Comments on Gilmour et al., “Comparative Toxicity of Size-Fractionated Airborne Particulate Matter”

Gilmour et al. (2007) have reported an analysis of the toxicity of a water-dilute acid extract of airborne particles in four different geographical areas, South Bronx, NY (SB), Sterling Forest, NY (SF), Salt Lake City, UT (SL), and Seattle, WA (S). All but one (SF) of the locations is urban in character. The extracts were instilled in mouse lungs; the animals were subsequently euthanized, for certain toxicological tests.

The goal of the study was to differentiate the toxicological reactions to particles by size and chemical composition from widely different locations. Sampling was done for 4 wk at each location to obtain material for extraction and use in mouse exposures. (The calendar dates by location were not listed.).

The study provides insight into the apparent toxicity of water-weak nitric acid-soluble extracts from ambient particles, and the apparent source of these differences. For example, it is of interest to observe that coarse particles (PM_{10 − 2.5}), particles sampled between 2.5 and 10 μ m aerodynamic diameter, from SB appear to be more toxic than either of the other size fractions (ultrafine [PM_0.1—particles less than 0.1 μ m aerodynamic diameter] or fine [PM_2.5—particles less than 2.5 μ m aerodynamic diameter] sampled in the study) from SB, based on polymorphonuclear leukocyte (PMN) values and macrophage inhibitory protein (MIP)-2 in bronchoalveolar lavage (BAL) fluid and creatine kinase in plasma. These results were not seen at other study locations. Two other potentially insightful results were described as well: The BAL measurements from the SB ultrafine PM extracts at a higher dose level were significantly more harmful than at the lower dose level (results not found for the other size fractions in SB or at other sites). The coarse PM extracts from SL were associated with elevated protein in BAL. Thus, in SB the PM extract seemingly produced a more toxic response than other sites, with most of the toxicity indexes due to the coarse fraction.

Unfortunately, the significance of these findings is obscured by incomplete reporting of (a) the sampling and extraction methods, (b) the analytical results of the extracted PM chemical composition, and (c) the source apportionment reported for each site. The study adopted a nonstandard method for collecting particles compared with conventional Teflon and quartz filter-based sampling used in U.S. air monitoring. Their method used a high-volume staged impactor-filter sampler, with collection of coarse and fine particles in a polyurethane foam, and ultrafine particles on a polypropylene filter. The particle size cut points of the sampler are not reported, and no reference is given to the comparability of this sampling method with the standard filter-based method currently used the national sampling networks.

The brief description of the analytical chemistry using water-dilute acid extraction of the foam and the filters glosses over the fact that this method will not dissolve much of the refractory portion of the collected PM, including many metal oxides, and tarry black or elemental carbon. The authors state that their recovery is about 80% except for iron, based on comparison with a water-dilute acid extract of a reference standard. This sidesteps this issue of the insoluble material that would make up the gravimetric mass concentration observed using direct filter analysis. Thus, it is not possible to compare the results of the toxicity experiments with complete filter-based chemical composition and gravimetric mass concentration as reported in contemporary air monitoring data.

The chemical composition of the aggregated samples of different particle size from each of the sites is not tabulated with a material balance for the reader to inspect, and only a small portion of the complete composition data expected from the instrumental analysis of the extracts is discussed in the study. Additionally, a comparison of particle composition reported at the same locations by the conventional filter-based air monitoring chemistry is not included. This prevents the reader from evaluating the similarities and differences in the PM extracts, which could lead to enhanced interpretation of the toxicity data.

The identification of PM source contributions using chemical mass balance (CMB) calculations modeling also is only summarized briefly (CMB only was applied to fine particles, because the coarse particle analysis was “unreliable” [reasons not stated]), and source profiles are “unavailable” for ultrafine particles. The CMB methodology is referred to an unpublished paper for more details (Duvall et al., 2007). Unfortunately, this manuscript as currently available from its authors (Fall of 2007) is no more explanatory of this CMB application than Gilmour et al.

Application of CMB requires that source profiles are chosen for each of the sites, based on chemical composition and knowledge of sources from emission inventories. This is not discussed; nor is it clear if the profiles were used uniformly for all sites. It is well known that each of the locations used has pollution from different local and regional sources, so that the preselection of relevant sources is important for the credibility of the results.

The application of CMB usually makes use of at least one tracer compound in the data fitting process, but the authors are unclear about which fitting species relate to which sources. Their fitting species are stated to be elemental carbon (EC; primarily mobile sources), potassium (primarily wood combustion), vanadium (residual oil), copper (mainly brake linings?), strontium (brake linings?), barium (lubricating oil?), and lead (street dust?). (Zn and Fe are referred to in the article, but evidently were not considered fitting species.) Strontium is an unusual element as a tracer; usually other elements are used, for example, Se for coal combustion (assuming no large local sources of residual oil), Si or Ca for soil dust, etc.

Secondary sulfate primarily from atmospheric oxidation of sulfur dioxide also is reported, and identified arbitrarily with coal combustion. It is well known that CMB cannot apportion secondary species of PM, so that a primary species indicator is required, which evidently was not used. Further, it is unclear from the description of analytical chemistry what “EC” actually represents in a water-dilute acid extract, since the extraction method will obtain water-soluble organic carbon, and perhaps some black-carbon-correlated soluble material. But extraction of the carbonaceous material is not quantitative in this dissolution method. Refractory EC will remain insoluble, and it is unclear whether this fraction is actually obtained quantitatively in the method.

