Associations of exposure to ambient PM2.5 and its major components with pulmonary tuberculosis in central and southwestern China: a spatiotemporal modeling study Authors: Xinqiang Wang, Leonardo Martinez, Lauren Linde, Qi Li, Xiaolu Wang, Jiayi Tang, XiaoHong Kan, Changhong Yang, Qu Cheng Published: 2026 ABSTRACT AbstractBackgroundChina, as well as many other populous countries in Asia, continues to face a dual burden of high pulmonary tuberculosis (PTB) incidence and severe air pollution, especially fine particulate matter with a diameter ≤ 2.5 μm (PM2.5). However, few studies have explored the effect of ambient PM2.5, especially its specific components, on the risk of developing PTB.MethodsWe obtained active PTB cases from 37 prefectures in China over a 7-year period. Prefecture-level monthly PTB counts were analyzed in relation to population-weighted mean concentrations of PM2.5 and its major components. Exposure data were derived from the Tracking Air Pollution in China (TAP) dataset and spatiotemporally linked to each case by prefecture of residence and month of diagnosis. To investigate the lagged exposure-response relationship, we applied an integrated spatiotemporal Bayesian hierarchical model within a distributed lag non-linear modeling framework. We further evaluated potential effect modification by sex, age, local PTB transmission intensity, greenness level, and temperature to assess effect heterogeneity and provide preliminary insights into potential underlying pathways. Mixed-effects meta-regression models were used to evaluate potential effect modification, incorporating interaction terms between the pollutant and each potential modifier.Results636,391 PTB cases were reported in our study area during the study period. We found positive associations between all components and PTB incidence and the effects are stronger and last longer at higher concentrations. At the 95th percentile of the concentrations the strongest effect occurring at a one-month lag and lasted for up to three months, the maximum incidence rate ratio were 1.069 (95% CI: 1.046–1.092 ) for PM2.5, 1.048 (95% CI: 1.029–1.068) for sulfate, 1.066 (95% CI: 1.040–1.092) for nitrate, 1.063 (95% CI: 1.039–1.087) for ammonium, 1.060 (95% CI: 1.038–1.083) for OM, and 1.065 (95% CI: 1.045–1.086) for BC. The peak incidence rate ratios (IRRs) per interquartile range (IQR) increase were comparable across components, although the cumulative effects of ammonium were slightly stronger than those of the other pollutants. Meta-regression analyses indicated that the effect sizes varied by sex, age group, PTB transmission intensity, greenness, and annual mean temperature. Stronger associations were observed among males, younger age groups, regions with lower transmission intensity and greenness, and those with higher annual mean temperature, although the magnitude and statistical significance of effect modification differed across components.ConclusionsExposure to ambient PM2.5 and its components may be associated with an increased risk of PTB. The observed short peak lag times and consistently larger incidence rate ratios in settings with lower local transmission intensity suggest that the increased risk may be more strongly related to disease reactivation or progression from recently acquired infections than to enhanced transmission.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12889-026-27341-0.Keywords: Pulmonary tuberculosis, PM2.5 components, Spatiotemporal Bayesian hierarchical model, Distributed lag non-linear model, Risk assessment Introduction Tuberculosis (TB) remains one of the leading causes of morbidity and mortality worldwide [1]. It has been the leading cause of death from a single infectious agent since 2015, with the exception of 2020 to 2022 when COVID-19 temporarily surpassed TB [2–4]. The global burden of TB has been further exacerbated since the COVID-19 pandemic, due to disrupted diagnostic services and treatment of TB during the pandemic [4]. From 2015 to 2023, the global incidence rate and death toll of TB decreased by only 8.3% and 23% respectively, far below the aimed reductions of 50% and 70% set by World Health Organization’s (WHO) End TB strategy for 2025 [5]. China plays an important role in achieving these goals; China ranked third globally in TB burden and contributed 6.8% (741,000 out of 10.8 million) of all new TB episodes worldwide in 2023 [5]. In addition to its high TB burden, China also experiences high levels of ambient air pollution, especially fine particulate matter with a diameter ≤ 2.5 μm (PM2.5) [6, 7]. In 2023, the annual average PM2.5 concentration in Chinese inland cities reached 32.5 µg/m3, nearly seven times higher than the WHO’s recommendation of 5 µg/m3 [8]. While associations between indoor PM2.5 from biomass combustion and increased TB risk have been previously shown [9–11], the effects of ambient PM2.5 have only more recently been examined and are currently an active area of research [12–14]. A spatiotemporal study in Lima, Peru, found that a 1 µg/m3 increase in annual mean PM2.5 concentration was associated with a 32% increase in TB risk after adjusting for poverty levels and long-term trends in TB incidence [15]. Another time series analysis in central China found that a 10 µg/m3 increase in daily PM2.5 concentration was associated with a 3.1% increase in TB risk [16]. However, most existing studies treat PM2.5 as a single chemical component, despite its complex