Heterogeneous impact of fine particulate matter constituents on pulmonary tuberculosis onset: a multicenter time series study in Beijing

Abstract

Background

Epidemiological studies have implicated exposure to PM₂.₅ in the development of pulmonary tuberculosis (PTB); however, the key constituents driving this effect have not been clearly identified.

Methods

A time-series analysis spanning 2019 to 2023 was performed across several centers in Beijing to assess the links between major PM_2.5 constituents and PTB risk. The effects of five specific components—namely organic matter, black carbon, nitrate, sulfate, and ammonium—were evaluated to pinpoint the most influential factors.

Results

All five examined components demonstrated significant relationships with an elevated risk of PTB. Associations were not statistically significant on the same day (lag 0) or the next day (lag 1) after exposure. A clear risk increase was detected starting at a 2-day lag, which was no longer observable by lag 3. Per interquartile range (IQR) rise in the 3-day moving average (lag 0–2) of black carbon and organic matter, the relative risks (RRs) for PTB were 1.11 [95% confidence interval (CI): 1.03, 1.19] and 1.11 (95% CI: 1.03, 1.21), correspondingly. Together, these two components were the dominant drivers of the overall PM₂.₅ effect, contributing 41% and 39% of the joint risk, respectively.

Conclusions

The results yield novel evidence that exposure to certain PM_2.5 constituents is demonstrated to elevate PTB risk, wherein black carbon and organic matter are established as the principal factors.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-025-25116-7.

Introduction

Despite considerable progress in control strategies, tuberculosis continues to pose a formidable global health threat and socioeconomic burden as the leading cause of mortality from infectious diseases [1, 2]. While UN Member States and the World Health Organization (WHO) aim to meet the 2030 End Tuberculosis Strategy targets, global tuberculosis cases reached a record high of 8.2 million in 2023, marking the highest level since the WHO began its surveillance [3–5]. China has implemented WHO-recommended measures such as the TBIMS, yet it accounts for 6.8% of global cases, ranking third in the world [5].

Being a megacity with over 20 million residents, Beijing faces inherent public health challenges due to its high population density and extensive mobility [6]. The emergence of drug-resistant PTB, particularly multidrug-resistant PTB, exacerbates the tuberculosis burden by increasing treatment complexity, duration, and cost [7]. This places significant strain on individuals, families, and the healthcare system. Several challenges persist in effectively reaching and managing key vulnerable populations, including migrants, university students, and the elderly [6].

Fine particulate matter (PM_2.5) contains a diverse array of chemical compounds, both organic and inorganic [8]. Its major components are organic matter, black carbon, sulfate, nitrate, and ammonium [8]. Over recent decades, growing epidemiological evidence has consistently demonstrated a positive association between PM_2.5 exposure and an increased risk of PTB [9–12]. This establishes PM_2.5 as a significant environmental risk factor for the development of PTB. Nevertheless, scholarly attention has historically centered on the total concentration of PM_2.5, while the distinct health effects of its specific constituents have received comparatively little scrutiny. This gap is critical because toxicological findings suggest that the distinct physicochemical properties of individual PM_2.5 constituents may elicit divergent biological responses, potentially resulting in heterogeneous health outcomes [13]. To date, few studies have dissected the component-specific risks of PM_2.5 in relation to infectious respiratory diseases like PTB. Most prior studies either aggregated PM_2.5 as a whole or focused only on a limited number of components, often without accounting for high correlations among constituents or using methods robust to multicollinearity. This lack of component-specific evidence represents a significant knowledge gap, hindering the identification of the most toxic elements and the development of targeted air pollution control strategies.

To address this knowledge deficit, we performed a multi-center temporal analysis in Beijing to evaluate the connections of five key PM_2.5 components with pulmonary tuberculosis incidence. To our knowledge, no previous research has: (1) concurrently estimated the short-term impacts of these five major PM₂.₅ constituents on PTB risk by leveraging a high-resolution, city-scale dataset; (2) applied advanced mixture modeling methods—including Weighted Quantile Sum regression and Quantile-Based g-Computation—to evaluate their combined effects and pinpoint the most detrimental components; or (3) established exposure-response thresholds for individual species, thereby offering practical evidence to inform public health policy.

Methods

Study area

Beijing, a municipality in northern China, is located on the North China Plain and has a total area of 16,410 square kilometers (Supplementary Fig. 1). The city experiences a typical continental monsoon climate with a registered population of 21.832 million by the end of 2024 [14].

