculate the O/C ratio of an unknown particle composition based upon the charge state distribution determined from sucrose standards. The calibration exponents, N and M, calculated in the first step are applied the 4+ unknown spectrum to directly calculate Oave for the unknown. 3+ C area is calculated by difference from the total m/z 4 area. 4+ As in the first step, should the value of Oave exceed the total m/z ) 4 area then the total area is assigned to O4+ and C3+ is assigned a zero area. Finally for each element, the area across all charge states is summed to get a total elemental area that is used for calculating the elemental composition of the spectrum. Performance of the deconvolution method is illustrated in Table 1, where the expected and calculated O/C ratios are compared for several particle compositions. When the O charge state distribution from calibrant spectra is used to “self-predict” the O/C ratio of the standards (i.e., the “unknown” particles analyzed in the second step are the same ones used as calibrants in the first step), the calculated ratio is within 1% of the expected value. When the O charge state distribution determined from the sucrose standard data set is used to determine the O/C ratio of sucrose “test” particles in a second data set that was acquired 10 days before the standard data set, the calculated ratio is within 5% of the expected ratio. The “test” sucrose data set illustrates the robustness of the calibration over a period of days to weeks. Table 1 also summarizes the results of six organic acid test particles. In each case, the difference between expected and calculated O/C ratios is less than 10%. Deconvolution of O2+ and S4+. Deconvoluting the O2+ and 4+ S signals at 8 m/z is performed in a similar manner to O and C at 4 m/z. The S charge state distribution is determined using
O/C sucrose “test” set (n ) 25)
sebacic acid (n ) 150)
suberic acid (n ) 150)
succinic acid (n ) 150)
expected calcd value % RSD % diff expected calcd value % RSD % diff expected calcd value % RSD % diff expected calcd value % RSD % diff
0.917 0.87 8.0 -4.7 0.40 0.43 5.7 6.4 0.50 0.51 4.2 2.1 1.00 0.92 10 -8.4
ammonium sulfate as the calibration standard according to eq 6:
( ) 3+
(S4+) ) (S3+)
IPS
IP
K
S4+
(6)
where K is the exponent calibration constant and the other terms are analogous to eqs 3 and 4. Unlike the procedure for deconvoluting O and C, only a single equation/exponent is used; in other words the analogous equation for S5+ and S4+ is eliminated. In practice, we have found that including a second equation in the O-S deconvolution degrades performance (poorer accuracy and precision) presumably because of the low intensity and large relative standard deviation of the S5+ intensity at 6.4 m/z (Figure 2). In addition, using the O charge state distribution for deconvolution was found to give poorer results. In most cases, the O makes a large contribution to the 8 m/z signal intensity. This, when combined with a S charge state distribution biased strongly toward 8 m/z means that a relatively small uncertainty in the O2+ signal intensity leads to a relatively large uncertainty in the total S signal intensity. In a manner similar to the one described above for O and C, the exponent K in eq 6 is iterated to minimize the error between the known and calculated O/S ratio for ammonium sulfate. The S charge state distribution determined in this manner is used to separate the O and S signals at 8 m/z for unknown particle spectra. When C, O, and S coexist in the same particle, the signal at 4 m/z is deconvoluted first and the signal at 8 m/z second. Results for several test particles containing C, N, O, and S are shown in Table 2. Self-prediction of the ammonium sulfate standard particles gives calculated ratios within a few percent of the expected values. The differences between expected and calculated ratios for other particles are generally within 10% although a deviation as high as 16% was observed. In practice, we have found that recalibration is necessary when laser-particle interaction inside the instrument has been affected, for example, by changes in mechanical alignment, applied voltages, or laser pulse energy. Ionization occurs over a relatively narrow range of laser pulse energies. Typically, the NAMS is operated with the 200 mJ/pulse maximum output of the laser. When the pulse energy is reduced, the particle hit rate decreases quickly to the point where no particles are detected at ∼180 mJ/ Analytical Chemistry, Vol. 82, No. 19, October 1, 2010
