In this study, we examined the in vitro pro-inflammatory and cytotoxic effects of PM of different sizes collected at different sites in the Netherlands. All samples, with exception of urban background and stop & go traffic PM, caused a concentration-dependent decrease in MTT-reduction activity and increase in production of TNF-α, IL-6 and MIP-2. The pro-inflammatory response was highest for fine PM collected at traffic locations whereas the MTT-reduction activity was most influenced by the underground train station site. Furthermore, MTT-reduction activity was negatively associated with PM oxidative potential for all PM size fractions. For pro-inflammatory response, a significant positive association with DTT consumption was only observed when only the outdoor sites were considered.
Air pollution characteristics
Per site, the sampling day with highest PM concentration (μg/ml) was chosen for in vitro toxicity testing since there had to be sufficient sample for testing the full concentration range. Our results should therefore be interpreted with caution as these single day samples may not be representative for the sites on a yearly basis. One sample per site, collected at different days, rather reflects differences between sampling days. This implicates that the observed differences in in vitro toxicity between our PM samples can be related to differences in PM composition, but cannot be contributed to site specific characteristics (for example, stop & go traffic emissions are less toxic than truck traffic emissions).
In figure S2 (Additional file 8) the air pollution characteristics of the selected sampling days are compared to site averages of the full sampling campaign. The most striking difference between the site averages and selected samples was observed for the stop & go traffic location. At the selected sampling day for this site, median particle number concentration was substantially higher (64,700 particles/cm3 compared to the site average of 53,000 particles/cm3). During this sampling day, PM10 concentrations were similarly elevated across the entire country, suggesting a regional pollution episode as the cause (data obtained from the Dutch National Air Quality Monitoring Network, not shown). Nonetheless, urban concentrations were well in excess of rural levels, and PM samples from this day represent local emissions imposed upon a substantial regional background.
Chemical composition and in vitro toxicity
The in vitro toxicity differed between PM of different sizes and collection sites in the Netherlands. In all PM size ranges, underground PM clearly caused the largest decrease on MTT-reduction activity expressed per μg PM. The effect on MTT-reduction activity might be linked to the high metal content, which was substantially higher compared to the other sites. At the underground train site, the most abundant metal species were iron and copper. In coarse PM, the amount of iron per μg PM was 7 to 50 times higher and the amount of copper was 12 to 385 higher compared to the other sites (Table 2). Similar contrasts were observed for fine PM: 11 to 113 times more iron and 24-315 times more copper per μg PM (Table 3). Several other studies showed that particulate samples with high metal content evoked inflammatory responses as well as cytotoxic responses in vitro and in vivo [17, 30–32]. For example, Gerlofs-Nijland et al.  compared the effects of PM samples collected in two cities that different in metal content in compromised rats. The results showed that both samples triggered inflammatory responses, which could be partially inhibited by addition metal chelator DTPA (diethylene triamine pentaacetic acid), but still the sample richer in metals revealed a greater enhancement in inflammatory responses. In addition, a recent in vitro study by Perrone et al.  revealed that the biological responses of a human lung epithelial cell line to urban fine PM were related to several (metal) elements: cell viability reduction was related with arsenic, zinc, chromium, copper and manganese; DNA damage with other elements like iron, chromium and cadmium (and possibly PAHs), and IL-8 production to arsenic, zinc and inorganic ions like sulfate.
In the fine size range, the pro-inflammatory activities (TNF-α and MIP-2 production) were highest for the truck traffic and continuous traffic site. This observation is in line with findings of Seagrave et al. , who exposed rats to fine PM collected at sites with different contributing PM sources. Seagrave et al. observed that the most toxic (/pro-inflammatory) samples were from the sites with the largest contributions of diesel and gasoline emissions. Our truck traffic and continuous traffic sites were similar in that they had the highest EC and OC content, which is characteristic (though not exclusive) for traffic emissions sources [34, 35]. It has previously been reported that a high EC and OC content is associated with pro-inflammatory activity [36, 37]. In most studies, the pro-inflammatory response was driven by compounds adsorbed at the carbon surface, such as transition metals or PAHs, rather than by EC or OC themselves [38–41].
Importantly, whereas fine PM from the continuous traffic and truck traffic sites did show pro-inflammatory activity, the sample from the other traffic related site, stop & go traffic, did not. In addition, this sample had a relatively low EC and OC content. This further strengthens the idea that the selected stop & go traffic sampling day was strongly influenced by other emissions sources, and thus did not represent average conditions at this site. In contrast, qUF PM from this site did induce a modest pro-inflammatory response (TNF-α and MIP-2). Since particles are known to grow in diameter as the distance from their source increases , the majority of qUF collected at this site therefore may still have originated from stop & go traffic.
