Purpose: Colorectal cancer is a worldwide cancer with rising annual incidence. Inflammation is a well-known cause of colorectal cancer carcinogenesis. Metabolic inflammation (metaflammation) and altered gut microbiota (dysbiosis) have contributed to colorectal cancer. Chemoprevention is an important strategy to reduce cancer-related mortality. Recently, various polypharmacologic molecules that dually inhibit histone deacetylases (HDAC) and other therapeutic targets have been developed.
Experimental Design: Prevention for colitis was examined by dextran sodium sulfate (DSS) mouse models. Prevention for colorectal cancer was examined by azoxymethane/dextran sodium sulfate (AOM/DSS) mouse models. Immunohistochemical staining was utilized to analyze the infiltration of macrophages and neutrophils and COX-II expression in mouse tissue specimens. The endotoxin activity was evaluated by Endotoxin Activity Assay Kit.
Results: We synthesized a statin hydroxamate that simultaneously inhibited HDAC and 3-hydroxy-3-methylglutaryl coenzyme A reductase (HMGR). Its preventive effect on colitis and colitis-associated colorectal cancer in mouse models was examined. Oral administration of this statin hydroxamate could prevent acute inflammation in the DSS-induced colitis and AOM/DSS–induced colorectal cancer with superior activity than the combination of lovastatin and SAHA. It also reduced proinflammatory cytokines, chemokines, expression of COX-II, and cyclin D1 in inflammation and tumor tissues, as well as decreasing the infiltration of macrophages and neutrophils in tumor-surrounding regions. Stemness of colorectal cancer and the release of endotoxin in AOM/DSS mouse models were also attenuated by this small molecule.
Conclusions: This study demonstrates that the polypharmacological HDAC inhibitor has promising effect on the chemoprevention of colorectal cancer, and serum endotoxin level might serve as a potential biomarker for its chemoprevention. Clin Cancer Res; 22(16); 4158–69. ©2016 AACR.
Colorectal cancer is a worldwide cancer, causing significant morbidity and mortality, and chemoprevention is one of the potentials to reduce cancer-related mortality. Here, we design a polypharmacologic small molecule, statin hydroxamate, that dually inhibits histone deacetylases (HDAC) and 3-hydroxy-3-methylglutaryl coenzyme A reductase. It attenuated dextran sodium sulfate (DSS)–induced colitis and azoxymethane/DSS (AOM/DSS)–induced colorectal cancer in mice through reducing proinflammatory cytokines, chemokines, expression of COX-II, and cyclin D1 in inflammation and tumor tissues, as well as decreasing the infiltration of macrophages and neutrophils in tumor-surrounding regions. Stemness of Colorectal cancer and the release of endotoxin in AOM/DSS mouse models were also attenuated by this small molecule. These results demonstrate that the polypharmacologic HDAC inhibitor has promising effect on the chemoprevention of Colorectal cancer, and serum endotoxin level might serve as a potential biomarker for its chemoprevention.
Colorectal cancer is a worldwide cancer, causing significant morbidity and mortality, and chemoprevention is one of the potentials to reduce cancer-related mortality by suppressing the initial phase of carcinogenesis or the progression of premalignant cells (1). Meta-analysis studies have shown that aspirin could reduce the incidence of colon polyps and cancer (2). The mechanisms were in part through inhibition of COX-II–related pathways and normalizing EGFR expression (3). However, adverse effects prohibit its clinical use. Thus, new strategy for chemoprevention of colorectal cancer meets an unmet medical need.
Inflammation through inflammation–dysplasia–carcinoma process is an important cause of carcinogenesis to induce colorectal cancer (4, 5). Inflammatory bowel disease (IBD) referring to both Crohn's disease and ulcerative colitis is a chronic disorder. Patients with IBD exhibiting a 2- to 8-fold higher risk of colorectal cancer, is a classic example to highlight the association between chronic inflammation (colitis) and colorectal cancer (4, 6). In addition, both sporadic and heritable colorectal cancer are also related to inflammation (6, 7). Metabolic syndrome (i.e., hypertension, type II diabetes, hyperlipidemia, and obesity) commonly resulted from excess nutrients and energy has been found to induce a low-grade, chronic, metabolically linked inflammatory state termed metaflammation (8). Patients having metabolic syndromes show higher incidence of colorectal cancer and higher recurrence rate of colon adenoma that demonstrated the close association between inflammation and colorectal cancer (8). Epidemiologic studies reported that anti-inflammatory medications effectively reduce the incidence of colorectal cancer among IBD patients (9, 10). Therefore, inhibition of inflammation is a potential strategy to prevent cancer initiation.
