Sub-weekly signatures relate ultrafine aerosols enriched in metals from intensive farming and urban pollution to Kawasaki disease

Air pollution (urban, industrial or rural) has been linked to a myriad of human ailments despite clear mechanistic associations that are often not thoroughly established. Daily variability of fine aerosols in a surveillance campaign in south Japan shows a striking coevolution between their trace elements (metal and metalloid, MM) content and Kawasaki disease (KD) admissions, suggesting a strong dynamical link. These aerosol MM could instigate an immune response that, along with genetic susceptibility, would lead to KD development. This association may account for over 40% of the total variability in the disease, being dominated by a clear sub-weekly cycle (SWC1). Thanks to both an unprecedented daily KD epidemiological record going back to 1970, light detection and ranging (LIDAR) atmospheric backscattering profiles for the interval 2010–2016 and HYSPLIT simulations with numerous sensitivity analyses, we can trace this SWC1 variability to occur concomitantly from sub-seasonal to interannual timescales in both KD and aerosols. This SWC1 appears to connect or disconnect Japan to air intrusions from above the planetary boundary layer (PBL), having their source in industrial and agricultural areas in NE Asia and points to a stronger case for an agricultural source for the exposure as opposed to urban pollution. The KD maxima always occur in full synchrony with the arrival of very small (<1 µm; PM1) particles showing that ultrafine aerosols appear as a necessary cofactor in the occurrence of KD and sets the field to associate other similar human diseases. Our study shows how signal-detection approaches can be useful to uncover hidden associations between the environment and human health, otherwise unnoticed and help set new early-warning systems for disease prevention.


Introduction
The causes conducive to Kawasaki disease (KD) pediatric syndrome are yet unknown, despite the intensive research on different fronts (genetic, immunological, environmental, epidemiological, etc Marrani et al 2018, Nagata 2019).Potential drivers include environmental, biological or chemical triggers (e.g.bacteria, fungi, viruses, toxins, dust, pollution, Leung et al 1993, Barton et al 2002, Matsubara et al 2006, Lee et al 2011).Unfortunately, no conclusive result has yet unequivocally solved this long-standing puzzle.On the etiology, the leading paradigm is that an unidentified agent enters through the upper respiratory tract and causes a dramatic immunologic response in certain genetically predisposed children (Rowley et al 2008, Onouchi 2009, Rodó et al 2014).Coronary artery aneurysms develop in 20%-25% of cases, with this outcome being at times clinically silent and often misdiagnosed with other more benign rash and fever syndromes, but that may lead, in rare occasions, years later, to sudden death or myocardial infarction (Burns and Glodé 2004).The presumed link in KD epidemiology to an environmental forcing factor is supported by a clear seasonal pattern, first described for Japan (Burns et al 2005) and then worldwide (Burns et al 2013).Increases in KD cases in the Northern Hemisphere winter in locations on both sides of the North Pacific Ocean were seen to occur associated with the seasonal enhancement of low-and high-tropospheric wind currents from the Asian continent (Rodó et al 2011).The ulterior identification of a fungal pathogen predominant in air masses over the planetary boundary layer (PBL) at times of KD maxima, led to the hypothesis that in Japan, KD could be promoted by a pathogen or environmental trigger carried by winds blowing from agricultural areas in the Asian continent (Rodó et al 2011, 2014, Frazer 2012, Maki et al 2019).This seminal study identified an area of intensive cropland production in NE China, known as the 'breadbasket of China' as the center of the source region of winds arriving in Japan a few hours to two days later.
At a time when PM attributed to automobiles and city origins have decreased in general, because of efforts to reduce air pollution, we are nonetheless still seeing large numbers of KD cases in Japan.If the source originated from urban pollution, we would expect the number of cases to lower, which is not the case.Instead, because of the growing resistance to pesticides among agricultural pests, pesticides may have become more potent and even more toxic, strengthening the hypothesis for an agricultural origin.In China (as in other regions), since the 1960s, the adoption of high-yield crop varieties and wide use of artificial fertilizers and pesticides has increased production markedly (Green and Grow 2020).These agricultural practices accumulate high levels of trace elements (in particular MM) in soils, affecting key biological processes with spoiling consequences on the quality of both soil and plant health.While in adequate levels metals, such as Fe, Mn, Mo, Cu, Zn, and Ni are necessary for plant growth, high concentrations of these MM can result in toxicity for humans (Peralta-Videa et al 2009).Phosphate fertilizers, for instance, are hazardous to the exposed rural communities, which may be a causative factor of chronic kidney disease (Jayasumana et al 2015).These phosphate rocks used in manufacturing fertilizers may contain micronutrients necessary for plant growth (such as magnesium (Mg), manganese (Mn), potassium (K), sodium (Na) and cobalt (Co)), but also environmental pollutants such as cadmium (Cd), copper (Cu), chromium (Cr), nickel (Ni), lead (Pb) and zinc (Zn, Tufail and Khalid 2008).Those MM are one of the major environmental problems around the world.MM constitutes an ill-defined group of inorganic elements, including transition metals, metalloids, lanthanides, and actinides.The most common MMs found at soil sites are Pb, Cr, arsenic (As), Zn, Cd, Cu and Ni.It is well-known that most metals may persist in soils for a long time after their introduction, resisting microbial or chemical mobilization.As for their inhalation from aerosols, data on the chemical composition of air masses and source transport processes are usually not available and, therefore, relationships to human health are extremely hard to substantiate.

