Source Apportionment of the Lung Dose of Ambient Submicrometre Particulate Matter

Urban ambient aerosols have been of much concern in recent decades due to their effects upon both atmospheric processes and human health. This study aimed to apportion the sources of submicron particles measured at an urban background area in London and to identify which sources are most responsible for particles deposited in the human lung. Particle number size distributions (PNSD) measured by a Scanning Mobility Particle Sizer (TSI, USA), covering the size range of 16.5–604 nm at the London North Kensington background sampling site during 2012 were used in a Positive Matrix Factorization (PMF) model to apportion to six dominant sources of particles. These included local traffic emissions (26.6% by number), aged traffic emissions (29.9%), urban accumulation mode (28.3%), nucleation (6.5%), inorganic secondary aerosol (1.7%) and mixed secondary aerosol (6.9%). Based on the ICRP model, the total deposition efficiencies of submicron particles for the aforementioned sources in the human respiratory tract were 0.57, 0.41, 0.24, 0.62, 0.24 and 0.24, respectively. In terms of source apportionment of particles deposited in the lung, traffic emissions represent the main source of particles deposited by number in both the regional and total lung, accounting for 59 to 71% of total deposited particles, followed by regional accumulation mode (17%) and nucleation (10%) particles. Secondary aerosols only account for 5.1% of total deposited particles by number, but they represent the main source of particles deposited in the lung expressed as surface area (44.6%) and volume (72.3%) of total deposition. The urban accumulation mode contributes 27.3% and 17.3% of total deposited particles by surface area and volume, while traffic emissions contribute 26.6% and 9.7%, respectively. Nucleation contributes only 1.6% and 0.7% of total deposited particles by surface area and volume.


INTRODUCTION
Particles in urban ambient air are released from various anthropogenic activities such as combustion of fuels and from secondary sources such as atmospheric nucleation (Ogulei et al., 2007).Exposure to ambient aerosols is consistently associated in numerous scientific studies with adverse health effects (Pope and Dockery, 2006).Very small particles such as ultrafine particles (Dp < 100 nm) are able to penetrate deep into the respiratory tract (e.g., reaching the pulmonary epithelium), causing serious health problems such as respiratory morbidity and mortality (Donaldson et al., † Also at: Department of Environmental Sciences/Center of Excellence in Environmental Studies, King Abdulaziz University, PO Box 80203, Jeddah, 21589, Saudi Arabia 1998).Moreover, different types of sources generate particles with different size distributions, chemical composition and concentration characteristics (Lighty et al., 2000).This results in particles from different sources behaving differently during the process of inhalation, showing different penetration through the respiratory tract and depositing with different efficiency in different regions of the lungs.For example, Löndahl et al. (2009) estimated the total lung deposition efficiency of particles by number emitted from traffic exhaust was 0.68 that was three times higher than those of particles released from biomass burning (0.22).Therefore a detailed identification of the most relevant sources of atmospheric particles and the association between a particle source and lung deposition could play a vital role not only for risk assessment of air pollution in epidemiological studies, but also for policymakers to introduce optimal legislation for air quality control.
In order to identify and apportion the most relevant sources of ambient particles, applying Positive Matrix Factorization (PMF) to particle number size distributions (PNSD) has become a widely used tool in recent years (Viana et al., 2008).PMF considers each size bin in the PNSD dataset as an input variable.Additional variables such as ion species, heavy metals, gaseous pollutants, meteorological parameters and traffic data can be very valuable to separate and identify the sources of particles as demonstrated in previous studies (Ogulei et al., 2006;Thimmaiah et al., 2009;Harrison et al., 2011).The advantage of this method is that it can identify specific sources of very small particles, such as nucleation.It can also separate sources of particles such as brake and tyre-wear, which are generally difficult to separate using PMF on particle composition data only.In addition to source identification and apportionment of concentrations measured in ambient air, and considering the different deposition efficiencies of particles generated from different sources, there is still the need for evaluation of which sources are most responsible for particles deposited in the human respiratory tract.
This study aimed to estimate lung deposition of particles generated from specific sources contributing to the particle size distribution in North Kensington, an urban background area of London, United Kingdom.In order to do so, we firstly apportioned the sources of submicron particles in North Kensington using PMF.Subsequently, we estimated the lung deposition efficiency of particles generated from each specific source by applying the ICRP model on the PMF factor profiles.Finally, we investigated which source may be most relevant to health outcomes by comparing the results of the source apportionment of particles deposited in the human respiratory tract.