Huge amount of bacteria flora inhabited in the intestine interacting with mucosa maintaining gut immunity and homeostasis (11). Recent studies suggest that imbalance of bacterial flora, termed dysbiosis, favors higher abundance of gram-negative bacteria (GNB; refs. 4, 12, 13). Endotoxin, which is a lipopolysaccharide (LPS) from the cell wall of GNB, is associated with various chronic diseases, such as atherosclerosis, diabetes, and cancers (13). The release of endotoxin by GNB is an important mechanism of dysbiosis-induced colorectal cancer (14). In addition to endotoxin, cancer stem cell (CSC) is another crucial factor for colorectal cancer initiation (15). Chronic gut inflammation leads to crypt damage and repeated regeneration, resulting in CSC expansion leading to colorectal tumorigenesis (16, 17).
Statins are 3-hydroxy-3-methyl glutaryl coenzyme A (HMG-CoA) reductase (HMGR) inhibitors treating hypercholesterolemia in patients with coronary artery and atherosclerotic diseases (18, 19). They have also shown promise in cancer prevention from observational and preclinical and certain aspects of randomized controlled studies (20). We first demonstrated statins with a carboxylic acid–containing long chain possessing the structure similar to hydroxamate that also inhibits histone deacetylase (HDAC) activity, which is an attractive target against cancer (21, 22). To further improve statins' activity against HDAC, three statin hydroxamates were designed to inhibit class I and II HDACs and HMGR (23). We named compound 12, 13, and 14 as JMF3086, 3171, and 3173, respectively, in this study. Recently, the strategy of developing polypharmacological molecules that dually inhibit HDACs and other therapeutic targets has been reported (24).
Oral administration of JMF3086 attenuated dextran sodium sulfate (DSS)–induced colitis and azoxymethane (AOM)/DSS–induced colorectal cancer in mice. A positive correlation between HMGR and HDAC activity and prevention of colorectal cancer was seen. JMF3086 exerted anti-inflammatory effect and inhibited the in vivo release of endotoxin. Furthermore, the stemness of colorectal cancer was inhibited in vivo. These results show that our polypharmacological molecule JMF3086 targeting HDACs and HMGR possesses great potential for the prevention of colitis and colitis-associated colorectal cancer.
Materials and Methods
Lovastatin was obtained from Lotus Pharmaceutical Co. SAHA was obtained and purchased from Merck. Anti-acetyl-histone H3 and anti-acetyl-histone H4 antibodies were obtained from Millipore. Anti-acetyl-tubulin antibody was obtained from Sigma. Anti-PARP1/2 antibody was purchased from Santa Cruz Biotechnology. Anti-caspase-3 antibody and anti-BAX antibody were purchased from Cell Signaling Technology. Anti-β-actin antibody was purchased from GeneTex. HDAC Fluorometric Assay/Drug Discovery Kit (AK-500) was purchased from Biomol. HMG-CoA reductase activity kit (CS-1090) and AOM were purchased from Sigma. DSS was purchased from MP Biomedicals. Endotoxin Activity Assay Kit was purchased from Spectral Diagnostics.
C57BL/6, BALB/c, and NOD/SCID mice were obtained from the National Laboratory Animal Center (Taipei City, Taiwan). Mice were performed in accordance with protocols approved by the Institutional Animal Care and Use Committee of the College of Medicine, National Taiwan University (Taipei, Taiwan).
Animal models of DSS-induced acute colitis
The DSS-induced colitis animal model exhibits many phenotypes that are relevant to human ulcerative colitis, including inflammation and ulceration of the colonic mucosa (4). Colitis was induced in seven-week-old male C57BL/6 mice by adding DSS to their drinking water for 5 days, followed by switching to regular drinking water. In the preventive model, test compounds were dissolved in corn oil and administered orally 7 days before, during, and after 3.5% (w/v) DSS treatment (Fig. 1A). In the therapeutic model, test compounds were dissolved in corn oil and administered orally 5 days following the 2.5% (w/v) DSS treatment (Fig. 2A).