mixture of organic matter (OM), black carbon (BC), sulfate, nitrate, and ammonium [17]. These components originate from different anthropogenic and natural sources. For example, OM and BC are directly released into the air, while sulfate, nitrate, and ammonium are formed through a series of chemical reactions in the atmosphere. Furthermore, each component may have varying levels of toxicity to human health at different temporal lags. For example, sulfate has a slightly higher hazard ratio than nitrate for all-cause mortality, with the effect of sulfate peaking at a 1-year lag, whereas the effect of nitrate is strongest in the concurrent year [18]. Such component-specific health impacts may extend to TB as well, as all five components were associated with upper and lower respiratory infections, such as nasopharyngitis, sinusitis, bronchiolitis and pneumonia, in early childhood [19]. Few studies have begun to examine the differential effects of individual PM2.5 components on PTB risk [20–26]. Among these studies, some focused on very limited spatial scales (e.g., a single prefecture [23, 24]), while others relied on coarse spatial aggregation of provincial- or state-level PM2.5 concentrations and PTB incidences [20–22, 25, 26] which may lead to substantial exposure misclassification, as within-province variability in PM2.5 exposure is considerable and the average exposure of cases may differ remarkably from the provincial mean. Moreover, none of these studies included a negative control exposure—an exposure independent of the outcome condition but related to the exposure factor and potential unmeasured confounders—to help address residual confounding [27]. Furthermore, effect modification analyses stratified by local transmission intensity may provide preliminary insights into potential underlying pathways, such as whether PM2.5 exposure is more likely to influence disease transmission, progression or reactivation. To address these gaps, we integrated a spatiotemporal Bayesian hierarchical model with a distributed lag non-linear model (DLNM) to examine the lagged effects of ambient PM2.5 and its five major components on PTB risk [18]. We further examined potential effect modification by sex, age, local PTB transmission intensity, greenness and temperature to explore potential underlying mechanisms. In addition, we incorporated negative lags (i.e., PM2.5 exposure after case reporting) as negative controls to evaluate potential residual confounding and strengthen the robustness of our findings. Materials and methods Study area We conducted a multi-center analysis across two populous Chinese provinces, Sichuan and Anhui (Fig. 1A), including a total of 37 prefectures. Both provinces are characterized by large population sizes and diverse landscapes. Sichuan has 21 prefectures with a total population size of 83 million, the 5th most populous province. Anhui has 16 prefectures with a total population size of 61 million, the 8th most populous province in China. The geographical area covered by this study is vast (approximately 29°N-34°N and 114°E-119°E for Anhui and 26°N-34°N and 97°E-108°E for Sichuan, spanning approximately 625,600 square kilometers), with diverse topography, climate zones, urbanization levels, and PTB incidence rates (Fig. 1B). Geographic location of the study area (A) and spatial variations in PTB incidence and PM2.5 concentrations averages in Anhui and Sichuan Provinces during the entire study period (B). Note: PM2.5, fine particulate matter with a diameter ≤ 2.5 μm; PTB, pulmonary tuberculosis Disease and population data PTB is a notifiable infectious disease in China, requiring all cases to be reported within 24 h after diagnosis to the web-based National Infectious Disease Reporting System, which covers approximately over 99% [28] percent of all healthcare facilities in China. We obtained the number of reported PTB cases by sex, age group, prefecture of residence, and month of diagnosis from 2013 to 2019 from Sichuan and Anhui Centers for Disease Control and Prevention. We included all reported cases residence in the study area that met the national diagnostic criteria for pulmonary tuberculosis (WS 288–2008 and its updated version WS 288–2017) based on clinical symptoms, sputum examination, and imaging findings, and had complete personal information (including sex, age, residential address, and date of diagnosis). We excluded cases with incomplete personal information and those non-definitive or suspected cases. We also obtained prefecture-level population sizes by year from Sichuan Statistical Yearbooks and Anhui Statistical Yearbooks to estimate incidence rates [29, 30]. We further obtained the 1 km population distribution dataset from LandScan [31] to calculate the population-weighted mean pollutant concentrations, normalized difference vegetation index, temperature and relative humidity for each prefecture as further described below. Environmental data PM2.5 data We obtained monthly mean concentrations of PM2.5 along with its five chemical components [i.e., sulfate, nitrate, ammonium, organic matter (OM), and black carbon (BC)] at a 10 km spatial resolution from the Tracking Air Pollution in China (TAP) dataset (http://tapdata.org.cn/). The dataset was generated by integrating multi-source information, including PM2.5 concentrations observed at ground monitoring stations, aerosol optical depth (AOD) derived from satellite remote sensing, outputs from a chemical