Site Description and Data Measurement
Particle number size distributions and concentrations of other air pollutants were measured at North Kensington in London, UK.This monitoring station, which is located in the grounds of Sion Manning School, is representative of the typical urban background area of London.The site location and pollution climate have been described by Bigi and Harrison (2010).
Particle number size distributions (PNSDs) were measured during 2012 by a Scanning Mobility Particle Sizer (SMPS) operated by the Department for Environment, Food and Rural Affairs (DEFRA) as part of the UK national network.) and meteorological data (wind speed and direction) were extracted from the DEFRA website (see http://uk-air.defra.gov.uk/ for more detailed information).

Data Handling, PMF and ICRP Models Data Handling
Data statistical analysis, polar plots and concentration weighted trajectories were performed in the R program (Version 3.1.5)using the "Open-air" package developed by Carslaw and Ropkins (2012).Missing data was linearly interpolated between the values from the nearest size bins.Air pollutant concentrations, SMPS data and meteorological data were averaged on an hourly basis.

PMF Models
In this study, a profile of 6,098 hourly PNSDs comprising 51 size bins ranging from 16.6 to 604 nm were input in the PMF US EPA model version 3.Each size bin in the PNSD was considered as an input variable.Since uncertainties were not provided by the experimental instruments, these were estimated based on an empirical method introduced by Ogulei et al. (2006Ogulei et al. ( , 2007)).The method for determining the correct number of factors in the PMF analysis has been described by Lee et al. (1999) and Yakovleva et al. (1999).PMF factors were interpreted based on (1) modal structure of number and volume size distributions, (2) the diurnal pattern of factor contribution, (3) the contribution of each factor to total number and volume, (4) the relationship with auxiliary information such as gaseous and chemical composition data and (5) source directionality by local wind trajectories and polar plots (Ogulei et al., 2007).The details of the PMF method and its use in our study are provided in the Supplemental Material.

ICRP Model
The International Commission on Radiological Protection (ICRP) model was developed to predict the deposition of particles with a wide size range -from 1nm to 100 µm -in the respiratory system consisting of three main regions: extrathoracic (ET), trachea-bronchial (TB) and pulmonary/alveolar (AL) (ICRP, 1994).In our calculations, we applied the ICRP model with Hind's parameterization of the grand average particle deposition to estimate the total and regional deposition efficiency (DE) due to inhalation by males and females at three exercise levels (Hinds, 1999).
The DEs calculated by the ICRP are for spherical hydrophobic particles.However, hygroscopic particles are typically dominant in the ambient environment (Asgharian, 2004;Montoya et al., 2004;Löndahl et al., 2009).To address this problem, we adjusted the ICRP curve for both hydrophobic and hygroscopic particles based on an assumption that particles generated from combustion sources (i.e., traffic emission or biomass burning) are nearly hydrophobic and inorganic/organic secondary aerosols are a mixture of less hygroscopic and more hygroscopic particles (Weingartner et al., 1997;Cruz and Pandis, 2000;Väkevä et al., 2002;Massling et al., 2005;Varutbangkul et al., 2006;Löndahl et al., 2009;Tritscher et al., 2011).The details of this method are given in Vu et al. (2015).The regional and total lung deposition efficiencies of particles from different sources identified by PMF was estimated based on an application of the ICRP model to each source's particle size distribution as adjusted by its expected hygroscopicity.

Overview of Data
The average total number concentration was 5.6 ± 3.3 × 10 3 particle cm -3 , of which 1.2 ± 0.9 × 10 3 , 3.3 ± 2.1 × 10 3 and 1.0 ± 0.9 × 10 3 particle cm -3 was accounted for the nucleation mode, Aitken mode, and accumulation mode, respectively.The majority of particle number was in the ultrafine region (Dp ≤ 100 nm) which represented 81.4% of total number concentration, whereas accumulation mode particles only accounted for 18.6% of total particle number, but presented 90.5% of total particle volume.
As shown in Fig. 1(a), the particle number size distribution shows a peak number mode at 36.6 nm and a peak volume mode at 294.3 nm.The diurnal pattern (Fig. 1(b)) shows two distinct peaks coinciding with the traffic rush hours, and the lowest concentration was found during the early morning.