Animal model of AOM/DSS–induced colorectal cancer
AOM in conjunction with the additional inflammatory stimulus of DSS results in tumor development that is restricted to the colon in mice (4, 6). This model is widely utilized to recapitulate human colorectal cancer because it results in inflammation and ulceration of the entire colon, similar to what is observed in patients (6). Colorectal cancer was induced in male C57BL/6 mice via the intraperitoneal injection of AOM (12.5 mg/kg) while the mice were maintained with a regular diet and drinking water for 7 days and then subjecting the mice to 3 cycles of DSS treatment, with each cycle consisting of the administration of 3.5% DSS for 5 days, followed by a 14-day recovery period with regular water (Fig. 3A). In the preventive model of AOM/DSS–induced colorectal cancer, test compounds were dissolved in corn oil and administered orally 5 days a week for 8 weeks, simultaneously with AOM exposure.
Clinical assessment of DSS-induced acute colitis and AOM/DSS–induced colorectal cancer
Body weight, the presence of occult or gross blood in the rectum, and stool consistency were determined daily in the mice. The weight change during the experiment was calculated as the percent change in weight compared with the baseline measurement. Bleeding was scored as 0, when there was no blood in the Hemoccult test; 1, for a positive Hemoccult result; 2, for slight bleeding; or 3, for gross bleeding. Regarding stool consistency, 0 points were given for well-formed pellets, 1 point for semiformed stools that did not adhere to the anus, 2 points for pasty stools, and 3 points for liquid stools that adhered to the anus.
HDAC activity assay
The HDAC activity was performed using HDAC Fluorescent Activity Assay Kit. Total lysate from mouse tissue was mixed with Fluor-de-Lys substrate for 10 minutes at 37°C, followed by adding developer to stop the reaction. Fluorescence was measured by fluorometric reader with excitation at 360 nm and emission at 460 nm. The IC50 values were calculated from SigmaPlot software.
HMG-CoA reductase activity assay
The HMG-CoA reductase activity was performed using HMG-CoA Reductase Assay Kit. Total lysate from mouse tissue was mixed with NADPH and HMG-CoA and incubated for 5 minutes at 37°C. The absorbance at 340 nm was measured. The IC50 values were calculated from SigmaPlot software.
IHC was performed with One Step Polymer-HRP Detection Kit (BioGenex) on sections from 10% paraffin-embedded samples according to the manufacturer's protocols. Pictures were acquired using TissueFAXS (TissueGnostics), and the results of DAB-positive cells were presented as scattergrams plot using HistoQuest software (TissueGnostics). The cut-off values for background staining were chosen manually using the forward/backward gating tools of the software.
Endotoxin activity assay
The endotoxin activity was performed using Endotoxin Activity Assay Kit. Serum from mouse whole blood was mixed with control standard endotoxin and limulus amoebocyte lysate reagent water at 37°C. The absorbance at 405 nm was measured.
The SPSS program (SPSS Inc.) was used for all statistical analyses. Statistical analysis was performed by two-sided Student t test. Data shown were representative of at least three independent experiments. Quantitative data are presented as means ± SD.
Detailed methodology is described in the Supplementary Material.
Statin hydroxamates prevent inflammation in the DSS-induced colitis
As inhibition of inflammation is a potential strategy to prevent cancer initiation, the acute colitis induced by DSS in C57BL/6 mouse model was employed to evaluate the anti-inflammation efficacy of JMF3086. JMF3086 (50 mg/kg) was orally administered for 7 days before treatment with 3.5% DSS began and was continued until the end of the experiment (Fig. 1A). It prevented the colitis symptoms (i.e., body weight loss, diarrhea, and rectal bleeding) and was more effective than the combination of lovastatin and SAHA (Fig. 1B).
The reduction in colon length that occurred in DSS-treated mice was prevented by JMF3086 (Fig. 1C). Histopathologically, DSS-treated mice exhibited complete destruction of the epithelial architecture with a loss of crypts and epithelial integrity and significant infiltration of neutrophils (Ly-6G) but not of macrophages (F4/80). In JMF3086-treated mice, the crypt architecture was maintained, and only a mild infiltration of neutrophils in the mucosa was detected (Fig. 1D and E). Furthermore, elevated COX-II expression was observed in the epithelial cells of DSS-treated mice, and this increase was prevented by JMF3086 (Fig. 1E). DSS-induced acute colitis is predominantly characterized by a cytokine response (25). TNFα, IFNγ, IL6, and IL12 secretion from colons in the DSS-treated mice were found to be markedly increased and prevented by JMF3086 treatment (Fig. 1F).
