Cross-Tissue Transcriptome-Wide Association Study Identifies Novel Genes Associated With POAG

Abstract

Purpose: Genome-wide association studies have identified numerous loci associated with POAG. However, functional insights remain limited owing to challenges from noncoding regions and complex linkage disequilibrium. We aimed to bridge these gaps in POAG by integrating genomic and multitissue transcriptomic data and identifying novel systemic regulatory genes.

Methods: We analyzed POAG genomic data from FinnGen and expression quantitative trait loci data from GTEx v8 for cross-tissue transcriptome-wide association studies. The Unified Test for Molecular Signature identified cross-tissue associations, complemented by single-tissue Transcriptome-wide association studies using Functional Summary-based Imputation for tissue-specific insights, and the Multi-marker Analysis of Genomic Annotation validated and refined results. Significant findings from the Unified Test for Molecular Signature, Functional Summary-based Imputation, and Multi-marker Analysis of Genomic Annotation were intersected to identify robust candidate genes, followed by summary data-based Mendelian randomization and colocalization analyses to explore their functional implications.

Results: Six candidate genes (AFAP1, CALCRL, KREMEN1, MTMR3, GFPT1, and TRIOBP) were identified with intersection evidence. Among these, CALCRL, MTMR3, and GFPT1 were novel. Summary data-based Mendelian randomization confirmed that AFAP1 (odds ratio [OR], 0.83; 95% confidence interval [CI], 0.78–0.88), CALCRL (OR, 0.86; 95% CI, 0.79–0.94), KREMEN1 (OR, 0.86; 95% CI, 0.77–0.97), and MTMR3 (OR, 0.77; 95% CI, 0.63–0.93) exhibited protective effects, and GFPT1 (OR, 1.34; 95% CI, 1.13–1.59) was identified as a risk role for POAG.

Conclusions: This study identified six genes associated with POAG, three of which were novel, offering novel insights into its genetic architecture and systemic regulatory mechanisms.

Glaucoma is the primary global cause of permanent blindness.¹ This chronic optic neuropathy is marked by the gradual deterioration of retinal ganglion cells, leading to a loss of peripheral vision that can progress to central vision impairment and, in advanced stages, total blindness.² Among its subtypes, POAG is the most prevalent, affecting approximately 1% to 2% of individuals over the age of 40.³ Family history is a well-established risk factor for POAG, strongly suggesting a significant genetic contribution to its pathogenesis.² 

Recent advances in genetic studies have laid the foundation of the POAG genetic architecture. Rare variants with large effects have been identified in genes such as MYOC and OPTN,^4,5 whereas common variants with smaller effects have been reported in loci such as TXNRD2, ATXN2, FOXC1, and so on.⁶ Despite these discoveries, the heritability of POAG remains largely unexplained, as mutations in these regions account for only a small fraction of cases.³ For instance, mutations in GLC1A are only linked to 2% to 4% of POAG, whereas mutations in OPTN are found in only 16.7% of families with hereditary POAG.^5,7 Furthermore, several loci discovered by genome-wide association studies (GWAS) are situated in noncoding areas, complicating the elucidation of their functional significance. The intricate nature of linkage disequilibrium (LD) also restricts the identification of causative variations, resulting in an improvement gap in our comprehension of the genetic determinants of POAG. 

Transcriptome-wide association studies (TWAS) offer a robust framework for linking expression quantitative trait loci (eQTL) data with summary GWAS statistics. This method facilitates the accurate identification of candidate genes and their functional roles in disease mechanisms.⁸ The Unified Test for Molecular Signature (UTMOST) improves gene-level association analysis by using eQTL data from various tissues among TWAS techniques.⁹ UTMOST applies a group lasso penalty to capture shared eQTL effects while preserving tissue-specific variations, thereby improving the accuracy and efficacy of genetic imputation models. Cross-tissue TWAS has proven effective in identifying genetic susceptibility genes linked to a range of complex disorders, including rheumatoid arthritis, migraines, and autism spectrum disorder.^10–12 As a multifactorial disease, POAG involves both localized ocular pathology, such as elevated IOP and progressive retinal ganglion cell loss,^13,14 as well as systemic factors with multiple tissues involved, such as vascular dysregulation and metabolic dysregulation,^15–17 necessitating a cross-tissue approach to elucidate the full spectrum of genetic signatures associated with POAG pathogenesis. Cross-tissue TWAS is particularly well-suited for POAG because it effectively integrates genetic signals from multiple tissues, capturing the systemic regulatory mechanisms that drive complex diseases with multiple tissues involved. Analyzing only a single tissue, such as the retina or optic nerve, may overlook genes with important functions in other key tissues, leading to an incomplete systemic genetic risk assessment. 