PMF Results
The profile of six resolved factors obtained by PMF is shown in Table 1 and Fig. 2. The first factor has a peak by number at a diameter of approximate 29 nm (Fig. 2).This factor represents 26.6% of the number concentration (Table 1), but has only a small contribution to the volume concentration (3%).The modality of this factor (Fig. 2) is similar to the shape of the nucleation mode of the particle number size distribution from road traffic emissions (Vogt et al., 2003;Zhang et al., 2004).These particles could be emitted directly from gasoline cars or from the growth of nucleation particles released from diesel cars (Ristovski et al., 2006;Wehner et al., 2009).This factor has a weak correlation with NO 2 (r 2 = 0.43) as shown in Table 2.In addition, the strong diurnal pattern shows two dominant peaks corresponding to morning and evening rush hours, suggesting that this factor represents the exhaust nucleation particle mode (Harrison et al., 2011).The polar plot shows no dominant direction, therefore this factor can be attributed to local traffic emissions.
The second factor shows the main number particle distribution in the size range between 20 nm and 100 nm with the peak number diameter around 52 nm corresponding to solid carbonaceous particles from diesel exhaust (Shi et al., 2000).It contributes 29.9% of the total number concentration and only 8.6% of total volume concentration.The diurnal variation shows an obvious morning and evening peak.In addition, this factor contribution has strong correlations with other air pollutants (CO, NO, NO 2 and BC) and a moderate correlation with PM 10 (see Table 2).Hence, this factor mainly comprises traffic emission sources, but unlike factor 1, the polar plot shows that this source dominantly comes from easterly and south-easterly directions.Therefore, it could be aged traffic emissions (Zhou et al., 2004), transported from central London.There was an increase of the factor contribution in winter and on weekend days, and therefore this factor may also contain domestic combustion sources such as biomass burning or cooking.4 Note: % N: percentage of particle by number; % V: percentage of particle by volume; N-peak: peak number mode; Vpeak: peak volume mode; Nu: Nucleation mode (Dp < 30 nm); Ait: Aitken mode (30 nm < Dp < 100 nm); Acc: Accumulation mode (100 nm < Dp < 600 nm)..280** Note: ** Correlation is significant at the p value < 0.01; * Correlation is significant at the p value < 0.05.
The particle number size distribution of factor 3 is dominated by nucleation size range particles with 87% of the total number concentration being within this range (see footnote to Table 1 for definitions on size ranges).This factor possesses small particle number concentrations (6.5%) and small volume concentration (1%).The diurnal pattern shows a major peak around noon that is attributed to regional photochemical nucleation events.The seasonal variation shows this factor is mostly found from June to September, and the polar plot shows marked directionality from the western sector as shown in Fig. S5.This could suggest that factor 3 is aged nucleation particles with an origin in the westerly sector.This is supported by a study on new particle formation at a rural site in southeast England which found that nucleation events occur predominantly with westerly maritime air masses, high wind speed, and low PM 2.5 and NO x concentrations (Charron et al., 2007(Charron et al., , 2008)).There is a smaller contribution of factor 3 from the south-easterly sector which is less easily explained.
The fourth factor accounts for only around 2% of total particle number, but accounts for nearly 25% of total particle volume.The size distribution of this factor shows a bimodal distribution with two peaks at 86 nm and 216 nm and more than 95% of total particle number in the Aitken and accumulation ranges while 99% of total particle volume is within the accumulation mode.Factor 4 is strongly correlated with PM 2.5 (r 2 = 0.86), PM 10 (r 2 = 0.85), NH 4 + (r 2 = 0.80), NO 3 -(r 2 = 0.76) and SO 4 2-(r 2 = 0.62), and has an inverse correlation with ozone.The obvious directionality shows that this factor originates from the east and northeast sectors, and the diurnal variation shows a significant decrease in the afternoon.This behaviour suggests that this factor represents regional inorganic secondary aerosol.This is consistent with results from cluster analysis conducted by Beddows et al. (2009), who associated regional air pollutant transport with clusters of concentrations showing a correlation with PM 10 , an inverse correlation with ozone and a decreasing concentration in the afternoon.
Factor 5 shows a main accumulation mode at 205 nm and a smaller peak at 27 nm.This factor has a similar diurnal pattern to factor 4. The main directionality is from the northeast, southeast and easterly sectors.This factor has also high and moderate correlations with other air pollutants (PM 2.5 (r 2 = 0.79), PM 10 (r 2 = 0.75), NH 4 + (r 2 = 0.63), NO 3 -(r 2 = 0.55) and SO 4 2-(r 2 = 0.51), although weaker correlations than those of factor 4. In addition, this factor shows a moderate correlation with total organic carbon (r 2 = 0.63).Therefore, this factor also suggests an association with secondary aerosol, but to a mixture of inorganic and organic aerosols.Fig. 3 shows the major directionality of factor 4 and factor 5 with PM 2.5 , PM 10 , NH 4 and SO 4 , suggesting that secondary aerosol could be originated from mainland Europe due to long-range transport.This is consistent with a study of receptor modelling of secondary particulate matter at UK sites (Charron et al., 2013).
The last factor (Factor 6) shows uni-modal number/volume size distribution with a peak around 93 nm by number and 165 nm by volume.This particle size distribution was attributed to combustion sources which mainly contain carbonaceous particles (Hildemann et al., 1991;Venkataraman et al., 1994).In addition, this factor accounts for 28.3% of the total number concentration and 29.7% of the total volume concentration.The diurnal pattern shows a peak in the morning rush hours and a peak later in the evening rush hours.This factor correlates with the second factor which has been identified as traffic emissions (r 2 = 0.6).Moreover, this factor has strong correlations with NO, NO 2 , NO x , and CO (Table 2) suggesting that it could also represent the solid particle mode from traffic emissions (Dall'Osto et al., 2012;Harrison et al., 2011).However, an increase of this factor contribution on weekend days and in the cold season and its strong correlations with both black and organic carbon shows that this factor could also be emitted by other combustion sources such as power stations or biomass burning which also show a peak around 100 nm (Janhäll et al., 2010;Wang et al., 2013).The polar plot shows the main directionality of this factor to be in the east and southeast sectors.Hence, this factor could be regional background accumulation mode.By using cluster analysis, Beddows et al. (2009) also found a large accumulation mode at 100 nm at the British Telecom Tower, which is also classified as in an urban background area of London.A recent study by Beddows et al. (2015) found there were four sources of particle number including traffic emissions, urban background, nucleation and secondary aerosols based on application of PMF to a two-year data set of particle number size distribution at North Kensington.In this study, we selected six factors because we found that the urban background factor consisted mainly of aged traffic emissions and wood burning emissions and the secondary aerosols contained inorganic secondary aerosols and mixed secondary aerosols, therefore these two factors (urban background and secondary) could split into four factors.This separation increased the best fit between modelled and measured data by reducing the total residual; however we noted that this separation was not totally clean since the traffic emission and other combustion such as wood burning could share a common size distribution profile.A comparison of the two results is provided in Supplementary Material.
In summary, the dominant source contribution of particles by number at a London urban background site was attributed to traffic emissions (56.5%).It was followed by the urban accumulation mode (28.3%) which is believed to originate largely from combustion sources such as wood burning.The contribution of traffic emissions to total particle number in urban areas was also found to be highest in the previous studies by Ogulei et al. (2007), Pey et al. (2009), Gu et al. (2011) and Wang et al. (2013).Gu et al. (2011) reported that traffic and combustion emissions represent 65.2 and 26.1% of total particles number in Augsburg, Germany.Particles from combustion-related sources (aged traffic and urban accumulation mode) have a strong correlation with black and organic carbon, suggesting that these sources mainly comprise carbonaceous particles.Secondary aerosols accounted for a small fraction of particle number (8.6%), but are the main contributor to submicron particles by volume (57.4%).These particles contain mainly ammonium nitrate and sulfate, with secondary organic compounds.The urban accumulation mode accounted for 29.7% of particles by volume while traffic emissions and nucleation accounted for 11.8 and 1.0%, respectively.In terms of particle surface area, the urban accumulation mode was the dominant fraction (39.2%), following by mixed secondary aerosol (26.9%), aged traffic emissions (14.2%), inorganic secondary aerosol (13.2%), fresh traffic emissions (6%) and nucleation (1.3%).

