Simulation of the transition metal-based cumulative oxidative potential in East Asia and its emission sources in Japan

The aerosol oxidative potential (OP) is considered to better represent the acute health hazards of aerosols than the mass concentration of fine particulate matter (PM2.5). The proposed major contributors to OP are water soluble transition metals and organic compounds, but the relative magnitudes of these compounds to the total OP are not yet fully understood. In this study, as the first step toward the numerical prediction of OP, the cumulative OP (OPtm*) based on the top five key transition metals, namely, Cu, Mn, Fe, V, and Ni, was defined. The solubilities of metals were assumed constant over time and space based on measurements. Then, the feasibility of its prediction was verified by comparing OPtm* values based on simulated metals to that based on observed metals in East Asia. PM2.5 typically consists of primary and secondary species, while OPtm* only represents primary species. This disparity caused differences in the domestic contributions of PM2.5 and OPtm*, especially in large cities in western Japan. The annual mean domestic contributions of PM2.5 were 40%, while those of OPtm* ranged from 50 to 55%. Sector contributions to the OPtm* emissions in Japan were also assessed. The main important sectors were the road brake and iron–steel industry sectors, followed by power plants, road exhaust, and railways.

www.nature.com/scientificreports/ carbon. In fact, dissolved oxygen causes interfacial catalytic oxidation of DTT in the presence of elemental carbon particles 12 . Verma et al. 13 revealed the importance of humic-like substances (HULIS), such as quinones and secondary organic aerosols, and Yu et al. 14 indicated that the interactions between HULIS and transition metals likely contribute to the DTT activity. In addition to catalytic redox reactions of transition metals and quinones, noncatalytic DTT-active organics such as organic hydroperoxides and electron-deficient alkenes have been highlighted 15 . Thus far, the relative importance of chemical compounds to the total DTT activity is not fully understood, but the importance of the coexistence of metals and organics is widely accepted [1][2][3]9,14,[16][17][18] . To date, several experimental studies have been performed to relate OP, chemical compounds, and health outcomes [4][5][6][7][8][9][10][11][12][13][14][15][16][17][18] . In terms of numerical simulations, a model has been proposed to determine the chemical reactions producing ROS in epithelial lining fluids 19 and a statistical model, called the land use regression model 20,21 , to predict the spatial variations in OP. However, none of these studies has derived the spatiotemporal variations in OP via the direct simulation using 3-dimensional numerical modeling. Very recently, Daellenbach et al. 22 derived the spatiotemporal variations in OP in Europe by combining the 3-dimensional simulations of organic aerosols and NO x and the statistical relationship between the measured OP and aerosol components in Switzerland and Liechtenstein. Daellenbach et al. 22 did not directly predict the redox active aerosol components such as transition metals and quinones but demonstrate that the simulated OP agreed very well with that observed in the measurement sites.
Within the context mentioned above, we developed a 3-dimensional model and emission inventories of the DTT-active transition metals in Asia (TMI-Asia) and Japan (TMI-Japan) and evaluated simulation results based on field measurements in Japan 23 . Before directly predicting the total OP, as a first step, we defined the cumulative OP based only on transition metals (OP tm *) and assessed its predictability in this study. OP tm * was defined as the summation of the (simulated or observed) DTT-active transition metal concentrations multiplied by the DTT consumption rate per unit mass (obtained by laboratory experiments). This study is the first trial to directly predict the oxidative potential, but it should be noted that the contributions of other components, such as the effects of organics 22 and the interactions between organics and metals, have not yet been considered. Many studies focused on OP only of PM 2.5 1-9,11-21 and so another feature of the current study is that we include the OP tm * contributions of coarse-mode particles and Asian dust because lung deposition of fine particulate matter (PM 10 ) may be nonnegligible in some human conditions 24 , and Asian dust particles containing metals may adversely affect health 25 . Nishita-Hara et al. 10 and Daellenbach et al. 22 also considered the coarse-mode particles.
The main objectives of this study are thus summarized as follows: (1) to assess the predictability of OP tm * by numerical simulations, (2) to show the differences of horizontal distributions and the source-receptor relationship between OP tm * and PM 2.5 , and (3) to identify the major emission sectors for anthropogenic OP tm *.