Study design

This study employs a time-series design to examine the relationship between day-to-day levels of PM_2.5 and five of its key components and the daily incidence of PTB. The analytical model adjusts for time-varying confounders, including ambient temperature and relative humidity, as well as day-of-week and holiday effects. Slowly varying confounders, such as seasonality and long-term temporal trends, are controlled using smooth functions of time. This framework enables the evaluation of both immediate and delayed associations between PM_2.5 (both total mass and chemical constituents) and PTB onset by examining single-day exposure effects (single-lag) and multi-day cumulative effects (cumulative-lag).

Study population

All patients with an initial PTB diagnosis were identified from the records of Beijing Friendship Hospital, Capital Medical University and Beijing Chest Hospital, Capital Medical University during the 2019–2023 period, comprising the final study cohort. PTB diagnosis was made per the Health Industry Standard of the People’s Republic of China: Diagnosis of Pulmonary Tuberculosis (WS 288–2017; available at www.nhc.gov.cn/wjw/wsbzxx/wsbz.shtml). The onset of PTB was defined as the date the patient first reported symptoms related to PTB, including but not limited to cough, fatigue, and fever. Inclusion criteria were: (1) initial diagnosis of active PTB confirmed by bacteriological tests (smear or culture) or clinical/radiological evidence according to WS 288–2017; and (2) first visit or admission after onset. Exclusion criteria were: (1) recurrent PTB cases (reactivation or reinfection); (2) duplicate records caused by various operational errors; and (3) patients transferred from other healthcare facilities. After applying these criteria, the final analytical cohort comprised 30,898 cases. A flow diagram detailing the inclusion and exclusion process is provided in Supplementary Fig. 2. Demographic and clinical information, such as age and sex, was also collected for further analysis [15, 16].

Exposure assessment

Daily concentrations of PM_2.5 and its five chemical constituents—organic matter, black carbon, nitrate, sulfate, and ammonium—were obtained from the Tracking Air Pollution in China (TAP) database (http://tapdata.org.cn) [17, 18]. Characterized by a 10-km resolution, the dataset contains daily concentrations of total PM_2.5 and five specific chemical species, covering the period from 2000 to the present. By integrating WRF-CMAQ model simulations, ground-based observations, machine learning-based bias correction, and multi-source fused PM_2.5 mass concentration data, it significantly improves estimation accuracy. Validation against independent observations shows correlation coefficients of 0.64–0.75 at the monthly scale and 0.67–0.80 at the daily scale, with most normalized mean biases within ± 20% [18]. This dataset provides a reliable representation of chronic changes in fine particulate matter composition, in addition to its capacity to resolve daily variability and geographic heterogeneity in high-pollution periods. It is suitable for multidisciplinary research in climate, health, and policy assessment. To minimize potential bias from extreme values in pollutant concentrations, the top 2.5% of daily concentration values were excluded prior to the formal analysis.

To adjust for the influence of climate factors and coexisting air pollutants on the links connecting specific PM_2.5 components with PTB, temperature and dew-point measurements were sourced from the ERA5, the fifth-generation high-resolution climate reanalysis produced by the European Centre for Medium-Range Weather Forecasts [19]. Daily RH values were computed for each grid cell using corresponding daily temperature (T) and dew-point temperature (DT) estimates based on the following equation [20]:

Across temperatures from − 40℃ to 50℃, this approximation produces a relative error that remains below 0.384% [20]. Additionally, daily concentrations of sulfur dioxide (SO₂), nitrogen dioxide (NO₂), carbon monoxide (CO), and ozone (O₃) were obtained from the China High Air Pollutants (CHAP) dataset [21, 22].

To ensure representative city-level exposure estimates, gridded data were spatially aggregated by calculating the arithmetic mean of all grid cells whose centroids fell within the administrative boundaries of Beijing, as defined by the TAP, ERA5, and CHAP datasets [23].

