Talaromyces marneffei causes life-threatening opportunistic infections, mainly in Southeast Asia and South China. T. marneffei mainly infects patients with human immunodeficiency virus (HIV) but also infects individuals without known immunosuppression. Here we investigated the involvement of anti–IFN-γ autoantibodies in severe T. marneffei infections in HIV-negative patients. We enrolled 58 HIV-negative adults with severe T. marneffei infections who were otherwise healthy. We found a high prevalence of neutralizing anti–IFN-γ autoantibodies (94.8%) in this cohort. The presence of anti–IFN-γ autoantibodies was strongly associated with HLA-DRB1*16:02 and -DQB1*05:02 alleles in these patients. We demonstrated that adult-onset acquired immunodeficiency due to autoantibodies against IFN-γ is the major cause of severe T. marneffei infections in HIV-negative patients in regions where this fungus is endemic. The high prevalence of anti–IFN-γ autoantibody–associated HLA class II DRB1*16:02 and DQB1*05:02 alleles may account for severe T. marneffei infections in Southeast Asia. Our findings clarify the pathogenesis of T. marneffei infection and pave the way for developing novel treatments.
Talaromyces (Penicillium) marneffei is an important intracellular fungal pathogen that can cause severe systemic infection. It is a thermally dimorphic fungus, presenting septate hyphae at 25°C and transforming to the pathogenic yeast morphology at 37°C during infection. It is endemic to Southern China, Taiwan, Thailand, Laos, Vietnam, Northeast India, and Hong Kong and is almost exclusively restricted to Southeast Asia (Vanittanakom et al., 2006). T. marneffei infections in humans are supposed to occur through inhalation of T. marneffei conidia in the environment. Upon entering the human body, this fungal pathogen can replicate in macrophages in yeast form to cause infection, ranging from local infection in skin and lungs to severe systemic infection (Cao et al., 2019). T. marneffei infections usually occur in immunocompromised individuals with impaired cell-mediated immunity, including secondary immunodeficiency due to HIV infection, cancer, and immunosuppressive therapy. However, T. marneffei infections can also occur in HIV-negative individuals with no obvious immunosuppression (Kauffman et al., 2014; Ramos-e-Silva et al., 2012). The factors underlying a host susceptibility to this infection are unknown, but a potential immunodeficiency is suspected. For example, children with certain inborn genetic disorders, such as cytochrome B-245 β chain (CYBB) and cluster of differentiation 40 ligand (CD40L) mutations, or signal transducer and activator of transcription-1 (STAT1) gain-of-function mutations, are susceptible to severe T. marneffei infection, suggesting T cell–macrophage immunity has a critical role in controlling T. marneffei infection (Kamchaisatian et al., 2006; Lee and Lau, 2017; Lee et al., 2014).
The production of neutralizing anti–IFN-γ autoantibodies is an emerging adult-onset immunodeficiency restricted to specific regions of the globe, including Hong Kong, Thailand, and Taiwan (Browne et al., 2012a; Chi et al., 2013, 2016; Döffinger et al., 2004; Lee et al., 2013; Patel et al., 2005). Sporadic cases have also been reported in Japan, the Philippines, Vietnam, Laos, and other Southeast Asian countries (Chan et al., 2014; Patel et al., 2005; Tanaka et al., 2007). Most patients with neutralizing anti–IFN-γ autoantibodies suffer from disseminated infections with nontuberculous mycobacteria (NTM; Browne et al., 2012a). Other opportunistic infections, such as T. marneffei infection, have also been observed in a few NTM infection patients with anti–IFN-γ autoantibodies (Pruetpongpun et al., 2016; Tang et al., 2010).
We showed previously that anti–IFN-γ autoantibodies in adults with disseminated NTM infections are strongly associated with two specific HLA class II alleles: HLA-DRB1*16:02/DQB1*05:02 and HLA-DRB1*15:02/DQB1*05:01 (Chi et al., 2013; Ku et al., 2016). HLA-DRB1*16:02/DQB1*05:02 is a specific haplotype commonly found in populations in South China and Taiwan, whereas -DRB1*15:02/DQB1*05:01 is more common in patients from Southeast Asia (Middleton et al., 2003). The specific distribution of these risk-associated HLA haplotypes accounts for the geographic/ethnic restriction of anti–IFN-γ autoantibody–related diseases. Given the high frequency of the DRB1*16:02/DQB1*05:02 risk haplotype in South China, we hypothesized that this part of the country, particularly Guangxi, would also be a region with high levels of anti–IFN-γ autoantibody production. To test this hypothesis, we analyzed the presence of anti–IFN-γ autoantibodies and HLA class II haplotypes in HIV-negative patients with T. marneffei infections from South China.
