Quantifying the community transmission of Mycobacterium tuberculosis in a rapidly growing Chinese city: a nine-year population-based genomic and spatial analysis Authors: Xi Zhu, Yunxia Wang, Liai Peng, Yu Lan, Jinzhou Mei, Meijuan Zhang, Weibin Zheng, Joshua Warren, Meng Li, Nan Li, Congyang Li, Yinglin Lao, Jingjie Luan, Junhong Chen, Xiaoyu Lu, Yanfang Guo, Ted Cohen, Chongguang Yang Published: 2026 ABSTRACT ABSTRACTTuberculosis (TB) remains a significant public health threat in urbanizing regions of China, where shifting population dynamics and migration may amplify TB transmission. We conducted a nine-year prospective epidemiological study of culture-positive TB patients diagnosed in Shenzhen, between 1 January 2014 and 31 December 2022, and employed whole-genome sequencing analysis to describe local transmission of Mycobacterium tuberculosis (Mtb). Mtb transmission hotspots were identified using a non-parametric distance-based mapping approach. We applied a spatially structured logistic regression and hierarchical Bayesian pairwise regression analysis to identify demographic, pathogen, and spatial factors associated with local transmission. A Bayesian phylogenetic analysis was used to infer probable transmission events. Among 4,560 individuals with culture-positive TB, 93.4% (4,261) were internal migrants in China. 21.8% (996/4,560) of individuals had Mtb isolates that belonged to genomic clusters, with multiple transmission foci detected in the central business district and suburban industrial areas. Transmission was more likely to occur among patients of similar age, close geographic proximity, and migrants with a shared provincial origin. Mtb isolates from communities with higher proportions of migrants had an increased risk of clustering. In this rapidly growing urban setting, a significant proportion of cases were documented to arise as a result of local transmission, which appears to be driven by specific social and geographical factors, particularly within migrant populations. Control strategies should therefore move beyond static models toward dynamic interventions that specifically target social networks, high-risk communities, and urban hotspots in dynamic urban environments. Introduction With an estimated 10.68 million cases and 1.25 million deaths in 2023, tuberculosis (TB) remains the world’s leading cause of death from a single infectious agent [1]. China, the third highest TB-burdened country globally with an estimated 740,000 new TB cases in 2023, has achieved a slow annual decline in incidence of 3.1% per annum [1]. This challenge is worsened by rapid urbanization and massive internal migration over the past two decades. In general, internal migration has flowed from less developed rural areas with historically higher TB incidence to rapidly growing southeastern cities with previously lower incidence. Shenzhen, for example, has grown from a small village of 30,000 residents in 1979 to a metropolis of over 17 million residents [2]. In Shenzhen, TB incidence increased from 39.7 to 51.26 per 100,000 between 2001 and 2010, followed by a slow decline to 43.4 per 100,000 in 2019; currently, more than 90% of incident TB cases occur amongst individuals who migrated from other parts of China [3,4]. This demographic shift underscores the urgent need to understand Mycobacterium tuberculosis (Mtb) transmission dynamics in these urban centres to inform more effective control strategies [5,6]. Disease among migrants may result from progression of imported latent infections (obtained in the higher incidence region of origin) or from more rapid progression from recent local transmission occurring in the lower incidence city of destination. While previous genomic epidemiological studies in Shenzhen provided initial insights [7], fully quantifying local transmission dynamics – specifically who transmits to whom and where – proved challenging. This difficulty is compounded by high population turnover, leading to an underestimation of the actual transmission burden. Modelling studies suggest that migrant-targeted interventions could reduce TB incidence by up to 20%, but quantifying the relative contributions of local transmission and associated risk factors is critical for designing effective, tailored control strategies [8]. Addressing this knowledge gap requires leveraging long-term data combined with whole-genome sequencing (WGS) and spatial analysis [5,9,10]. Previous models suggest targeting transmission foci could be a key strategy for TB control [11]. Studies have shown that individual Mtb strains may spread over short distances creating local foci of transmission [12,13]. However, it is also clear that spatial heterogeneity in TB notification rates does not always accurately reveal the location of these transmission hotspots [9,12], highlighting the need to integrate genomic sequencing to more precisely identify areas of active transmission. Here we present results from a nine-year prospective study of Mtb transmission, in which we combined pathogen sequencing with epidemiological and spatial data to investigate the roles of importation and local transmission in TB epidemiology in Baoan, the most populous district in Shenzhen. We used WGS data to infer transmission networks, which were then integrated with geospatial data to identify local transmission hotspots. We applied hierarchical Bayesian pairwise regression analysis to identify both the