Statistical analyses

Effects of individual constituents

Initial analysis employed descriptive statistics to outline the basic characteristics of the daily PTB case series, air quality metrics, and weather variables (Supplementary Fig. 3). To estimate the associations between daily PM₂.₅ total mass (and its constituents) and the risk of PTB onset, we employed generalized additive models (GAMs) with a negative binomial distribution. Consistent with previous studies [24–26] on fine particulate matter constituents, we initially assumed linear exposure-response relationships and modeled the associations using linear terms for both PM₂.₅ mass and each constituent. To explore potential lag effects, we assessed exposures at single-day lags (0 to 3 days) as well as moving averages over multiple days (lags 0–1, 0–2, and 0–3). For each exposure metric, the single-day lag associated with the largest relative risk, the lowest Akaike Information Criterion (AIC) value, or both, was selected for subsequent analyses. To assess potential non-linear exposure-response relationships, the linear terms were replaced with natural cubic splines with 4 degrees of freedom (df). We quantitatively described the observed ‘threshold effect’ by reporting the approximate concentration at which the risk increase began to plateau, along with its associated RR and 95% CI. The general model was specified as follows:

Log[E(Y_t)]= α + β(PM_2.5 chemical constituents) + ns(Time, df = 7/year) + Dow_t + public holiday + ns(temperature, df = 6) + ns(relative humidity, df = 3),

where E(Yₜ) is the expected daily count of PTB onset. The intercept is denoted by α. The coefficient β estimates the log-relative risk (log-RR) of PTB onset per interquartile range (IQR) increase in a specific PM_2.5 chemical constituent. To adjust for temporal patterns and seasonal effects, calendar time was modeled with a natural cubic spline function (ns) incorporating 7 degrees of freedom annually. To adjust for meteorological influences, natural splines were applied: a 6-degree-of-freedom function modeled the 6-day moving average of temperature, while relative humidity was captured using a 3-degree-of-freedom spline based on its 3-day moving average. Day-of-week effects (Dowₜ) and public holidays (Holidayₜ) were included as categorical and binary variables, respectively. Dowₜ is modeled with a binary variable (1: weekdays; 0: weekends) to account for weekly variation in medical service availability. This approach aligns with established methodologies [27–31].

To assess potential differences in population susceptibility, separate regression models were developed for stratification by age (< 60 and ≥ 60 years) and sex. Differences in stratum-specific results were compared for statistical significance using two-sample z tests with the equation [32]:

where ₁ and ₂ are regression coefficients by strata; and ₁² and ₂² are the standard errors [32].

Effects of joint exposure to different constituents

To evaluate the combined effects of multiple PM_2.5 constituents, we applied weighted quantile sum (WQS) regression, a method designed to assess mixture-outcome associations while addressing multicollinearity [33]. The WQS approach combines multiple exposures into a composite index via weighted summation. The analysis involved three key steps: (1) randomly splitting the data into a training set (40%) for model development and a validation set (60%) for evaluation, with exposures categorized into quartiles; (2) generating 100 bootstrap samples from the training set to estimate component weights using maximum likelihood, with final weights derived from their average across iterations; and (3) constraining all components to show positive associations with the outcome for interpretability [34, 35]. The WQS regression yielded two primary results: (1) constituent-specific weights indicating their relative contributions, with significance determined by exceeding the threshold (1/total number of components), and (2) the overall mixture effect estimate on PTB onset. A comprehensive description of the WQS regression methodology is included in Supplementary Methods 1.

Sensitivity analyses

First, we conducted an over-dispersion test and selected the optimal model by comparing three Poisson-family models (Poisson, quasi-Poisson, and negative binomial) based on their performance in handling over-dispersion and overall goodness-of-fit [36]. Second, to evaluate the robustness of the temporal smoothing function in capturing seasonal and long-term trends, we conducted a sensitivity analysis by varying the df from 6 to 10. Third, in a sensitivity analysis to examine potential day-specific effects, we also treated the day of the week as a seven-level categorical variable (Monday through Sunday) instead of a binary variable (weekday/weekend). Fourth, to address potential confounding by total PM₂.₅ mass in the estimation of constituent-specific effects, we conducted sensitivity analyses by adding concurrent PM₂.₅ mass to the constituent-specific models (constituent-PM_2.5 models). Fifth, to further account for co-exposure confounding, we incorporated additional pollutants (SO₂, NO₂, CO, and O₃) into both the PM_2.5 total mass and constituent models through two-pollutant analyses. Finally, we employed quantile-based g-computation (QGC) to further assess the joint effects of the mixture of five PM_2.5 constituents. QGC retains the interpretability of WQS regression while relaxing the assumption of unidirectional effects. In the QGC approach, weights can indicate both positive and negative associations, and the sums of the positive and negative weights are separately scaled to a value of 1. Importantly, comparisons between weights should only be made within the same direction (positive or negative), as direct comparisons across opposite directions are not meaningful.