In this study, we performed a cross-tissue TWAS analysis to identify genes associated with POAG. By combining GWAS summary statistics for POAG with eQTL data from the Genotype-Tissue Expression Project (GTEx) v8, we further used Functional Summary-based Imputation (FUSION) to analyze tissue-specific associations and Multi-marker Analysis of Genomic Annotation (MAGMA) to quantify gene–trait relationships.^18,19 To strengthen the reliability of our findings, we followed the framework proposed by Gui et al.¹² and integrated UTMOST cross-tissue results with significant genes identified through FUSION and MAGMA, discovering potential candidate genes. Taking the intersection ensures that the identified genes are supported by complementary evidence from cross-tissue and single-tissue analyses, as well as functional annotations. This intersection not only reduced false positives, but also mitigated LD-related confounding and addressed the challenge of interpreting noncoding GWAS loci by linking regulatory variations to gene expression through eQTL integration, bridging a critical gap in GWAS studies and strengthening the biological plausibility of the results. Summary data-based Mendelian randomization (SMR) and colocalization analyses were further used to investigate the detailed biological functions and possible causal roles of candidate genes (Fig. 1). 

Figure 1.

 

Study design.

Methods

Data Source

The POAG GWAS data were sourced from the latest FinnGen Study R12, released in November 2024, which included a total of 484,589 participants, comprising 10,832 POAG cases and 473,757 controls.²⁰ Diagnoses of POAG within the FinnGen dataset were established based on the criteria outlined in the International Classification of Diseases. The data were adjusted for age, sex, and genetic principal components.²⁰ Additionally, we used the POAG GWAS dataset from Gharahkhani et al., which included 216,257 participants, for replication purposes.⁴ The eQTL data were sourced from the GTEx v8 dataset, encompassing comprehensive gene expression profiles across 49 different tissues collected from 838 deceased donors.¹⁹ The Medical Ethics Committee of the Zhongshan Ophthalmic Center at Sun Yat-Sen University, Guangzhou, China, approved this study and waived the requirement for obtaining informed consent (approval number: 2023KYPJ350). 

Cross-Tissue TWAS Analyses

We conducted cross-tissue TWAS analyses using UTMOST to assess gene-POAG associations at the organismal level.⁹ This approach leverages penalized multivariate regression to train cross-tissue expression imputation models, accounting for the varying directions and effect sizes of eQTL signals across tissues.¹⁰ This method facilitated the discovery of a larger quantity of genes in tissues exhibiting heightened trait heritability and improved imputation precision.¹² To combine these associations across tissues, we applied the generalized Berk–Jones test, using covariance from single-tissue statistics.⁹ A false discovery rate (FDR) threshold of less than 0.05 was used to determine statistical significance. 

Single-Tissue TWAS Analyses

We further conducted a single-tissue TWAS using the FUSION program, combining POAG GWAS data with eQTL data from GTEx v8 across 49 tissues to assess the relationship between each gene and POAG.¹⁹ Initially, LD between the prediction models and single nucleotide polymorphisms (SNPs) at each GWAS locus was analyzed using samples from the 1,000 Genomes European population. FUSION applied multiple prediction models to evaluate the overall influence of SNPs on gene expression, selecting the model with the best predictive accuracy to assign weights to gene expression.¹² 

Next, these gene weights were combined with the genetic effects of POAG (GWAS Z-scores) to perform the TWAS analysis for POAG. To refine the results further, we used the conditional and joint (COJO) module within FUSION to identify independent genetic signals within each locus.¹⁹ This step was critical for distinguishing features that were conditionally independent while accounting for LD among genetic markers.²¹ Genes identified as independently associated through COJO were classified as jointly significant, whereas those with only marginal significance were excluded from further analysis. 

