Review of Measurement Methods and Compositions for Ultrafine Particles

Impactor, virtual impactor/aerosol concentrator, and aerodynamic lenses are used to separate the ultrafine particle (UP) fraction from other particle sizes for chemical analysis. Cascade impactors, such as the Micro-Orifice Uniform Deposit Impactor (MOUDI), are most commonly used in field studies, with sampling onto substrates amenable to different chemical analyses. Impactors need sufficient sampling flow rates and homogeneous deposits on the impaction surfaces for multiple chemical analyses. Mass, elements, ions, and carbon fractions can be measured on these substrates by several analytical methods. Specific organic compounds measured by solvent extraction require substantial mass loadings that can only be obtained by compositing samples from several measurement periods unless aerosol concentrators or high-volume sampling devices are used. Thermal desorption-gas chromatographic/mass spectrometry has potential to obtain organic speciation with small sample sizes. Studies of UP composition began in the late 1990s, with 25 ambient studies surveyed here. These are mostly from urban areas. Organic material, including polycyclic aromatic hydrocarbons (PAHs), usually constituted the most abundant portion of UP, with high elemental concentrations found near industrial sites. Much of the UP < 50 nm appears to be semi-volatile, consistent with it being composed by organic materials such as hopanes from engine oils or condensed secondary organic aerosol such as organic acids.


INTRODUCTION
Ultrafine particles (UP) are loosely defined as those with diameters in the range of ~1 nm to 100 nm (nanometers). UP are bigger than air molecules (~0.3 nm), but smaller than the upper limits of PM 2.5 or PM 10 (particles with aerodynamic diameters less than 2.5 and 10 micrometers [µm], respectively) regulated by the U.S. National Ambient Air Quality Standards (NAAQS; U.S. EPA, 2006). Although UP do not contribute large quantities to PM 2.5 or PM 10 mass, they dominate the number concentration and most of the surface area. UP are produced by condensation of hot vapors in fresh combustion emissions, or form naturally when gases oxidize to compounds with lower vapor pressures and spontaneously nucleate or condense on other small particles (Kulmala et al., 2004). Owing to their small sizes and high mobilities, UP diffuse rapidly and may combine with each other, with larger particles, and with nearby deposition surfaces within minutes to hours. UP may contain organic material, transition metals, sulfuric acid (H 2 SO 4 ), and free radicals. UP often consist of volatile components at atmospheric temperatures, and the high curvature of the smallest particles (< 10 nm) favors evaporation over larger particles of the same composition (the Kelvin effect). Gases evaporated from small particles may re-condense on larger ones, thereby shifting the distribution toward larger particle diameters (Zhang and Wexler, 2002).
UP are often considered deleterious when they are inhaled or ingested into the human body (Oberdörster et al., 1995). They penetrate to the lower parts of the lung (Daigle et al., 2003) where their large numbers can defeat defensive mechanisms. Thereafter, they can transport through the bloodstream or lymphatic system to vital organs (Oberdörster et al., 2004). However, UP can also be lifesavers. The same inhalation and transport properties can rapidly deliver medicines to specific locations. Iron (Fe) nanoparticles can be magnetically directed to specific locations without invasive procedures.
UP are produced by either pollution sources or for commercial applications and can result in adverse health effects in the ambient and workplace environments if they are not appropriately controlled Chow et al., 2005c;Chow and Watson, 2006). Control methods are more complex and cost more than those for larger particles.
This review examines the state of knowledge concerning UP chemical composition. Specific objectives are to explain and evaluate measurement methods for UP chemical composition, summarize UP chemical composition, compare them among different times and places, and specify gaps and uncertainties in the current knowledge base regarding UP composition and potential methods of filling those gaps.

MEASUREMENT METHODS FOR ULTRAFINE PARTICLE CHEMICAL COMPOSITION
PM composition measurements require collection of materials on substrates that can then be analyzed by precise laboratory methods (Chow, 1995). Since UP mass concentrations are low compared to other size fractions (e.g., PM 2.5 and PM 10 ), large sample volumes or sensitive analytical methods are needed. Inertial size-selective inlets for UP present design and application challenges. In situ single particle mass spectrometers are emerging technologies for UP chemical characterization (Middlebrook et al., 2003).

