SEARCH

SEARCH BY CITATION

Keywords:

  • microRNA;
  • oropharyngeal SCC;
  • prognosis;
  • gene signature;
  • human papillomavirus

Abstract

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Note Added in Proof
  8. REFERENCES

BACKGROUND:

Oropharyngeal squamous cell carcinoma (SCC) rates have been increasing significantly in recent years, despite a decreasing incidence of head and neck cancer in general. Oropharyngeal SCC has many characteristics that are distinct from other head and neck cancers, and thus it is important to focus specifically on cancers arising in this region, with the goal of improving patient outcomes. One important goal is to identify those patients who are likely to fail standard therapy and who could potentially benefit from alternative or targeted treatments.

METHODS:

In the current study, the prognostic value of microRNAs (miRNAs) was evaluated in patients with oropharyngeal SCC. miRNAs are small, noncoding RNAs that are master regulators of many important biological processes. In total, 150 oropharyngeal tumors were analyzed using the recently developed quantitative polymerase chain reaction-based method for miRNA expression profiling. In addition, the expression of miRNAs was also compared with human papillomavirus (HPV) transcriptional activities.

RESULTS:

The current study identified 6 miRNAs that were found to be significantly associated with cancer survival. A combined expression signature of these miRNAs was prognostic of oropharyngeal SCC, independent of common clinical features or HPV status.

CONCLUSIONS:

This new miRNA signature was experimentally validated in an independent oropharyngeal SCC cohort. Furthermore, 5 HPV-related miRNAs were identified, which may help to characterize HPV-induced cancers including both oropharyngeal and cervical SCC. Cancer 2013. © 2012 American Cancer Society.


INTRODUCTION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Note Added in Proof
  8. REFERENCES

Unlike most other types of head and neck cancer, oropharyngeal squamous cell carcinoma (SCC) has many unique characteristics, including its strong association with human papillomavirus (HPV) infection.1 Epidemiologic studies have indicated that HPV-positive oropharyngeal cancer occurs commonly in patients who are of younger age and have a higher numbers of sexual partners, more exposure to oral sexual practices, and lower smoking rates.1-4 Previous studies have also shown that HPV-infected patients with oropharyngeal cancer generally have a more favorable outcome compared with those without HPV infection.3-7 However, to the best of our knowledge, the molecular mechanisms underlying this divergence in disease outcome are largely unknown.

Although the overall incidence of head and neck cancer has decreased steadily within the past decades, the number of reported cases of oropharyngeal cancer has increased significantly.8, 9 Thus, there is an urgent need to focus specifically on oropharyngeal cancer to determine its unique characteristics with the goal of developing specific and targeted treatments. Currently, oncologists rely primarily on tumor stage to make treatment decisions for patients with oropharyngeal cancer. In recent years, the HPV status of patients with oropharyngeal cancer has also been proposed as a promising prognostic marker and treatment factor.3-7

In the current study, we investigated the prognostic value of microRNAs (miRNAs) for oropharyngeal SCC. miRNAs are a family of small, noncoding RNAs that suppress the expression of their gene targets.10 Approximately 1000 human miRNAs have been identified to date. Both computational and experimental studies have indicated that thousands of human protein-coding genes are directly regulated by miRNAs.11-13 Thus, miRNAs function as master regulators for many important biological processes, including cell growth, apoptosis, viral infection, and cancer initiation and progression.10, 14-17

Previous studies have demonstrated that miRNA expression signatures are promising biomarkers for the diagnosis and prognosis of a wide array of human cancers.18, 19 However, to the best of our knowledge, the prognostic value of miRNAs in patients with oropharyngeal cancer has not been investigated to date. Relevant to the current study, recent studies have shown that miRNAs are aberrantly expressed in head and neck cancer cell lines as well as in tumor tissues compared with normal tissues.20-24 In addition, altered miRNA expression has also been reported in relation to HPV expression in cancer cell lines.25 These miRNA studies focused primarily on the early diagnosis of head and neck cancer, but not the prediction of disease outcome. More importantly, these studies were all performed using heterogeneous tumor tissues or cell lines from multiple types of head and neck cancer. It is now evident that oropharyngeal cancer has many unique characteristics that are distinct from other head and neck cancers. Thus, these published results may not be applicable to oropharyngeal cancer. To avoid these pitfalls, we focused our miRNA analysis exclusively on patients with oropharyngeal SCC.

MATERIALS AND METHODS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Note Added in Proof
  8. REFERENCES

Patients and Tumor RNA Samples

The current study was approved by the Human Research Protection Office of the Washington University School of Medicine in St. Louis, Missouri. A total of 150 oropharyngeal SCC cases were included in this study, including 101 cases for training and 49 cases for validating a new miRNA-based prognostic model (Table 1).26, 27 All the tumor tissues were collected from patients treated by a single radiation oncologist (W.L.T.) at the Washington University School of Medicine. All the patients were treated with either definitive radiotherapy or surgery followed by postoperative radiotherapy. In addition, approximately one-half of the patients also received chemotherapy. Clinical data were collected prospectively from the patients and were then updated retrospectively after follow-up review.