Methods and datasets
When associating meteorological and air quality factors to health outcomes, the transient nature of such relations should be considered (Bragazza 2008, Andersen et al 2009, Lindegren and Eero 2013).We used statistical techniques aimed at characterizing these couplings, often involving thresholds that have proven effective.For instance, the singular spectrum analysis (SSA, Dettinger et al 1995) (Fontal et al 2021).
Air pollution from Asian dust and aerosols are routinely monitored at around 20 locations in East Asia at two-wavelengths (1064 nm and 532 nm) using polarization-sensitive LIDARS.The data from the Asian Dust and Aerosol Lidar Observation Network (AD-Net) are processed and published in near real time (Nishizawa et al 2016).Our study uses this rich database of vertical profiles that contains high resolution (<10 m) optical measurements of the atmospheric column.Measurements can trace the changing structure and physics of the air column and of particles.As no routine detailed data exists for MM in air quality stations other than a few metals, we used a unique daily survey of aerosol's chemical composition collected in Kumamoto (Japan) for 37 consecutive days in spring 2011 covering nearly 60 major and trace elements (mostly MM) determined from air filter samples (see Moreno et al 2013 for a full description of the methodology, figure 1).
SSA was then used to separate the variability components of KD and the total amount of MM (figures 1(B) and 2(D); see Methods).
As in previous studies, the origin of these aerosols was traced to the NE Asia region with the use of Lagrangian particle model simulations.To this end, we used a similar approach to seek for the origins and trajectories of the air masses in this particular event.With the HYSPLIT particle model Version 5.2.0;Stein et al (2015) (downloaded from www. arl.noaa.gov/hysplit/),we compared all air trajectories and emerging sources between peak KD days and lowest KD days (see Methods and supplementary information).and 19) in the case of total MM (red), total carbon (blue), PM10 (green) and KD (black).Right plot displays the same as the left one but for the weekly reconstructed components.Note consistencies and differences among phases and convoluting amplitudes of the reconstructed signatures in the different variables.Embedding dimensions for the SSA analysis were always varied to test for signal stability (p < 0.05).Concentration is in ng m −3 for METtot and in µg m −3 for Ct and PM10 but were shown in the same graph for comparison.As KD03 has accumulated cases from day 1 to day 3, no intra-weekly variability is present.

Results
Figure 1(A) displays a clear temporal covariation between MM measurements in the 37 days interval and the evolution of KD-admitted children in the group of Kumamoto prefecture hospitals (supplementary figure 1).The chemical results for the main elements analyzed are in full agreement with measurements routinely obtained in that period at 33 air quality monitoring stations in urban, suburban and rural areas operated by the Kumamoto Prefectural Government (see also www.pref.kumamoto.jp).The most abundant MMs were Zn, Pb, barium (Ba) and Cu, despite a very similar evolution which also manifests for many other trace elements in aerosols pointing to the same behavior.We used a cluster analysis to group the MM along ten different clusters (supplementary figures 3 and 2(A), (B)), with cluster 1 selecting 38 MM (figures 1(A) and (B) red curves).When compared to the aggregated KD evolution in the same interval (e.g. with KD 03 being KD cases added from day 0 to day 3 as the response time is established to vary between hours and 3 d, see Methods and Rodó et al 2014), a remarkable degree of covariation shows up (figure 2(B)).A linear doseresponse relationship arises in this interval between the new KD cases and total MM' concentration (e.g. with bootstrapped low-high µ difference being significant, p = 0.0001, figure 2(C)).A Poisson linear regression model based on total metal concentrations accounts for over 40% of the variability in KD (figure 2(D) showing a yellow line with 95% confidence intervals; p = 4.35 * 10 −4 ; see Methods).As an example, figure 1(A) displays the highest individual coevolution of KD 03 with eight of these MM species (e.g.As, Zn, Pb, Bi, Tl, Ba, Mn, Cu) and Se.The percentage of variance accounted for by individual MM in this 37 days atmospheric event reaches around 50% on average, with values nearing 60% in the case of Zn (figure 1(A)).The statistical model also denotes how an increment of around 70 ng m −3 in the concentration of total MM is associated with the appearance of one new KD case.