Interestingly, the PM samples from the underground site caused the largest decrease in MTT-reduction activity, but not the highest pro-inflammatory activity. Similar effects were observed in another in vitro study by Karlsson et al. . They compared the genotoxicity and the ability to induce inflammatory mediators of particles from different sources (wood combustion, tire-road wear, an urban street and a subway station). Karlsson et al. found that particles from subway were most genotoxic of all particles tested. However, particles from street level were the most potent to induce inflammatory cytokines. Furthermore, in vitro differences in toxicity versus pro-inflammatory activity were also observed by Gualtieri et al. . They observed that urban PM samples collected in winter were more cytotoxic than summer samples, whereas the summer PM samples exhibited a higher pro-inflammatory potential. A possible explanation could be that the observed toxicity led to reduced responsiveness to pro-inflammatory triggers. Another possible explanation might be that different PM characteristics are responsible for inducing cytotoxicity and causing pro-inflammatory effects [6, 44–46]. In the study of Jalava et al.  mouse RAW 264.7 macrophages were exposed to coarse, fine and qUF PM from six European urban background sites. Their data showed considerable heterogeneity between the PM samples with regard to both pro-inflammatory activity as well as cytotoxic activity, but there was no statically significant correlation between these parameters. In addition, Guastadisegni et al.  showed that the metal chelator DTPA inhibited arachidonic acid release in RAW 264.7 cells exposed to traffic-related PM, whereas recombinant endotoxin-neutralizing protein partially inhibited TNF-α production, demonstrating that different PM components triggered pro-inflammatory responses through separate pathways.
In the fine and qUF size range, there were a few PM samples causing a small, non-linear decrease in MTT-reduction activity (farm fine, urban background fine and qUF PM). The absence of a linear concentration-dependent effect does not necessarily imply that there is no effect at all. A possible explanation could be that these PM samples contain a compound with high "potency", but low "efficacy" to influence MTT-reduction activity. In that case, even the lowest concentration used for the in vitro exposures would be within the maximum effect range. Unfortunately, there is no chemical composition data available for the sampling days at the farm and urban background site. Therefore, it is not possible to relate these finding to specific components of ambient PM.
Oxidative potential and in vitro toxicity
In all PM size ranges, the underground site had very high oxidative potential (Additional file 2, table s1). This might be very well caused by high metal levels, since those were much higher in the underground samples then for the other sites. In line with these findings, Ntziachristos et al.  found high correlations of transition metals and DTT consumption for PM2.5 and PM0.15 samples collected at different sites in California, USA. Furthermore, metals are known to catalyze the oxidation of DTT as measured by the DTT assay [48, 49]. Li et al.  observed a significant correlation (r2 0.98) between DTT consumption and PAH content of coarse and fine PM. Conversely, when we performed the same (statistical) analysis on our data, we found no significant association between PAH and DTT consumption (correlation < 0.25, p-value > 0.1; data not shown). However, in our study, PM chemical composition (including PAH) was determined on filter extracts, whereas (concentrated) impinger samples were used for the in vitro studies. Analysis of the association between PAH and DTT was thus not performed on the same type of samples. Also, the number of samples available for this analysis was limited (n = 11, Table 2 Table 3 and Additional file 2, table s1), since we have examined only one sampling day per site. An in-depth analysis on associations between oxidative potential and PM composition (including all sampling days for each site) will be published elsewhere (Godri et al., manuscript in preparation).
A more general approach is to correlate DTT consumption to the observed in vitro toxicity. In our study, PM samples that had high DTT consumption caused a decrease in MTT-reduction activity. This association was statistically significant even after excluding the underground sample (highest data point for both decreases in MTT-reduction activity as well as oxidative potential). In addition, samples with a high DTT consumption also had a high pro-inflammatory activity except for the underground sample. As mentioned above, this might be attributable to the fact that different elements may activate different cellular responses or that the observed toxicity hampers the cellular responsiveness.
To date, several other in vitro studies have been published that used the DTT assay to determine the oxidative potential of (ambient) PM [21, 50–52]. Most of these studies focused on the relationship between chemical composition and DTT-activity of PM. So far, limited research is performed on the association between the in vitro DTT-activity and the in vitro or in vivo toxicity of PM. For example, Li et al.  demonstrated that cellular heme oxygenase 1 (HO-1, an enzyme responsive to oxidative stress) expression was correlated to DTT-activity. However, to our best knowledge, there is currently no clear relationship with adverse health effects in humans.
PM size fraction related effects
In the present study, PM size fraction did not seem the principle determinant of MTT-reduction activity. Although there were significant differences between the coarse, fine and qUF fractions within three sites (harbor, truck traffic and stop & go traffic), these were not pointing towards one direction (e.g. decrease in MTT-reduction activity). Considering all sites together, there were no statistically significant differences between the size fractions (Additional file 4, table s3). The observation that PM size fraction is not the most important PM characteristic in determining PM effect on MTT-reduction activity has been previously reported [6, 53].
