2.1 CRF59_01B sequences

181 CRF59_01B pol gene region sequences (HXB2: 2253−3821 nt, 1568 bp) were collected from newly diagnosed people living with HIV (PLWH) between 2008 and 2020 through HIV molecular surveillance in Guangzhou, China. Sample collection, sequencing and subtyping have previously been described.^{22, 23} Additional 110 CRF59_01B sequences (HXB2: 2253−3821 nt) were retrieved from the Los Alamos National Laboratory (LANL) HIV sequence database. CRF59_01B sequences without sampling location or sampling year information were excluded from the complete data set (Figure 1). According to the sampling regions, the CRF59_01B sequences were classified into 10 populations, that is, Baiyun district, Panyu district, the city centre of Guangzhou (GZ centre), the outskirts of Guangzhou (GZ outskirts), Shenzhen (a nearby city to Guangzhou within the Pearl River Delta), North China (North CN), Central China (Central CN), East China (East CN), South China (South CN) and Southwest China (Southwest CN) (Figure 1; Table S1). This classification divided Guangzhou into four regions (city centre, city outskirts, and the other two districts between them) to obtain a high-resolution interpretation of spatio-temporal history regarding CRF59_01B circulation. When assessing the overall profile (e.g., the overall export of CRF59_01B), Guangzhou was analyzed as a whole region. We termed these two classifications as the small-scale and large-scale analysis, respectively.

Demographic (age at diagnosis, gender, marital status, education, and occupation), social (HIV infection route, and the total number of reported contacts through sexual behavior or needle sharing) and clinical (HIV/AIDS status at diagnosis and current treatment status) characteristics were extracted from the electronic follow-up records from China Information Management System for Comprehensive Prevention and Control of AIDS, with the permission of Guangzhou Center of Disease Control and Prevention (CDC). This study was approved by the Institutional Review Board of School of Public Health, Sun Yat-sen University (No. 2023073).

2.2 Molecular network inference

To profile the geographic transmission pattern, molecular network of CRF59_01B was constructed at a pairwise genetic distance (GD) threshold of 0.012 substitution/site using HIV-TRACE (TRAnsmission Cluster Engine).²⁴ CRF59_01B pol sequences were initially aligned by MAFFT v7.²⁵ Pairwise GD was then calculated by employing Tamura-Nei 93 model.²⁶ Each PLWH was assigned according to the sampling region and represented by a node in the molecular network while nodes were linked if the pairwise GD is 0.012 substitution/site or less. The molecular network was visualized using Cytoscape 3.7.0²⁷ and spatialized to generate a transmission intensity matrix according to the numbers of links between regions.

2.3 Temporal structure test

We conducted regression of sampling time versus root-to-tip GD to examine the temporal structure in the data set. The tree topology and branch lengths were inferred using maximum likelihood (ML) method in IQ-TREE 1.6.12.²⁸ The nucleotide substitution model was selected for phylogenetic reconstruction based on the Bayesian information criterion using ModelFinder.²⁹ Node support in the ML phylogeny was estimated using an ultrafast bootstrap with 1000 replicates and a Shimodaira-Hasegawa approximate likelihood-ratio test with 1000 replicates.³⁰ The regression analysis was performed on the ML tree using TempEst 1.5.3.³¹ The outliers with poor temporal signals were subsequently removed from the final data set (Figure 1).³¹

2.4 Bayesian phylogeographic analysis

Bayesian discrete phylogeographic analysis was performed on the final data set using BEAST v1.10.4 to investigate the spatiotemporal dynamic of CRF59_01B in China.³² Bayesian stochastic search variable selection (BSSVS) procedure incorporating an asymmetric substitution model was used to infer viral migration.³³ Additionally, a robust counting (Markov jumps) approach was applied to estimate the expected location-state transitions along the phylogenetic branches.³⁴ Simultaneously, Bayesian coalescent phylogeny construction was done by using Markov Chain Monte Carlo (MCMC) integration,³² to estimate the evolution rate and the time to the most recent common ancestor (tMRCA). We selected a general time-reversible nucleotide substitution model with a gamma distribution prior on each relative substitution rate.³⁵ A relaxed uncorrelated lognormal molecular clock model was used to infer the timescale of CRF59_01B evolution, further specified with a gamma distribution prior on the mean clock rate (shape = 0.001, scale = 1000).³⁶ A skygrid demographic model was implemented to infer the changes in effective population size of CRF59_01B over time.³⁷ The MCMC chain length was set to 3 × 10⁸ steps, logging every 5000 iterations.