transport model (CMAQ), meteorological reanalysis data, and geographic auxiliary variables such as land use. A two-level machine learning framework was applied to ensure data accuracy. The first-level model, based on a random forest algorithm combined with the synthetic minority over-sampling technique, focused on optimizing the identification and prediction of high-pollution events. The second-level model further refined the concentration fields by modeling the residuals between CMAQ-simulated values and ground-based observations [32, 33]. For PM2.5 components, under the constraint of total PM2.5 concentration, an extreme gradient boosting model was used to correct the relative contribution spectrum of CMAQ-simulated components, thereby deriving the concentrations of individual components. Out-of-bag cross-validation demonstrated that the TAP model’s predicted PM2.5 concentrations during 2013–2020 were in high agreement with ground-based observations (R² = 0.80–0.88, root-mean-square error (RMSE) = 13.9–22.1 µg/m³). The predicted concentrations of individual chemical components were also consistent with observations, with R² > 0.73, normalized mean bias within ± 7.07%, RMSE < 7.58 µg/m³ [32, 33]. To account for the unequal distribution of population across space, we used population-weighted mean, instead of the simple mean, concentrations of PM2.5 and its components for each prefecture. Meteorological data To control for potential confounding effects of temperature and humidity on TB risk, we obtained monthly mean temperature (MT) and dewpoint temperature from the ERA5 Monthly Aggregates dataset (https://developers.google.com/earth-engine/datasets/catalog/ECMWF_ERA5_MONTHLY). These rasters were then weighted by the LandScan 1 km raster population dataset to calculate population-weighted mean values by prefecture. These two temperature measures were further used to calculate monthly mean relative humidity (RH) based on the equation proposed by Singh et al. [34]. Greenness We used Normalized Difference Vegetation Index (NDVI), a widely adopted proxy for greenness level in previous studies [35], to examine if greenness level modifies the effects of PM2.5 and its components on the risk of PTB. We first obtained the 1 km monthly mean NDVI from Terra Moderate Resolution Imaging Spectroradiometer (MODIS) Vegetation Indices Monthly (MOD13A3) Version 6.1 (https://developers.google.com/earth-engine/datasets/catalog/MODIS_061_MOD13A3), then weighted them by population to obtain the monthly population-weighted mean NDVI for each prefecture during the study period. Statistical analyses Model structure To account for potential lagged nonlinear effects of PM2.5 and its components on the risk PTB, as well as the spatial and temporal autocorrelation between PTB incidence rates in different prefectures at different time points, we incorporated a distributed-lag nonlinear model (DLNM) into the Bayesian spatiotemporal hierarchical model. All prefecture-level regions were analyzed simultaneously within a single model, which enable robust estimates of the exposure-response relationships through borrowing information across regions and over time. Prefecture-specific random effects were included to account for spatial heterogeneity in PTB risk, while temporal trends and seasonal patterns were modeled hierarchically, allowing them to vary across prefectures but share information through the hierarchical structure. This approach pools information across regions and yields overall effect estimates informed by data from all prefectures. To address potential overdispersion in monthly PTB case counts, we modelled the number of cases in year and month for prefecture using a negative binomial distribution: where is the expected PTB case count, and is the dispersion parameter of the negative binomial distribution [34, 36]. The expected case count was further modeled as: where represents prefecture-specific intercept; represents prefecture-specific yearly random effects; represents prefecture-specific monthly effects to account for seasonal variations in PTB risk; represents the effect of Chinese New Year (CNY) on healthcare seeking and case reporting since fewer patients seek healthcare and are reported during Chinese New Year ( = 1 when the Chinese New Year of year fell in month , and 0 otherwise); represents the lagged nonlinear impact of PM2.5 and or its components, where represents the concentration of pollutants (r = 1, 2, …, 6 represents total PM2.5 and its five components, respectively) and represents the cross-basis function in DLNM; represents the nonlinear impacts of climatic factors, where represents mean temperature (MT), relative humidity (RH), or both, and represents the natural spline function; and is the population size of prefecture in year . We explored various prior specifications for , and , as well as different combinations of climatic factors , and selected the best-fitting specification based on the lowest deviance information criterion (DIC) (Table S1). In the DLNM component (i.e., ), natural cubic splines were used for both the exposure-response and the lag-response relationships. For the exposure-response relationship, three knots were included and were placed at the quartiles of the pollutant’s distribution dimensions [37]. Given the prolonged latency and slow progression of PTB, which typically occur over months, short-term lag structures may not be biologically meaningful. Accordingly, we