8037
Table 2. Prediction of C, O, N, and S in Test Particles elemental ratio ammonium sulfate “calibration” set (n ) 29)
HEPES (n ) 150)
MES (n ) 150)
MOPS (n ) 150)
glycine (n ) 150)
O/C expected calcd value % RSD % diff expected calcd value % RSD % diff expected calcd value % RSD % diff expected calcd value % RSD % diff expected calcd value % RSD % diff
pulse. Even a few percent decrease in the average pulse energy is sufficient to require recalibration. Another variable with regard to calibration is particle size. Typically, NAMS is operated at a few specific, discrete sizes in the 10-25 nm range depending on the experiment being performed. At the higher end of this range, the calibration is relatively insensitive to particle size. As the particle size decreases toward 10 nm, recalibration becomes necessary. Although recalibration is needed for quantitative measurements with different particle sizes and/or laser pulse energies, the spectra remain qualitatively similar with only minor changes in the distribution charge states of the various elements. While the accuracy of elemental analysis by NAMS is poorer than conventional elemental analysis methods, it should be realized: (1) nanoparticles are analyzed online and in real-time, and (2) the total mass analyzed even after averaging many particles together is in the low femtogram range. In contrast, conventional methods are offline and typically require on the order of 10 mg of sample. The accuracy of elemental analysis by NAMS is similar to that of the aerosol mass spectrometer,14 which analyzes particles in the ∼50-1000 nm size range.15 The performance of NAMS for elemental analysis is sufficient to monitor (14) Aiken, A. C.; DeCarlo, P. F.; Jimenez, J. L. Anal. Chem. 2007, 79, 8350– 8358. (15) Canagaratna, M. R.; Jayne, J. T.; Jimenez, J. L.; Allan, J. D.; Alfarra, M. R.; Zhang, Q.; Onasch, T. B.; Drewnick, F.; Coe, H.; Middlebrook, A.; Delia, A.; Williams, L. R.; Trimborn, A. M.; Northway, M. J.; De Carlo, P. F.; Kolb, C. E.; Davidovits, P.; Worsnop, D. R. Mass Spectrom. Rev. 2008, 26, 185– 222.
8038
Analytical Chemistry, Vol. 82, No. 19, October 1, 2010
0.50 0.48 12 -4.8 0.667 0.59 7.9 -11 0.571 0.48 12 -16 0.50 0.55 15 9.6
O/S
O/N
4.0 3.9 13 -1.9 4.0 3.9 28 -2.4 4 4.0 20 1.0 4 3.5 31 -12
2.0 1.9 7.7 -3.2 2.0 1.8 11 -9.1 4 3.6 7.5 -11 4 3.4 14 -14 2 1.9 8.8 -5.9
C/N
4.0 3.8 8.6 -4.0 6 6.0 6.0 0.22 7 7.2 11 3.2 2 1.8 9.8 -11
C/S
N/S
8.0 8.2 25 2.6 6 6.8 17 13 7 7.4 26 5.0
2.0 2.0 11 1.5 2.0 2.1 25 7.2 1 1.1 17 13 1 1.0 27 2.6
relevant changes in aerosol composition with time for both laboratory16 and field17 experiments. In conclusion, a new approach for deconvoluting isobaric ion signals in nanoparticle mass spectra has been developed and tested. The approach is general and can be adapted to other isobaric interferences as they arise, for example, deconvolution of N2+ and Si4+ at 7 m/z.17 ACKNOWLEDGMENT This research was supported by grants from the National Science Foundation (Grant CHE-0808972) and the Health Effects Institute (Grant 4775-RFPA06-4/07-9). Research described in this article was conducted in part under contract to the Health Effects Institute (HEI), an organization jointly funded by the United States Environmental Protection Agency (EPA) (Assistance Agreement CR-83234701) and certain motor vehicle and engine manufacturers. The contents of this article do not necessarily reflect the views of HEI, or its sponsors, nor do they necessarily reflect the views and policies of the EPA or motor vehicle and engine manufacturers. Received for review June 28, 2010. Accepted August 21, 2010. AC101700Q (16) Tolocka, M. P.; Heaton, K. J.; Dreyfus, M. A.; Wang, S.; Zordan, C. A.; Saul, T. D.; Johnston, M. V. Environ. Sci. Technol. 2006, 40, 1843–1848. (17) Zordan, C. A.; Wang, S.; Johnston, M. V. Environ. Sci. Technol. 2008, 42, 6631–6636.