The MCMC output was inspected with Tracer v1.7.2.³⁸ After discarding the first 10% of samples as burn-in using LogCombiner v1.10.4, we checked the convergence by ensuring that effective sample sizes for parameters exceeded 200. A maximum clade credibility tree (MCC) with discrete geographical traits was inferred from the posterior set of trees in TreeAnnotator v1.10.4, with a 10% burn-in of measured states. The MCC tree was visualized in the ggtree R package.³⁹ To reveal when and from where CRF59_01B entered a particular region, we defined an introduction as the oldest node on a branch which transitioned into this region and counted all of its descendants as part of the same introduction irrespective of further exports or re-introductions. The location of the parent node of the introduction branch indicated where CRF59_01B imported, while the time of the transition node gave a conservative time estimation of introduction.⁴⁰ To avoid an accident introduction, we required at least 3 local infections present in the branch. In addition, statistical support for each pairwise diffusion between regions was evaluated by Bayes factors (BF) summarized in SpreaD3 0.9.6.⁴¹ Migration pathways were considered significant if supported by BF ≥ 3 and posterior probability ≥ 0.50.

2.5 Associating geographic traits with inferred tree topology

Association of geography with phylogeny was tested using a Bayesian Tip-association Significance Testing (BaTS).⁴² By assigning a geographic trait to each PLWH and subsequently reconstructing the ancestral states of nodes in the phylogeny, we determined whether a geographic trait is more likely to be shared by PLWH whose viruses are closely related than expected by chance alone, that is, when compared to random distribution on the same phylogeny. To address phylogenetic uncertainty, 1000 trees were subsampled every 300 000 iterations from the posterior set of trees using LogCombiner and analyzed using 100 null replicates per each tree. The phylogeny-trait association was measured by three statistics: association index (AI) statistic, parsimony score (PS) statistic, and monophyletic clade (MC) statistics for each discrete-trait. In brief, the AI statistic is an observed-to-expected ratio that measures the frequency of the most common branch tip trait subtended by internal nodes.⁴³ The PS statistic calculates the PSs for the traits supplied by the operator.⁴⁴ The MC statistic evaluates the number of tips with a given trait value in the largest MC identified in subsampled trees.⁴⁵ The statistics were calculated and tested using BaTS v0.9 beta.⁴² p < 0.05 means statistically significant.

2.6 Statistical analysis

Joinpoint regression analysis was conducted to examine the trend in counts of newly diagnosed CRF59_01B infections in Guangzhou using Joinpoint v5.0.2.⁴⁶ Chi-square test was performed to compare the clustering rates between groups of different infection routes. PLWH in Guangzhou and those involved in the introductions of CRF59_01B to Guangzhou were mapped according to their available residence, respectively. Descriptive analysis was used to characterize PLWH that involves in the largest introduction to Guangzhou. Univariate and multivariate logistic models were used to identify potential factors related to the largest introduction of CRF59_01B to Guangzhou. Statistical analysis was performed using R v4.2.3.

3.1 Rapid increase of CRF59_01B infections

During 2008 and 2020, a total of 181 HIV-1 CRF59_01B pol sequences were obtained from individuals newly diagnosed with HIV-1 infection in Guangzhou. Joinpoint regression analysis of these CRF59_01B sequences indicated that the number of people infected with HIV-1 CRF59_01B significantly increased in Guangzhou with an average annual percent change of 25.63 (95% confidence interval [CI]: 15.14−36.55) while the annual percent change [APC] between 2011 and 2015 was as high as 94.72 (95% CI: 45.71−187.63, [Figure S1A]). To further investigate the change of CRF59_01B epidemic, a data set of 265 HIV-1 CRF59_01B pol sequences were established, including the above 170 HIV-1 CRF59_01B sequences obtained from Guangzhou and 95 CRF59_01B sequences retrieved from LANL database, and the detailed information of CRF59_01B infections were collected (Figure 1 and Table S1). The data set was used to construct the molecular transmission network with a pairwise GD threshold of 0.012 substitutions/site. 78.9% of these sequences formed 25 transmission clusters of 2-100 sequences, comprising 1131 potential transmission links (Figure 2A and Figure S2A). Of note, the clustering rate for non-MSM was significantly higher than that for MSM (87.8% vs. 77.7%, χ² = 28.8, p < 0.001). Most of the clustered sequences were sampled from Shenzhen (61/209, 29.2%), or the city centre (47/209, 22.5%) and Panyu district (32/209, 15.3%) of Guangzhou. Furthermore, the transmission intensity matrix shows that the transmission links were predominantly from CRF59_01B-infected PLWH in Guangzhou and Shenzhen city (1117/1131, 98.8%).