Which Source is Most Responsible for Particles Deposited in the Human Respiratory Tract? Total and Regional Lung Deposition Efficiencies (DEs)
Table 3 shows the regional lung deposition efficiencies (DE) of particles from different sources.These were estimated by apply a modified ICRP model (Vu et al., 2015) to the PMF factor profiles.In terms of particle number, particles released from nucleation (F3) and local traffic emission sources (F1) were found to have the highest deposition efficiencies in the total lung, with fractional efficiencies of 0.62 and 0.57, respectively, followed by aged traffic particles (F2, DE = 0.41).In contrast, the volume deposition efficiency of nucleation particles (F3) is only 0.17, while the volume deposition efficiency for secondary aerosol was found to be the highest, ranging between 0.25 (F5) and 0.41 (F4).By surface area, 22-37% of total secondary aerosol surface concentration can deposit into the lung, while the surface area deposition efficiency for accumulation mode particles (F6) is only 16%.The total surface area deposition efficiency fraction for local (F1), aged traffic emission (F2) and nucleation (F3) is 0.37, 0.28 and 0.27, respectively.
In terms of regional lung deposition, traffic particles and regional nucleation particles have higher deposition efficiencies in the alveolar (AL) region rather than in the extra-thoracic (ET) or tracheo-brochial (TB) regions in all three number, surface area and volume metrics.For example, the deposition efficiency of local traffic particles (F1) by number is 0.40, 0.11 and 0.06 for AL, TB and ET regions.But for secondary aerosols, the highest deposition efficiency values are found in the ET region by volume and surface area.The regional and total lung deposition efficiencies of particles generated from different sources are mainly controlled by their size distribution.Smaller particles such as nucleation or local traffic emission are found to penetrate deeper into the respiratory tract.Table 4 reports a compilation of deposition efficiencies reported in previous studies.