applied monthly data with extended lag periods to capture potential delayed effects of air pollution and reporting delays [38, 39]. For the lag-response relationship, lags from − 3 (i.e., pollutant concentrations three months after the diagnosis) to 12 (i.e., pollutant concentrations 12 months before the diagnosis) months were included, with two knots placed equally across this lag period. The negative lags (i.e., cases diagnosed before the exposure) served as negative controls to help assess whether the estimated effects were caused by unmeasured confounding, under the assumption that exposures after diagnosis should not affect the outcome [40]. The lower quartiles of pollutant concentrations were used as the reference levels and the effect sizes were reported for the 50th, 75th, and 95th percentile concentrations across selected lags. Assessment of multicollinearity To assess potential multicollinearity within the single-pollutant models, we calculated the generalized variance inflation factor (GVIF) for the cross-basis matrix representing the lagged exposure to each pollutant. Given that the cross-basis functions have multiple degrees of freedom, we adjusted the GVIF by taking it to the power of 1/(2×df), making it comparable to the standard VIF metric for univariate terms [41]. All estimated adjusted GVIF were well below the common threshold of 5 (Table S2), indicating that multicollinearity was not a substantial concern within the single-pollutant models. Effect modification We assessed effect modification by sex, age groups, local transmission levels, greenness, and temperature. For categorical variables (i.e., sex and age groups), we repeated the analyses after stratifying cases by sex and age groups (0–24, 25–44, 45–64, and ≥ 65 years) [36, 42–44]. For continuous variables, such as local transmission levels, greenness (NDVI), and temperature, we included a linear interaction term between each pollutant and included one modifier at a time in the model. We reported the exposure-response relationships at low (10th percentile), median (50th percentile), and high (90th percentile) levels of each effect modifier. Local PTB transmission intensity was represented by annual mean incidence rate in individuals younger than 24 years old. Effect modification was assessed by sex and age group, reflecting established heterogeneity in tuberculosis epidemiology across demographic groups [45, 46]. For continuous modifiers, pollutant–modifier interaction terms were included, and exposure–response associations were reported at the 10th, 50th, and 90th percentiles of the modifiers to characterize variation across the distribution while avoiding instability from extreme values [47]. Additionally, to formally evaluate effect modification of the association between pollutant exposure and PTB risk, we conducted mixed-effects meta-regression analyses, which allowed us to account for the variances of the estimated effect size. For each potential effect modifier, a separate model was fitted using pollutant concentration, the modifier, and their interaction term as the independent variable, and the estimated log IRR as the dependent variable. Statistical significance of effect modification was assessed using omnibus tests of moderators. Sensitivity analyses We examined the robustness of the results to changes in the minimum lag time for negative control (-6 months), maximum lag time (6 and 9 months), the number of nodes in the lag-response relationship (2, 4, and 5 nodes), and the degree of freedom for the ns function when controlling for the effects of climatic factors (2, 4, and 5 degrees of freedom). All data processing and statistical analyses were performed using Google Earth Engine (https://earthengine.google.com) and R 4.3.1. All data and code for this study are available at https://github.com/qu-cheng/PM2.5_TB. Results Descriptive results Between 2013 and 2019, a total of 636,391 PTB cases were reported in Sichuan (407,166, 64.0%) and Anhui (229,225, 36.0%), corresponding to annual mean incidence rates of 6.1 and 4.1 per 100,000 population, respectively. In both provinces, more cases were reported among individuals older than 25 years compared to younger individuals (Table S3), with the incidence rate in the elder group being approximately three times that of the younger group. Higher incidence rates were observed in Hefei and Maanshan prefectures in Anhui, and in Aba, Ganzi and Liangshan in Sichuan (Fig. 1B). Over time, the overall incidence rates of PTB declined in both provinces and displayed seasonal peaks in spring (Fig. 2A); however, these patterns varied across prefectures (Figures S1–S2). The concentration of PM2.5 and each component also decreased gradually over time, with noticeable seasonality, reaching their lowest levels in summer and highest in winter (Fig. 2B; Figures S3–S4). The monthly mean PM2.5 concentration during the study period was 43.8 µg/m3 for Sichuan and 54.5 µg/m3 for Anhui, both exceeding the WHO guideline of 5 µg/m3 and China’s national ambient air quality standard (GB3095-2012) of 35 µg/m3 for residential areas [48]. Among the five components, nitrate and OM had relatively higher concentrations, while BC had lowest levels (Fig. 2B; Figures S3–S4). Region-specific percentile distributions (5th–95th) of each pollutant are provided in Table S5. Notably, both PTB incidence and PM2.5 concentrations exhibited pronounced seasonal and cyclical patterns, although their