individual and paired levels that contribute to local transmission. By integrating genomic sequencing with spatial analysis, we aim to map spatial transmission clusters and identify key risk factors associated with local transmission. Materials and methods Study setting and design Shenzhen, the most densely populated city in China, experiences significant population mobility, with notable inflows and outflows. Migrants comprise 71.0% of the population, higher than in Beijing (38.5%) and Shanghai (42.1%) [14]. Due to high living costs, most migrants stay temporarily, with nearly half considering Shenzhen a “transit point” and only a few intending to settle permanently [15]. The average time of residence in Baoan for migrants is approximately five years [16]. Baoan, the most populous district in Shenzhen, had an estimated population of over 4.54 million in 2022, with 85.0% of residents born outside the city [2]. Individuals with TB-like symptoms (e.g. persistent cough, fever, chest pain, weight loss, night sweats) or abnormal chest radiographs are referred to TB-designated hospitals in Baoan for diagnosis. Confirmed TB cases are registered in the China Tuberculosis Information Management System (TBIMS) and managed by the local TB control department at the Baoan Centre for Chronic Disease Control. WGS was performed on all pre-treatment culture-positive Mtb isolates from individuals diagnosed in Baoan between January 2014 and December 2022 (n = 5,075). Demographic, clinical, and residential address data were extracted from the TBIMS, and social factors were collected from structured questionnaires routinely administered to TB patients at treatment initiation. Migrants are defined as individuals without a Shenzhen household registration status through the Chinese hukou system. The study protocol was approved by the Research Ethics Committee of the School of Public Health (Shenzhen), Sun Yat-sen University (protocol number: 2023084). Whole-genome sequencing Mtb strains were sequenced as previously described [7]. Raw reads were trimmed with Sickle [17] and aligned to the Mtb complex reference genome H37Rv (NC_000962.3) using BWA-MEM [18]. Single nucleotide polymorphisms (SNPs) were called using SAMtools [19] and Varscan [20]. Strains with sequencing depth <20× or genome coverage <95% were excluded. Fixed SNPs (frequency ≥90%) were used to calculate pairwise SNP distances, excluding those in drug-resistance genes and repetitive regions (e.g. PPE/PE-PGRS family genes, phage sequences, insertions, or mobile elements). Genomic clusters were defined as those within twelve or fewer SNPs. Strains were classified into lineages according to Coll et al [21]. L2 strains were further classified into L2.1, L2.2, and L2.3 [22]. TB-Profiler v4.4.2 was used for resistance prediction [23]. Multidrug-resistant tuberculosis (MDR-TB) was defined as Mtb with mutations conferring resistance to at least isoniazid and rifampicin. A maximum-likelihood phylogeny was constructed using IQ-TREE [24]. The best-fit model was selected using ModelFinder, with the General Time Reversible model (GTR + F + R4) chosen based on the Bayesian Information Criterion. Tree visualization and editing were performed using ITOL (https://itol.embl.de/). We also used the R package phybreak (version 0.5.2) [25] to infer transmission direction and infection time for pairs of cases within each genomic cluster (Supplementary Methods). Spatial aggregation of transmission networks We referred to genomic clusters of Mtb isolates as transmission networks to clarify the different meanings of “clusters” in spatial and genetic analysis. For each putative transmission network with at least five isolates, a non-parametric distance-based mapping (DBM) [26] approach in the R package hotspotr [27] was used to identify spatial transmission foci. We divided the study area into a 100*100 grid with a cell size of 0.0001 degrees (approximately 11.1 m x 10.3 m) and performed a DBM analysis for each large transmission network. Each cell was assigned a score between 0 and 1, with scores ≥0.95 considered a spatial focus of transmission for the network. A narrow smoothing window (0.01) was applied to avoid potential over-smoothing. The identified spatial focus represents a zone of aggregation and does not necessarily encompass all members of the corresponding transmission network. Spatially structured logistic regression model Baoan comprises 10 administrative areas and 125 communities (Figure 1(C)). The spatially structured logistic regression model identifies community-level clustering risk, accounting for individual, regional, and spatial factors. The binary outcome, Yij, indicates whether individual j in region i belongs to a transmission network, modelled using logistic regression: (1) Yij∼Bernoulli( pij),i=1,…,n;j=1,…,mi, Where the probability of being in a transmission network, pij, is modeled as a function of the covariates and a spatial random effect. The model is expressed as: (2) logit( pij)=xijTβ+ziTγ+ϕi where xij represents individual-level covariates (e.g. gender, age, residence time, strain lineage), zi represents community-level covariates (e.g. proportion of migrant population, industrial land use, number of subway stations, per capita GDP, population density, and number of designated medical institutions; Supplementary Table S1 and Figure S1), and ϕi is the spatial random effect at the region level, capturing spatial