Particle measurement systems
A generic particulate sampling system includes size-selective inlet, sampling surface, denuders, filter holders, flow controllers, and pumps . Although simple in concept, practical implementation requires a careful integration of the components specific to the sampling objectives. The nature of the aerosol being sampled, environmental sampling conditions (e.g., temperature and relative humidity), and the types of chemical analyses applied to the filter deposit must be evaluated before the sample is taken.

UP size-selective inlets
Impactors, virtual impactor/aerosol concentrators, and aerodynamic lenses have been devised to separate the UP fraction from larger particles.

Impactors
Stacked or cascade impactors obtain particle size ranges in series. At near-ambient pressures, the lower size ranges of cascade impactors were once limited to ~100 nm because it was impractical to make jet widths small enough and flow rates high enough to permit impaction of smaller particles (Marple, 2004). A filter located after the final stage would collect all of the smaller particles, but this was insufficient to characterize the UP fraction.
UP sizes are similar to the mean free path of air molecules (66.4 nm at 293°K and atmospheric pressure), the distance a typical air molecule travels before encountering another molecule. The mean free path decreases at lower pressures. The Cunningham slip correction factor is used in impactor design equations to account for differences in particle movements as they become less equally bombarded by air molecules. By operating impactor stages at lower pressures, size cuts can be reduced. Smaller nozzle widths decrease downstream pressures as well as increasing the velocity through the nozzle, which also lowers the cut-point. UP sampling on substrates became practical in the late 1970s and early 1980s with the perfection of low pressure impactors (Berner, 1972;1976a;1976b;1984;Hering et al., 1979a;1979b;Marple et al., 1981;Wang and John, 1988;Hillamo and Kauppinen, 1991). Table 1 summarizes the cut points for different cascade impactors.
The 13-stage Low Pressure Impactor (LPI; Hering et al., 1979a;1979b) uses a single nozzle for each stage to concentrate collected particles into a spot on the impaction surface. It was designed to use metal strips as impaction substrates that could be analyzed for sulfur by flash volatilization (Roberts and Friedlander, 1976). The small amount of deposit collected by the UP stages, the non-uniformity of the deposit, and the collection substrate are not amenable to a broad range of chemical composition measurements.
Variation in the number of nozzles per stage creates uncertainties for chemical analysis. Halder et al. (1999) modified their elemental detection system to rotate the samples in front of the excitation beam. This is not satisfactory for other methods where a portion of the deposit is used to extrapolate results for the entire sample.
The 13-stage Electrical Low Pressure Impactor (ELPI, Dekati Instruments, Finland) uses a unipolar corona discharge to impact an unit charge on each particle (Marjamäki et al., 2000). The charged particles then travel through a series of impactors and are deposited onto stages based on their aerodynamic diameters. Each impactor stage is electronically isolated and the accumulated charge on each substrate is proportional to the number of particles deposited on that stage. ELPIs have been used in a variety of source characterization studies for diesel (e.g., Arnold et al., 2006;Mamakos et al., 2006), gasoline (e.g., Maricq et al., 1999), wood combustion (e.g., Hays et al., 2003), and power plant (e.g., Yi et al., 2006) emissions, as well as for indoor (e.g., Mosley et al., 2001) and outdoor (e.g., Gouriou et al., 2004) studies and characterization of pharmaceuticals (Glover and Chan, 2004).
The 12-stage Small Deposit Low Pressure Impactor (SDI; Maenhaut et al., 1996) was developed for compatibility with Proton Induced X-ray Emission (PIXE) spectroscopy to measure elemental composition. It is called "small deposit" because the sample is focused within an 8 mm diameter spot.
Although cascade impactor substrates are sometimes greased or oiled to minimize reentrainment and bounce from one stage to the next, these coatings interfere with chemical analyses (Fujitani et al., 2006;Wang et al., 2005). This is more of an issue for larger, and drier, soil particles than for PM 2.5 which are often inherently oily. Serious particle bounce is indicated by soil-related elements in the lower impaction stages.