Results
Temporal variations in OP tm * based on simulations and observations and their comparison. Figure 1  for the simulation of the transition metals and cumulative OP (OP tm *) based on the top five key transition metals, namely, Cu, Mn, Fe, V, and Ni, as described in Eq. (2) and Table 2 in Method section. D01 covers East Asia with a horizontal grid resolution of 30 km to simulate the long-range transport from the Asian continent to Japan via synoptic-scale disturbances such as the fronts of cyclones and migrating anticyclones. D02 covers the densely populated and industrial areas of Japan with a complex topography, and a fine grid resolution (i.e., 6 km) is necessary to accurately predict the domestic and transboundary contributions to the surface concentrations of air pollutants.
OP tm * based on the simulated transition metals (referred to as the simulated OP tm * hereinafter for simplicity) was compared to OP tm * based on the observed transition metals (hereinafter referred to as the observed OP tm *) of the total suspended particles (TSP) collected at the Yonago site ( Fig. 1). As described below, the r i values of Eq. (2) substantially varied among laboratories, as experimental methods such as buffer usage, pH, reaction time, and control volume differed 2 . To consider the above experimental variability, we derived two different OP tm * values based on r i retrieved from two different experiments, namely, Charrier and Anastasio 9 and Fujitani et al. 17 , which are referred to as OP tm *(CA) and OP tm *(F), respectively. Figure 2 shows a time series of the simulated (PM 10 ) and observed (TSP) OP tm *(CA) and OP tm *(F) at Yonago, as well as the fractions of anthropogenic, fine-mode, and domestic components and the contributions of various elements to the simulated OP tm *. Statistical metrics for the comparison are summarized in Table 1. Note that the OP tm * simulated with Eq. (2) includes correction factor f i based on measurements (the nationwide PM 2.5 survey conducted by the Ministry of Environment, Japan (MOEJ); http:// www. env. go. jp/ air/ osen/ pm/ monit oring. html; last accessed: 6 November 2020), which were independent from the Yonago data.
As indicated in Table 1, OP tm * was suitably predicted by the numerical simulations with correction factors based on independent MOEJ nationwide observations, even though OP tm * was primarily contributed by Cu and the discrepancies between the simulated and observed Cu were large (Tables 4 and 5 of Kajino et al. 23 ). Nevertheless, it is not surprising because the approach was analogous to the application of a multimodel ensemble, which generally reduces the uncertainty in each model. The summation of Eq. (2) reduced the uncertainty in the simulation of each metal element. In fact, the normalized root mean square errors (NRMSE; RMSE divided by observation average) for OP tm *(CA) and OP tm *(F) in TSP at Yonago (0.47 and 0.48) were smaller than those for Cu (4.7), Fe (1.0), Mn (0.87), Ni (1.4), and V (1.4). The median values of OP tm *(CA) were almost one order of magnitude smaller than those of OP tm *(F) due to the experimental variations and thus the discussion on the absolute values of OP tm * is not a scope of this study. Therefore, the relative magnitudes in time and space (i.e., the temporal and spatial variations, respectively) are mainly discussed. The simulated relative contributions of Scientific Reports | (2021) 11:6550 | https://doi.org/10.1038/s41598-021-85894-z www.nature.com/scientificreports/ metals to OP tm * were consistent with those based on the observations, while those to OP tm *(CA) and OP tm * (F) differed. OP tm *(CA) primarily consisted of Cu, followed by Mn and Fe, which was similar to measurements obtained in California. However, due to the relatively high r i values for Ni and Fe(II) and relatively low r i value for Mn, the major contributors of OP tm *(F) were Fe and Cu, followed by Ni. Although the relative contributions of each metal were different between the two methods, the relative magnitudes of anthropogenic compounds vs. Asian dust, anthropogenic fine-mode vs. coarse-mode particles, and anthropogenic domestic vs. transboundary OP tm * values were similar. Their variations were consistent with those in the metals, as shown in Fig. 6 of Kajino et al. 23 . Specifically, the contribution of Asian dust was large in spring, and the transboundary contribution was large in colder seasons except from late July to early August, while the fine-mode fractions were inversely correlated with the domestic contributions, which are explained in the next subsection.