Results and discussion
Characteristics of patients with
T. marneffei infections
We enrolled 58 patients suffering from severe T. marneffei infections. Their demographic characteristics, proven pathogens, sampling sites (listed in order of occurrence), and treatment outcomes are summarized in Table 1. The patients’ mean age was 54.2 yr (range, 22–77), and the study population consisted of 34 men and 24 women. The mean age of the control group was 49.5 yr (range, 23–65), and this control group consisted of 59 men and 48 women. No prior medical condition was reported for 47 of the cases (81.0%). Comorbidities were identified in 11 patients (19.0%), mainly type 2 diabetes mellitus (n = 7), followed by hypertension (n = 3), renal insufficiency (n = 2), thalassemia (n = 1), coronary atherosclerotic heart disease (n = 1), embolism of the pulmonary artery (n = 1), and acute kidney injury (n = 1). None of the patients had a history of cancer, autoimmunity, or any form of immunosuppressive treatment.
Detection of anti–IFN-γ autoantibodies
We evaluated possible adult-onset immunodeficiency due to anti–IFN-γ autoantibodies by performing an indirect ELISA on serially diluted plasma from healthy donors and patients to test for the presence of anti–IFN-γ autoantibodies. Anti–IFN-γ autoantibodies were detected (OD >0.5, in 1:100 dilution) in most patients (55/58 cases; Fig. 1 A). In the inhibition assay, we incubated a fixed concentration (100 pg/ml) of recombinant IFN-γ with serially diluted plasma from the patients. Recombinant human IFN-γ was almost undetectable in the 55 patients defined as anti–IFN-γ autoantibody–positive, even though the plasma was diluted significantly (1:200,000 dilutions for 16 patients; Fig. 1 B). Further analysis showed that most of these autoantibodies were of the IgG1 and IgG4 types (Fig. S1).
We then evaluated these anti–IFN-γ autoantibodies’ biological effect by assessing whether they could neutralize the IFN-γ-induced HLA-DR expression and STAT-1 phosphorylation in THP-1 cells (Fig. 1, C–E). Plasma from all anti–IFN-γ autoantibody–positive patients decreased IFN-γ–induced HLA-DR expression and STAT-1 phosphorylation, but no inhibitory effect was observed with plasma from healthy donors or the three antibody-negative patients. IFN-γ had been shown to enhance the clearance of T. marneffei in myeloid cells (Sisto et al., 2003). Consistent with this idea, plasma from anti–IFN-γ autoantibody–positive patients could impair the IFN-γ–mediated clearance of T. marneffei in THP-1 cells (Fig. S2).
We used a bead-based multiplex system to screen for the presence of autoantibodies on five other cytokines (GM-CSF, IL-6, IL-17A, IL-12, and IL-23), which have been linked to immune dysregulation and infection (Ku et al., 2020). Anti–GM-CSF autoantibodies were found in one patient (case 28), but these autoantibodies to GM-CSF were nonneutralizing, and this patient has no anti–GM-CSF autoantibody–related disease (Table 1 and Fig. S3). No other anti-cytokine autoantibodies were observed in the remaining T. marneffei–infected patients. 39 HIV patients (19 of whom had T. marneffei infections) were screened for anti–IFN-γ autoantibodies, and all were negative. In addition, we performed whole-exome sequencing with the DNA from the 50 anti–IFN-γ autoantibody–positive and 3 anti–IFN-γ autoantibody–negative cases. Neither reported pathogenic nor homozygous/compound heterozygous-predicted deleterious single-nucleotide polymorphisms (SNPs) were identified in Mendelian susceptibility to mycobacterial disease (MSMD)–associated genes (IL12RB1, IL12B, ISG15, SPPL2A, IRF8, TYK2, IFNGR1, IFNGR2, STAT1, NEMO, CYBB, IL23R, IFNG, or IL12RB2) and chronic mucocutaneous candidiasis-associated genes (IL17F, IL17RC, IL17RA, ACT1, CARD9, STAT3, RORC, and MAPK8; Bustamante, 2020; Kerner et al., 2020; Li et al., 2017, 2019). The three heterozygous SNPs (IL12RB2 c.G1003T, p.D335Y; SPPL2A c.T995C, p.L332P; and IFNGR2 c.C115T, p.R39C) identified in the three patients (cases 1, 11, and 33, respectively) were predicted as probably damaging. Nevertheless, the immunodeficiency by autosomal dominant inheritance of IL12RB2 or SPPL2A has never been reported, and the frequency of IFNGR2 c.C115T is relatively common (∼0.07%) in East Asians. Thus, these SNPs are unlikely to cause the immunological defect observed, and no unifying genetic theory could be found for this patient cohort. Together, we demonstrated that neutralizing anti–IFN-γ autoantibodies were present in the plasma of 55/58 (94.8%) patients with severe T. marneffei infections.