Aerosol concentrators
Virtual impactors are often used as aerosol concentrators (e.g., Sioutas et al., 1995) to obtain aerosol concentrations higher than those found in ambient air. Several UP concentrators have been developed (Gordon et al., 1999;Gupta et al., 2004a;Kim et al., 2000a;2000b;Misra et al., 2004;Sioutas et al., 1999) where UP are drawn through an inlet that removes the large particles (e.g., > 0.15 µm), then into a chamber saturated with water vapor over a warm water reservoir. UP subsequently pass through a condensing area at lower temperature where the particles grow as water vapor condenses on them. These grown particles then pass through a 1.5 µm cut-point virtual impactor where they are separated from most of the airstream.
They then pass through a drier where the water is evaporated and they return to their UP size ranges. The implementation of Kim et al. (2000b) operates at 120 L/min and the minor flow can vary from 3 to 12 L/min. This provides for an enrichment of 10 to 40 times the ambient UP concentration. Although constructed primarily for animal exposure studies, this type of concentrator inlet can be used to obtain large quantities of UP on filters amenable to different chemical analysis methods. Owing to the different hygroscopic properties of UP, the sampled composition and size distribution may differ from that of the ambient air.

UP aerodynamic lenses
The aerodynamic lens (Liu et al., 1995a;1995b;Middha and Wexler, 2005;Petrucci et al., 2000) consists of apertures of varying sizes in a series. A particle beam is produced when a particle-laden gas expands through a nozzle into a vacuum. Particles move closer to the axis when their aerodynamic diameters are less than a critical value and experience small radial drag forces. They stay close to the axis during nozzle expansion and therefore form a narrow particle beam downstream. The major effects that limit the minimum beam width are Brownian motion and lift forces on particles during the nozzle expansion. Aerodynamic lenses are used almost exclusively on single particle spectrometers for particle sizes < 100 nm. Because the lower size range of particle mass spectrometers is limited, and quantification is less than 100% efficient, these instruments underestimate particle counts.

Sampling substrates
Sampling substrates must be matched to the analysis purpose (Chow, 1995). For impactors such as the MOUDI, the substrates must be thin enough that they do not interfere with the narrow gap between the nozzle exit and the impaction plate. The substrates cannot contain the substances being measured, so it is often necessary to operate several instruments in parallel to accommodate a number of filter media. MOUDIs can be obtained for either 37 mm or 47 mm diameter substrates.
Ringed Teflon-membrane filters have been found to pop up the retainer ring on the MOUDI filter holder and have been used after the last impactor stage that allows the flow for the impactors to be drawn through. Because filter porosity is not needed in an impactor such as the MOUDI, 37 or 47 mm disks of non-porous Teflon fluorinated ethylene propylene copolymer (FEP) films of 0.002 thickness (DuPont) are used. These disks are soaked in methanol overnight, rinsed with distilled-deionized water (DDW) and dried in a vacuum chamber to remove contaminants. These substrates are amenable to mass, elemental, and ion determinations.
Aluminum (Al) foil in 37 or 47 mm disks (Reynolds Aluminum; Gresham, OR) are often used as substrates for carbon analysis. The disadvantage of foil is that it is highly reflective and melts at ~600°C when heated in an inert atmosphere. This makes the separation of elemental from organic carbon (EC, OC) uncertain when measured by thermal/optical methods (Chow et al., 1993;. One solution is to apply the pyrolysis correction on the quartz backup filter to the total carbon (TC) measured on the Al substrates.
Other substrates (e.g., Pallflex TX40HI20 and T60A20 Teflon-coated glass-fiber filters) can be used for ion analyses and for specific organic compounds, but not for TC analyses owing to their Teflon coating. Glass-fiber filters contain borosilicate glass filaments and should not be considered for particle sampling (Coutant, 1977;Spicer and Schumacher, 1979;Witz et al., 1983;Lin and Friedlander, 1988). Fujitani et al. (2006) showed that mode diameter varies from 56 nm for ELPI with Al foil, to 100 nm for a Nano-MOUDI with Al foil, to 260 nm for ELPI with quartz-fiber filters for diesel exhaust particles. Heavy loadings on ELPI-Al (similar to those of polycarbonate substrates) experienced particle blow-off, resulting in a smaller mode diameter.
Etched polycarbonate-membrane filters have low elemental blank levels and are appropriate for elemental and ion analyses but not for thermal evolution carbon analysis .