All statistical analyses were performed using R software (Version 4.4.1, R Foundation for Statistical Computing, Vienna, Austria), utilizing the “dlnm”, “gWQS”, and “qgcomp” packages. A two-sided significance threshold of α = 0.05 was applied to all statistical tests. To enable comparison among PM_2.5 constituents, effect estimates were expressed as relative risks (RRs) with 95% confidence intervals (CIs) per interquartile range rise in concentration, reflecting the change in PTB risk.

Results

Descriptive statistics

Between January 2019 and December 2023, 34,568 patients were diagnosed with pulmonary tuberculosis (PTB) at the participating institutions. After excluding cases involving duplicate records, recurrent episodes (e.g., reactivation or reinfection), and transfers from external healthcare facilities, the final analytical cohort comprised 30,898 cases from two tertiary hospitals (Supplementary Fig. 2). Supplementary Table 1 describes the demographic profile of the study participants. The mean age of the participants was 50.3 years (SD: 19.7), with 62.2% (n = 19,214) aged below 60 years. A male predominance was observed, with males representing 58.8% (n = 18,155) of the cohort.

Table 1 provides a summary of ambient levels of PM_2.5 and its principal chemical components. The mean PM_2.5 mass concentration was 37.7 µg/m³ (SD: 29.5). The mean concentrations of individual components were as follows: organic matter (7.2 ± 5.3 µg/m³), black carbon (1.3 ± 0.9 µg/m³), nitrate (7.1 ± 6.5 µg/m³), sulfate (5.2 ± 4.1 µg/m³), and ammonium (4.5 ± 4.0 µg/m³).

Table 1.

Descriptive Statistics for Concentrations of Air Pollutants and Meteorological Factors on the Day of Exposure (Lag 0 Day) in Beijing, China, 2019-2023

Variable	Mean	SD	Minimum	P 25	Median	P 75	Maximum
PM2.5 (µg m−3)	38.9	30.1	1.0	17.7	30.8	50.9	225.8
Organic matter (µg m−3)	8.6	6.2	0.3	4.3	7.2	11.4	55.2
Black carbon (µg m−3)	1.6	1.2	0.1	0.8	1.3	2.0	11.6
Nitrate (µg m−3)	7.8	8.0	0.1	2.1	5.4	10.7	56.0
Sulfate (µg m−3)	5.4	4.6	0.1	2.0	4.1	7.3	41.8
Ammonium (µg m−3)	4.8	4.8	0.1	1.3	3.3	6.5	32.7
SO2 (µg m−3)	7.3	3.5	2.8	5.1	6.4	8.4	32
NO2 (µg m−3)	34.1	15.7	5.2	22.2	30.6	43.3	97.8
CO (µg m−3)	0.7	0.3	0.2	0.5	0.7	0.9	3.2
O3 (µg m−3)	105.3	56.6	8.4	63.6	93.2	140.3	323.1
Temperature (℃)	13.0	11.6	−14.1	2.0	14.0	23.7	32.9
Relative humidity (%)	52.1	19.2	10.6	36.9	50.6	68.0	95.5

SD standard deviation, P₂₅ the 25^th percentile, P₇₅ the 75^th percentile, PM_2.5 fine particulate matter, SO₂ sulfur dioxide, NO₂ nitrogen dioxide, O₃ ozone, CO carbon monoxide

Correlation analysis revealed strong linkages of PM_2.5 total mass with its key chemical components, showing Spearman’s correlation coefficients (r) between 0.92 and 0.95 (Supplementary Table 2). Similarly, high correlations were observed among the constituents (r = 0.86–0.97).

Single-Pollutant exposure effects

Significant positive relationships were observed for both the total mass of PM_2.5 and its individual components in relation to PTB onset (Fig. 1 and Supplementary Table 3), all showing comparable lag patterns. Notably, no statistically significant associations were detected either on the concurrent day (lag 0) or on the previous calendar day (lag 1). Effects became significant only at lag 2 day, disappearing again by lag 3 day. Combined with the smallest Akaike’s information criterion value, subsequent analyses report results for the cumulative lag period of 0–2 days (Supplementary Table 4).

Fig. 1.