3.2 Origin and evolution of CRF59_01B

Root-to-tip regression excluded 11 CRF59_01B sequences that are incongruent in their genetic divergence and phylogenetic position and 254 CRF59_01B sequences with a root-to-tip correlation coefficient of 0.47 were included in the molecular clock model to infer time-scaled tree.

Bayesian phylogenetic analysis showed that the tMRCA of CRF59_01B was 1998.6 with a 95% highest probability density [HPD] interval of 1994.6−2001.9 in a small-scale analysis where Guangzhou was coded into 4 regions (see Section 2). The mean substitution rate was estimated as 2.75 × 10⁻³ (95% HPD interval: 2.22 × 10⁻³−3.29 × 10⁻³) substitutions/site/year. The large-scale study where Guangzhou was coded as a whole region (see Section 2) revealed the estimated tMRCA of 1997.5 [1992.9−2001.5], along with the mean substitution rate of 2.75 × 10⁻³ [2.23 × 10⁻³−3.30 × 10⁻³] substitutions/site/year. The time-scaled Bayesian skygrid demographic plot illustrated a rapid increase of CRF59_01B population since its emergence although a slight decline was observed after 2018 (Figure 3B).

Both small and large scale BSSVS analysis indicated that the root of the phylogenetic trees of CRF59_01B was in Shenzhen city with a posterior probability of 0.937 (Figure 3A and Figure 4A) and 0.721 (Figure 5A), respectively. Furthermore, the global phylogeny-geography association for CRF59_01B was statistically supported by AI (15.227, p < 0.001) and PS (114.708, p < 0.001, Table S2), indicating that CRF59_01B sequences were phylogenetically clustered by region. The association tested by the MC statistic was significant for the outskirts (p = 0.010), Baiyun (p = 0.050) and Panyu (p = 0.010) district of Guangzhou, Shenzhen city (p = 0.010), as well as North (p = 0.010), East (p = 0.010), and Southwest China (p = 0.010). The geographic structure further confirmed the circulation of CRF59_01B in these regions (Table S2). Taken together, the Bayesian phylogeographic analysis indicated that CRF59_01B originated in Shenzhen around 1998, and rapidly spread in Shenzhen and transmitted to other regions in China.

3.3 CRF59_01B migration between Shenzhen and Guangzhou

Molecular network analysis exhibited 300 inter-city transmission links between Shenzhen and Guangzhou while 35.3% (106/300) of the links associated with GZ centre (Figure 2B and Figure S2B). The small-scale phylogeographic analysis revealed the migration of CRF59_01B from Shenzhen to the city centre (BF = 2581, posterior probability = 0.997, mean rate = 3.57 events/year) and Baiyun district (BF = 82, posterior probability = 0.908, mean rate = 0.98 events/year) of Guangzhou while the migration of CRF59_01B was also identified from the centre of Guangzhou to Shenzhen (BF = 19, posterior probability = 0.694, mean rate = 0.93 events/year, Figure 4A,B and Table S3). The migration pathways between Guangzhou and Shenzhen were strongly supported in the large-scale analysis (Figure 5A,B and Table S3). By counting the location transitions along the phylogeny, we confirmed the major import of CRF59_01B from Shenzhen to Guangzhou (Figure 4C and Figure 5C). The time-scaled MCC tree showed that eight introductions (see Section 2) occurred from Shenzhen to the centre of Guangzhou during 2004 and 2011 (Figure 3A and Table S4). Notably, six of them appeared around 2007 or 2010. And the largest introduction occurred around 2004.2 [2001.2−2007.4], and transmitted back to Shenzhen around 2010.8 [2009.2−2012.5]. This introduction caused HIV-1 infection of 71 PLWH in Guangzhou, 3 in Shenzhen, 1 in Central CN, 1 in East CN, 1 in South CN and 1 in Southwest CN. Our analysis confirmed the frequent transmission of CRF59_01B between Shenzhen and Guangzhou since 2004 while the city centre of Guangzhou served as the entry point of CRF59_01B.