The SSA yielded a clear separation of this variability into four pairs of significant components, with two pairs showing a 15 d periodicity that varied fully in line with the oscillations reconstructed for KD 03 (figure 1(B)).Two other paired components recovered a marked weekly cycle (WC), being the 15 d cycle super-harmonic of the latter.Strikingly, coarse particulate matter (denoted by PM 10 ) and total carbon content (C tot ) also display the same two periods above, albeit they both peak with slight delay in the latter part of the survey (e.g. 1 d after) and manifest a clearly different convoluting amplitude.Both C tot and PM 10, in the coarser aerosol fraction, are therefore not fully concurrent with neither MM (heareafter MET tot ) nor KD 03 , whereas instead, figure 1(B) shows how the same evolving amplitudes of the reconstructed components (RCs) of MET tot and KD 03 (with at times a one-day lead for the former).
To obtain regional source estimates, we ran from Kumamoto city, 4 d back trajectories four times every day (at 00:00, 6:00, 12:00 and 18:00 h locally), for each event with either 5 cases (maxima) or days with 0 cases (minima) (figure 3(A)).A display of the trajectories inferred for all 37 d is included in supplementary figure 2. For comparison with former results (generated up to 2010), we also ran similar back trajectories for all extreme 5% percentile days in the interval 2010-2016 (figure 3(B)).Figures 3(A Back trajectories for the days with either maximum or minimum in KD cases in Kumamoto.(A) 37 d when the air was sampled in Kumamoto and HYSPLIT back trajectories for both the days with maximum (red) and minimum (blue) KD cases (with cases showing the rolling sum of KD cases up to +3 d for each of these days).Definition of KD Maxima is for days with five cases (n = 3, red) and the KD minima as the days with 0 cases (n = 3, blue).Air trajectories associated with these maxima and minima events are depicted in the lower panel as four daily back trajectories of the 96 previous hours starting in Kumamoto at 10 m of altitude (at 00:00, 6:00, 12:00 and 18:00 h).(B) Same as in (A) but for all the daily recorded KD cases in all of Japan for the period from 2010 to 2016, selecting the extreme 5% percentiles as maxima/minima (32 d, 128 trajectories for each).Grid-point representations of the maxima-minima differences are also shown below to allow a better comparison between air trajectories.the NE China region as previously identified in Rodó et al (2014), but with the Kumamoto event showing the source location also slightly southern with respect to the 2010 results.A detailed sensitivity analysis on this back tracking of the source has also been conducted as well as an estimation of the associated errors and uncertainties (see Methods and supplementary figures 10-12).The rich record of epidemiological incidence of KD in Japan going back to the 1970s enables a very detailed and unprecedented statistical approximation to the long-term evolution and structure of the WC shown above for the Kumamoto event (Rodó et al 2011, Moreno et al 2013).As cases were there grouped from day 0 to day 3, inspection of intra-weekly variation is not possible in this short record.Instead, this motivated the analysis of the epidemiological records of daily KD admissions for the entirety of Japan (figure 4) and for nine regions grouping prefectures (not shown).Applying an SSA decomposition (M = 100; see Methods and Ghil et al 2002) to study temporal variability, a prominent SWC along with marked seasonality shows up as the dominant variability components (figure 4(A)).An autoregressive (AR) spectrum with a backward-forward singular value decomposition algorithm was computed for KD (figure 4(B)) to further determine the sources of spectral variability and assess the nature of these weekly oscillations.First two RC recovered seasonality (RC 12 , 70% of the variability, p < 10 −5 ), and the WC showed up strongly in RC 34 with the latter accounting altogether for a constant 20% of the remaining variance (figure 4(A)).This SWC is composed of two clearly distinguishable sub-weekly periods (figure 4(B)) and not period 7 d, with the fundamental one being 3.5 d (SWC 1 > 10.6%; p < 10 −5 ), and its harmonic at 2.33d (SWC 2 = 9.2%; p < 10 −5 , see Methods and supplementary figure 4).