correlation in the outcome variable. The spatial random effects ϕ1,ϕ2,…ϕn follow an intrinsic conditional autoregressive (iCAR) model [28], defining spatial correlation based on neighbouring regions (i.e. regions with touching boundaries). The model was fitted using the Integrated Nested Laplace Approximation (INLA) package [29]. Individual-level variables (Supplementary Table S2) were first screened using univariable logistic regression (Supplementary Table S3). Variables with a P value < 0.20 were considered for inclusion in the spatially structured logistic regression models. Models incorporating different combinations of spatial covariates (Supplementary Table S1) and selected individual-level factors were evaluated, and the final model was chosen based on the lowest Watanabe-Akaike Information Criterion (WAIC) (Supplementary Table S4) [30]. Detailed variable selection procedures are described in the Supplementary Methods. Figure 1.Temporal trends and spatial distribution of TB in Baoan, 2014–2022. (A) TB cases and notification rate in Baoan, 2014–2022; (B) Community-level culture positive TB case notifications per 100,000 population per year over the study period; (C) Distribution of Baoan administrative area and community. Pink shaded areas represent industrial zones, and the red circle indicates the central business district (CBD) in the south.Three visuals: one clustered column and line chart and two maps showing tuberculosis cases and notification patterns in Baoan, 2014 to 2022.The figure shows temporal and spatial views of tuberculosis in Baoan from 2014 to 2022. The top visual is a clustered column and line chart. The horizontal axis is year from 2014 to 2022. The left vertical axis is number of cases from 0 to 1500, and the right vertical axis is notification rate per 100000 population from 0 to 60. For each year, 3 adjacent columns represent all cases, migrant cases, and resident cases. Superimposed lines show notification rates for all cases, migrants, and residents, with values labeled above points. Total notification decreases from about 41.3 in 2014 to about 25.7 in 2022 with a local rise around 2021. Migrant notification declines from about 44.6 to about 29.4, and resident notification declines from about 23.5 to about 13.7, with a trough near 2019 and an increase around 2021. All data are approximate. The lower left visual is a choropleth map of Baoan communities, shaded by notification rate of culture positive tuberculosis per 100000 people per year, grouped into 6 categories from 0.0 to 6.0 up to 16.0 to 35.0, with a 0 to 5 kilometer scale bar. The lower right visual is a reference map of Baoan outlining 10 administrative areas and 125 communities. Labeled areas include Yanluo, Songgang, Shajing, Xinqiao, Fuhai, Fuyong, Hangcheng, Jichang, Xixiang, Xinan, and Shiyan. Industrial areas are shaded, and a circle in the south marks the central business district labeled CBD. A 0 to 5 kilometer scale bar appears at the bottom. All data are approximate. Temporal trends and spatial distribution of TB in Baoan, 2014–2022. (A) TB cases and notification rate in Baoan, 2014–2022; (B) Community-level culture positive TB case notifications per 100,000 population per year over the study period; (C) Distribution of Baoan administrative area and community. Pink shaded areas represent industrial zones, and the red circle indicates the central business district (CBD) in the south. Pairwise regression analysis We applied the R package GenePair [31,32], a hierarchical Bayesian logistic regression method with spatially structured individual-level random effects to account for network dependence, to assess associations between genetic relatedness of case pairs and individual- and pair-level factors. For each unique pair of TB patients in Baoan, the binary outcome is defined as one if they belong to the same transmission network, and zero otherwise. The model included five risk factors: the combined age of the two individuals, the age difference of the two individuals, the spatial distance in kilometres (km) between their residential addresses, an indicator of migrant status (i.e. pairs of migrants from the same province, and any other pairs as the reference group), and gender pairs (i.e. both male, both female, and mixed gender as the reference group). The default weakly informative priors were used in the analysis. Given the computation-intensive nature of pairwise analysis in a Bayesian setting, the full dataset was randomly divided into six subsets, and the model was run individually for each. We then used R metafor [33] to combine results to approximate estimates for the entire dataset. Statistical analyses Continuous variables, presented as medians (IQR), and categorical variables, as counts (%), were analyzed using the Mann-Whitney U and chi-square tests, respectively. Statistical significance was set at P < 0.05. Changes in temporal trends of clustering rates were analyzed using Joinpoint (version 4.9.0.0) [34]. For all possible tuberculosis (TB) patient pairs, we calculated the geographic distance between residences at diagnosis using the geopy.distance function in Python. The relationship between genetic distance of Mtb isolates and patients’ geographic distance was then assessed. Using various SNP thresholds (3, 6, 12, and 25), we compared the odds of close genetic relatedness among isolate pairs across different levels of geographic proximity