peak timings were not identical. To minimize potential confounding by long-term trends and seasonality, these temporal structures were explicitly controlled for in the models using seasonal terms and time-varying random effects. Strong positive correlations were observed between pollutants, while negative correlations were observed between pollutants and other environmental variables, especially mean temperature and NDVI (Figure S5). Temporal changes in (A) the incidence rate of PTB and (B) concentrations of PM2.5 and its components in the two provinces. Note: PM2.5, fine particulate matter with a diameter ≤ 2.5 μm; OM, organic matter; BC, black carbon; PTB, pulmonary tuberculosis Effects of PM2.5 and its major components According to the spatiotemporal DLNM, compared to the 25th percentile concentrations of PM2.5 and its components, the risk of having PTB increased at the 75th and 95th, but not 50th percentiles (Figs. 3 and S6). Higher concentrations were associated with longer-lasting effects, which are up to 2 months at the 75th percentiles and up to 3 months at the 95th percentiles. At the 95th percentiles, the maximum incidence rate ratio occurred at a lag of one month for all pollutants and were similar across components, which was 1.069 (95% confidence interval: 1.046–1.092) for total PM2.5, 1.048 (1.029–1.068) for sulfate, 1.066 for nitrate (1.040– 1.092), 1.063 (1.039–1.087) for ammonium, 1.060 (1.038–1.083) for OM, and 1.065 (1.045–1.086) for BC, respectively. Weak or lack of significant positive associations during the negative lag periods (i.e., lags of -3 to -1 months, when PTB cases were reported before the pollutant exposure) suggested a low likelihood of bias from unmeasured confounders (Fig. 3). Lag-response association for 50th, 75th and 95th percentile concentrations relative to 25th percentile concentrations of PM2.5 and its components. Note: PM2.5, fine particulate matter with a diameter ≤ 2.5 μm; OM, organic matter; BC, black carbon; IRR, incidence rate ratio At the exposure-response dimension, since the adverse effects of PM2.5 and its components on PTB risk lasted up to 3 months at the 95th percentile concentrations of PM2.5 and its components, we present cumulative effects aggregated over lag 0–3 months across pollutant concentrations (Fig. 4). The IRRs demonstrated non-linear increases for all components, with steeper slopes at higher concentrations. The magnitude of the effects was comparable across components, although they were slightly higher for ammonium (Fig. 4). Cumulative exposure–response relationships between PM2.5 and its components and pulmonary tuberculosis over lag 0–3 months. Note: PM2.5, fine particulate matter with a diameter ≤ 2.5 μm; OM, organic matter; BC, black carbon; IRR, incidence rate ratio Effect modifiers IRRs increased with pollutant concentrations in both males and females, but the risks were consistently slightly higher among males than females (Figs. 5A and S7-S9). When stratifying by age groups, younger age groups had higher IRRs (Fig. 5B). The differences were more pronounced for BC, OM, and sulfate, while almost the same for all age groups for nitrate and ammonium (Figures S10-S12). IRRs were higher at lower transmission intensities for all components (Figs. 5C and S13-15), suggesting that the impact of PM2.5 on PTB risk may be less likely to be driven by recent transmission and more likely attributable to the reactivation of M. tuberculosis infection. When examining the interaction between NDVI and pollutant concentrations, consistently larger IRRs were observed at lower NDVI levels across all PM2.5 components, especially for BC, OM and ammonium (Figs. 5D and S16-S18). The effects of PM2.5 and its components on PTB risk were slightly greater at higher temperatures across all components, although the differences were not statistically significant, with more notable differences observed for nitrate and OM (Figs. 5E and S19-S21). Meta-regression analyses further supported these findings (Table S4). Cumulative exposure–response relationships between PM2.5 and pulmonary tuberculosis over lag 0–3 months, stratified by sex, age, local PTB transmission intensity in the younger age group, level of NDVI and MT. Note: PM2.5, fine particulate matter with a diameter ≤ 2.5 μm; OM, organic matter; BC, black carbon; IRR, incidence rate ratio; IR, incidence rate; NDVI, normalized difference vegetation index; MT, mean temperature Sensitivity analysis Positive associations between PM2.5 and its components and the risk of PTB were consistently observed across various model specifications, including changes in the number of nodes in the lag dimension (Figures S22-S24), the degree of freedom for the ns (MT) function (Figures S25-S27), the maximum lag time (Figures S28-S30) and the minimum lag time for the negative control (Figures S31-S32). Although the magnitude of these associations varied slightly, the overall consistency supports the robustness of our findings. Discussion To the best of our knowledge, this is the first study to estimate the association between individual PM2.5 components and the risk of PTB. We found that exposure to total PM2.5 and each of its five components was associated with an increased risk of PTB, with similar cumulative effect sizes from 0 to 3 months and peaking at a one-month lag. The associations were stronger among males, younger individuals, in areas with lower local transmission intensity, lower greenness and