Gene Analysis

We further used MAGMA to aggregate SNP-level association data into gene scores, enabling the evaluation of each gene's association strength with POAG. The methodology involves two main components: a gene analysis phase and a gene-set analysis phase. In the first phase, MAGMA quantifies the degree of association each gene has with POAG by calculating gene P values based on SNP data. This is achieved through a multiple linear regression model that accounts for LD between SNPs. The second phase of MAGMA further involves the actual gene-set analysis using gene P values and gene correlation matrix.^18,22 

SMR Analysis

To investigate the biological characteristics of candidate genes, we performed SMR analysis. This method has strong statistical power when top cis-eQTLs are included, provided large, independent samples supply exposure and outcome data.²³ Cis-eQTLs were identified within ±1,000 kb of target genes using a significance threshold of P < 5.0 × 10^–8, while SNPs with allele frequency differences exceeding 0.2 or minor allele frequencies below 0.01 were excluded.^24,25 The heterogeneity in dependent instrument (HEIDI) test was used to distinguish pleiotropy from linkage, with P < 1.57 × 10^–3 as the cutoff for heterogeneity and P < 0.01 suggesting potential pleiotropy.^24,25 Analyses were conducted using SMR software,²³ applying a significance threshold of P < 0.05 for SMR results. 

Colocalization Analysis

The purpose of this analysis was to ascertain whether the causal variant shared by the connections between eQTL and POAG that were found. The Bayesian model used in colocalization analysis considers five possible hypotheses with associated posterior probabilities: (1) neither trait is associated (H0), (2) only trait 1 is associated (H1), (3) only trait 2 is associated (H2), (4) the two traits are associated with different causal variants (H3), and (5) both traits share the same causal variant (H4).²⁶ We used the default parameters, with a prior probability of 1 × 10^–4 for an SNP being associated with trait 1, 1 × 10^–4 for association with trait 2, and 1 × 10^–5 for association with both traits. A low PPH3 and PPH4, coupled with a high PPH0, PPH1, and/or PPH2, indicates restricted power in the colocalization.²⁷ Referring to previous study, we considered the relationship between the eQTL and POAG to be colocalization at PPH3 + PPH4 ≥ 0.8.²⁷ 

Results

TWAS Analyses in Cross-Tissue

The cross-tissue TWAS analysis identified 364 genes with P < 0.05, of which 58 remained significant after FDR correction (P_FDR < 0.05) (Supplementary Table S1). In the replication dataset, 317 genes met the P < 0.05 threshold, with 44 retaining significance following FDR correction (P_FDR < 0.05) (Supplementary Table S2). Among the 58 identified genes, 13 were replicated in the replication dataset, including HMX1, AFAP1, SORCS2, AC012066.1, AC016730.1, RPRM, AC012501.2, FMNL2, MICALL1, GCAT, DNAJB14, CALCRL, and MN1 (Supplementary Table S2). Among the 58 identified genes, 44 were protein-encoding genes, including CALCRL, ZC3H15, LNX1, TNS1, OSM, SF3A1, SLC35E4, TRIOBP, GCAT, ANKRD54, MICALL1, PICK1, XBP1, FBXO7, C22orf31, AP1B1, CABP7, MTMR3, USP46, STAM2, GLS, TNFAIP6, AAMP, FIP1L1, SLC25A4, FMNL2, ARL5A, DNAJB14, MN1, TADA2B, HORMAD2, CCDC121, RPRM, CACNB4, KREMEN1, CHEK2, UQCR10, RBM43, SORCS2, SYN3, CCDC157, AFAP1, GFPT1, and HMX1. 

TWAS Analyses in Single Tissue

The single-tissue TWAS analysis identified 833 genes with P_FDR < 0.05 in at least one tissue (Supplementary Table S3). In the replication sample, 745 genes met the same significance threshold across at least one tissue (Supplementary Table S4). After COJO analysis to account for potential false positives owing to LD in the discovery dataset, 328 genes remained jointly significant in at least one tissue (Supplementary Table S5). In the replicated sample, 290 genes were still jointly significant in at least one tissue (Supplementary Table S6). Among the 328 genes identified, 66 were replicated in the replication dataset. 

Gene Analysis of MAGMA

The MAGMA gene-based test identified 514 genes significantly associated with POAG (P_FDR < 0.05) in the discovery sample and 263 significant genes in the replication sample (Supplementary Tables S7 and S8). Of 514 significant genes identified, 124 were replicated in the replication dataset. 