Spatial distribution and seasonal variation in OP tm *. Figure 3 shows the seasonal mean surface air concentrations of the anthropogenic PM 2.5 -OP tm *(F), anthropogenic coarse-mode PM c -OP tm *(F) (simulated PM 10 minus PM 2.5 ), and OP tm *(F) of Asian dust. The simulated OP tm *(CA) is not shown because the horizontal distributions were very similar. Generally, PM 2.5 -OP tm *(F) is higher than PM c -OP tm *(F). These anthropogenic surface concentrations were the highest in the winter under stable meteorological conditions. However, due to the presence of surface snow, the emissions of Asian dust were suppressed in the winter. The surface concentrations of Asian dust were the highest in the spring over the Gobi Desert and were almost equivalent to those of PM 2.5 -OP tm *(F). As shown in Fig. 2, the long-range transport of PM 2.5 was more prominent than that of PM c .
The long-range transport of air pollutants from the Asian continent to Japan is influential during the cold seasons such as the spring and winter. A westerly wind prevails in the winter, while northerly and westerly winds prevail in the spring, which results in high concentrations in the respective downwind areas. The long-range The observation site (Yonago) and regional names adopted in the analysis are shown. The national and prefecture borders are depicted in D01 and D02, respectively. This figure is identical to Fig. 1 of Kajino et al. 23 . The map was generated using the Generic Mapping Tools v4.5.7 (https:// www. gener ic-mappi ng-tools. org). www.nature.com/scientificreports/ transport of Asian dust was influential in the spring. Due to the presence of a Pacific high, the long-range transport was generally insignificant in the summer. The summer of 2013 was an exception. An anticyclone persisted over the southwestern part of the Japanese archipelago from late July to mid-August, which continuously carried pollutants from the Asian continent to Japan via the marginal flow along the northern edge of the anticyclone. The seasonal mean wind pattern exhibited features of the Pacific high, but the high-surface concentration areas , (upper-right panels) 10-d mean simulated (D02) fractions of (red) anthropogenic compounds to the total compounds, (green) anthropogenic fine-mode particles to anthropogenic total particles, and (blue) anthropogenic domestic contributions to the total OP tm *(CA) and OP tm *(F), (lower-left panels) relative contributions of each metal to the observed and simulated OP tm *(CA) and (lower-right panels) same as the lower-left panels but for OP tm *(F) at Yonago. Note that the simulated concentration of TSP is equivalent to the simulated PM 10 concentration. Domestic contributions of OP tm * and differences from those of PM 2.5 . The spatial distribution of the simulated anthropogenic OP tm *(F) in PM 2.5 over D02, together with its domestic contributions, is shown in Fig. 4. A contrast between the domestic and transboundary components and their seasonal differences are clearly observed in the figure. Transboundary air pollution dominated in the spring and winter, and there was a clear horizontal gradient in the surface PM 2.5 -OP tm *(F) concentrations from west to east during that season. However, high surface PM 2.5 -OP tm *(F) concentrations were found in Kanto, including the Tokyo Metropolitan Area, and the domestic contribution exceeded 50% throughout the year. In addition to the Kanto region, highconcentration areas were observed around large cities, such as Nagoya (in Chubu) and Osaka (in Kansai), where the domestic contributions were as large as those in Kanto (even though the areas were smaller). The domestic contributions were large over the inland seas and their surroundings in the western part of Japan, such as the Seto Inland Sea between Chugoku, Shikoku, and Kyushu and the Bungo Channel between Kyushu and Shikoku. Under the strong influence of the transboundary transport in the spring and winter, the concentrations over the areas were higher than those in the other areas at the same longitudes. The Seto Inland Sea is a major route of vessels in Japan, and thus, large industrial regions are located along the coast, and as a result, the transition metal emissions from ships and industries are high in this region, as shown in Fig. 6.