Organic speciation
The most common method used for speciated organic compounds (such as PAHs, alkanes, alkenes, and polar organics) in PM samples is solvent extraction (SE), followed by gas chromatography (GC)/MS, time-of-flight (TOF)/MS, or flame ionization detection (FID).
Combined Fourier transform infrared (FTIR)/MS techniques or high performance liquid chromatography (HPLC)/MS are also used. HPLC is a form of IC that uses columns and eluents specific to water soluble carbon, especially organic acids. Large sample deposits are required for these analyses, much more than is available in the UP fraction unless acquired from high-volume sampling, impactors, or aerosol concentrators.
Thermal desorption (TD)-GC/MS (e.g., Ho and Yu, 2004;Hays and Lavrich, 2007) is an emerging technology in which a small section of the substrate can be placed directly into the GC injector and heated. Organic materials are volatilized and detected by GC/MS or GC/MS-FID.
Very small samples, such as UP samples from impactors, can be analyzed by this technique.
Thermal denuders can also be used with other UP detection devices, such as the scanning mobility particle sizer (SMPS), to infer aerosol composition (Hasegawa et al., 2004). Particles are drawn through an inlet that is cycled between ambient and a selected higher temperature and the SMPS size distributions are compared. Temperatures in the range of 300°C are often used to include sulfates, nitrates, and many OC compounds. The difference between the heated and unheated measurements separates the stable from the "semi-volatile" fraction of the aerosol.
Another approach is to use a humidifier between two SMPSs to determine the hygroscopicity of the sampled aerosol. This is termed the Tandem Differential Mobility Analyzer (TDMA; Rader and McMurry, 1986). Sodium chloride, ammonium sulfate, ammonium nitrate, and some organic compounds grow to larger sizes when humidified. Other materials, such as EC, often do not grow as much. This gives a qualitative indication of the potential chemical composition.

Individual particle analysis
Single particles are characterized by optical or electron microscopy. Optical microscopy (Lee and Kelly, 1980;Janocko et al., 1982;Casuccio et al., 1983;Dattner et al., 1983;Lucas et al., 1988) is useful for coarse particles with sizes much larger than the wavelength of light (0.3 to 0.7 µm). Electron microscopy (EM; such as scanning EM [SEM] or transmission EM [TEM]) is needed to characterize smaller particles and their size distribution. These methods provide information on particle color, shape, size, and composition. Estimated from pie charts in Figure 5 • Used particle density of 1.7 g/cm 3 to convert equivalent number concentration to mass distribution.