We conducted a two-way SDC (TW-SDC) analysis (Rodó 2001, Rodó and Rodriguez-Arias 2006) to further confirm the base frequencies present in KD daily records.This technique is specifically aimed at uncovering transitory associations between joint variability structures in time series (see Methods and Fontal et al 2021).We used the daily series of KD in the recent interval for which LIDAR atmospheric profiles are available (from 2010 in supplementary figure 4, and January 2010-December 2016 in figures 4 and 5) and built a synthetic time series of 3.5 d (SWC 1 ), 2.33 d (SWC 2 ) and 7 d (SWC 3 ) periods to seek for similar signatures in KD (supplementary figure 3 and see Methods).As an example, supplementary figure 4 displays for the year 2010, the TW-SDC results for SWC 1 (3.5 d period; panels a, d), SWC 2 (2.33 d period; b, e) and SWC 3 (7 d period; c, f) at different window sizes (S) to capture the periodicities whenever they appear (namely S = 6 d in a, b, c and S = 22 d in d, e, f ).The 3.5 d cycle (SWC 1 ) stands out strongly, both in terms of its high correlation magnitude as well as for being present throughout the interval 2010-2016 (e.g.compare a vs c, d vs f in supplementary figure 4).In contrast, a very faint SWC 3 (7 d) that is linked to the 3.5 d cycle, correlates mostly to periods of very low KD incidence (supplementary figures 4(c) and (f)).A sensitivity analysis with an intermediate S of 10 -to maximize a potential 7 d cycle-applied to the entire 2010-2016 interval, similarly states the dominant role of SWC 1 in KD (not shown).Summing up, two maxima and two minima clearly appear in a week (figure 4(D)), with this SWC 1 also showing up strongly in all the three large epidemics (e.g.see its maximum in supplementary figure 5 for the largest KD epidemic in 1982 and a marked contribution in 1979 and 1986).SWC 1 exists throughout the entire KD record since the surveillance was initiated in the early 1970s in Japan (and until the end of 2016).The SWC 1 cycle amplifies towards the present concomitantly with the appearance of a more marked seasonality in the epidemiological time series (supplementary figure 5).
We further checked whether bias-reporting could have generated those persistent periodicities.Oneway SDC analysis of daily KD per prefecture were grouped annually for the entirety of Japan, resulting in strong positive correlations close to the diagonal that we summed up (supplementary figure 5, red line; see Methods).The results show a clear amplification of the SWC 1 towards the present, with a rising trend and also an intriguing 4 year oscillation.The total yearly aggregated KD incidence for Japan is here shown for comparison (supplementary figure 5, black line), showing a steeper upward trend towards the present.These long-term patterns are incompatible with a mere weekend bias in reporting, which would have only had a dominant presence at those very short time scales of intra-weekly variation.Supplementary figure 6 (supplementary figure 7) displays the different distribution of KD occurrences (significances) throughout the week, when reported data is considered as either days of admission (DoA) or days of onset (DoO).DoO is, as a consensus, estimated by the medical community to be 5 d before the day of admission.However, our study shows a strong median value centered around 2-3 d.Furthermore, the results reveal some slight discrepancies between the WC when considered from DoA (supplementary figure 6(A)) and DoO (supplementary figure 6(B)).When split in years, the patterns found are quite consistent within the two categories, displaying the aforementioned Sunday and Wednesday-Thursday minima and the Monday-Friday maxima in DoA (compare supplementary figures 6(A) and 4(D)).When DoO is analyzed, a similar SWC emerges, despite having different magnitudes and a much weaker Sunday minimum.DoO is, however, a much more volatile value due to its subjective nature, albeit consistentcy in the appearance of the SWC 1 in both cases provides further strength to the real nature of this type of variability in KD.
As a comparison, similar representations -both of the WCs and the fundamental frequencies-were derived for other concurrent diseases in Tokyo (namely heat strokes, HST; supplementary figures 8(A), (C) and 9(A)) and influenza A; supplementary figures 8(B), (D) and 9(B)).Additionally, a number of environmental covariates were also tested for comparison (namely, temperature, absolute humidity, AQ, PM 10 , NO x and ozone; supplementary figure 9(C).None of the alternate diseases nor environmental covariates displayed a similar SWC signature to KD (supplementary figure 9).
The rate of daily presence of the SWC 1 in the LIDAR aerosols datasets was similarly extracted with the aid of SDC and is displayed in figure 4(C) for Toyama, Tsukuba and Tokyo (red bars), aligned with the base value measured from the LIDAR (absc532 as black series, serving as an analogue of the fine aerosol fraction; see Methods).These three locations are unique in that they all have routine LIDAR instruments being operated from 2010-2016 (www-lidar.nies.go.jp/AD-Net/index.html),as well as long-term KD epidemiological records.The SWC 1 presence in aerosols is increasing as we move closer to the air intrusion vertical coordinates denoted by synopticscale meteorology.The SWC 1 is roughly constant throughout the record, but displays large oscillations in its relative contribution, ranging from near 0 to around 60% and with an increasing value as we approach Tokyo that is consistent with atmospheric deposition.Clear seasonal oscillations also appear, but with maxima centered in the winter and early spring (figure 4(C)).
The two WCs are overlaid for Tokyo in figure 4(D) (blue line for columnar LIDAR up to 6 km and black for KD new cases) and a clear co-variation and slight anticipation consistently manifests.Expansion of this same analysis to cover all the available LIDAR data period (2010-2016) also displays high agreement when compared to KD new cases grouped for all Japan (supplementary figure 10(A)).The LIDAR data is again an average of the atmospheric layers from near-surface up to 6 km high and a striking and consistent 1 month lead time in both maxima and minima appears between this fraction of aerosols and KD new cases.The amount of these aerosols in particulate matter (PM) clearly grows during events associated with KD maxima (supplementary figure 10(A)) as a striking monthly positive co-evolution exists between the SWC 1 -PM 1 cycle (blue) and KD incidence (black).The presence of the 3.5 d cycle shortly anticipates KD (supplementary figure 10(A)).Synchronous annual cycles show up in both variables, with coincident maxima in the winter and minima in the summer, as expected.This large seasonal coherence among wind and SWC 1 -PM 1 further appears to reinforce the central role of aerosols in this disease.Again, source-tracking errors and sensitivities were further assessed, both by increasing and reassorting the number and type of HYSPLIT simulations conducted (see Methods and supplementary figures 11-13).Further confirmation of a higher altitude of air circulation from NE Asia is obtained for high-KD days in comparison to low-KD days (supplementary figure 13).