The assignment of secondary sulfate to regional coal combustion in the eastern United States (affecting SB and SF) is arbitrary since there are local or regional sources of sulfate, both primary and secondary, that do not derive from coal combustion (e.g., diesel fuel; residual oil used in power plants and commercial or residential buildings in some Eastern cities; and shipping). Coal combustion as a sole source of sulfate is problematic for S since only one coal plant that we know of exists nearby the sampling location (∼ 60 miles south), while a number of sulfur-emitting paper mills, vehicles, and shipping are also present the Puget Sound area. A local coal combustion source of sulfate exists in SL, along with a nearby copper smelter and a steel mill, both sulfur sources, aside from transportation. If a tracer like Se is not used in the CMB, then the only means for establishing a regional coal source (other than inference from factor analysis) is the application of air mass trajectory analysis, which apparently was not done.

We also note that there appears to be a difference in reporting of “sulfate from power plants” cited in the abstract (31% of extracted [?] SB mass vs. 35% as stated elsewhere). The abstract also implies the importance of power plant emissions in SB from the 31–35% sulfate thought to originate with these plants. It is difficult to reconcile this implication with two other findings. First, no effects were found from the PM extracts at the nearby SF location, which was reported to have greater sulfate content than SB (Figure 7), both of which are presumably affected by the same regional coal plant sulfate. Second, the SB effects were found almost entirely from the extracts of the coarse PM fraction (without source attribution), whereas the sulfate from sulfur dioxide oxidation of coal plant emissions is known to be mainly in the fine and ultrafine fractions.

Like the reporting of the details of the chemical composition of the various particle samples, the fine particle source apportionment tabulation for the different sites is not provided; nor is a sense of a material balance discussed to insure that the composition extract mass is accounted for the sources chosen. This information appears to be essential to support the interpretation of the PM chemistry as asserted in the article. The source fractions reported in the article give only an incomplete picture of source contributions. Sulfate, for example, at SB is about 31–35% as noted earlier, and is 45% at SF. What are the apparent sources for the remainder of the material?

We believe that it is potentially “dangerous” to exclude minor sources of PM in interpreting toxicity or health hazard. Certainly the establishment of the hazardous air pollutants (HAPs) component of the Clean Air Act reflects this concern. Gilmour et al. (2007) reflect that some sources such as residual oil combustion contributed less than 5% to extracted PM mass, implying that such sources are unimportant for toxicity, and perhaps small source contributions were excluded from consideration. Yet Maciejczyk and Chen (2005) found in an earlier SF study, for example, that only the 2% of fine particle mass associated with residual oil combustion was identified with inflammatory processes in laboratory animals.

Gilmour et al. state that an interleukin (IL)-6 indicator was measured, but no IL-6 associations are noted. Even if any such associations were insignificant, reporting such findings is important for comparison with other studies. For example, Mutlu et al. (2007) instilled PM₁₀ from Dusselfdorf, Germany, air sampling in mouse lungs and found that the increased clotting was significantly mediated by IL-6 increases. If instillation of extracts of coarse and fine PM from sampling at the four locations in the United States, including one highly polluted area (SB), failed to increase IL-6, this suggests important differences between the response to exposure of German and U.S. PM samples. This difference may be rationalized considering the details of obtaining the PM chemical data, or possibly the laboratory methods for animal exposure.

An apparent toxicity of PM in SB seems to be identified with elevated (soluble) Fe and Zn concentrations in coarse particle fraction relative to other locations. Zn evidently is elevated even more in the ultrafine particle fraction in SB, in contrast with Fe, which appears to be more elevated in the coarse fraction. Elevated concentrations of metals such as Fe and Zn likely reflect different sources, perhaps brake lining debris or road dust for both, but Fe also may derive from soil dust or industrial processing. Zn in the ultrafine fraction, on the other hand, is likely to be associated with additives for diesel fuel or lubricant. These metals also may be indirect indicators for other unmeasured species affecting toxicity. For example, Maciejczyk et al. (2004) found that polycyclic aromatic hydrocarbon (PAH) concentrations (carcinogen indexes) were highly elevated in the SB air, relative to other areas in and near New York City. (Potentially soluble) PAH species apparently were not investigated in the extracts reported for the Gilmour et al. study, although they are known to be important toxic species present in urban air.

In the last paragraph of their article, the authors opine that “either this bioassay approach and its form of analysis has limited sensitivity to detect size and location-driven differences in biological effects, or that the … responses reflect those of PM mass-based increases in health effects with little influence of the noted differences in chemistry.” This conclusion may be overly pessimistic. In other recent work, Seagrave et al. (2006) found sharp differences in 10 separate measures of toxicity among PM sources and constituents. In that article, the sources found to be linked with toxicity were related primarily to emissions from gasoline or diesel combustion. Thus it is possible either that there wasn't measurement for as many relevant chemical species for comparison, or the methods of source profiling used by Gilmour et al. may not have been sufficiently robust, or both, notwithstanding the notable differences reported.

We agree with Gilmour et al. that in this relatively new and challenging field of PM-health effects studies, more work is needed, “to quantify the relative potency of source materials within ambient PM.” Our caveat of this conclusion proposes that the work be done with closer integration with the atmospheric chemists, emissions experts, and other workers to promote precise source identification and intercomparisons of results with different bioassay methods. Creation and use of source apportionment models, in order to produce unique source identification, is very difficult to do unambiguously and accurately (Grahame & Hidy, 2007). In particular, source apportionment models could better utilize details of ambient PM chemistry and emission characteristics, while also using monitoring which better expresses exposure to emissions. Further, some emissions are produced by more than one source category, and steps should be taken to identify carefully which source(s) of such an emission are significantly associated with health endpoints.