Clinical manifestations in the 55 patients with anti–IFN-γ autoantibodies
In the 55 patients with anti–IFN-γ autoantibodies (cases 1–55; Table 1), common clinical features of T. marneffei infections included fever (85.5%), cough (80.0%), weight loss (78.2%), and lymphadenopathy (76.4%; Table 2). Multiple organ involvement was observed in these patients. The lungs were affected most frequently (100%), followed by the lymph nodes, skin, bones/joints, liver, and spleen (78.2%, 47.3%, 23.6%, 14.5%, and 9.1%, respectively; Table 2). Seven patients presented reactive skin lesions, five patients had Sweet’s syndrome, and two had generalized pustulosis. Laboratory investigations revealed leukocytosis, thrombocytosis, anemia, and high CD4+ and CD8+ counts (Table 3). Before and after T. marneffei infections, 30 patients (54.5%) were also infected with other opportunistic pathogens, including NTM (n = 7; 12.7%), varicella zoster virus (VZV; n = 8; 14.5%), CMV (n = 3; 5.5%), Candida albicans (n = 3; 5.5%), Mycobacterium tuberculosis (tuberculosis [TB]; n = 3; 5.5%), Salmonella spp. (n = 4; 7.3%), Klebsiella pneumoniae (n = 3; 5.5%), EBV (n = 2; 3.6%), Burkholderia cepacia (n = 1), Aeromonas veronii (n = 1), Acinetobacter baumannii (n = 1), Aspergillus (n = 1), and Enterobacter cloacae (n = 1; Table 1). Based on computed tomography and/or biopsy findings similar to those for TB, anti-TB antibiotics were given to some patients before T. marneffei infections were confirmed. Except for 10 cases of confirmed mycobacterial infections, TB infections were excluded by additional computed tomography scans, staining for acid-fast bacilli, and/or nonresponse to anti-TB treatment.
We followed disease progression in 43 patients with anti–IFN-γ autoantibodies. 13 died despite antimicrobial treatment (Table 1), and 19 suffered recurrent T. marneffei infections or subsequently developed other opportunistic infections, both requiring long-term antimicrobial treatment and severely affecting the patients’ quality of life. Only eight patients were cured after antimicrobial treatment (cure was defined as sustained resolution 6 mo after the end of treatment; Ouyang et al., 2017), and three patients sustained resolution for >3 mo after the end of treatment. Mortality was higher in these patients than it was in the HIV-positive patients investigated in a previous study (Kawila et al., 2013).
HLA alleles associated with anti–IFN-γ autoantibodies
We showed previously that the HLA-DRB1*16:02/DQB1*05:02 and DRB1*15:02/DQB1*05:01 haplotypes are strongly associated with the presence of anti–IFN-γ autoantibodies (Ku et al., 2016). Therefore, we analyzed the HLA–class II molecules DRB1 and DQB1 in patients with T. marneffei infections. In total, 16 DQB1 alleles and 25 DRB1 alleles were detected in patients and healthy controls (Table 4 and data not shown). We showed that DRB1*16:02 (n = 47; 85.4%) and DQB1*05:02 (n = 48; 87.3%) were more frequent in patients with anti–IFN-γ autoantibodies than in the control population (22.4% in controls for DRB1*16:02, and 43.9% for DQB1*05:02). Additionally, we found that 98.2% of the anti–IFN-γ autoantibody–positive patients carried the DRB1*15:02 or DRB1*16:02 alleles. The presence of the DRB1*16:02 or DQB1*05:02 allele was strongly associated with severe T. marneffei infections, and also associated with the production of anti–IFN-γ antibodies with odds ratios (ORs) of 20.32 (95% CI, 8.46–48.81; P = 1.92 × 10−14; corrected P value [Pc] = 4.80 × 10−13) and 8.75 (95% CI, 3.63–21.11; P = 1.13 × 10−7; Pc = 1.8 × 10−6), respectively (Table 4).