This surrogate variable representing the presence and movement of the SWC 1 in particles in the Tokyo air column (absc532 standing for 532 nm attenuated backscattering coefficient) is inferred from LIDAR optical measurements, and shown averaged at both the surface and above the PBL for Tokyo, Toyama and Tsukuba (supplementary figure 10(B)).A clear co-variation again shows up among the three locations when changes are studied at the seasonal scale.Month-to-month average synchronic evolution in the transport of this SWC 1 fraction of aerosols (here followed by comparing changes in the SWC 1 at different height layers in the troposphere), can be seen plotted together with monthly KD average (brown bars).This way, surface (green bars) and above the PBL (oPBL, blue bars), appear to coherently covary, showcasing the top-to-bottom movement of the SWC 1 in fine aerosols and the associated KD changes (supplementary figure 10(B)).The former might indicate an all-year round similarity, a consistent source region and the same mechanisms associated with KD at all times, with only the maritime intrusions cleansing air during times of KD minima.In fact, when inspecting in detail such a multi-scale consistency between the arrival and entry of the SWC 1 and KD occurrence in each of the three locations, an abnormally leptokurtic distribution centered at zero lag emerges (supplementary figure 14).While it is expected that this distribution of events takes place centered at zero lag, it is significantly different from a similarly computed random model with the same amount of scale comparisons (blue line in supplementary figure 14).This indicates a fast response in KD following strong air intrusions, but we should stress that DoA accounts for a potential delay of 0-5 d since exposure.The attenuated backscattering ratio between 1064 nm and 532 nm (absc1064/absc532) has indeed been described as an appropriate analogue for tracing the sizes of aerosol particles (Raut andChazette 2009, Grainger 2022), yielding values well below 1 (supplementary figure 15(A)); and, therefore, clearly denoting entrainment and dominance of particles of size below 1 µm (supplementary figure 15).This is again fully consistent among the three locations studied (supplementary figure 15(A)) and the movement of air can be tracked among different layers of the atmosphere and down to locations (supplementary figures 15(B) and 16, 17).
To expand on the above results, we applied a SDC correlation analysis in supplementary figure 16 between the same two variables for the interval 2010-2016.A strong link emerges between them that enhances towards the present, denoting a leading role for airborne particles (top series) on KD (left series; see Methods).This lead time is clearly evident as strong positive associations (red dots) denoting correlation values above +0.6 and up to +0.8 (p < 0.001) (see black box in lower panel) are present at and below the main diagonal, with the distance to the main diagonal representing the response time.A strong but non-linear relationship clearly emerges (e.g.see red striped area wandering below the main diagonal and intensifying around 2012).This same atmospheric aerosol transport over Tokyo was studied for an independent assessment also in the air column up to 6 km height and the direction of the movement traced (figure 5 and Bourgeois et al 2018).In this way, downward movement of weekly particles in the air column for 2010-2016, that arrive at Tokyo (e.g.difference calculated as before between the SCW above and below the PBL denoting the aerosol's arrival to the surface of the SWC 1 ; blue line) is shown in figure 5 (top panel) (see Methods).Wind speed (red line) and KD incidence (black line) are also depicted for inspection.A clear co-variation is evident between the SWC 1 and KD maxima cooccurring at times of maximum wind speed, and not conversely, with air column stability (supplementary figures 15-17).These maxima therefore correspond to air intrusions and rapid downward motion over Tokyo, as indicated by the bottom panel in figure 5 showing the monthly averaged LIDAR atmospheric profile for 2010-2016 (see also supplementary figure 15).High (low) seasonal concentration of aerosols exists trapped in the surface due to strong (weak) downward wind advection typically occurring in the winter (summer) since 2010 (figure 5).This weekly evolution in the three aforementioned variables further reinforces their strong co-variation (figure 5).