to those whose residences were separated by ≥10 km. All statistical analyses was performed in R (version 4.5.1). Missing data was not included in the analyses. Results Characteristics of the study population Between 1 January 2014, and 31 December 2022, a total of 9,995 TB cases were notified in Baoan District, Shenzhen; the vast majority of individuals with TB were internal migrants (91.4%, 9,135). The average annual notification rate in Baoan was 31.4 cases per 100,000 population, showing a slight annual decline of 2.7%. Internal migrants consistently had a higher notification rate (average 35.3 per 100,000) compared to residents (average 14.3 per 100,000) (Figure 1(A)). Among 9,995 notified TB cases, 5,075 (50.8%) were culture-positive pulmonary TB. Figure 1(B) shows the community-level notification rate of culture-positive TB over the study period. After excluding strains with failed re-culture or poor sequencing quality, genomic data were obtained for 4,560 (89.9%) strains. Of these, 4,261 (93.4%) were from migrants originating from 29 provinces across China (Supplementary Figure S2). Approximately 12.6% migrant pairs shared the same home province, whereas only 1.4% originated from the same city or county (Supplementary Table S5). Among 4,560 patients with genomic data, 66.8% (n = 3,045) were male, with a median age of 32 years (IQR 25–45), and 94.2% (n = 4,299) were new TB cases (Supplementary Table S2). To assess TB onset after migration to Baoan, residence time at diagnosis was analyzed. Among 3,570 migrants with available data, 25.2% (899/3,570) developed TB within one year of arrival, and 52.1% (1,861/3,570) within three years (Supplementary Figure S3). Phylogenetic and transmission cluster analysis Among the 4,560 strains with genomic data, 73.3% (n = 3,347) belonged to lineage 2 (L2), 25.9% (n = 1,183) to lineage 4 (L4), and 0.6% (n = 28) to lineage 1 (L1). The most prevalent sublineage was L2.3 (“modern” Beijing) (49.5%, n = 2,261), followed by L2.2 (“ancient” Beijing) (22.1%, n = 1,009), and L4.4 (10.9%, n = 498) (Figure 2). Drug resistance prediction identified 19.6% (n = 896) strains resistant to at least one anti-tuberculosis drug, including 3.2% (n = 148) multidrug-resistant (MDR) strains. Compared with isolates from new TB patients, a higher proportion of isolates from retreated patients were drug-resistant (33.7% vs 18.8%, P<0.001) or MDR (14.6% vs 2.6%, P<0.001) (Supplementary Table S6). Figure 2.Phylogenetic tree of 4,560 Mycobacterium tuberculosis strains from Baoan. Branch colours indicate lineages and sublineages. The inner ring (red) highlights clustered samples, the middle ring (green) represents residents, the outer ring shows drug resistance profiles.A circular phylogenetic tree diagram of 4560 Mycobacterium tuberculosis strains from Baoan with rings for clustering, residency, and drug resistance.The figure shows a circular tree diagram of 4560 Mycobacterium tuberculosis strains from Baoan. Branches radiate outward from the center and are grouped by lineage and sublineage, including L1, L2.1, L2.2, L2.3, L3, L4.2, L4.4, L4.5, and other L4 sublineages. A scale bar at the top indicates a tree scale of 0.001. Around the outer circumference, three concentric annotation rings summarize sample features. The innermost ring marks whether each strain is part of a genomic cluster, with one color indicating clustered and another indicating not clustered. The middle ring marks population status, distinguishing migrants from residents. The outer ring marks drug resistance profile with three categories: pan susceptibility, non multidrug resistant drug resistance, and multidrug resistant. Most branches belong to lineage L2, especially sublineage L2.3, which occupies a large continuous sector of the circle. Smaller sectors correspond to lineages L1, L3, and various L4 sublineages. Clustered strains and drug resistant strains appear intermittently along the circumference rather than in a single contiguous block. No numeric axes or data labels are displayed. All data are approximate. Phylogenetic tree of 4,560 Mycobacterium tuberculosis strains from Baoan. Branch colours indicate lineages and sublineages. The inner ring (red) highlights clustered samples, the middle ring (green) represents residents, the outer ring shows drug resistance profiles. A total of 996 (21.8%, 996/4,560) isolates were classified into 400 transmission networks, each comprising 2–13 patients, the majority (75.0%, 300/400) consisting of only one pair (Supplementary Figure S4). As expected, over the nine-year period the cumulative clustering rate steadily increased as more strains could be included. While clustering rates for sliding 3-year windows remained stable at first and then increased, no significant changes were observed (Supplementary Figure S5). We inferred transmission direction between pairs of cases within each transmission network. Of 864 transmission events with a posterior probability >0.5, excluding the putative index case in each network, 477 (55.2%, 477/864) events were identified. The majority (62.5%, 298/477) occurred between individuals residing in different administrative areas. Sensitivity analyses confirmed that, regardless of the SNP threshold, most transmission events occurred between individuals residing more than 10 km apart (Supplementary Figure S6). The most frequent transmission sources and recipients were identified in Central Business