higher mean temperature. Our results indicated a delayed effect of total PM2.5 as well as its five components on PTB risk lasting up to 3 months, with the strongest effect occurring at the one-month lag. This relatively short lag period suggests that the underlying mechanism may involve the reactivation of M. tuberculosis infection or the facilitation of disease progression among individuals with recently-acquired infections [49–51]. The absence of effect modification by local transmission intensity further supports this hypothesis, since if the pollutants affected PTB mainly through facilitating transmission, a stronger effect in areas with higher transmission intensity would be expected. Previous cohort studies also support the role played by PM2.5 in facilitating reactivation of M. tuberculosis infection. For instance, a study in China found that exposure to PM2.5 may speed up endogenous reactivation through accelerating the mutation rate of M. tuberculosis during the reactivation process in the host [52]. Similarly, a multicenter cohort study found that PM2.5 exposure was associated with an increased risk of progression from M. tuberculosis infection to PTB, but not tuberculosis transmission [53]. Previous studies have reported inconsistent findings regarding the relative risks of different PM2.5 components in relation to tuberculosis incidence. For example, nitrate was identified as the most important contributor in spatially coarse provincial-level ecological analyses in China [21, 26], possibly through mechanisms involving neutrophilic inflammation and declines in lung function, as suggested by animal experiments [54]. In contrast, a population-based study in the United States found that sulfate was the primary driver of tuberculosis incidence, accounting for approximately 76% of the population attributable fraction [20], and a multicenter study in mainland China also reported a stronger association between sulfate exposure and PTB [22]. Meanwhile, a multicenter study in Beijing identified BC and OM as the dominant components, contributing 41% and 39% of the overall combined PM2.5-associated risk, respectively [23]. Ammonium was highlighted in a multi-province study in mainland China and in another study conducted in Pakistan as being associated with the largest excess risk of tuberculosis [24, 25]. This heterogeneity across studies may stem from key contextual differences, including regional variation in pollution source characteristics (affecting component mixtures and co-pollutant profiles), atmospheric conditions influencing secondary aerosol formation (e.g., temperature and humidity) [55–58], and socioeconomic factors that modify population susceptibility [15]. In our study, the five major PM2.5 components exhibited broadly comparable associations with PTB incidence in the study area, with ammonium showing only a slightly stronger association. By providing high-resolution, component-specific assessments within a consistent spatiotemporal framework, our study suggests that under the specific pollution mixture and environmental conditions examined, no single secondary component overwhelmingly dominates the overall PM2.5-related PTB risk. These findings underscore the importance of region-specific analyses and suggest that the attributable risk of PM2.5 may be distributed across multiple co-occurring components rather than being driven by a single constituent. When stratified by sex, meta-regression analyses indicated significantly stronger associations between PM2.5 (and its components) and PTB risk among males, consistent with some prior studies [45]. The observed sex difference may reflect a combination of biological susceptibility and unmeasured behavioral factors. Biologically, sexual dimorphism in respiratory anatomy (e.g., larger airway diameters and lung volumes in males) and immune function could influence the deposition, clearance, and physiological response to inhaled pollutants [59]. Behavioral differences, such as higher smoking prevalence among males in China, may also contribute; however, this cannot be directly assessed in our study due to the lack of smoking prevalence data [60–62]. Meta-regression analyses showed that the associations between PM2.5 (and most of its components) and PTB risk were generally stronger among younger individuals, particularly those individuals younger than 25 years old which may be explained by several reasons. First, children and adolescents typically spend more time with outdoor activities and are less likely to take protective measures against air pollution due to limited awareness of its health risks; this may result in higher cumulative exposure compared to older adults [63]. Secondly, children and adolescents have smaller body sizes but higher breathing rates, which is likely to result in more inhaled pollutants per unit of body weight than adults [64]. Thirdly, their lungs and immune systems are still developing and are not as robust as those of adults, making them more susceptible to the harmful effects of air pollutants [65]. We also observed that the impact of OM and BC varied more significantly among age groups compared to sulfate, nitrate, and ammonium, with stronger effects among younger individuals. Possible explanations include, first, BC is mainly composed of extremely fine carbon cores (often existing in the form of nanoscale aggregates) and OM also contains a large number of ultra-fine particles. These particles