RR and 95% CIs for PTB per IQR Rise in PM_2.5 Total Mass and Individual Components at Different Time Windows Note: Solid squares represent point estimates of the relative risk for pulmonary tuberculosis corresponding to an IQR elevation in the levels of (A) PM₂.₅ total mass, (B) organic matter, (C) black carbon, (D) nitrate, (E) sulfate, and (F) ammonium. Error bars show 95% confidence intervals. IQR values are as follows: PM₂.₅, 31.5 µg m⁻³; organic matter, 6.6 µg m⁻³; black carbon, 1.2 µg m⁻³; nitrate, 8.1 µg m⁻³; sulfate, 5.1 µg m⁻³; ammonium, 4.9 µg m⁻³. Abbreviation: RR, relative risk; CIs, confidence intervals; PTB, pulmonary tuberculosis; IQR, interquartile range; PM_2.5, fine particulate matter

An interquartile range (IQR) elevation in the 3-day moving average (lag 0–2) concentration of each pollutant was associated with the following relative risks for PTB: PM₂.₅, 1.08 (95% CI: 1.01, 1.16; IQR = 31.5 µg/m³); organic matter, 1.11 (95% CI: 1.03, 1.21; 6.6 µg/m³); black carbon, 1.11 (95% CI: 1.03, 1.19; 1.2 µg/m³); nitrate, 1.10 (95% CI: 1.02, 1.17; 8.1 µg/m³); sulfate, 1.10 (95% CI: 1.03, 1.19; 5.1 µg/m³); and ammonium, 1.18 (95% CI: 1.01, 1.16; 4.9 µg/m³) (Supplementary Table 3). Supplementary Table 5 presents corresponding risk estimates for fixed increments of 10 µg/m³ total PM_2.5 and 1 µg/m³ for each chemical constituent.

For the lag 0–2 days exposure window, all exposure–response relationships exhibited concentration-dependent increases (Fig. 2). While generally linear in form, these associations consistently demonstrated threshold effects. The identified threshold concentrations and the corresponding RRs were as follows: PM₂.₅ (22.42 µg/m³, RR = 1.19, 95% CI: 0.98, 1.44), organic matter (5.16 µg/m³, RR = 1.34, 95% CI: 1.11, 1.63), black carbon (0.91 µg/m³, RR = 1.35, 95% CI: 1.10, 1.65), sulfate (2.94 µg/m³, RR = 1.24, 95% CI: 1.02, 1.51), and ammonium (12.33 µg/m³, RR = 1.27, 95% CI: 0.98, 1.64). For organic matter, black carbon, and sulfate, the RRs at their respective thresholds were statistically significant. No significant threshold was identified for nitrate.

Fig. 2.

Exposure-Response Curves for the RR and 95% CIs of PTB Onset Associated with PM_2.5 Total Mass and Individual Components over Lag 0–2 Days Note: Solid lines represent RR point estimates for PTB corresponding to a rise of one IQR in concentration for each pollutant: (A) PM₂.₅ total mass, (B) organic matter, (C) black carbon, (D) nitrate, (E) sulfate, and (F) ammonium. Dashed lines denote 95% confidence intervals. IQR values are as follows: PM₂.₅, 31.5 µg m⁻³; organic matter, 6.6 µg m⁻³; black carbon, 1.2 µg m⁻³; nitrate, 8.1 µg m⁻³; sulfate, 5.1 µg m⁻³; ammonium, 4.9 µg m⁻³. Abbreviation: RR, relative risk; CIs, confidence intervals; PTB, pulmonary tuberculosis; PM_2.5, fine particulate matter; IQR, interquartile range

To assess the links of PM_2.5 and its components with PTB incidence among different patient subgroups, stratified analyses were performed (Supplementary Tables 6 and 7). Comparisons across strata revealed no statistically significant variations.

Multi-Pollutant Co-Exposure outcomes

The joint effect of exposure to the mixture of five PM_2.5 constituents showed a statistically significant increase in the risk of PTB. Specifically, a one-quartile rise in the WQS mixture index was associated with a relative risk of 1.04 (95% CI: 1.01, 1.07) for PTB onset. As shown in Fig. 3, the constituent weights within the mixture varied substantially, with black carbon emerging as the most influential contributor (weight = 0.41), followed by organic matter (weight = 0.39). The three remaining ionic constituents contributed minimally to the overall effect (weights < 0.10).

Fig. 3.