ULTRAFINE PARTICLE COMPOSITION
• OC ( • Two ATOFMS at the three urban sites. • One ten-stage MOUDI (Model 100) and non-rotating MOI (Model 110) report the bottom six stages (0.056 < d p < 1.8 µm) and after filter.
• Low-volume total and fine PM sampler. UP composition for the MOUDI stacked bar charts were too small to quantify UP composition.
• Particle density of 1.3 g/cm 3 used to convert equivalent particle number concentration to mass distribution.
• Nominal cut points for MOUDI and MOI were not reported.
• Different numbers of impactor samples were taken at the different sites on different days, precluding site-to-site comparison.
• Unimodal particle mass distribution found at Riverside, with the peak around 0.5 µm, larger than the mode at Long Beach.
• UP mass was ~1-1.5 µg/m 3 , with organic compounds being the largest contributor at urban sites. Santa Catalina Island UP was mostly inorganic, but not well identified.
• UP composition varied by location and season.
• The three UP stages are 35, 67, and 93 nm.
• OC was predominant in UP at both sites, accounting for 50-80% of the total mass.
• Distinct mode of UP mass and OC in the 32-56 nm range was independent of time of day at Downey.
• OC drove mass size distribution at the diesel-dominated Downey site. Morning peaks (32-56 µm) resulted from condensation of vehicle-emitted organic vapors.
• Average UP OC/EC ratios were 2.4, 2.7, and 3.4 at Downey, and 7.5, 9.9 and 7.5 at Riverside for morning, midday, and evening periods, respectively. Higher OC/EC ratios at Riverside were attributed to gasoline-fueled vehicle emissions.  • Unimodal size distribution with mode 30-40 nm at Downey and bimodal at Riverside with an increase in accumulation mode (0.1-2.5 µm).
• An increase in sub-100 nm during 1400-1600 PST was found at Riverside.
• MOUDI quartz-fiber after filter OC may have positive or negative artifacts.
• Fe was most abundant of the transition metals in UP, followed by Cu and Zn (relatively high V, Cr, and Ni were also found).
• UP V was 11 times higher at Downey than at Riverside, reflecting the impact from refineries at Downey.  other metals (specific elements were not given), and unknown categories.
• Obtained approximately 15,000 spectra for individual particle composition but no quantitative chemical compositions were given.
• No distinguishable differences were found for minerals, SiO 2 , Fe, and other metals.
• No S or SO 4 -were found in compound classes.
• Many of the major compound classes appeared in same size and/or wind direction ranges, indicating emissions from specific sources. • No quantitative UP mass or chemical composition were reported.
• About 7% of TC was organic for d p < 0.5 µm. • UP and fine PM were not elevated during the study.
• No acidic or fibrous particles were found in UP fractions.
• Morphology of fine PM collected soon after the WTC collapse with major fires showed agglomerates of small particles under incomplete combustion. • Use of specific ng compounds at specific time intervals and location to imply source origins. * Average concentrations were approximated from histograms.
• No quantitative data was presented.
• Particle mass distributions suggested photochemical origins during summer at both sites and during winter at the Riverside site.
• Particle growth was not significant during dry and hot summer, owing to the hydrophobic nature of EC and nonpolar organics in UP.
• Abundant hopanes (0.2-1.1 ng/m 3* ) in UP during the morning at the USC site suggested fresh vehicle emissions (i.e., from motor oil).
• Abundant BaP (0.05-1.5 ng/m 3* ) and COR found in UP suggested poor spark-ignition combustion in gasoline-fueled vehicles. • Data presented in mass and 3-D volume distribution using 11-point second order Savitzky-Golay smoothing function, TDMA growth factor distributions, and size-resolved hygroscopicity. • Divided UP composition into four categories based on their hygroscopicity: − pure and mixed water solubles: -inorganic salts, acids, and oxidized organics − pure and mixed water insolubles: -soot, organic compounds, metals, and dust.
• By combining size distribution with aerosol hygroscopicity, the size-resolved aerosol composition was inferred.
• Vehicle related OC and EC emissions during morning traffic hours resulted in simultaneous increase in pure insoluble material throughout the size ranges up to 344 nm.
• Particles > 0.1 µm exhibited bimodal growth, with increasing importance of the hygroscopic mode with increasing dry particle size.
• During two episodes, pronounced increase in small particle concentrations were followed by particle growth. It appeared that condensation of organic compounds was responsible for the initial growth, while condensation of inorganic compounds was responsible for the continued growth.
• Morning traffic resulted in an increase in pure insoluble material throughout the 700 nm size range. • Increase in UP corresponded to total traffic and high concentration of diesel engine vehicles.
• Particle concentration for 18.4 < d p < 50 nm increased in the morning under calm winds, which corresponded to the increased EC and NO.
• The 30 nm peaks at the suburban Tsukuba site in the afternoon implies continued impacts from vehicle exhaust as well as photochemical reactions.
• UP was of 1.5% of PM 11 .
• Highest PAHs, 2-NF and mutagenicity per unit PM mass were found in UP fractions, suggesting UP were efficient at carrying mutagenic compounds which cause adverse health effects.
• Based on size distribution of PM between 0.11 and 11 µm, 2.9-5.8% of PAHs and 2-NF, and 5.1-5.9% of mutagenicity were found in UP.
• Attributed PAH to emissions from combustion sources. and Cr, indicative of gas-to-particle conversion in the atmosphere and by the condensation of gas vaporized in the combustion process.
• Only one LPAI stage contained UP.
• Peak EC concentration was in the range of 0.12-0.29 µm, while peak OC was found at 0.29-0.67 µm.
• High OC/EC ratio suggested that organic compounds were emitted as primary particles from combustion sources in Kansai.
• No specific source apportionment results were given for UP.
• Factor analysis indicated automobile exhaust, fossil fuel contribution, refuse incinerations, iron industry, and geological material contributed to PM 2.5 . • PHE-FLT and BAA-IND in UP mode accounted for 30% and 40% of the PAH mass.
• EC size distribution did not vary over the study period with abundant UP and accumulation modes.
• Correlations with temperature were high for SO 4 = and EC in UP mode and OC and NO 3 -in coarse mode.
• As temperature decreased, particle-phase PAH concentration (i.e., the BAA-IND group) increased with decreasing volatile species, suggesting the increased partitioning from vapor phase and increased photostability during the winter.
• Similar size distributions were found for monthly composite target PAHs from October to February.
• Coarse fraction PAHs were found from March to July. into "pure" and "internally mixed" classes.
• Identified events on the basis that concentration increased 10 fold over the baseline concentration levels.
• Reported pure or fresh (50-90 nm) and aged (110-220 nm) NO • Obtained approximately 75,000 size particle spectra in the positive ion mode.
• UP NO 3 -events were found during low temperatures and high RH.
• Partitioning of NH 4 NO 3 to particle phase was influenced by the particle number concentrations and chemical composition during NO 3 -events.
• V, a marker for residual oil combustion, was observed from all wind directions.
• Fe and Pb were observed from the east-northeast, whereas As and Pb were observed from the south-southeast. Fang et al., (2005; 2006b) • One traffic sampling site, west of Taichung, Taiwan.
• • Used cumulative fractions to interpret the UP fraction.
• Ion concentrations were reported as µg/g, and metal concentrations as mg/g or as µg/m 2 for filter density, inconsistent with mass concentrations of µg/m 3 . • Daytime UP concentrations were ~50% lower than nighttime concentrations.
• UP concentrations did not increase during multi-week stagnation in the SJV (consistent with coagulation as a dominant removal mechanism for UP) though PM 2.5 increased by 7 fold.. • Carbon and hygroscopic particles exist separately in the SJV until coagulation mixes them in the accumulation mode.
• Converted SMPS number to mass concentrations by assuming density of 1.6 g/cm 3 .
• Highest UP and EC mass found at USC, probably due to the impact of nearby freeway traffic with diesel vehicle emissions.