Discussion
Previous studies on the causes of KD have essentially addressed variability at seasonal and longer time scales while discarding any detailed analyses on the day-to-day disease and aerosol changes.The discovery of a consistent sub-weekly (SWC 1 ) signature in all the KD epidemiological records, at both the prefecture level and for the entirety of Japan, which is present continuously since the early 1970s, sheds new light on the necessary concurrent factors leading to the development of KD events.This variability has not been previously considered, or was simply interpreted as a weekend bias in reporting, as only high-frequency Rossby wave (RW SHG) events approaching 8 d have sporadically been described that generate via cascade from low-frequency waves.They occur mainly in the subtropics and therefore are extremely rare and identified in the middle atmosphere (He and Forbes 2022).Now our results clearly show that the fundamental cycle of 3.5 d does not match a mere weekend effect (figure 5, supplementary figure 10).Such co-variability cannot simply arise by chance and sustains a strong factual demonstration.
The implications of this finding are considerable as they point either to the chemical composition and/or the physical size of particles as factors important to KD epidemiology.Both features have been hypothesized -as shown above-as leading to the exacerbation of human immune responses that are also typical in KD.
The existence of a WC in human diseases -e.g. in clinical data-has long been attributed to the weekday-weekend dynamics in hospital admissions (e.g.worse prognosis for patients admitted during weekends in poor health care centers with reduced staffing levels or less experienced staff; Zare et al 2007, Cavaliere et al 2008, Uematsu et al 2016), or to lower number of admissions for mild diseases.Also, to biases in reporting (e.g.admissions being assigned to the following Monday or working day; Barnett et al 2002).Lower admission rates also occur over the weekend for asthma and mild diseases (Rosselló-Urgell et al 2004) and in prevalence surveys of nosocomial infections, due to the reduced effects of air pollution on hospital admissions and mortality (Fung et al 2003).The association of mortality with X Rodó et al day and time of admission has been tackled by many studies, with considerable heterogeneity in the results.
It is well-known that no natural WCs exist, being the weekly period one with a unique anthropogenic signature (Cerveny and Balling 1998, Gong et al 2007, Xia et al 2008), other than the atmospheric long waves, but having mostly a subtropical origin.Human activity in industrialized countries, instead, largely follows a 7 d cycle, where fossil fuel combustion is expected to be reduced during weekends (Cleveland et al 1974, Martin et al 2003, Kanda 2007, Sanchez-Lorenzo et al 2012).This weekend effect is well known from local, ground-based measurements, and may even translate into a small temperature signature, associated with a reduction in transportation (Beirle et al 2003).In fact, the influence of pollution on weather patterns is long known (going back to at least Ashworth 1929), and illustrates how weekdays are usually warmer than weekends (Cleveland et al 1974, Gordon 1994), and how atmospheric pollution shows substantial weekly variability (Wu et al 2022).
Trace elements detected in the 37 d record analyzed are enriched to varying degrees in the different particle size distributions, and varying concentrations of Zn and Pb and to a lesser degree also Mn, Ba, Se and Cu are evident with coherent PM 10 and C tot loads.This composition denotes a mixture of dust and aerosol of small sizes.A mixture of dust and pollution aerosols from different parts of East Asia (e.g.corresponding to air stagnation over central and the eastern polluted China regions, a Gobi Desert plume from the NW with agricultural debris) creates a complex air quality plume.Considering the distinct air quality phases described in this of 37 d, days with maximum KD cases occur during high continental pollution episodes.These events are mainly characterized by fine sulphatic transboundary aerosol intrusions driven by the established anticyclone, accompanied by higher ambient concentrations of metallic trace elements, such as Pb, As, Zn and Bi.In contrast, the advection of cleansing marine air, recorded by NaCl peaks, removes the Asian mainland influence and results in days with minimum KD cases.These marine aerosols were associated with minimum levels of the most toxic elements, such as As, Pb, Bi and Cd.The coarsening of aging recirculating sulphate aerosols, with low toxic metal loading, seems to be associated with minimum KD occurrence.
These trace elements in airborne PM indicate anthropogenic combustion processes (Dordevic et al 2014).Factor analysis in a study of the deposition of MM from the atmosphere in the most polluted city in China (Chengdu) indicated that As, Pb, Zn, Cu, Cd, and antimony (Sb) mainly originated from anthropogenic sources.Also, another type of sulphatic aerosol (rich in Zn and Cu) has been recognized in the finer fraction.Although As, Zn, and Cu could all result from coal combustion, as and the other MM have potentially different sources (Cheng et al 2018).
Overall, the patterns found are extremely consistent with an agricultural origin, somewhat enriched by local events adding air pollution from adjacent urban areas.Interestingly, when inspecting whether the sampled atmospheric event in Kumamoto corresponds to a high or low KD incidence episode, supplementary figure 1 clearly indicates a low-to-moderate KD incidence (see also supplementary figure 2).
Previous studies addressing a relationship between urban air pollution and KD have shown negative or inconclusive results and controversy yet remains in the scientific literature (Zeft et al 2016).However, recently, Buteau et al (2020), using a population-based cohort comprising 505 336 children and including 539 with KD, found that both prenatal exposure to ambient and industrial air pollution were associated with the incidence of KD in childhood.Similarly, Jung et al (2017) analyzing hospital admissions in Taiwan showed that O 3 was positively associated with KD in the same day of admission, as well as during earlier seasons.Our results harmonize all the former studies in this direction, thanks to the unprecedented analysis of this high-resolution 46 yr daily record of KD admissions.This should therefore stimulate further research on a potential link between KD -or with other similar diseases-and the chemical nature of aerosols as a contributing or concomitant factor.