Districts (CBD, i.e. Xixiang administrative area). While the highest intra-administrative area transmission occurred within Xixiang (51 events), the primary cross-administrative area routes flowed from industrial areas Shajing and Fuyong to the CBD, with 19 events each (Supplementary Figure S7). To assess whether transmission occurred before or after migrants’ arrival, 676 events (78.2%, 676/864) with available arrival data were analyzed. As the arrival dates for 311 events overlapped with the 95% credible interval (CrI) of estimated infection time, they were not applied to the classification. For the remaining 365 events, 359 (98.4%, 359/365) occurred after arrival, and 6 (1.6%, 6/365) occurred before arrival. In addition, a sensitivity analysis using median estimated infection time for all 676 events supported this, showing 91.2% (619/676) of clustered migrants were infected after arrival (Supplementary Table S7). Detection of transmission foci To investigate spatial aggregation of Mtb transmission, we analyzed 16 large transmission networks that included at least five individuals with TB. We identified statistically significant spatial foci for eight of these networks (Figure 3). Seven of these foci were associated with Beijing lineage Mtb strains, and one with L4.4 Mtb strain. These foci were primarily located in densely populated areas, including Xixiang and Xinan administrative areas (both within the CBD), and industrial zones such as Shajing and Shiyan administrative areas. To better understand the transmission dynamics of these eight networks, transmission pathways inferred from the phybreak model were mapped onto the map (Figure 4). The largest transmission network (#216, n = 12) had a spatial aggregation focus, including seven patients residing in Xinan administrative area. A shared provincial origin was observed among migrants in network #216: six individuals with TB were from different cities within Guangdong Province, and two from Hunan Province. Similarly, the second-largest network (#19, n = 11) included four patients (36.4%) who formed a spatial focus in the Anle community of Xinan administrative areas. This network comprised migrants mainly from Sichuan (n = 3) and Hunan (n = 5) provinces, with the Hunan patients exhibiting high residential proximity. Notably, among the eight networks with identified spatial foci, seven involved migrants sharing the same provincial origin (detailed in Supplementary Results). Another network (#68, n = 6) consisted mainly of local residents from Shenzhen (5/6). Overall, small and dispersed transmission foci were detected in Baoan, primarily among migrant populations from the same provinces. Figure 3.Transmission foci identified using distance-based mapping (DBM) on large putative transmission networks with at least five cases (A), and all large putative TB transmission networks by size, lineage, and fraction of cases within the transmission focus (B).Two visuals: a map of Baoan with 8 numbered TB transmission foci, and a bar chart of genomic cluster size and focus fraction by lineage.The figure shows two visuals summarizing tuberculosis transmission foci in Baoan. The first visual is a district map outlined with internal administrative boundaries and a scale bar from 0 to 4 kilometers. Eight shaded clusters mark inferred spatial foci labeled 19, 45, 51, 62, 68, 96, 110, and 210 or 216, positioned over different subareas of Baoan, including regions that correspond to Xixiang, Xinan, Shajing, Shiyan, and Anle. Most highlighted foci correspond to Beijing lineage Mycobacterium tuberculosis, with one focus linked to lineage 4 point 4. The second visual is a vertical bar chart titled along the bottom axis as Genomic Cluster number. The horizontal axis lists cluster identifiers including 216, 18, 19, 51, 49, 199, 210, 35, 45, 68, 10, 107, 110, 146, 62, and 96. The left vertical axis is labeled Cases, ranging from 0 to 12 in increments of 2. Bars show cluster sizes; the largest cluster 216 reaches 12 cases, and several others range from 6 to 11 cases. Each bar color indicates lineage categories labeled in a legend as lineage 2 point 2, lineage 2 point 3, and lineage 4 point 4. A line graph overlays the bars, using the right vertical axis labeled Percentage of cases within foci, scaled from 0 percent to 100 percent in approximately 25 percent increments, showing how the proportion of mapped cases varies by cluster; a few bars are marked with an X indicating no spatial focus detected. All data are approximate. Figure 4.Transmission dynamics of eight clusters with transmission hotspots. Transmission pathways between individuals drawn based on the Phybreak inference. Connecting lines represent inferred transmission events: solid lines indicate links with a posterior probability ≥ 0.5, while dashed lines indicate those with a probability <0.5. Red shaded areas represent transmission foci identified using distance-based mapping (DBM). Patient labels are colour-coded by origin: purple backgrounds indicate local residents, while distinct shared background colours (e.g. yellow, green, blue) denote migrants originating from the same province.Eight maps showing tuberculosis cluster routes in Baoan, with solid and dashed arrows linking labelled cases and highlighting hotspots.The figure shows eight maps of Baoan district illustrating inferred tuberculosis transmission networks labeled A to H. Each map outlines the same administrative