Which Sources Are Most Responsible for Particles Deposited in Different Regions of the Lung?
The contribution of each source to submicron particles deposited (%) by number, surface area and volume in the extra-thoracic (ET), trachea-bronchial (TB), and alveolar (AL) regions and total lung are shown in Table 5.

By Number
A majority of particles by number deposited in the total and regional lung is found to be related to combustion sources, in which traffic emissions were found as a main contributor.Local and aged traffic particles are responsible for 67.8% of particle number deposited in the total lung.In terms of lung regions, traffic emissions are dominant and account for 68.5% of particle deposition in the AL region and up to 70.8% in the TB region.Regional accumulation mode and nucleation particles account for approximately 17% and 10% of particle number deposited in the total lung while secondary aerosol only account for around 5.1%, respectively.The dominant contribution to particles deposited in the total lung is traffic emissions because it is the main source of particles by number in an urban area.In addition, traffic particle size distributions were found predominantly in the ultrafine size range (< 100 nm), with high deposition efficiencies.Therefore traffic emission particles can easily penetrate and deposit into the human respiratory tract.

By Surface Area
Combustion and secondary aerosols account for the majority of particles by surface area deposited in the lung.Secondary aerosols contribute nearly half of the particle surface area deposited in total lung (44.6%), while regional accumulation mode particle and traffic particles account for 27.3% and 26.6%, respectively.In the AL and TB regions, regional accumulation mode particles, traffic emissions and secondary aerosols share around one third of total deposited particles.However, in the ET region secondary aerosols account for more than 76% total particle surface area while accumulation mode particles and traffic emissions account for 12.7 and 9.5%.Only 1.6% of particle surface area deposited in the total lung is attributed to nucleation particles.