The preventive effect of JMF3171 and JMF3173 on DSS-induced colitis was also examined and found to exert nice efficacy in colitis symptoms, as well as the reductions of colon length, infiltration of neutrophils, destruction of the epithelial architecture, and cytokine releases (Supplementary Fig. S1).
JMF3086 treats DSS-induced colitis
We further examined the efficacy of JMF3086 in treating DSS-induced colitis in mouse models. The colitis symptoms were inhibited by JMF3086, which was more effective than the combination of lovastatin and SAHA (Fig. 2B and C). Microscopically, JMF3086 reduced DSS-induced colonic ulceration (Fig. 2D). Furthermore, infiltration of neutrophils as well as the secretion of proinflammatory cytokines and chemokines in the colons was attenuated by JMF3086 (Fig. 2E and F). Elevated serum endotoxin level found in DSS mice was reduced by JMF3086 (Supplementary Fig. S2).
JMF3086 attenuates the colitis symptoms and tumor growth in AOM/DSS–induced colorectal cancer mouse models
As JMF3086 could prevent and treat acute inflammation in the DSS-induced colitis in mice, the chemoprevention of JMF3086 was further evaluated in AOM/DSS–induced colorectal cancer mouse models. JMF3086 (25 or 50 mg/kg) was administered orally 5 days per week concurrent with AOM injections (Fig. 3A). The effect of lovastatin (50 mg/kg) and SAHA (50 mg/kg) either individually or in combination was compared. Symptomatic parameters, such as loss of body weight, diarrhea, and rectal bleeding, were observed in mice after AOM/DSS treatment, and JMF3086 prevented the occurrence of these symptoms (Fig. 3B). Colon length was decreased in AOM/DSS–treated mice, but not in those treated with JMF3086 (Fig. 3C).
Polyps and colon tumors were macroscopically observed and counted, and H&E staining revealed adenocarcinomas with dysplasia in AOM/DSS–treated mice (Fig. 4A and B and Supplementary Fig. S3A). JMF3086 at a dose of 50 mg/kg decreased tumor numbers and sizes and protected against AOM/DSS–induced dysplasia/adenocarcinoma lesion to surface tumor necrosis (Fig. 4A and B).
To examine whether HMGR and HDAC were inhibited by JMF3086 in vivo, their activity was analyzed in cell lysates from colon tissue and found to be increased in AOM/DSS–induced colorectal cancer. JMF3086 inhibited these effects and induced acetylations of histone H3, H4, and tubulin. Its accumulation in colon tissues detected by LC/MS-MS and ESI-MS analyses was seen (Fig. 4C and Supplementary Figs. S3B and S4A–S4C). There was a positive correlation between HMGR and HDAC activity and tumor number and size (Fig. 4D), indicating that inhibition of both activities contributed to the prevention of JMF3086 on AOM/DSS–induced colorectal cancer in mice.
JMF3086 reduces the mRNA expression of proinflammatory cytokines/chemokines and intracolonic infiltration of macrophages and neutrophils in AOM/DSS–induced colorectal cancer mouse models
JMF3086 potently decreased the mRNA expression of proinflammatory cytokines and chemokines in the colons of AOM/DSS–induced colorectal cancer mice, and the mRNA expression of COX-II, which is an important tumor marker for colorectal cancer, as well as the proliferation marker cyclin D1 (Fig. 5A). JMF3086 also reduced the intracolonic infiltration of inflammatory cells, such as macrophages and neutrophils (F4/80 and Ly-6G), and decreased COX-II expression around the sites of tumors (Fig. 5B and Supplementary Fig. S5). The efficacy of JMF3086 was superior to the combination of lovastatin and SAHA (Figs. 5B, 4A and B). In addition, JMF3086 did not induce in vivo apoptosis despite of protecting AOM/DSS–induced colorectal cancer (Fig. 5C and D).
JMF3086 treatment did not exhibit any toxicity based on biochemical examinations and H&E staining of various organs (Supplementary Fig. S6A and S6B). However, combination of lovastatin and SAHA increased serum creatinine levels and reduced the concentration of albumin (Supplementary Fig. S6A), indicating renal toxicity.