Intersection Evidence

To enhance the reliability of our findings, we integrated evidence from UTMOST, COJO, and MAGMA, ultimately demonstrating six prospective candidate genes for future investigation (AFAP1, CALCRL, KREMEN1, MTMR3, GFPT1, and TRIOBP) (Fig. 2). 

Figure 2.

 

Venn diagram illustrating the intersection evidences from UTMOST, FUSION, and MAGMA.

SMR and Colocalization Results

According to the COJO results, the GFPT1 gene was associated with POAG in colon sigmoid. SMR confirmed a causal relationship between GFPT1 and POAG risk in colon sigmoid (odds ratio [OR], 1.34; 95% confidence interval [CI], 1.13–1.59; P = 0.001), which was also supported by high confidence of colocalization (PPH₃ + PPH₄ = 0.959) (Figs. 3, 4). 

Figure 3.

 

The results of SMR analysis between candidate genes and POAG.

Figure 4.

 

The results of colocalization analysis between candidate genes and POAG.

The AFAP1 gene is associated with POAG in brain cerebellar hemisphere, cells cultured fibroblasts, colon sigmoid, muscle skeletal, pituitary, and vagina. After excluding pairs without appropriate SNPs, a total of three pairs (brain cerebellar hemisphere, cells cultured fibroblasts, and muscle skeletal) were included in the SMR analysis. After removing the pairs with high probability of pleiotropy (P_HEIDI < 0.01), SMR confirmed a causal relationship between AFAP1 and POAG in brain cerebellar hemisphere (OR, 0.83; 95% CI, 0.78–0.88; P < 0.001), which was supported by high confidence of colocalization (PPH₃ + PPH₄ = 1.000) (Figs. 3, 4). 

Similarly, the CALCRL gene was associated with POAG in artery aorta and nerve tibial. After excluding the pair without appropriate SNPs, another pair (artery aorta) was included in the SMR analysis. SMR confirmed a causal relationship between CALCRL and POAG in artery aorta (OR, 0.86; 95% CI, 0.79–0.94; P < 0.001), which was supported by high confidence of colocalization (PPH₃ + PPH₄ = 0.975) (Figs. 3, 4). 

The KREMEN1 gene was associated and had a causal relationship with POAG in breast mammary tissue (OR, 0.86; 95% CI, 0.77–0.97; P = 0.016) and esophagus mucosa (OR, 0.76; 95% CI, 0.66–0.87, P < 0.001); both of them were supported by high confidence of colocalization (PPH₃ + PPH₄ = 1.000) (Figs. 3, 4). 

The MTMR3 gene was associated with POAG in lung, minor salivary gland, and testis. After excluding pairs without appropriate SNPs, one pair (lung) was included in the SMR analysis. SMR confirmed a causal relationship between MTMR3 and POAG in lung (OR, 0.77; 95% CI, 0.63–0.93; P = 0.007), which was supported by high confidence of colocalization (PPH₃ + PPH₄ = 1.000) (Figs. 3, 4). 

The TRIOBP gene was associated with POAG in brain spinal cord cervical c-1; however, this association lack of appropriate SNPs for SMR analysis. 

Discussion

This study used a cross-tissue TWAS approach, integrating GWAS summary data with predictive model from GTEx v8 dataset across 49 tissues, to identified genetic associations with POAG. By combining intersection evidence from UTMOST, FUSION, and MAGMA, six candidate associated genes, namely, AFAP1, CALCRL, KREMEN1, MTMR3, GFPT1, and TRIOBP were identified. Further biological properties exploration through SMR and colocalization analyses confirmed causal relationships for AFAP1, CALCRL, KREMEN1, and MTMR3 of the protective roles in POAG pathogenesis, and GFPT1 as the risk role. Previous publications have not documented the potential roles of CALCRL, MTMR3, and GFPT1 in POAG, which are novel findings. 

This investigation contributes to the understanding of POAG's genetic mechanisms by identifying novel genes and discovering their potential systemic and tissue-specific regulatory effects. By applying COJO, we mitigated LD-related confounding, and MAGMA further provided functional validation, strengthening the biological plausibility of the findings. Moreover, we bridges the gap between noncoding GWAS loci and their functional impact by leveraging eQTL data through TWAS methods (UTMOST and FUSION) to link regulatory variants to gene expression. The identification of six genes underscores the capability of this integrative framework to refine genetic findings, elucidate the genetic architecture of POAG, and address previous gaps in understanding its underlying mechanisms. 