are generally smaller than those found in secondary inorganic salts (i.e., sulfate, nitrate, ammonium), and ultra-fine particles (< 0.1 μm) have stronger penetration ability, allowing them to penetrate deeper into the developing and less defended respiratory tracts, and even circulatory system, of children [66]. Secondly, BC and the majority of OM (especially in urban areas) primarily originate from the incomplete combustion of fossil fuels (particularly diesel vehicles) and biomass, making them key markers of traffic-related emissions [26, 67]. A substantial proportion of children live or attend school near high-traffic areas, where they are disproportionately exposed to this complex mixture of traffic pollutants [26].An exception was observed for nitrate, which showed relatively higher risks among older age groups. This divergence may reflect differences in exposure settings and physicochemical properties, as nitrate is a secondary inorganic aerosol with more homogeneous regional distribution and stronger indoor infiltration, making older populations—who typically spend more time indoors—more susceptible [68, 69]. We found that higher NDVI levels were associated with a reduced impact of PM2.5 and its component exposures on PTB. Green exposure in the environment has received extensive attention in air pollution-related research as a beneficial natural indicator. It is known to have functions such as reducing noise, mitigating the hazards of high temperatures and air pollution, and promoting healthy activities [70]. Multiple studies have shown that green spaces can reduce respiratory-related mortality rates [71, 72]. A study in China had found that higher NDVI levels reduced the impact of PM2.5 on tuberculosis incidence rates [73]. Among the five components, nitrate appeared to be more strongly modified by NDVI, which might be related to its strong hygroscopicity. In areas with high vegetation coverage, surface moisture on plant leaves may further promotes the hygroscopic growth of nitrate, accelerating its deposition and removal from the air [74]. We observed that the effects of PM2.5 and its components on PTB risk appeared slightly stronger at higher mean temperature (MT). Several previous studies have found a direct correlation between PTB incidence and temperature changes [75–78]. Higher temperatures may enhance the replicative capacity of pathogens, and increased air circulation in warmer conditions may facilitate the spread of airborne pathogens [79, 80]. Among the five components, we found that nitrate and OM were more significantly affected by temperature. One possible explanation is that increased temperature accelerates the reaction between nitrogen oxides and hydroxides in the atmosphere, thereby promoting the formation of nitrate [81]. Additionally, under high temperature conditions, the rate of photochemical reactions accelerates, leading to the reaction of volatile organic compounds with oxidants to form secondary organic aerosols, thereby increasing the concentration of organics in PM2.5 [82, 83]. Besides altering pathogen properties and the composition of PM2.5, temperature also modifies human behavior. A study has shown that when being exposed to high temperatures, individuals were more likely to stay indoors, which may also increase the transmission of airborne pathogens [84]. When interpreting the lag-response associations, particularly at the temporal boundaries, consideration of methodological artifacts is important. As is known in DLNM, estimates at the extremes of the lag period can be subject to edge effects, potentially inflating significance [36, 43]. Our sensitivity analysis, which extended the minimum lag to -6 months, supported this interpretation, as the associations previously observed at the far ends of the lag period were substantially attenuated and no longer significant. While a very modest signal persisted around the − 3 months in this sensitivity analysis, it may represent residual statistical noise; nevertheless, its presence in a pre-specified negative control period cautions against a causal interpretation. Therefore, we focus our biological inference on the more stable estimates within the core lag interval. There are limitations to this study. A key limitation of this study is the lack of individual-level data on important potential confounders, such as smoking behavior, socioeconomic status, and occupation, which limits our ability to interpret the effect modification results. For example, the stronger effects observed in males may reflect true biological differences in susceptibility, but they could also be influenced by the higher prevalence of smoking and outdoor occupations among men, which we could not directly adjust for in our models. Secondly, the TB data for this study came from a passive surveillance system that relies on the existing health care system to report cases. There may have been underreporting because not all cases sought medical help or not all medical providers reported cases. We used the report date instead of the onset date, because there was a delay from the onset of the disease to seeking medical care, so the actual delay time from exposure to disease onset might have been shorter than the time we determined. The exposure assessment of pollutants was performed at the prefecture-level, with potential misclassification. The average area of cities in Anhui Province and Sichuan Province is approximately 8,756 square