JMF3086 prevents cancer stemness and endotoxin release in AOM/DSS–induced colorectal cancer mouse models
It has been suggested that chronic inflammation causes crypt damage and regeneration, resulting in stem cell expansion and leading to colorectal tumorigenesis (17, 26). CD166, EpCAM, CD44, and ALDH1 are putative cell-surface markers associated with colorectal CSCs (26, 27). The mRNA expression of these markers was found to be increased in AOM/DSS–induced colorectal cancer mice (Fig. 6A). JMF3086 prevented the induction of these markers, whereas statins and SAHA were ineffective or weaker (Fig. 6A).
The microbiota and the immune system mutually interact to maintain homeostasis in the intestine. However, endotoxin released from the microbiota can alter this balance and promote chronic inflammation and lead to intestinal tumor development (28). To evaluate whether JMF3086 maintained gut homeostasis, endotoxin level in the serum of AOM/DSS–induced colorectal cancer mice was detected. High level of endotoxin was detected in AOM/DSS–treated group, and this effect was prevented by JMF3086 with the extent correlated with the reductions in tumor number (Fig. 6B).
Chemoprevention is a rational and appealing approach to reduce the risk of cancer by using pharmacologic means, such as aspirin for colorectal cancer and tamoxifen for breast cancer (29, 30). Aspirin has been reported to prevent carcinogenesis of colorectal cancer in ApcMin/+ mice and prevent 44% tumor formation at a dosage of 200 mg/kg through attenuating inflammation (31, 32). However, its clinical use for chemoprevention is prohibited due to a significant incidence of gastrointestinal upset and bleeding. Epigenetic modulators, such as HDAC2 overexpression, were associated with colorectal cancer progression, and a selective HDAC2 inhibitor was reported to prevent colorectal cancer tumorigenesis in both AOM/DSS mouse models and ApcMin/+ mice (33). A recent study shows that gut microbiota could ferment dietary fiber into butyrate, a class I and II HDAC inhibitor, to exert chemopreventive effect on colorectal cancer in gnotobiotic mouse models (34), suggesting that HDACs are potential targets for colorectal cancer prevention. In addition, epidemiology study suggests that statins attenuated the risk of colorectal cancer (35). In this study, our polypharmacological small molecule, statin hydroxamate (JMF3086), dually inhibits HDACs and HMGR exerting anti-inflammatory effect to prevent colitis and colitis-associated colorectal cancer. HMGR and HDAC activities in colon lysates of AOM/DSS–induced colorectal cancer were increased and inhibited by JMF, indicating that both inhibitions contributed to its chemoprevention. The advantage of JMF being superior to lovastatin plus SAHA is like a two-in-one antibody with superior inhibitory activity, compared with two monospecific antibodies (36). Additional mechanisms beyond the inhibitions on HDAC and HMGR activities or due to different pharmacokinetic property are also possible and awaited for further investigation. However, we reveal a new strategy for colorectal cancer chemoprevention.
IBD is an inflammatory disorder of the gastrointestinal tract (6). Epidemiologic studies suggest that the incidence and prevalence of IBD are increasing around the world. However, the exact etiology remains unknown and disease pathogenesis is not fully understood. DSS-induced colitis has been shown to be similar to human IBD in multiple aspects and has become an important tool to investigate its pathophysiologic mechanisms and immunologic processes (37). DSS-induced colitis is triggered by disrupting the colon epithelial barrier, allowing intestinal bacteria to penetrate the injured mucosa and induce mucosal inflammation, which is characterized by increased infiltration of inflammatory cells and an excessive production of proinflammatory cytokines/chemokines, leading to colitis exacerbation (38). Therefore, inhibition of inflammation is a potential strategy to prevent cancer initiation. In this study, we found JMF3086 could prevent and treat acute inflammation in the DSS-induced colitis mouse models by inhibiting the release of proinflammatory cytokines/chemokines and the infiltration of neutrophils into colitis regions. AOM in conjunction with the additional inflammatory stimulus of DSS results in tumor development that is restricted to the colon in mice (12). This model is widely utilized to recapitulate human colorectal cancer through inflammation and colon ulceration, similar to what is observed in patients (39). JMF3086 is demonstrated to prevent AOM/DSS colorectal cancer in this study, indicating that both initiation and progression of colorectal cancer are inhibited by JMF3086.