AFAP1 has been identified as a gene associated with POAG in GWAS.²⁸ Previous studies have identified a specific variant within the AFAP1 gene, rs4619890[G], which is significantly associated with an elevated risk of POAG (OR, 1.20; P = 7.0 × 10^–10).²⁸ AFAP1 is more commonly described as encoding an adaptor protein that participates in various signaling pathways mediated by the Src kinase family,²⁹ while Src kinase plays a role in TGF-β–induced IOP elevation.³⁰ Therefore, variations in AFAP1 may contribute to POAG pathogenesis by influencing IOP regulation and aqueous humor outflow.³¹ 

The CALCRL encodes a G-protein-coupled receptor characterized by a seven-transmembrane domain structure, which plays a pivotal role in both normal physiological processes and various disease states.³² For this receptor to be expressed on the cell surface and to effectively bind its peptide ligands, it must be co-expressed with one of three single transmembrane domain coreceptors, known as receptor activity modifying proteins (RAMPs). These include RAMP1, RAMP2, and RAMP3.³³ Specifically, the CALCRL/RAMP1 complex interacts with the 37-amino acid neuropeptide, calcitonin gene-related peptide, mainly sourced from sensory neurons,³⁴ whereas the CALCRL/RAMP2 and CALCRL/RAMP3 complexes are associated with the binding of adrenomedullin.^32,33,35 Upon ligand binding, there is a subsequent increase in cyclic AMP (cAMP) levels, leading to the activation of protein kinase A.^32,35,36 Calcitonin gene-related peptide, a multifunctional peptide, is notably involved in the regulation of blood pressure and induces vasodilation, which potentially offers protective effects for POAG.^37,38 

The CALCRL gene can also interact with RAMP2 to trigger downstream cAMP signaling, a pathway has been demonstrated to be crucial for retinal ganglion cell survival and documented to play a significant role in POAG pathogenesis.^39,40 Studies have shown that cAMP not only modulates inflammatory responses in glial cells during endoplasmic reticulum stress, but also promotes neuroprotection.⁴¹ Furthermore, cAMP has been demonstrated to prevent endoplasmic reticulum stress–induced apoptosis by reducing p53 accumulation.^42–44 Endoplasmic reticulum stress–induced apoptosis is common in neurodegenerative diseases and is also implicated in the pathogenesis of glaucoma.⁴⁵ Through these mechanisms, CALCRL-mediated cAMP signaling contributes to the survival of RGCs and potentially counteracts the apoptosis observed in glaucomatous neurodegeneration, supporting its protective role in POAG. 

KREMEN1 encodes a transmembrane protein that acts as a high-affinity receptor for Dickkopf1, forming a complex that inhibits Wnt/β-catenin signaling.⁴⁶ This signaling pathway plays a significant role in oxidative stress, inflammation, and cellular senescence,^47–49 all of which are involved in POAG pathogenesis. By promoting the endocytosis of Wnt receptors (LRP5/6) and suppressing Wnt/β-catenin signaling,⁵⁰ KREMEN1 may protect against the deleterious effects of oxidative stress and inflammation. 

Myotubularin-related protein 3 (MTMR3) is a member of the myotubularin family and possesses phosphoinositide 3-protein tyrosine phosphatase activity. Through its N-terminal phosphatase domain, MTMR3 dephosphorylates phosphatidylinositol 3-phosphate (PtdIns3P) and PtdIns(3,5)P₂.^51,52 PtdIns3P plays a crucial role in recruiting effectors to autophagic membranes; thus, MTMR3 is known to inhibit autophagy.⁵³ Excessive activation of autophagy has been associated with glaucoma phenotypes, where autophagy exacerbates trabecular meshwork cell damage and reduces their viability. Conversely, inhibition of autophagy has been demonstrated to rescue trabecular meshwork cell survival and prevent apoptosis.⁵⁴ By modulating autophagy and maintaining trabecular meshwork cell homeostasis, MTMR3 may have a protective role in POAG. 