3.4 CRF59_01B spread within Guangzhou

The molecular transmission network showed that 51.1% (578/1131) of the potential transmission links of CRF59_01B were within Guangzhou (Figure S2B). Among them, a lot of inter-region links were observed between GZ centre and other regions of Guangzhou (Baiyun [80], Panyu district [106] and GZ outskirts [60]) while few links were within GZ centre (54) (Figure 2B). The phylogeographic analysis further revealed inter-region migration pathways from GZ centre to Baiyun (BF = 497095, posterior probability > 0.999, mean rate = 2.73 events/year), Panyu (BF = 99412, posterior probability > 0.999, mean rate = 2.69 events/year), and GZ outskirts (BF = 82842, posterior probability > 0.999, mean rate = 2.22 events/year), indicating the major migration pathway of CRF59_01B from the centre to the other regions of Guangzhou, rather than its transmission within the city centre.

Eight introductions of CRF59_01B from Shenzhen resulted in 123 infections in Guangzhou and primarily located in the city centre (36/123, 29.3%) and Panyu district (33/123, 26.8%, Figure S1C). Logistic regression analysis identified the factors associated with CRF59_01B infections, including residence, marital status (married vs unmarried: odds ratio [OR] = 2.33, 95% confidence interval [95% CI]: 1.18−4.67), number of closed contact and engagement in commercial sex (OR = 2.75, 95% CI: 1.12−7.23, Table S5). Compared with the city centre of Guangzhou, the multivariate model indicated that residence of Panyu district (OR = 6.90, 95% CI: 1.81−30.40) and GZ outskirts (OR = 4.76, 95% CI: 1.21−20.86) was more likely to be a risk factor related to CRF59_01B infection. Both univariate and multivariate models confirmed that fewer closed contacts were less likely to be infected with CRF59_01B since OR was 0.89 (95% CI: 0.79−0.98) and 0.84 (95% CI: 0.70−0.97), respectively.

3.5 CRF59_01B expansion nationwide

The molecular network showed that CRF59_01B transmission was more frequently observed between Guangzhou and Central CN (2 vs. 1), East CN (1 vs. 0), South CN (46 vs. 1) and Southwest CN (64 vs. 16) than that between Shenzhen and these regions in China (Figure S2B) while CRF59_01B infections in North CN exclusively linked with Shenzhen. The Markov jump method confirmed the exports of CRF59_01B from Guangzhou and Shenzhen to other regions in China (Figure 5C). The 7 nationwide migration pathways included migration from Shenzhen to Guangzhou (mean rate = 5.08 events/year, BF = 158, posterior probability = 0.968), from Guangzhou to Shenzhen (mean rate = 1.89 events/year, BF = 15820, posterior probability > 0.999), from Guangzhou to South CN (mean rate = 0.76 events/year, BF = 796, posterior probability = 0.993), from Shenzhen to North CN (mean rate = 0.59 events/year, BF = 428, posterior probability = 0.988), from Guangzhou to Southwest CN (mean rate = 0.76 events/year, BF = 156, posterior probability = 0.967), from Guangzhou to Central CN (mean rate = 0.47 events/year, BF = 77, posterior probability = 0.936) and from Guangzhou to East CN (mean rate = 0.42 events/year, BF = 37, posterior probability = 0.874) (Figure 5A,B and Table S3). The results of tMRCAs showed that CRF59_01B spread from Shenzhen to North CN since 2004.3 [2001.7−2006.8] and from Guangzhou to Southwest CN and East CN since 2009.0 [2007.7−2010.2] and 2013.6 [2010.0−2017.1], respectively. Thus, CRF59_01B spread from Shenzhen to North CN, while it expanded from Guangzhou to Central, East, South, and Southwest China.