Tumors are tightly connected with microbiota and immune system (40). The link between microbiota and colorectal cancer in patients and mice has been studied and found that germ-free animals showed fewer tumors compared with those of normal microbiota. Antibiotic cocktail depleting the intestinal microbiota in mice reduced tumor growth in liver and colon, suggesting that microbiota promotes tumor progression (41). Specifically, Fusobacterium nucleatum compared with healthy people is more common in the gut of IBD patients to promote colorectal cancer (42), and clinical isolate of F. nucleatum promoted carcinogenesis in ApcMin/+ mice (43). The gut microbiota modulates carcinogenesis through three mechanisms (28). First, dysbiosis induces bacterial translocation, leading to induction of inflammation mediated by microorganism-associated molecular patterns and Toll-like receptors in several cell types (41). Intestinal inflammation promotes colorectal cancer development through release of proinflammatory cytokines/chemokines and the infiltration of macrophages and neutrophils into tumor regions (28, 44). Bacterial translocation was detected at tumor sites in ApcMin/+ mice, and antibiotic eradication reduced colorectal cancer development (45). Second, microbiota-derived metabolites could induce DNA damage, genotoxicity, and chromosome instability to induce barrier failure of gastrointestinal tract, which also trigger colorectal cancer carcinogenesis. For example, GNB-produced LPS and cytolethal distending toxin directly trigger double-strand DNA damage responses, leading to barrier disrupt. Third, microbiota could convert bile acid in host into secondary metabolites, such as deoxycholic acid, which in turn promotes the development of hepatocellular carcinoma and colorectal cancer (46, 47). In this study, increased mRNA expressions of proinflammatory cytokines and chemokines, infiltrations of macrophage (F4/80) and neutrophil (Ly-6G), as well as increased serum endotoxin (LPS) level were seen in AOM/DSS colorectal cancer mouse models. All these events were prevented by JMF3086, suggesting that JMF3086 might prevent colorectal cancer development through modulating gut microbiota–induced bacterial translocation and barrier failure of gastrointestinal tract. It is important to find a reliable biomarker to detect the effect of chemopreventive agents. Serum endotoxin level might serve as a potential biomarker to detect the prevention of JMF3086 in colorectal cancer.
The collection of bacteria, viruses, and fungi that live in the human body was known as the “microbiome” (48). It is estimated that less than 1% of all microorganisms in the natural world have been identified (49). It was largely unknown and regarded as a black box. Recently, the advent of new technologies, including high-throughput DNA sequencing and bioinformatics, has increased the capacity and launched a revolution in the microbiome (40). In this study, we found that JMF3086 prevented endotoxin release in AOM/DSS–induced colorectal cancer mouse models as well as the treatment model of DSS-induced colitis. It is possible that JMF3086 acts through modulating the population of gut microbiota. High-throughput microbial 16S rRNA gene sequencing combined with mouse models is awaited to discover functional genes of microbiota impacted by JMF3086.
Colorectal CSCs have been reported to initiate colorectal cancer and to play an important role in the genesis, maintenance, recurrence, and metastasis of cancer (50, 51). The expression of colorectal CSC markers, CD166, EpCAM, CD44, and ALDH1, in AOM/DSS–induced colorectal cancer mouse models was prevented by JMF3086, indicating that inhibition of CSCs might also contribute to its chemoprevention on colorectal cancer. Statins and HDAC inhibitors have been reported to reduce the stemness of colorectal CSCs (52–54).
In conclusion, our findings indicated that JMF3086, which is a potent dual inhibitor targeting HDACs and HMGR, has potential benefit and is a promising lead for the prevention of colitis-associated colorectal cancer.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Conception and design: T.-T. Wei, C.-C. Chen
Development of methodology: T.-T. Wei, Y.-T. Lin, R.-Y. Tseng
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T.-T. Wei, Y.-T. Lin, R.-Y. Tseng
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T.-T. Wei, Y.-T. Lin, R.-Y. Tseng, C.-T. Shun
Writing, review, and/or revision of the manuscript: T.-T. Wei, Y.-C. Lin, M.-S. Wu, C.-C. Chen
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T.-T. Wei, M.-S. Wu, J.-M. Fang, C.-C. Chen
Study supervision: C.-C. Chen
Other (obtained funding): C.-C. Chen
This work was supported by a research grant from the National Science Council of Taiwan and the Institute of Biomedical Sciences, Academia Sinica (IBMS-CRC103-P02).
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.