To determine the differences between the health hazard based on OP and the conventional health hazard, i.e., the PM 2.5 mass concentration, the domestic contributions of PM 2.5 and OP tm * are shown and compared in Fig. 5. The horizontal distributions of the domestic contributions of OP tm *(CA) and OP tm *(F) were very similar. The distributions of the domestic contributions of PM 2.5 were much broader than those of the domestic contributions of OP tm *. This result occurred due to the difference in the contributions of the secondary aerosols. OP tm * is composed only of primary aerosols, i.e., metal elements, while PM 2.5 is composed of both primary and secondary aerosols. Generally, the relative contributions of secondary aerosols are larger in downwind regions (after long-range transport). Consequently, the domestic contribution of OP tm * is the largest near the source regions, while that of PM 2.5 is larger in the downwind regions. As a result, the areal mean values of the PM 2.5 domestic contributions are larger than those of PM 2.5 -OP tm *, but the areal maximum values of PM 2.5 -OP tm * are as large or even significantly larger than those of PM 2.5 especially over the Kyushu-Okinawa and Chugoku-Shikoku regions, where long-range transport is predominant. In these regions, more than 60% of PM 2.5 was attributed to transboundary contributions and 40% was attributed to domestic contributions. However, in regard to OP, the domestic contribution was as high as 50%. The domestic contributions of PM 10 -OP tm * were generally larger than those of PM 2.5 -PM tm * by up to 5% in terms of the areal average or approximately 10% in terms of the areal www.nature.com/scientificreports/ maximum. As previously mentioned, this result occurred because the lifetime of PM 10 is generally shorter than that of PM 2.5 . Hence, the relative contributions of long-range transported PM 10 are lower than those of PM 2.5 .

Contributions of the emission sectors to OP tm * in Japan. The relative contributions of each emission
sector to OP tm *(CA) and OP tm *(F) of PM 2.5 and PM 10 are shown in Fig. 6. While large differences occur in the metal contributions to OP tm *(CA) and OP tm *(F), the sector contributions based on these two methods are not very different both in terms of PM 2.5 and PM 10 . However, the sector contributions to PM 2.5 and PM 10 -OP tm * are very different. For example, the road brake sector is the top contributor to PM 10 -OP tm * but not to PM 2.5 -OP tm *. It should be noted that the size distribution of the current inventory has not yet been evaluated. In fact, a recent laboratory experiment 28 has demonstrated that most brake wear particles occur in the fine mode, i.e., PM 2.5 . The size apportionment of the emission inventory certainly requires further improvement. The results provided by TMI-Japan are presented below in this section. The most important sector in regard to PM 10 -OP tm * was the road brake sector, followed by the iron-steel industry sector. Regarding PM 2.5 -OP tm *, the other sectors such as other industries (nonmetals), navigation, incineration, power plants, road exhaust, and railways attained almost equal contributions (ranging from only a few percent up to 10-20%). As shown in Fig. 2, the source contributions mainly reflected those of Cu and Fe followed by those of Mn and Ni. The large contribution of the road brake sector to PM 10 -OP tm * originated from Cu and Fe and that to PM 2.5 originated from Fe. The iron-steel industry contribution to OP tm * originated primarily from Fe and Mn, while it originated from Ni and Cu in regard to PM 2.5 . The contribution of the other industry (nonmetals) sector to PM 2.5 -OP tm * originated from Ni and Cu, and the contributions of the metal industry sectors other than the iron-steel and incineration sectors originated from Cu and Fe. The power plant The contribution of the navigation sector was the largest in Chugoku-Shikoku, originating from V and Ni. The contribution of the navigation sector was large in regard to PM 2.5 -OP tm *(F) due to the high r i value for Ni. Vanadium (V) and Ni achieved almost equal contributions to PM 2.5 -OP tm *(F). The metal emission amounts were the largest in Kanto, the most populated region of Japan, in terms of Cu, Mn, and Fe, while the V and Ni emissions were the largest in Chugoku-Shikoku. As previously described, this result occurred due to the aerosols generated during heavy fuel oil combustion emitted from vessels in the Seto Inland Sea and industrial factories along the coast.
The railway sector contributed approximately 10% to PM 2.5 -OP tm * in Kanto, which was primarily attributed to Fe (Fe stemming from the railway sector accounted for approximately 15% of the total Fe emission) and Cu. This sector also emitted Mn. Because railway emission data were available only for Kanto, the OP tm * and metal emissions in the other regions could be underestimated at similar fractions.

Discussion
Toward the effective emission control measures. Daellenbach et al. 22 and this study indicated that the emission sources for PM 2.5 and OP could be very different. There may be a possibility that an emission control measure to reduce PM 2.5 surface air concentrations may not necessarily reduce their health risk, if OP is essential to the health hazard of aerosols and if the selected control measure does not reduce OP. For example, reduction of ammonia emission may lead to decrease in surface air concentrations of PM 2.5 , but not OP because the emission sources of transition metals and quinones are different from those of ammonia. Besides, decrease of ammonia in the air can lead to increase of aerosol acidity and thus solubility of metals, which could enhance OP of aerosols. On the other hand, even though an emission control measure does not significantly decrease the PM 2.5 concentration, it may be effective if it decreases OP efficiently.