Data reporting
Inconsistent definitions are used for the terms "ultrafine particle" and "nanoparticle." Many of these definitions are operational, depending on the measurement method. In Table 2, Kawanaka et al. (2004) define UP as particles < 110 nm, while Geller et al. (2002), Fine et al. (2002), Miguel et al. (2004) and Sardar et al. (2005) define UP as < 180 nm. Chow et al. (2005c) recommended the use of < 10 nm for nanoparticles and < 100 nm for UP in the physical sciences.
Many of the references report their results graphically rather than in tables, thereby making it difficult to obtain quantitative concentrations for the different chemical species. Several reporting methods include stacked bar charts of the composition for the different size fractions in which only the major components could be identified. Some references report pie charts with percentage contributions of each chemical component. These had to be deconvoluted to obtain the absolute concentrations reported in Table 2, and this was only possible when the UP mass was reported with the pie chart. Several of the studies report only a few of the UP size fractions rather than the total for all particles less than 100 nm. The most straightforward references were those that reported the results in a table for each of the species measured and each stage of size-segregation, such as that of Pakkanen et al. (2001b).
Data are qualitative or semi-quantitative for the aerosol mass spectrometer studies Rhoads et al., 2003;Tolocka et al., 2004a;. Zhang et al. (2004) show that OC is a major component of small particles, demonstrated by the percentage of nucleation events. Erdmann et al. (2005) classify single particle spectra, but this is often subjective and it is not possible to compare the results among different investigators, even when they are using the same instrument. Tolocka et al. (2004a) shows a clear directionality of UP metals using the short-time resolution available from the single particle mass spectrometer, in which most of the UP Fe is coming from a source to the northeast, whereas most of the UP arsenic (As) derives from sources to the southeast. This type of data can be correlated with source locations to better identify and quantify contributors to excessive concentrations.
Other U.S. cities include Houston, TX , Atlanta, GA , and Baltimore, MD (Tolocka et al., 2004a;, where single particle spectrometers were applied. Gasparini et al. (2004) reported TDMA measurements from Houston to evaluate hygroscopic properties. MOUDI measurements were also taken by Cohen et al. (2004)  Most of the studies were performed during the late 1990s and early 2000s. The number of samples taken ranges from two to ~50, which is a small data set from which to draw general conclusions. Sampling locations were largely within highly populated urban centers, with a few in non-urban areas for contrast. Sometimes example results from a few interesting samples were reported, while at other times only the all-sample averages were reported.