T. marneffei is considered an opportunistic fungal pathogen specifically found in South China and Southeast Asia (Vanittanakom et al., 2006). Here, we showed that, in addition to the previously reported NTM infections, anti–IFN-γ autoantibodies were highly prevalent in the patients with severe T. marneffei infections from the Guangxi region. We further demonstrated that these patients carried the HLA-DRB1*16:02/DQB1*05:02 haplotype, which was associated strongly with anti–IFN-γ autoantibodies, as we showed previously (Chi et al., 2013). Overall, our results suggest that autoantibodies against IFN-γ are the main etiology of T. marneffei infections in HIV-negative individuals, in which the risk HLA–class II alleles HLA-DRB1*16:02 and -DQB1*05:02 are highly prevalent.
The strong association between anti–IFN-γ autoantibodies and HLA-DRB1*16:02/DQB1*05:02 observed in this study and our previous work (Chi et al., 2013) suggests a pathogenic role of this HLA haplotype in anti–IFN-γ autoantibody generation. The geographic distribution of HLA-DRB1*16:02/DQB1*05:02 and T. marneffei largely overlaps in Guangxi, Guangdong, and Yunnan (provinces in South China; Ku et al., 2016). The high epidemicity of T. marneffei infections in these regions, in which bamboo rats serve as a reservoir of this species (Cao et al., 2011), is likely increased markedly by the high prevalence of anti–IFN-γ autoantibodies and the underlying HLA haplotype (Deng et al., 1988; Jiang et al., 2019; Lee and Lau, 2017). Indeed, severe T. marneffei infections are generally considered opportunistic infections found in a few patients with anti–IFN-γ autoantibodies (Chan et al., 2013; Tang et al., 2010). We further provided a large cohort of patients with T. marneffei infections, demonstrating that anti–IFN-γ autoantibody production is the major etiology of T. marneffei infections in previously healthy adults.
The discovery of this acquired IFN-γ deficiency in patients with T. marneffei infection suggests IFN-γ has a role in combating this fungal infection in humans. The IFN-γ’s biological role in humans has mostly been elucidated by studies of children with MSMDs (Bustamante et al., 2014; Casanova and Abel, 2002). MSMD patients with defects in the IFN-γ circuit mainly present with severe to lethal infections, and saprophytic nontuberculosis mycobacteria is one of those opportunistic pathogens. Similar to mycobacteria, T. marneffei is capable of replicating in human macrophages (Roilides et al., 2003). IFN-γ plays a key role in activating phagocytes to clear engulfed pathogens. Impairment of IFN-γ function may disturb the clearance of intracellular pathogens from various phyla, not only intracellular bacteria. Consistent with this view, a severe T. marneffei infection was recently reported in a Thai child with IFNGR1 deficiency (Lee and Lau, 2017). Therefore, IFN-γ-mediated macrophage activation and microbicidal activity are critical to control T. marneffei infections. Nevertheless, the detailed mechanism by which IFN-γ contributes to the control of T. marneffei in vivo remains to be determined. Further studies in humans and animal models are warranted to address this question.
In our cohort, only 12.7% of patients (7/55) with anti–IFN-γ autoantibodies experienced NTM infections (Table 1). NTM infections are the most common clinical manifestations of anti–IFN-γ autoantibody disease; however, the low NTM infection rate in our cohort might be important for future consideration. NTM are common saprophytes that are found in diverse natural ecosystems. Therefore, the rate of microbial exposure is unlikely to account for the lower prevalence of NTM infections compared with T. marneffei infections in patients with anti–IFN-γ autoantibodies in Guangxi (Claeys and Robinson, 2018). We speculate that T. marneffei is more pathogenic than NTM, and anti–IFN-γ autoantibodies’ neutralizing effects might be less potent in patients with T. marneffei infections. Whereas, residual IFN-γ activity may be sufficient to control the dissemination of NTM infections in these patients. Further studies are needed to test this hypothesis.
Persistent, life-threatening infections have been observed in a significant portion of our patients, even in those undergoing anti-fungal antibiotic treatment. B cell–depletion treatments, such as anti-CD20 monoclonal antibody therapy, were applied in anti–IFN-γ autoantibody patients who failed to response to antibiotics (Browne et al., 2012b). By eliminating the antibody-producing B cells, the titer of autoantibody decreased, and the endogenous IFN-γ–mediated immunity against pathogens was restored. This indicates that administering anti-CD20 antibody therapy may be useful for treating anti–IFN-γ–positive patients who suffer from recurrent T. marneffei infections.