Certainly, OP is not a sole and perfect health hazard of aerosols, but concerning only PM 2.5 may mislead the emission control strategy. In order to seek for the better and effective control measures, OP must be directly simulated by numerical models in addition to PM 2.5 and other conventional aerosol components, because numerical models are one of the most powerful tools to quantitatively evaluate the impacts of emission control on the surface air concentrations.
Future research. The current study has the following limitations, which should be resolved in the future.
We only simulated the total (water-soluble and water-insoluble) metal concentrations and assumed a constant solubility over time and space. Water solubility of metals depends on their chemical forms and aerosol acidity. As acidity is higher, some of transition metals such as Cu, Mn, and Fe become more water soluble. We should consider specific water solubilities of metals for each emission source and simulate any changes in water solubility of metals due to changes in aerosol acidity occurring during transport 29 . We only considered metals but organics, www.nature.com/scientificreports/ especially quinones, are important DTT-active agents 9,[11][12][13] . The interactions between metals and organics can also enhance DTT consumption and thus should be considered 14,19 . Metals are primary species, but quinones are secondary and produced in the air. During transport from emission source to downwind regions, the relative contributions of quinones to OP may increase, which can cause differences in dominant emission sectors affecting OP in the two regions. There are species that are not redox active but cause a notable oxidative stress, such as endotoxin 30 . These species should also be taken into account in the simulation. Finally, epidemiological studies are required to assess the applicability of the new health hazards [6][7][8]31 . Further integration of meteorology, chemistry, toxicology, and epidemiology is indispensable in the next study stages.

Methods
Numerical simulations of the transition metals. A transition metal version 23 of the Japan Meteorological Agency (JMA) regional-scale meteorology-chemistry model (NHM-Chem 32 ) was adopted in this study. We retrieved simulation results of the considered transition metals from Kajino et al. 23 , and details are not presented here.
The transition metal version of NHM-Chem employs three aerosol categories: anthropogenic submicron particles (SUB), anthropogenic coarse-mode particles (COR) and mineral dust (MD). The full chemistry version of NHM-Chem fully considers aerosol microphysical processes such as nucleation, condensation, coagulation, and deposition, but the transition metal version only considers deposition processes, such as wet deposition (in-cloud and below-cloud deposition), fog deposition (contact of cloud droplets in the bottom layers of the model to the ground surface 33 ), and dry deposition. Upon emission, the size distributions of the above three aerosol categories are prescribed 23 , which change during transport only due to advection, turbulent diffusion, and removal processes. A prescribed hygroscopicity is assumed for each category 23 , which was adopted to obtain cloud condensation nuclei activity used for in-cloud scavenging and fog deposition and hygroscopic growth used for below-cloud scavenging and dry deposition.
As shown in Fig. 1, two model domains were applied to simulate the long-range transport from the Asian continent to Japan and the contributions of local emissions and local transport in Japan. D01 covers East Asian countries with a grid spacing (Δx) of 30 km and 200 × 140 grid cells on the Lambert conformal conic projection to resolve the transport of air pollutants due to synoptic-scale disturbances. D02 covers the Kyushu, Shikoku, and Honshu (only the Chugoku, Kansai, Chubu, and Kanto regions and part of the Tohoku region) islands of Japan with Δx = 6 km and 226 × 106 grid cells on the Lambert conformal conic projection. The horizontal grids of the meteorological and transport parts of NHM-Chem are the same, but the vertical grids differ. There are 38 vertical levels up to 22,055 m above sea level (a.s.l.) for the meteorological simulations and 40 levels up to 18,000 m a.s.l. for the transport simulations. A JRA-55 global analysis 34 (1.25° × 1.25°, 6 h) was applied to the initial and boundary conditions of the meteorological simulations in D01. The JMA meso-regional objective analysis (MANAL; 5 km × 5 km, 3 h) was adopted in D02. These analysis data were also used for the spectral nudging of the meteorological simulations.