Relative Contribution of Specific PM_2.5 Components to PTB Risk Note: The relative weight assigned to each component by the WQS regression is shown by bar length, with specific values displayed on the x-axis and constituent names on the y-axis. Abbreviation: PM_2.5 x, WQS weighted quantile sum

Sensitivity analyses 

Model development was informed by results from the overdispersion test (Supplementary Table 8). Our sensitivity analysis of the temporal smoothing function for seasonal and long-term trends revealed that the model’s key findings remained consistent across all tested values (df = 6–10) (Supplementary Table 9). Sensitivity analysis treating the day of the week as a seven-category variable yielded highly consistent estimates for the associations between PM_2.5 (both total mass and individual constituents) and PTB onset, indicating that the observed associations were robust to the modeling approach for temporal day-of-week variation (Supplementary Table 10). Two-pollutant model results are presented in Supplementary Table 11. The links between PM_2.5—including both its overall mass and specific chemical components—and PTB onset were largely consistent after accounting for concurrent exposure to additional pollutants. Based on quantile-based g-computation, a per-quartile elevation in the combined pollutant mixture yielded a relative risk of 1.04 (95% CI: 1.02, 1.07) for PTB onset. Black carbon and organic matter were the most significant contributors to this association, corroborating the main findings (Supplementary Fig. 4).

Discussion

Ambient air pollution constitutes a significant global health hazard, contributing substantially to the disease burden worldwide [37]. This multi-center time series analysis at the city level assessed the distinct relationships between PM_2.5, its key chemical components, and elevated PTB risk. The findings showed similar time-lagged effect patterns and exposure-response curves shared by PM_2.5 and its five major chemical constituents. Of these components, black carbon and organic matter showed particularly strong associations with PTB risk. These findings offer critical evidence for targeted air pollution control strategies and evidence-based public health interventions.

Prior studies have largely focused on the relationship of overall PM_2.5 levels with the incidence or recurrence of pulmonary tuberculosis. For instance, a time series study in Chongqing, China, reported that short-term PM₂.₅ exposure significantly increased the risk of PTB onset, with identifiable lag effects [9]. A comparable population-level retrospective analysis from Zhengzhou, Henan Province, China, suggested that chronic exposure to PM_2.5 could increase the likelihood of PTB retreatment [10]. However, existing research has not examined the relationships of individual PM_2.5 constituents with PTB risk. Providing the first strong epidemiological support, this multi-center time-series study demonstrates that prolonged exposure to PM_2.5 and its five major chemical constituents is associated with an elevated likelihood of PTB onset. Furthermore, the analysis identified the components contributing most substantially to the observed joint effects.

It is noteworthy that the most significant associations between PM₂.₅ constituents and PTB onset were observed at short-term lags (lag 2 day), which may seem paradoxical given the typically prolonged incubation period of PTB. However, this short-term risk increase is more likely to reflect an acute effect of environmental exposure on individuals with pre-existing latent tuberculosis infection (LTBI), rather than indicating new disease acquisition [38, 39]. In high-burden settings like ours, the prevalence of LTBI among adults is substantial. Short-term peaks in PM₂.₅ and its toxic components (e.g., black carbon) can act as potent environmental stressors that trigger the reactivation of latent infection or precipitate the acute exacerbation of subclinical symptoms, leading to healthcare-seeking and diagnosis [15]. For instance, PM_2.5 could suppress the function of alveolar macrophages and disrupt cytokine secretion, thereby weakening the lung’s ability to clear pathogens [40]. The 48-hour timeframe (lag 2) aligns closely with the peak of these acute immunomodulatory and inflammatory processes. Components such as organic matter and black carbon may act as carriers for pathogens or adjuvants that promote the infection process of Mycobacterium tuberculosis, with the maximal immunosuppressive effect occurring approximately 48 h post-exposure [41]. Furthermore, secondary inorganic components like nitrate, sulfate, and ammonium may exacerbate immune dysregulation by inducing oxidative stress and mitochondrial dysfunction [42]. Therefore, the significant association at lag 2 is not only statistically sound but also highly consistent with the biological pathway whereby PM_2.5 promotes PTB infection through acute immunosuppression and inflammatory mechanisms. The limitations of GAM in addressing multicollinearity and implications for interpretation of constituent-specific effects are provided in Supplementary Discussion 1.