GFPT1 plays a crucial role in synthesizing UDP-GlcNAc, a nucleotide sugar essential for various glycosylation pathways, including N- and O-linked glycosylation.⁵⁵ UDP-GlcNAc serves as a key donor in the N-glycosylation of ECM proteins,⁵⁶ and enhanced protein N-glycosylation promotes ECM protein secretion, resulting in excessive ECM accumulation.⁵⁷ In patients with POAG, abnormal ECM accumulation is observed in the trabecular meshwork.^14,58,59 This accumulation reduces extracellular spacing within the TM, increases aqueous outflow resistance, and subsequently elevates IOP, contributing to glaucomatous pathology.⁶⁰ 

The TRIOBP gene is responsible for encoding a variety of proteins, each of which contributes significantly to the regulation and assembly of the actin cytoskeleton.⁶¹ One of these proteins, TRIOBP-1, is predominantly structured and exhibits widespread expression across different cell types.⁶¹ TRIOBP-1 interacts with F-actin, effectively inhibiting its depolymerization. Research has demonstrated that this protein is essential for numerous cellular functions, including the regulation of the cell cycle, the maintenance of adhesion junctions, and the facilitation of neuronal differentiation.⁶¹ Although TRIOBP lacks suitable SNPs for SMR analysis, TWAS results suggest that TRIOBP may exhibit a protective function in POAG. Prior research demonstrates that the knockout of TRIOBP enhances the granularity and intensity of actin and cell membrane,⁶² thereby increasing mechanical resistance in the trabecular meshwork. This resistance subsequently reduces aqueous humor outflow, resulting in elevated IOP, a characteristic feature of POAG. 

Our research has some limitations. First, the study was conducted exclusively on individuals of European ancestry, which may restrict the applicability of the findings to other ethnic groups. Second, we used three genetic analysis tools—UTMOST, FUSION, and MAGMA—to identify genes associated with POAG by integrating GWAS data with multitissue transcriptomic datasets. To ensure the robustness of our findings, we focused on genes consistently identified by all three methods, refining our analysis through their intersection. This strategy reduced false positives but may lower sensitivity. As a consequence, our findings may ignore certain previously identified POAG-associated genes, especially those with borderline significance or lack of intersection evidence. Third, our approach does not directly determine the change direction of gene expression but infers the roles of specific genes in POAG. Future studies combining systemic transcriptomic comparisons when available may offer additional insight. In addition, owing to the limitations of the dataset, we were unable to assess or validate the expression levels of the identified genes in tissues more directly related to POAG. Further investigations will be required to confirm the proposed pathophysiological mechanisms. 

Our findings suggest that AFAP1, CALCRL, KREMEN1, and MTMR3 play protective roles in POAG pathogenesis. In contrast, GFPT1 acts as a risk factor. These insights provide a deeper understanding of the genetic architecture contributing to POAG and highlight potential targets for therapeutic intervention. A systematic strategy is required to advance these findings toward clinical application: first, validating their precise systemic roles through advanced in vivo models; second, developing targeted therapeutic strategies such as CALCRL agonists or GFPT1 inhibitors and conducting early phase clinical trials to assess safety and efficacy; and third, integrating multi-omics approaches to identify biomarkers for early diagnosis and stratified treatment. Advancing these findings toward clinical application requires integrated efforts across genetics, pharmacology, and ophthalmology, paving the way for novel therapeutic strategies for POAG. 

Acknowledgments

The authors thank BioRender (biorender.com); Figure 1 for this manuscript was created using this software. 

Supported by the National Key R&D Project of China (2020YFA0112701); Guangdong Basic and Applied Basic Research Foundation (2024A1515013058); National Natural Science Foundation of China (General program), No. 8217040283; and Science and Technology Program of Guangzhou, China (202206080005). The funders of the study had no role in the study design; data collection, analysis, or interpretation; writing of the report; or in the decision to submit the paper for publication. 

Author Contributions: Conceptualization: J.C., X.Z., Y.L., Ye.Z., H.T., and L.L.; formal analysis: J.C., X.Z., and Y.L.; funding acquisition: Yi.Z., Ye.Z., and L.L.; investigation: J.C., X.Z., and Y.L.; methodology: J.C., X.Z., and Y.L.; validation: Yi.Z., Z.L., and X.S.; visualization: J.C.; writing-original draft: J.C., X.Z., and Y.L.; writing-review & editing: Yi.Z., Ye.Z., H.T., and L.L. 

Disclosure: J. Chen, None; X. Zhuo, None; Y. Li, None; Y. Zhu, None; Z. Li, None; X. Shen, None; Y. Zhuo, None; H. Tan, None; L. Lei, None