In conclusion, we show that the adult-onset acquired immunodeficiency, mediated by anti–IFN-γ autoantibodies, is the most important factor underlying severe T. marneffei infections in HIV-negative patients. Based on our findings, we strongly recommend screening for anti–IFN-γ autoantibodies in HIV-negative patients suffering from severe T. marneffei infections, even if they do not have NTM infections. These findings not only improve our understanding of the pathogenesis of this disease but may also point out the anti–IFN-γ autoantibody–producing B cells as novel therapeutic targets for these patients in the near future.
Materials and methods
We conducted a 5-yr (from January 2013 to June 2018) prospective, cross-sectional, case-control study at the First Affiliated Hospital of Guangxi Medical University. The inclusion criteria for cases were (1) over the age of 18 yr; (2) negative for anti-HIV antibodies and no obvious immunosuppressed condition, such as malignant tumors and organ transplantation; (3) presented with histologically and culture-proven T. marneffei infections; and (4) provided informed consent for participation. Among the 174 patients with T. marneffei infection identified during this period, 108 cases were excluded due to HIV infection (101 cases) or other immunosuppressed conditions (7 cases). Informed consent was obtained from 58 of the 66 remaining cases. We followed these patients after first diagnosis and/or during the active stage of T. marneffei infection. Complete histories were obtained, and physical examinations, including routine clinical laboratory tests, were performed on all patients. The healthy control group was recruited from the physical examination center from the same hospital. The Institutional Ethics Review Board of the First Affiliated Hospital of Guangxi Medical University approved the study, which was conducted in accordance with Good Clinical Practice and the Declaration of Helsinki.
Determination of autoantibodies against IFN-γ
Anti–IFN-γ autoantibodies in plasma were measured by indirect ELISA according to a modified version of a previously described procedure (Chi et al., 2013). Briefly, 100 µl of 2 µg/ml recombinant human IFN-γ (R&D Systems), in carbonate buffer (pH 9.5), were added to each well in a 96-well flat-bottomed MaxiSorp plate (Nunc), and the plate was incubated overnight at 4°C. The next day, the wells were washed three times with wash buffer (0.05% Tween-20/PBS); 100 µl of blocking buffer (5% human albumin in PBS) were then added to each well, and the plate was incubated for 2 h at room temperature. Plasma samples were diluted serially with blocking buffer. Diluted plasma (100 µl) was added to each well, and the plate was incubated for 2 h at room temperature. The plate was washed four times with wash buffer, and 100 µl of horseradish peroxidase–conjugated goat anti-human IgG (1:5,000 diluted; Invitrogen) were added to each well. The plate was incubated for 1 h at room temperature and then washed four times with wash buffer. Next, 100 µl of tetramethylbenzidine (Sigma-Aldrich) were added to each well and incubated for 15 min at room temperature. Tetramethylbenzidine stop solution (100 µl; Southern Biotech) was added to each well to stop the reaction, and OD was determined at 450 nm. Samples with OD values >0.5 were considered positive for antibodies against IFN-γ, and this result was confirmed by inhibition assays and functional tests. ELISA was performed in duplicate for all samples.
IFN-γ inhibition assay
Plasma samples were serially diluted (1:20; 1:200; 1:2,000; 1:20,000, and 1:200,000) and incubated with recombinant human IFN-γ at a final concentration of 100 pg/ml for 3 h at room temperature. IFN-γ levels were determined with a human IFN-γ ELISA kit (BD Biosciences) according to the manufacturer’s instructions. Inhibition assays were performed in duplicate for all samples.