UP composition
Organic material (OM = OC×1.4 to account for the unmeasured hydrogen, oxygen, nitrogen, and other material that is associated with organic molecules) was the major component of the UP fraction. Hughes et al. (1998) found that OM accounted for 40% to 53% of the UP in Los Angeles during 1996. This was consistent with the results of Cass et al. (2000), Geller et al. (2002), Kim et al. (2002), and Sardar et al. (2005) for southern California samples. However, Miguel et al. (2004) found that 43% of UP was composed of EC while only 24% came from OC.
For northern California, Herner et al. (2005; found that OC and EC constitute ~98% of UP mass. OC and EC in UP were also reported by Huang et al. (2006) for a southern China tunnel. Different OC and EC levels could be due to the different sampling sites, sampling times, UP cut points, and measurement methods. In any case, OC and EC were major portions of UP in nearly all of the summarized studies.
When the OC fraction was examined for more specific organic compounds, Miguel et al. (2004) found that phenanthrene/fluoranthene and benzo[a]anthracene/indeno[1,2,3-cd]pyrene in UP mode accounted for large fractions of PAH mass in Southern California. These compounds indicate contributions from fossil fuel combustion sources, especially gasoline-and dieselpowered vehicle exhaust. In Saitama, Japan, Kawanaka et al. (2004) found PAH in the range of 3-30 ng/g PM and 3-6% of PAH were found in UP. They also found that these samples were highly mutagenic. Mutagenicity was not observed at urban and suburban locations in Guangzhou, South China, especially for heavy PAHs (> 5 rings; Duan et al., 2005). PAH are semi-volatile in nature and can easily condense on other particles. The larger surface-to-volume ratios in the UP fraction probably encourage this condensation. Cohen et al. (2004) measured PM 2.5 PAH as high as 1,500 ng/m 3 , reflecting the intensity of emissions from fires soon after the 9/11/2001 World Trade Center attack. Fine et al. (2004) sought contributions from cooking and wood burning by using levoglucosan and cholesterol as organic markers for these sources. Their concentrations were low, with levoglucosan undetected in many samples, although it did achieve a value as high as 50 ng/m 3 in one UP sample. Cholesterol was also low (< 0.2 ng/m 3 ), which indicated a negligible contribution from meat cooking to UP in Southern California during the study period.
Mass, elements, and ion concentrations were reported by Pakkanen et al. (2001a), while only carbon was reported by Viidanoja et al. (2002b). Therefore, mass closure was not achieved, although Pakkanen et al. (2001a) found organic acids that could indicate biogenic origins. Phares et al. (2003) found that amines constituted 2.4% of the UP mass in Houston, TX. Hasegawa et al. (2004) found that nearly all of the UP < 30 nm disappeared when heated to 250°C with a thermal denuder for roadway samples. This is consistent with UP being composed of semi-volatile OM that can evaporate from small particles and condense on larger ones, thereby enhancing particle growth (Zhang and Wexler, 2002).
EC was also present in most UP fractions, but not in the lowest size ranges as indicated by the thermal denuder results of Hasegawa et al. (2004) and Ishizaka and Adhikari (2003). This is consistent with the laboratory tests reported by Kittelson (1998) that showed most of the UP < 50 nm fraction evaporating after thermal denuding. This might happen with low-sulfur diesel fuels and after-engine soot removal owing to evaporated lubrication oil that condenses after emission and cooling to ambient temperatures (Zielinska et al., 2004;Vaaraslahti et al., 2005). This hypothesis is somewhat supported by the hopane measurements of Fine et al. (2004), as hopanes are believed to originate in engine oils.
Non-carbonaceous substances were found to dominate some of the samples, but this may depend on the measurement method. Single particle spectrometers, for example, draw materials into a vacuum, and some of the semi-volatile organic compounds may evaporate prior to detection. These spectrometers collect UP < 50 nm with < 100% efficiency, even with aerodynamic lens inlets, and it appears that much of the OC resides in this fraction. Phares et al. (2003) found that silicon oxides constitute 30% of UP, with potassium (K) constituting another 31% of UP. Carbon was only 16% of the UP in Houston, TX. This may be an artifact of the particle spectrometer measurement system, but it might also indicate that some very fine geological material penetrates to the UP fraction. Using an impactor, Chung et al. (2001) found large quantities of calcium (Ca) in the UP fraction at Bakersfield, CA. Ca is also an engine oil additive, and this may be the source. Kim et al. (2002) found that Fe was the largest metal component in Los Angeles, ranging from 10 to 130 ng/m 3 . This may be evidence of particle reentrainment from previous impactor stages, as mentioned above. Ma et al. (2004)  Sulfur (S) was an important, but minor, UP component in most of the studies. This is surprising, since sulfuric acid (H 2 SO 4 ) nucleation is considered to be a major formation mechanism of small particles. Once it begins, in fresh combustion exhaust or as the result of photochemical activity, it appears that organic or metal vapors rapidly condense and constitute most of the UP mass.