Clear WCs of anthropogenic origin, such as for NO 2 and PM x have been described by satellite measurements (e.g.GOME; Cleveland et al 1974, Beirle et al 2003).Peaks and troughs, however, also vary among regions for cultural and even religious traditions.Some studies have equally shown WCs of tropospheric NO 2 for different regions of the world, with a clear minimum in the US, Europe and Japan and even with 35% lower NO 2 levels on Sundays than on working days in Germany (Wickert 2001).Therefore, variability at this intra-weekly scale can only be linked to an anthropogenic effect on air quality, with two maxima and two minima in both KD and aerosols.The former is clearly evident for the three locations studied (Tokyo, Toyama and Tsukuba).These same patterns are effectively traced to exist for air particles at all levels in the lower troposphere up to 6 km, being the SWC 1 enhanced at times of air intrusions (and concurrent with KD maxima).However, the fact that SWC components altogether account at most for around 20% of the overall KD variability, denotes that the factors associated with SWC 1 are, however, not sufficient to determine alone a KD outcome.The former is in vast agreement with other studies where seasonal-to-interannual changes were studied as the main source of KD variability (Ballester et al 2013, Buteau et al 2020), but our study sets the numerical basis to substantiate this assertion.For instance, while the three large epidemics might be related to climate extremes in the source region (e.g.floods or droughts), the trend in KD is possibly mostly related X Rodó et al to other processes distinct to the SWC here described (e.g.increases in cropland yield and cover; Rodó et al 2016), despite being somewhat also present in supplementary figure 5. MM contamination accumulating in soils is also a major environmental concern that affects large areas worldwide.Agricultural practices have been the main source of trace elements in soil such as Pb, Cr, As, Zn, Cd, Cu and Ni as plants can uptake these toxic MM and nitrogen fertilization also increases Cd concentrations in soil and plants (Wangstrand et al 2007).Phosphate fertilizers are also a significant source for contaminated soils with potentially toxic elements, such as Cd, F, Hg and Pb (Guo andZhou 2006, Moraes 2009) and As.Our results also confirm that the anthropogenic component of suspended PM in the atmosphere usually builds up more slowly and does not fully decline as rapidly as the mineral and organic coarse fraction dust (figure 1(B)).
Exposure to some of the MM found (e.g.Zn, Cu and Pb) have shown a high correlation with pulmonary inflammation by inhalation toxicological studies using animal models (Park et al 2018).The long-and short-term exposure to these hazardous air pollutants, individually or collectively, may cause respiratory (eye, nose, throat and sinus) irritation, allergic dermatitis and adverse cardiovascular effects (Mills et al 2009).Inhalation of air pollutants thus induces pulmonary oxidative stress (ROS) and inflammation and the presence of soluble transition metals in aerosols enhances the inflammatory responses via increased oxidative stress and the release of ROS (Davis et al 2021).Fine (PM 2.5 ) and ultrafine particles (those <0.1 µm) induce innate immune response via ROS generation by transition metals and/or polyaromatic hydrocarbons (Roberts et al 2003).They modulate both the innate and adaptive immune responses (Hollingsworth et al 2007, Mikerov et al 2008, Miyata and van Eeden 2011, Zhao et al 2014) in ways that can be similar to those observed in KD pathogenesis, although specific research along these topics is needed.

Conclusions
Our study is the first to successfully demonstrate how the close scrutiny of LIDAR profiles can be used as effective sentinels for alerting of high-risk KD events.The highly-informative real-time data that the wide AD-Net network in Japan and nearby countries provide, should be promptly used for developing this important air quality service for health.
This study establishes that KD exacerbation is associated with external air intrusions having an anthropogenic signature that might also be associated to aerosol's MM.The analysis of metal contents can also be used as an early-warning system helping in preventing high KD episodes despite this link to MM requires further research and longer-term datasets.Although causality cannot be directly inferred, strength in these results definitely point towards an unequivocal link among them all and to the involvement of tropospheric fine aerosols in the epidemiology of KD.This aerosol facilitation can at times account for around half of the total variability in this disease, despite it being most often seen to act as a secondary -albeit necessary-cofactor in KD epidemiology.While we do not dispute that a portion of the variability in disease epidemiology can be associated with other local factors increasing individual susceptibility, what is clear is that these would, in any case, be very minor when compared to this atmospheric transport from NE Asia.Transiently, different air mixing apportions take place from both highly polluted urban centers and intensive cereal cropland areas.These results point to the chemical (and possibly also the biological) composition of agricultural aerosols as a likely cause for KD and to a dose/response relationship to MM conducive to cell inflammation.However, it also remains to be investigated as other studies have suggested, whether embedded in aerosols and affected by their chemistry, or other factors, such as microbes or organic by-products (e.g.coming from pesticides or fertilizers), in genetically predisposed children, might also be playing a key role in KD pathology and in other similar vasculitis diseases.economic and enthusiastic support.X R acknowledges support from the grant CEX2018 000806S funded by MCIN/AEI/10.13039/501100011033and support from the Generalitat de Catalunya through the CERCA program.