area with internal boundaries and a 10 kilometer scale bar. In map A, titled number 216 with N equals 12, labels such as TB191881, TB17700, TB22462, TB22539, TB22027, TB190688, TB20403, TB220144, TB21250, TB20626, TB220146, and TB21343 are placed at different residential locations. Solid arrows and dashed arrows connect these labels across the district, and a shaded cluster marks a transmission focus. Map B, number 19 with N equals 11, shows labels including TB193373, TB181471, TB16669, TB21122, TB219184, TB21250, TB20350, TB15517, TB21167, TB17118, and TB14367 with similar arrow connections. Map C, number 51 with N equals 10, displays labels TB17642, TB20074, TB22464, TB15112, TB18686, TB17474, TB20004, TB193221, TB190669, and TB20469 linked by mixed arrow types and a shaded focus. Map D, number 45 with N equals 6, maps labels TB15065, TB18071, TB18176, TB192628, TB20245, and TB190843. Map E, number 68 with N equals 6, concentrates labels TB15411, TB17687, TB17428, TB22193, TB17678, and TB22149 tightly around a shaded area. Map F, number 210 with N equals 6, shows labels TB22350, TB21702, TB17637, TB18332, TB17675, and TB21359. Map G, number 110 with N equals 5, includes TB16194, TB21262, TB18454, TB192750, TB17412, and TB18006. Map H, number 62 with N equals 5, places labels TB21425, TB15582, TB20156, TB20515, and TB15521 near another shaded focus. Legend boxes indicate resident tuberculosis cases and migrant tuberculosis cases from the same province, as well as arrow style for higher or lower posterior support probability and shading for transmission foci. Transmission foci identified using distance-based mapping (DBM) on large putative transmission networks with at least five cases (A), and all large putative TB transmission networks by size, lineage, and fraction of cases within the transmission focus (B). Transmission dynamics of eight clusters with transmission hotspots. Transmission pathways between individuals drawn based on the Phybreak inference. Connecting lines represent inferred transmission events: solid lines indicate links with a posterior probability ≥ 0.5, while dashed lines indicate those with a probability <0.5. Red shaded areas represent transmission foci identified using distance-based mapping (DBM). Patient labels are colour-coded by origin: purple backgrounds indicate local residents, while distinct shared background colours (e.g. yellow, green, blue) denote migrants originating from the same province. Association of individual and spatial covariates and membership in any transmission network After accounting for individual and spatial factors, results from the spatially structured logistic regression model suggest that younger patients, as well as those with longer residence in Baoan, were at a higher risk of belonging to any genetic cluster. For every 2-year increase in age, the odds decreases by 3% (odds ratio, OR 0.97, 95% CrI 0.96–0.98). Additionally, longer residence in Baoan was associated with higher clustering risk (7–12 years: OR 1.27, 95% CrI 1.03–1.56; >12 years: OR 2.06, 95% CrI 1.66–2.55, each compared with 0–3 years). Individuals infected with the Beijing lineage Mtb (OR 1.67, 95% CrI 1.40–1.99, compared with non-Beijing lineage Mtb) showed an increased odds of clustering. Compared to pan- susceptible strains, non-MDR drug-resistant strains were associated with a higher odds of clustering (OR 1.33, 95% CrI 1.11–1.60), whereas no significant association was observed for MDR-TB strains (OR 0.87, 95% CrI 0.57–1.32). Furthermore, a higher proportion of the migrant population in the community was associated with an elevated risk of clustering (OR 1.04, 95% CrI 1.01–1.07, per additional 10%) (Table 1). The spatial random effect showed only a small difference (Supplementary Figure S8). Table 1.Covariates associated with belonging to any transmission network, based on a spatially structured logistic regression model. Coefficient of covariateOdds ratio β95% CrIOR95% CrIAge (each two years increase)−0.03(−0.04, −0.02)0.97(0.96, 0.98)Residence time in Baoan, years ≤3Ref 4∼60.04(−0.19, 0.26)1.04(0.83, 1.30) 7∼120.24(0.04, 0.44)1.27(1.03, 1.56) >120.72(0.51, 0.93)2.06(1.66, 2.55)Beijing strain NoRef Yes0.51(0.34, 0.69)1.67(1.40, 1.99)Drug resistance profile Pan-susceptibleRef Other drug resistance0.29(0.10, 0.47)1.33(1.11, 1.60) Multi-drug resistance−0.14(−0.56, 0.28)0.87(0.57, 1.32)Migrant population proportion (per additional 10%) in the community0.04(0.01, 0.06)1.04(1.01, 1.07)Industrial land use (per additional 10%)−0.05(−0.12, 0.01)0.95(0.89, 1.01)Number of subway stations (each additional station) in the community−0.05(−0.12, 0.02)0.95(0.89, 1.02)OR: Odds Ratio; CrI: Credible Interval. Table 1.Covariates associated with belonging to any transmission network, based on a spatially structured logistic regression model. Coefficient of covariateOdds ratio β95% CrIOR95% CrIAge (each two years increase)−0.03(−0.04, −0.02)0.97(0.96, 0.98)Residence time in Baoan, years ≤3Ref 4∼60.04(−0.19, 0.26)1.04(0.83, 1.30) 7∼120.24(0.04, 0.44)1.27(1.03, 1.56) >120.72(0.51, 0.93)2.06(1.66, 2.55)Beijing strain NoRef Yes0.51(0.34, 0.69)1.67(1.40, 1.99)Drug resistance profile Pan-susceptibleRef Other drug resistance0.29(0.10, 0.47)1.33(1.11, 1.60) Multi-drug resistance−0.14(−0.56, 0.28)0.87(0.57, 1.32)Migrant population proportion (per additional 10%) in the community0.04(0.01, 0.06)1.04(1.01, 1.07)Industrial land use (per additional 