CONCLUSIONS AND KNOWLEDGE GAPS
UP chemical composition measurements are most commonly acquired with laboratory measurements from the lower stages of cascade impactors. The MOUDI is usually used for these measurements. XRF, PIXE, and INAA are applied for elements, with thermal/optical methods used for carbon. Ions are measured by IC, AC, or AAS. Single particle mass spectrometers are used for semi-quantitative chemical measurements.
Data are reported in different ways by different authors and are not completely comparable across studies. In the U.S., data have been reported for: the Los Angeles area, central (Fresno, Bakersfield), and northern California (Sacramento, Modesto); Houston, TX; Atlanta, GA; Baltimore, MD; and New York, NY. In Finland, data are available from Helsinki and Luukii, Espoo. In Japan, data are available from Nagoya, Mikuni, Kawasaki, Tsukuba, and Saitama. In Asia, data are available for Taiwan and Southern China. Most of these are urban areas, with some results from roadside, industrial, non-urban, coastal, and regional background sites.
Organic material is the most abundant portion of UP in most, but not all samples. Some have high elemental concentrations, especially from industrial sites. K, Ca, and Fe were found to be important elements in some samples. K originates from biomass burning, and Ca is used as an oil additive. Condensed Fe vapors are often found in industrial processes. Much of the UP < 50 nm appears to be semi-volatile, consistent with it being comprised of organic materials such as hopanes from engine oils or condensed secondary organic aerosol such as organic acids. PAH were abundant in the UP fraction in a few studies.
Sampling inlets and upper and lower size cuts for UP fractions are not consistent or welldefined. Methods are needed to establish UP fractions, then to translate the results from different sampling methods into a common particle size range. Differences in the analytical methods applied for the measurement of OC, EC, and organic compounds need to be better quantified and standardized. More comparison studies are needed among laboratories for particles specific to the UP fraction.
Standardized concentration units, reporting methods, and common data bases are needed to better compare UP composition measurements among sampling sites. These data bases would allow more efficient and consistent comparisons of results from different studies.