By Volume
Secondary aerosols are identified as the main source of submicron particles by volume deposited in the total and regional lung.They represent approximately 72.3% of particle volume deposited in the total lung and up to 82.6% in the ET region.This can be explained by a major contribution of secondary aerosol to particle volume.In addition, due to their more hygroscopic properties, their lung deposition efficiencies are much higher than those from hydrophobic particles with the same initial dry diameter (Vu et al., 2015).Accumulation mode particles account for 7.4%, 25.9%, 27.5% and 17.3% of total particle volume deposited in the ET, TB, AL and total lung, respectively.Traffic emissions only represent around 10% of the particle volume deposited in the total lung.Traffic emission particles mainly penetrate and are deposited into the TB and AL regions where they account for around 15-16%, while they only contribute to 3.5% of particle volume deposited in the ET region.Nucleation particles do not make a significant contribution to total submicron particle deposition by volume in the lung (< 1%).
Some studies suggest that ultrafine particles, most commonly measured in terms of their number concentration, may have higher toxicity compared to the corresponding mass of fine particles due to their larger surface area, oxidative capacity and potential to form radical species that can lead to cellular DNA damage or induce inflammatory effects (Kreyling et al., 2004;Atkinson, 2010).In addition, Harrison et al. (2010) indicated that relating health outcomes to measured particle mass concentrations can underestimate the public health impacts and emphasized that the regional lung dose, not pollutant exposure, probably drives health outcomes.Therefore, our findings over the contributions of particle sources to regional lung deposition for different particle metrics may help address the question of what is most important metric linked to health effects in epidemiological studies.It may also help to inform policy makers over decisions in controlling particulate matter as measures to reduce particle mass may not be very effective in relation to particle number.Reduction of traffic emissions would be most effective in controlling particle number because the contribution (59.0-70.8%) to deposition in both the total and lung regions is large.But if the target is to reduce submicron particles by mass, the most effective target is the secondary aerosols and urban accumulation mode (mostly urban combustion).

CONCLUSIONS
This study found that traffic emissions were the most relevant ambient source of submicron particles deposited by number into the human respiratory tract, accounting for 67.8% of particles deposited in the total lung by number at the North Kensington site.This can be explained by considering local and aged traffic sources as a major source of submicron particles by number (56.5%) in this urban background area in conjunction with their high lung deposition efficiency (0.41-0.57).Moreover, traffic particles can penetrate deeper into the lung with deposition efficiencies of 0.3-0.4 in the AL region and 0.07-0.11 in the TB region.Considering the high concentrations by number attributed to traffic emissions in the PMF analysis, local and aged traffic emissions represent 68.5% and 70.8% of the total particle number deposited in the AL and TB region, respectively.Urban accumulation mode particles and regional nucleation particles were also found to contribute significantly to the number of particles deposited in the lung (16.9% and 10% respectively).Secondary aerosols contribute only approximately 5.1% of submicron particles deposited in the total lung by number, but they represent a major source of particles deposited in the regions and total lung by volume (56.2-82.6%),followed by urban accumulation mode (7.4-27.5%),traffic emissions (3.5-16.2%)and nucleation (0.4-0.9%)In terms of deposited particles measured by surface area, secondary aerosols were found to be dominant in the ET region (76.9%), while the main contribution in the TB region was traffic emissions (40.1%) followed by the urban accumulation mode (33.6%).In the AL region, urban accumulation mode was the highest contributor to particle surface area deposition (35.0%), followed by traffic emissions (34.3%), secondary aerosol (28.9%) and nucleation (1.8%).

Fig. 1 .
Fig. 1.Particle number/volume size distribution (A) and diurnal pattern of particle number in terms of nucleation mode (NU), Aitken mode (AIT), Accumulation mode (ACC) and total particle (Total) in North Kensington, UK during 2012.

Table 1 .
Statistics of the particle number and volume concentrations for six factors.

Table 2 .
Correlations between contributions of six factors with other chemical species.

Table 3 .
Total and regional lung deposition of each source (i.e., factors from PMF analysis).Sec.: secondary aerosol, Acc.: accumulation mode; DE total = DE ET + DE TB + DE AL .

Table 4 .
Lung Deposition Efficiencies in previous studies.

Table 5 .
Source apportionment of submicron particles deposited in the regional and total lung (%).