In contrast, the pro-inflammatory activity of the fine size fraction was significantly higher than the qUF size fraction. Numerous studies have attempted to link PM size to pro-inflammatory activity [5, 54–56]. For example, Hetland et al.  showed that coarse PM was more potent at inducing induce pro-inflammatory cytokines than fine PM in vitro, whereas de Haar et al.  showed that ultrafine but not fine particulate matter causes airway inflammation in vivo. However, several recent (in vitro) studies have shown that both PM size as well as PM composition contributed to adverse effects of PM exposure [57, 58]. Ramgolam et al. exposed human airway epithelial cells to size-fractionated PM collected at an urban background site in Paris. They used isomass exposures (same particle concentrations for each size fractions) as well as isovolume exposures (same volume of particles in suspension; to respect the particle proportions observed in ambient air) and characterized the pro-inflammatory response by measuring the pro-inflammatory cytokine Granulocyte Macrophage-Colony Stimulating Factor (GM-CSF). Ramgolam et al. found that the pro-inflammatory response decreased with aerosol size when cells were exposed to isomass of particle suspension. However, when cells were exposed to isovolume of particle suspensions, the GM-CSF release was maximal with the fine fraction. The authors concluded that this could be related to the chemical composition of each size fraction. In another in vitro study, by Osornio-Varga et al , urban PM10 and PM2.5 was investigated in three different biological systems. For most, but not all, parameters they found a significant effect of PM size: hemolysis and induction of DNA degradation were pre-dominantly induced by PM10 whereas the inhibition of cell proliferation was significantly stronger by PM2.5 (not size dependent: TNF-α and IL-6 release). In addition, results of PM elemental composition principal component analysis were used in associating cellular effects. This revealed that different elements triggered specific biological parameters. Taken together, it is most likely that the combination of PM characteristics, such as source-dependent PM composition and oxidative potential, determines the toxicity rather than a single PM characteristic as PM size. In this study, the observed differences in pro-inflammatory response after in vitro exposures of fine and qUF PM was likely caused by the differences in chemical composition rather than the particle size per se.
Limitations of this study
The PM samples used for in vitro toxicity testing were collected using a liquid impinger system (VACES), whereas ambient PM is usually collected on filters. Impingers have the advantage over filter-based systems that PM does not have to be extracted from the filter before use, and might therefore better represent ambient particle characteristics. However, HVS filter extracts were used to determine PM chemical composition. This hampers the interpretation of the chemical composition data in relation to the observed in vitro toxicity.
In addition, despite the enrichment of the aerosols using a VACES, insufficient mass was collected to perform toxicity studies at the desired concentrations and PM samples thus had to be concentrated in the laboratory. We tried to avoid changing particle properties by using a mild concentration procedure. Therefore, we concentrated the samples by evaporating the water at 25°C under a nitrogen flow to avoid biological contamination and avoid heating samples to a high temperature or freeze-drying. Nevertheless, the endotoxin content of six concentrated samples turned out to be extremely high. A possible explanation could be that biological material present in the sample multiplied during the concentration procedure. Additional measurements in the corresponding un-concentrated samples showed that the endotoxin levels were indeed 5.5-57 times lower before concentration (Additional file 3, table s2). Consequently, the endotoxin levels in the concentrated samples as used for the in vitro exposures most likely do not reflect ambient endotoxin levels.
However, although the endotoxin levels in our concentrated impinger samples may not reflect ambient endotoxin levels, the biological material most likely originates from the ambient PM itself, and not from laboratory contamination. Firstly, because endotoxin in ambient PM is associated with the coarse fraction  and five out of the six outliers in the present study were coarse samples. The sixth sample containing high endotoxin levels was a fine sample from the farm site were endotoxin levels were expected to be high. Second, the endotoxin levels in our un-concentrated samples are within the same range as measured by others who used VACES to collected ambient PM [9, 29]. For example, the endotoxin concentration PM collected in Mexico City ranged from 16 to 895 EU per mg PM . Finally, most importantly, because we have also measured endotoxin in PM10 filter extracts. This data is available for all RAPTES sampling days and is recently published by Strak et al. . The average PM10 endotoxin concentration at our sampling sites (Strak et al., Table 2: farm 26.15 EU/m3, other sites ≤ 1.71 EU/m3) are in the same order of magnitude as observed by others at outdoor sites [12, 60, 61]. For future in vitro or in vivo exposure studies with either impinger or filter extracts, we recommend to treat PM samples in such way that endotoxin effects can be either avoided or assessed in a controlled manner. For example, to irradiate samples before use or to combine PM exposures with an endotoxin blocker/inhibitor such as polymyxin or recombinant endotoxin-neutralizing protein (rENP).
Another point of consideration is that all the comparisons between toxic activities of the PM samples used in this study have been made on mass basis. This does not take into account the exposure levels at each site. For example, PM from the steel work site had relatively high pro-inflammatory activity, but the ambient PM concentration at this site was relatively low compared to the other sites. In other words, although the hazard for induction of inflammation by the steelworks site would be large, the risk would be small since the PM exposure at this site is low. In addition, the deposited dose in the lung and respiratory tract depends on the aerodynamic size of PM, but this is not included in this in vitro study. For risk management, not only the potency but also the exposure and dose have to be taken into consideration.