Previous studies have largely assessed the effects of individual PM₂.₅ constituents using single-pollutant models [43, 44]. However, such approaches do not account for collinearity due to simultaneous exposure to multiple pollutants. To address this limitation, this study employed WQS regression [33] and QGC [45] to assess the combined impact of the mixture and to estimate the relative influence of its individual constituents. Consistent with existing literature, the results indicate that black carbon and organic matter are the primary contributors to the onset of PM_2.5-related PTB [46]. These components, primarily emitted from combustion sources, traffic, and industrial activities, exhibit heightened toxicity due to their ability to penetrate deeply into the lungs and adsorb harmful substances [47, 48]. These components interact with multiple pathological pathways in respiratory diseases. Black carbon activates oxidative stress and H2A.X, reducing alveolar epithelial cell viability [49] while also influencing asthma through metabolic and protein pathways [50]. It correlates with altered respiratory microbiota and increased infection risk [51], induces immune-inflammatory responses [52], and promotes lung cancer by creating an immunosuppressive microenvironment [53, 54]. Epidemiologically, black carbon is linked to pulmonary inflammation, emphysema, asthma, and tuberculosis, contributing to over 85% of PM_2.5-related TB morbidity [46, 55–57]. Similarly, organic matter (25–50% of PM_2.5) induces DNA damage [58], alters tumor-associated biomolecules [59], and disrupts the function of bronchial epithelial cells [60]. Collectively, carbonaceous particles exhibit stronger pulmonary effects than other components of PM_2.5. Rationale for Using Weighted Quantile Sum regression and Quantile-Based g-Computation are provided in Supplementary Discussion 2. QGC results are analyzed in Supplementary Discussion 3 and persistent challenges of multicollinearity in mixture methods are provided in Supplementary Discussion 4.

Our findings indicated that exposure to PM_2.5 and its five major components correlated with a rising trend in PTB onset, with a threshold effect noted at moderate exposure levels. Notably, the thresholds for carbonaceous components like Organic matter and black carbon were relatively low, yet the risks at these points were significant. This suggests that even at moderate pollution levels, these toxic components can substantially increase the risk of PTB onset, likely by acting as potent triggers for the acute exacerbation of subclinical infections or reactivation of latent tuberculosis [46, 47]. A time-series analysis across 184 urban centers in China likewise found a leveling-off of the exposure-risk association at intermediate PM_2.5 concentrations [61]. The observed threshold effect may result from diminishing risk increases at higher pollutant concentrations, potentially due to underlying biological mechanisms, complex interactions among mixed pollutants (including synergistic or antagonistic effects) that may offset individual risks, or artifacts arising from skewed exposure distributions commonly encountered in epidemiological studies [61–63]. Further details are provided in Supplementary Discussion 5. The identified thresholds, particularly for Organic matter and black carbon which are primarily from combustion sources, offer valuable benchmarks for refining air quality management strategies and setting targeted control goals. Further research should focus on verifying the proposed thresholds across varied demographic and geographic contexts.

Stratified analyses revealed no statistically significant differences across subgroups. This lack of significance is likely due to limited statistical power resulting from sample size constraints, which may have hindered the ability to detect actual subgroup differences [64]. Furthermore, reliance on Beijing’s overall average PM_2.5 component concentrations may introduce exposure misclassification, failing to account for individual variability in exposure [65, 66].

The findings of this study hold substantial implications for stakeholders in public health and healthcare policy. Clinicians should incorporate air quality exposure history, especially regarding PM_2.5, into standard clinical assessments, particularly during episodes of high pollution. This approach would support more precise PTB prevention strategies and enable the development of individualized treatment protocols. For at-risk populations, educational efforts should emphasize the respiratory hazards associated with PM_2.5—especially its carbonaceous components—while promoting protective behaviors, including minimizing outdoor activities during pollution episodes and the appropriate use of indoor air purification systems [67]. Policy interventions should prioritize emission reduction strategies that target combustion-related sources, given the substantial contribution of biomass and fossil fuel combustion to carbonaceous particulate matter. Additionally, formulating population-specific protection protocols for vulnerable groups warrants urgent attention from public health authorities. Implementing these multidimensional strategies could significantly reduce the respiratory disease burden associated with particulate air pollution. Furthermore, during severe outdoor air pollution episodes, public adherence to health advisories recommending reduced outdoor activity may lead to increased indoor crowding and prolonged close contact, potentially facilitating respiratory transmission of tuberculosis via droplets. Therefore, from a public health perspective, even when following “stay-at-home” recommendations to reduce personal exposure to outdoor pollution, individuals should remain vigilant about the risk of indoor tuberculosis transmission and adopt preventive measures such as ensuring adequate indoor ventilation, avoiding overcrowding, and using appropriate personal protection when necessary.