Functional test for autoantibodies against IFN-γ
The autoantibodies’ neutralizing activity against IFN-γ was assessed by evaluating their ability to reduce the IFN-γ–induced HLA-DR expression on THP-1 cells. THP-1 cells were cultured in complete RPMI-1640 medium containing 10% FBS (Gibco) and 1% penicillin/streptomycin (Gibco). THP-1 cells (2 × 105 cells/ml) were incubated with human recombinant IFN-γ (10 ng/ml, R&D Systems), in the presence or absence of plasma (10 µl), from anti–IFN-γ autoantibody–positive patients or healthy donors for 24 h at 37°C. HLA-DR expression was measured with a PE–anti–HLA-DR antibody (555812; BD Biosciences) by flow cytometry (FACS Canto II; BD), and the results were analyzed with FlowJo VX software. To measure STAT-1 phosphorylation, 2 × 105 peripheral blood mononuclear cells were placed in 200 µl of RPMI-1640 with 10% FBS (Geneteks) and 1% penicillin/streptomycin. Cells were then stimulated with 20 IU or 200 IU IFN-γ in RPMI-1640, which were preincubated with 20% normal plasma or patient plasma for 20 min at room temperature. After 30 min of stimulation in a 37°C incubator, cells were fixed and permeabilized using lysing solution (BD). The PE-phospho-STAT1 (pY701) antibody (BD PharMingen) was applied, and data were collected and analyzed with a FACSVerse flow cytometer and FACSuite software (BD Biosciences).
Blood from all participants was collected into EDTA for DNA extraction with the DNeasy Blood & Tissue Kit (Qiagen) according to the manufacturer’s instructions. The HLA-DQB1 and HLA-DRB1 polymorphisms in exon 2 of the target genes were investigated using a sequence-based typing method (Al-Hussein et al., 2002; Perz et al., 2007). All primers were synthesized by Shanghai Health (Table S1). The sequencing results were compared with the latest known allele sequences from an authoritative database (the human major histocompatibility complex section of the international immunogenetics database, ImMunoGeneTics/HLA) to identify the allele that was present in each subject (Robinson et al., 2001).
The proportions of HLA allele carriage were determined by direct counting. The allelic OR, 95% confidence interval, and P values for two-tailed tests were obtained with SPSS (version 22.0, IBM). We corrected for multiple testing by calculating Pc using the Bonferroni method. A value of Pc < 0.05 was considered significant.
Online supplemental material
Fig. S1 shows IgG subtypes of anti–IFN-γ autoantibody in our cohort. Fig. S2 provides evidence through T. marneffei killing assay that patients’ plasma with anti–IFN-γ autoantibody obviously blocked the IFN-γ clearance of T. marneffei. Fig. S3 shows the autoantibodies against GM-CSF found in case 28 plasma. Table S1 provides the primer set we used to amplify HLA-DQB1 and -DRB1 exon 2.
We thank the clinicians who referred their patients to us and the patients for agreeing to participate in the study.
This study was supported by grants from the National Natural Science Foundation of China (81571971 and 81271804) and the Natural Science Foundation of the Guangxi Province of China (2017GXNSFAA198004). C.-L. Ku was supported by grants from Chang Gung Memorial Hospital (CMRPD1A0681-3, CMRPD1F0171-3, CMRPD1I0151-2, and BMRP) and the Taiwan Ministry of Science and Technology (105-2628-B-182-002-MY3 and 108-2628-B-182-005). The funders had no role in study design, data collection and analysis, the decision to publish, or preparation of the manuscript.
Author contributions: J. Guo, X.-Q. Ning, J.-Y. Ding, H.-P. Shih, C.-L. Ku, and C.-W. Cao designed the experiments. J. Guo, X.-Q. Ning, and J.-Y. Ding conducted the experiments. J. Guo, X.-Q. Ning, Y.-Q. Zheng, J.-Y. Ding, H.-P. Shih, H.-T. Ting, Y.-F. Lo, I.Y.-F. Chang, C.-F. Yeh, K.-H. Tu, Y.-H. Tsai, C.-L. Ku, and C.-W. Cao analyzed and/or interpreted the data. N.-N. Shi, F.-Y. Wu, Y.-K. Lin, G. Liang, X.-C. Lu, J.-L. Kong, K. Wang, Y.-B. Lu, Y.-J. Fu, R. Hu, T.-M. Li, K.-S. Pan, X.-Y. Li, C.-Y. Huang, and C.-W. Cao provided samples and performed clinical diagnoses and patient follow-ups. C.-L. Ku and C.-W. Cao supervised the study. X.-Q. Ning, J. Guo, J.-Y. Ding, Y.-H. Tsai, C.-L. Ku, and C.-W. Cao wrote the manuscript and designed the figures, to which all other authors contributed. All of the authors revised the manuscript.
Disclosures: The authors declare no competing interests exist.
J. Guo, X.-Q. Ning, J.-Y. Ding, and Y.-Q. Zheng contributed equally to this paper.
C.-L. Ku and C.-W. Cao contributed equally to this paper.