Figure 1 .
Figure 1.Aerosol's metal concentrations and synchrony with the KD incidence in Kumamoto (Japan, see methods).(A) Coevolution between the most correlated individual MM (red) contained in aerosols and the number of accumulated KD03 cases (integrated from 0 to 3 d (black) in a sequence of 37 d) in Spring 2011(8).(B) SSA reconstructed variability at the bi-weekly timescale (left, see methodsand 19) in the case of total MM (red), total carbon (blue), PM10 (green) and KD (black).Right plot displays the same as the left one but for the weekly reconstructed components.Note consistencies and differences among phases and convoluting amplitudes of the reconstructed signatures in the different variables.Embedding dimensions for the SSA analysis were always varied to test for signal stability (p < 0.05).Concentration is in ng m −3 for METtot and in µg m −3 for Ct and PM10 but were shown in the same graph for comparison.As KD03 has accumulated cases from day 1 to day 3, no intra-weekly variability is present.

Figure 2 .
Figure 2. Cluster analysis of the metal content in aerosols and its synchrony with KD incidence in Kumamoto (A) Dendrogram generated when using Pearson's correlation as distance metric and the nearest point algorithm to compute distances across the newly formed clusters.Colors denote the different clusters assigned when setting a distance threshold of 0.3.(B) Variation during the 37 sampled dates of the (min-max normalized) concentration of MM of each cluster.For clusters with more than a single member, the thick line represents the median value of all cluster members at each time-point, with the individual metal contributions shown in the shaded thin lines.(C) Boxplots for the distribution of daily KD cases in the Kumamoto prefecture as a function of the daily concentration of Cluster A MM are categorized into three groups: low (0-114.00ng m −3 , n = 16), medium (114.01-227.00ng m −3 , n = 11) and high (227.01-341.00ng m −3 , n = 10).On the other hand, the distribution of the difference of means for the pairwise comparison of groups performing a bootstrap test, showing significant differences in KD cases at an alpha =0.05 for days with high concentrations compared to those, in both medium, and low concentrations.(D) Reported (true) KD cases in the Kumamoto prefecture (blue) and predicted cases with a Poisson model using the concentration of Cluster A MM as predictor.The shaded area represents a bootstrapped 95% CI.
Figure 3.Back trajectories for the days with either maximum or minimum in KD cases in Kumamoto.(A) 37 d when the air was sampled in Kumamoto and HYSPLIT back trajectories for both the days with maximum (red)

Figure 4 .
Figure 4.The long-term evolution of Kawasaki disease (KD) incidence for the entire of Japan (1970-2016) and contribution of the constituents of the weekly cycle.(A) Top: aggregated KD incidence for all of Japan (black) and reconstructed variability of the seasonality (red, corresponding to RC12 of the SSA decomposition; see Methods for spectral decomposition and reconstruction (19).Bottom: same as before but for the weekly cycle variability.(B).Top panel corresponds to the AR spectrum of the KD time series with FB (forward-backward) singular value decomposition algorithms.Bottom panel displays the constituent sub-weekly frequencies obtained (namely 3.5 d SWC1 and 2.33 d SWC2, respectively), and an exploration of different orders in the decomposition for stability of results.(C) Temporal evolution of the absc532 coefficient and the SWC in the three stations.Bars display the monthly percentage of days with 'SWC presence' (with an r > .4),while the black line displays the daily readings of the absorbance coefficient at 532 nm for each station.(D) Co-evolution of the weekly cycle in both the SWC of aerosols and KD in Tokyo for the interval 2010-2016.

Figure 5 .
Figure 5. Association of PM arrival (absc532) to KD cases in Japan at a weekly scale.(A) Weekly association of KD in Japan (black) with PM arrival with the SWC1 dynamics in Tokyo (blue) and wind speed (red) for the 2010-2016 period.The maxima synchronization between the PM arrival and wind speed emphasizes the external source for the aerosols.(B) Attenuated backscatter coefficient at 532 nm (absc532) vertical profile displaying the aerosol intrusions during the winter months when KD, PM arrival and wind speed have a peak.The absc532 scale has been adjusted to highlight the vertical transport from high altitudes (circa 6 km) towards the surface, which tends to be between 0-3 [Mm −1 s −1 ].The absc532 values below 2 km are of higher magnitude (∼4-8 [Mm −1 s −1 ]), greyed in this figure.