10%)−0.05(−0.12, 0.01)0.95(0.89, 1.01)Number of subway stations (each additional station) in the community−0.05(−0.12, 0.02)0.95(0.89, 1.02)OR: Odds Ratio; CrI: Credible Interval. Covariates associated with belonging to any transmission network, based on a spatially structured logistic regression model. OR: Odds Ratio; CrI: Credible Interval. Association of individual- and pairwise-covariates with membership in the same transmission network To further assess pairwise factors associated with being in the same transmission network, we applied pairwise regression with GenePair to data from TB cases with recorded residential addresses (n = 4,190, resulting in 8,771,955 unique pairs). Due to computational limitations, the dataset was divided randomly into six non-overlapping subsets: five subsets of 699 individuals and one of 695. We collected 5,000 samples from the joint posterior distribution, discarding 10,000 iterations prior to convergence and thinning the remaining samples by a factor of 5 to reduce autocorrelation. We found that individuals with TB who were closer in age and geographic proximity, and pairs of migrants from the same province, had higher odds of belonging to the same transmission network. A two-year increase in age difference was associated with 5% decreased odds (OR 0.95, 95% CrI 0.91–0.98), and each 5 km increase in spatial distance decreased the odds by 26% (OR 0.74, 95% CrI 0.64–0.85). Moreover, migrants from the same province were more likely to belong to the same transmission network (OR 1.90, 95% CrI 1.25–2.88) (Table 2). Other pairwise factors, including combined age and gender, were not significantly associated with membership in a single transmission network. Table 2.Individual- and pair-wise covariates associated with belonging to the same transmission network, based on pairwise regression with GenePair.EffectsEstimate OR (95% CrI)Age differences (each two years increase)0.95 (0.91, 0.98)Combined age (each two years increase)1.00 (0.98, 1.02)Distance between patients (for each five kilometres increase)0.74 (0.64, 0.85)Sex Both female vs mixed pair0.73 (0.38, 1.38) Both male vs mixed pair1.29 (0.87, 1.93)Migrants sharing the same province of household registration1.90 (1.25, 2.88)OR: Odds Ratio; CrI: Credible Interval. Table 2.Individual- and pair-wise covariates associated with belonging to the same transmission network, based on pairwise regression with GenePair.EffectsEstimate OR (95% CrI)Age differences (each two years increase)0.95 (0.91, 0.98)Combined age (each two years increase)1.00 (0.98, 1.02)Distance between patients (for each five kilometres increase)0.74 (0.64, 0.85)Sex Both female vs mixed pair0.73 (0.38, 1.38) Both male vs mixed pair1.29 (0.87, 1.93)Migrants sharing the same province of household registration1.90 (1.25, 2.88)OR: Odds Ratio; CrI: Credible Interval. Individual- and pair-wise covariates associated with belonging to the same transmission network, based on pairwise regression with GenePair. OR: Odds Ratio; CrI: Credible Interval. Discussion This nine-year, population-based study provides a comprehensive understanding of Mtb transmission dynamics in Baoan, Shenzhen, a rapidly urbanizing megacity with a predominantly migrant population. Our findings indicate that despite a substantial decline in overall TB incidence, local transmission persists, accounting for approximately one-fifth of all cases (21.8%). Transmission was primarily driven by a combination of shared demographic and social characteristics, and was most commonly identified among migrants of similar age, geographic proximity, and, most notably, a shared provincial origin. Spatial analysis further identified multicentric transmission hotspots in both CBD and suburban industrial zones. These findings challenge traditional assumptions about TB in urban settings in China and underscore the necessity of moving toward tailored, dynamic intervention strategies. Our study found that only one-fifth of TB cases were clustered, consistent with other migrant-dominated Chinese cities, such as Shanghai (25.2%) [13]. While a substantial proportion of unclustered cases likely stem from endogenous reactivation – particularly among migrants harbouring latent infections from high-burden origins – we assume a subset also results from undetected recent transmission. Transmission inference remains challenging in this highly mobile population, as only a fraction of transmission events leave detectable genomic footprints owing to sampling limitations. For instance, index cases may be missed shortly after arrival, and secondary cases may return to their hometowns after infection. Cross-regional transmissions beyond our surveillance scope could also explain the lower clustering rate [6]. Quantifying the exact ratio of reactivation to missed transmission is challenging. Our pairwise SNP analysis against diagnosis intervals (Supplementary Figure S9) reveals numerous clustered pairs with short intervals (≤12 months) but large genetic distances (>5 SNPs). Given the slow mutation rate of Mtb (0.04–2.2 SNPs/genome/year), these cases are likely connected through unsampled intermediate cases, highlighting the challenge of fully reconstructing transmission chains in this context. Our study highlights the role of migrant populations in TB transmission. Communities with higher proportion of migrants and migrant patients with longer residence in Baoan, showed increased genomic clustering (Table 1). As Shenzhen continues to urbanize, TB transmission driven by migrants remains a significant concern. Although TB screening is mandatory for formal enterprise, housekeeping workers and the unemployed – who constitute over half of migrant TB cases – often fall outside these programmes. Strengthened surveillance and targeted case-finding for long-term migrants are essential to disrupt transmission chains and reduce disease burden. Beyond individual covariates, we identified pairwise covariates associated with transmission. Pairs of individuals within the same transmission network often had a small age difference (Table 2), consistent with a population-based contact survey showing that individuals tend to interact more frequently with those of similar age [35]. Notably, we also found that migrant pairs from the same province were 1.9 times more likely to share a genetic cluster than those from different provinces. However, the small fraction of migrants from the same city or county (1.4%) and our transmission inference (Supplementary Table S7) suggest that most provincial-origin transmissions occurred after arrival in Baoan. This finding links TB transmission to a deeply rooted sociological phenomenon in China known as “LaoXiang” (from the same origin) networks [36], a critical support system helping migrants find housing, jobs, and social connections in unfamiliar urban environments [37]. Furthermore, clustered patients from the same province also lived in closer proximity (an average of 4.63 km) than non-“LaoXiang” pairs (7.60 km, P < 0.001), suggesting these social networks may drive spatial aggregation that facilitates transmission. Given this context, expanding close-contact investigations to include migrants from the same origin offers a novel public health intervention for Shenzhen and similar cities, though this active screening strategy requires practical validation. By integrating spatial data, we identified multicentric transmission foci (Figure 3) aligning with urbanization and industrial patterns. The CBD-centered southern coastal region (Xixiang and Xinan administrative areas), has concentrated commercial industries, while manufacturing areas (Fuyong, Shajing, and Shiyan) feature overcrowded urban “villages” with low-cost migrant housing. Geographical hotspots are similar for both the genomically clustered and unclustered cases (Supplementary Figure S10; P = 0.370). Furthermore, most inferred transmission (62.5%) occurred between different administrative areas (Supplementary Figure S7), possibly reflecting frequent movement between homes and workplaces. Consistent with previous studies in Chinese cities, these findings underscore the role of casual contacts within highly mobile urban populations [13,38]. However, China’s TB reporting system lacks sufficient individual-level data to capture the full extent of casual, long-distance transmission events. Future TB elimination efforts in middle-burden urban areas should integrate social network analysis with expanded contact investigations to improve the efficiency of case finding. Moreover, closer geographic proximity was a significant risk factor for transmission (Table 2), consistent with previous studies [5,9,13]. Geographically concentrated transmission patterns, driven by “LaoXiang” culture and revealed through genomic and spatial analysis, underscore the need to expand close-contact investigations in urban settings. A study in Botswana extended TB contact investigations to the nearest and second-nearest neighbours, resulting in a 146% increase in case detection [39]. Similarly, expanding the investigation radius to 50 m around patients’ homes in Pakistan increased case notifications by 7.9% [40]. Our findings, along with these studies, suggest that residents living near an index TB patient may be high-risk groups that would benefit from contact investigations. Our study has several limitations that may have led to an underestimation of transmission. First, the reliability of our transmission inferences depends on the completeness of sampling. However, our WGS data were only obtained from culture-positive Mtb isolates. Second, we likely missed cases among migrants who returned to their home provinces for diagnosis or treatment, and the potential bias associated with these missing cases is difficult to quantify. Third, due to high population mobility across different regions of Shenzhen and nationwide, only patients diagnosed in Baoan were included. This means cross-regional transmission events were not considered, potentially underestimating the local TB clustering rate. Finally, geographical data only included patients’ residential locations, without accounting for their movement patterns within the city. In conclusion, we employed genomic and spatial analyses to quantify the contributions of local TB transmission in an emerging, migrant-dominated urban city. Our findings indicate that populations sharing similar social or geographical characteristics play an important role in Mtb transmission, with multicentric transmission foci concentrated in the CBD and industrial areas. Integrating genomic and spatial approaches is essential for understanding localized transmission and implementing tailored case-finding interventions, such as expanding contact investigations to include nearby neighbours or migrants from the same origin. Supplementary Material 4Supplementary material_520.docx