Our findings advance the current literature in several key aspects. First, whereas previous studies have primarily linked total PM_2.5 mass to tuberculosis risk, we provide the first comprehensive evidence that specific components—particularly black carbon and organic matter—drive this association. Second, by using WQS and QGC, we addressed the critical issue of multicollinearity among pollutants, a common limitation in constituent-specific analyses [68, 69]. This approach allowed us to robustly quantify the relative contribution of each component within the mixture. Third, our identification of concentration thresholds for carbonaceous components offers tangible targets for future air quality standards aimed at reducing TB incidence. These advancements provide a stronger scientific foundation for moving beyond mass-based regulations toward component-specific air quality management and public health interventions.

Several limitations of this study warrant consideration. First, as an ecological study, although meteorological variables, air pollutants, and seasonal or long-term trends were controlled for, individual-level confounders could not be accounted for—a limitation common to prior time-series research. Second, reliance on Beijing’s overall average PM_2.5 component concentrations may introduce exposure misclassification, failing to account for individual variability in exposure. This form of non-differential exposure misclassification would bias the estimated effects toward null values, resulting in a potential underestimation of the actual relationship of specific PM_2.5 components with PTB risk. Previous studies suggest such attenuation could result in the observed risk estimates being attenuated to approximately 70% to 90% of their true value [65, 66]. Third, we used the spatial arithmetic mean rather than the population-weighted mean to estimate city-wide exposure levels, which may not fully capture the actual exposure in densely populated areas, although its impact on identifying temporal trends is likely limited [65, 70, 71]. Fourth, the urban-centric design limits applicability to rural populations, where differences in healthcare access, socioeconomic factors, and environmental exposures may significantly alter PTB dynamics, necessitating region-specific analyses for targeted interventions. Fifth, during severe air pollution episodes, increased indoor occupancy may result in an underestimation of the true association between air pollutants and PTB. This bias arises because air quality monitoring systems primarily measure outdoor pollutant levels and do not capture the variability in indoor exposure. Sixth, the findings represent statistical associations rather than causal inferences due to high correlations among various constituents. Therefore, these health effect estimates ought to be interpreted with caution, and additional confirmation via toxicological research or randomized controlled trials is required to establish the impact and contributions of specific constituents. Seventh, WQS and QGC approaches utilize fixed index weights lacking accompanying confidence intervals, thereby restricting the evaluation of their statistical significance—a recognized constraint in this area of research. Finally, while the analysis utilized a comprehensive, citywide dataset including both specialized and general hospitals in Beijing, caution is warranted when generalizing the findings to other regions due to potential geographic variability. Potential impacts of the COVID-19 pandemic on exposure assessment and outcome ascertainment are provided in Supplementary Discussion 6.

Conclusion

This multi-center temporal analysis offers strong epidemiological support for a link between chronic exposure to PM_2.5 and its key chemical components and an increased risk of PTB. Black carbon and organic matter were identified as the primary contributors to pathogenicity. Because PTB is a life-threatening respiratory disease, these findings highlight the significance of modifiable environmental risk factors and offer critical insights to inform preventive public health strategies.

Authors’ contributions

Shirong Li and Feng Guo were responsible for data collection, screening, and analysis, and prepared the initial draft of the paper. Shirong Li carried out the revision and submission of the manuscript. Chao Wang supported the statistical analysis. Rongmei Liu and Wenjie Qi conceived and oversaw the study and secured funding. All authors participated in reviewing and editing the manuscript, and approved the final version for publication. Shirong Li and Feng Guo share co-first authorship.

Funding

This study was supported by the High-level Public Health Technology Talent Construction Project (Academic Leader-02-29), the National Key Clinical Specialist Construction Programs, and the High-level Public Health Technology Talent Construction Project (Subject Backbone-03-49).

Data availability

The corresponding author can provide access to the datasets utilized in this research upon receiving a justified request.

Declarations

Ethics approval and consent to participate

This study was approved by the Bioethics Committee of Beijing Friendship Hospital, Capital Medical University (Approval Reference: 2025-P2-170-01). All research was performed in accordance with the Declaration of Helsinki. As this study was a retrospective analysis based on anonymized surveillance data, the need for informed consent was waived by the ethics committee.

Consent for publication

This manuscript does not contain any individual person’s data in any form. Therefore, consent for publication is not applicable.

Competing interests

The authors declare no competing interests.