kilometers and 23,143 square kilometers, respectively. The internal population density and urbanization of cities in Anhui Province and Sichuan Province vary greatly. Cities in Anhui Province generally show a typical pattern of strong center and weak periphery, but the extent varies among cities, with the greatest difference between the provincial capital Hefei and Fuyang. The internal differences within cities in Sichuan Province are more complex, superimposing geographical constraints and the suctorial effect of a super-strong core. Taking the provincial capital Chengdu as an example, its ring-like structure exhibits world-class differences. In this study, we tried to address this problem by using population-weighted average concentrations. Last, due to the high correlations among PM2.5 components, we included only one component at a time to avoid multicollinearity and ensure stable estimates. While this approach allows us to assess the association of each component individually, it cannot identify their independent contributions after controlling for others, nor can it examine the potential interactions between components. Future studies employing advanced mixture modeling techniques (e.g., Bayesian Kernel Machine Regression) are needed to address this question. As a large-scale ecological study involving two populous provinces and 636,391 cases, our findings suggest that the increased PTB risks may be more strongly related to disease reactivation or progression from recently acquired infections than to enhanced transmission. However, due to the inherent limitations of ecological studies, future individual-level epidemiological or experimental studies will be essential for establishing causality and elucidating underlying biological pathways. Building on these findings, several avenues warrant further investigation. First, to evaluate the generalizability of our results, this analytical framework can be replicated in diverse geographic and socioeconomic settings with varying PM2.5 composition and TB transmission intensity. Second, to examine the complex joint effects and potential interactions among PM2.5 components, advanced mixture analysis methods, such as Bayesian Kernel Machine Regression, could be applied in future studies. Third, to overcome the inherent limitations of ecological studies, individual-level studies incorporating personal exposure assessment, together with detailed information on socioeconomic status, occupation, and behavioral factors are needed. In particular, future studies could attempt to differentiate primary and reactivated PTB cases through whole genome sequencing and examine the effects of PM2.5 components separately in these groups, which would provide stronger evidence regarding potential underlying pathways. Finally, given the high correlations among PM2.5 components in real-world data, experimental studies using cellular or animal models may help disentangle their independent biological effects. Conclusion In summary, we found that exposure to PM2.5 and its components was significantly associated with an increased risk of PTB. Stronger associations were consistently observed among males and younger individuals, indicating potential vulnerability in these subpopulations. Higher levels of greenness appeared to attenuate the adverse effects of air pollution, whereas higher mean temperature was associated with slightly stronger associations. The observed patterns are more consistent with mechanisms related to the reactivation of Mycobacterium tuberculosis infection or the acceleration of disease progression among infected individuals, rather than with enhanced recent transmission, although causal pathways cannot be definitively established in this observational study. Overall, these findings suggest that reducing ambient air pollution, increasing urban greenness, and prioritizing protective strategies for susceptible populations may contribute to lowering PTB burden. However, well-designed intervention and longitudinal studies are needed to confirm these potential public health benefits. Supplementary Information Supplementary Material 1 Acknowledgements We thank the Sichuan and Anhui Provincial Centers for Disease Control and Prevention for their support in data acquisition. Abbreviations Black carbon Chinese New Year Deviance information criterion Distributed lag non-linear model Mean temperature Organic matter Fine particulate matter with a diameter ≤ 2.5μm Pulmonary tuberculosis Tuberculosis Tracking Air Pollution in China World Health Organization’s Authors’ contributions **Xinqiang Wang** : Visualization, Software, Data curation, Formal analysis, Writing – original draft. **Leonardo Martinez** : Writing – review and editing. **Lauren Linde** : Writing – review and editing. **Qi Li** : Writing – review and editing. **Xiaolu Wang** : Writing – review and editing. **Jiayi Tang** : Writing – review and editing. **Xiaohong Kan** : Data curation, Resources. **Changhong Yang** : Data curation, Resources. **Qu Cheng** : Supervision, Methodology, Writing – review and editing. Funding This work was supported by National Natural Science Foundation of China [grant number: 82204110]. Data availability The datasets analyzed during the current study are available at https://github.com/qu-cheng/PM2.5_TB. Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement The datasets analyzed during the current study are available at https://github.com/qu-cheng/PM2.5_TB.