Kajino et al. 23 developed emission inventories of eight DTT-active metals, namely, Cu, Mn, V, Ni, Pb, Fe, Zn, and Cr, in anthropogenic PM 2.5 and PM 10 in Asia and Japan, referred to as Transition Metal Inventory (TMI)-Asia v1.0 (Δx = 0.25°; monthly, 2000-2008; 9 sectors) and TMI-Japan v1.0 (Δx = 2 km; hourly, weekday/weekend; monthly, 2010; 29 sectors), respectively. Kajino et al. 23 also simulated metals originating from Asian dust. TMI-Asia and TMI-Japan were used for the transport simulations over D01 and D02, respectively. Anthropogenic PM 2.5 and PM 10 emissions were allocated to the SUB and COR categories, and those originating from Asian dust were allocated to the MD category. The simulated metal concentrations of SUB were compared to MOEJ PM 2.5 concentration measurements, and the simulated concentrations of COR plus MD were compared to measurements of TSP reported in Kajino et al. 23 and this study. A part of mineral dust mass should be included in PM 2.5 in reality, but it was neglected in this study.

Surface air concentration measurements of the transition metals.
To derive the observed OP tm *, we used the same observation datasets as that reported in Kajino et al. 23 . The measurement data were collected in Yonago city, Tottori Prefecture, Japan (Fig. 1) 43°N, 133.33°E), approximately 20 m above ground level. The inorganic elements were analyzed using the fundamental parameter quantification method of energy-dispersive X-ray fluorescence spectrometry (EDXRF-FP), which was developed and evaluated by Okuda et al. 35,36 . Definition and derivation of the cumulative oxidative potential based on transition metals (OP tm *). In this section, we assessed the model predictability of the observed cumulative OP based on the top five DTT-consuming metals in air determined via reagent experiments. It should be noted that this parameter is not the realistic OP in the atmosphere but the idealized OP. The realistic OP can be expressed as follows: where r i and r j are the specific OP values (DTT consumption rate per unit of mass) of metals i and organic compounds j, respectively, C i and C j are the surface air concentrations, and i and j are the metal ions and organic compounds, respectively. The last term is an interaction term between the metal ions and organic compounds 14 . www.nature.com/scientificreports/ However, because water solubility data of the metals and DTT-active organic compounds are not available in the inventory and model, we defined the cumulative OP based only on the total (soluble + insoluble) metals (OP tm *) as follows: where r i , χ const,i , T i , and f i are the specific OP, water solubility (assumed constant in time and space), total surface air concentration, and simulation bias correction factor, respectively (i: top five DTT-consuming metals, namely, Cu, Mn, Fe, V, and Ni). The values used in Eq. (2) are listed in Table 2. It is known that the r i values vary among laboratories, as experimental methods such as buffer usage, pH, reaction time, and control volume are different 2 .
To consider the variability in experiments, two different OP tm * values were derived using the r i values of Charrier and Anastasio 9 and Fujitani et al. 17 , referred to as OP tm *(CA) and OP tm *(F), respectively. Charrier and Anastasio 9 and Fujitani et al. 17 provided fitted r i values by using exponential functions for specific species, but the r i values of all species were fitted with linear functions in this study. Because there was no water solubility information available in the measurements or emission profiles, constant values of χ i were assumed and applied, which were obtained from Okuda et al. 37 . Moreover, f i was equal to 1 for the observed OP tm *, while the inverse of the Sim:Obs ratio obtained from the comparison of the D02 simulations and nationwide PM 2.5 measurements of the MOEJ was adopted for f i in regard to the simulated OP tm *. The same f i was applied for PM 2.5 -OP tm *, PM 10 -OP tm *, and TSP-OP tm *. Although f i was derived based on simulated and observed PM 2.5 without consideration of simulated PM 2.5 fraction of mineral dust, it was proved to be reasonable from the comparisons of simulated and observed TSP-OP tm * at Yonago as shown in Table 1. In this paper, OP tm * based on the simulated and observed transition metal concentrations was simply referred to as the simulated OP tm * and observed OP tm *, respectively.

Code availability
The NHM-Chem source code is available subject to a license agreement with the JMA. Further information is available at http:// www. mri-jma. go. jp/ Dep/ glb/ nhmch em_ model/ appli cation_ en. html (last accessed: 2 November 2020).