Table 1. Characteristics of the Patients With Oropharyngeal SCCa
CharacteristicTraining Cohort (n=101)Validation Cohort (n=49)
  • Abbreviations: K SCC, keratinizing squamous cell carcinoma; NK SCC, nonkeratinizing squamous cell carcinoma; SD, standard deviation.

  • a

    The SCCs were histologically typed as keratinizing (K SCC), nonkeratinizing (NK SCC), and nonkeratinizing with maturation (NK SCC with maturation), as previously published.26, 27

  • b

    Smoking was defined as any lifetime smoking use versus no history of smoking.

   
Age at diagnosis (mean ± SD), y56.1 ± 9.156.2 ± 9.3
Sex  
Male94 (93%)43 (88%)
Female7 (7%)6 (12%)
Race  
White85 (84%)46 (94%)
Other16 (16%)3 (6%)
Smokingb  
Yes74 (74%)33 (67%)
No26 (26%)16 (33%)
T classification  
T125 (25%)7 (15%)
T238 (38%)17 (35%)
T313 (13%)12 (25%)
T424 (24%)12 (25%)
N classification  
N07 (7%)4 (8%)
N115 (15%)6 (13%)
N271 (70%)37 (77%)
N38 (8%)1 (2%)
Histologic type  
K SCC19 (19%)5 (10%)
NK SCC with maturation29 (29%)11 (22%)
NK SCC53 (52%)33 (67%)

For all the patients, formalin-fixed, paraffin-embedded (FFPE) tumor tissues were collected for pathological analysis before radiotherapy or chemotherapy. Sections from each case were stained with hematoxylin and eosin and reviewed independently by 2 study pathologists at Washington University to confirm the diagnoses. Tumor regions from each section were identified and macrodissection was performed. Finally, total RNA was extracted from the identified tumor regions with the miRNeasy FFPE Kit (Qiagen Inc, Valencia, Calif) according to the manufacturer's protocol. In this way, we were able to focus on the profiling of the tumor tissues with minimal contamination from adjacent normal tissues.

miRNA and HPV Expression Profiling

miRNA expression profiling was performed using our recently developed method, which is based on real-time reverse transcriptase-polymerase chain reaction (RT-PCR).28 The details of the experimental procedure have been described previously.28 In brief, the RT reaction was performed with the High Capacity cDNA Reverse Transcription Kit (Applied Biosystems, Foster City, Calif). Each RT reaction included 150 ng of tumor RNA and a pool of miRNA-specific RT primers. Real-time PCR was performed with Power SYBR Green PCR Master Mix (Applied Biosystems) and miRNA-specific PCR primers. Raw profiling data based on PCR threshold cycles (Ct) were normalized using a quantile-based scaling method as described previously.28

The expression profiles of E6 and E7 from 13 oncogenic HPV types were also determined by real-time RT-PCR. New HPV assays were designed based on our previous algorithm29 and the details of the new design are described elsewhere (unpublished data). In brief, primer sequences were selected from the E6 and E7 coding regions of the high-risk HPV types 16, 18, 31, 33, 35, 39, 45, 52, 56, 58, 59, 66, and 68. The expression profiles of glyceraldehyde 3-phosphate dehydrogenase (GAPDH) and β-actin were used as reference controls for data normalization.

Survival Analysis

Overall survival (OS) and disease-specific survival (DSS) were used as the endpoints to represent disease outcome, defined as the time interval between treatment start date and the date of death from any cause (OS) or the date of death with cancer disease (DSS). Statistical data analyses were performed using the R statistical package (http://www.r-project.org/). Univariate Cox proportional hazards regression analyses were performed to evaluate the correlation between miRNA or HPV signatures with disease outcome. The P values for outcome correlation were calculated using the Wald test, and further adjusted by permutation tests as previously described.30 Multivariate Cox proportional hazards regression analyses were performed to evaluate the independent prognostic value of the miRNA signature after controlling for common clinical variables. Residuals from Cox models were examined graphically and tested for proportional hazards assumption. The Kaplan-Meier estimator was used to estimate the empirical survival probabilities, and P values from the log-rank test indicated the significance of the miRNA outcome prediction model.

RESULTS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Note Added in Proof
  8. REFERENCES

miRNA Expression Profile Was Correlated With Oropharyngeal Cancer Survival

miRNA expression profiling was performed for 101 oropharyngeal SCC cases in the training cohort. The characteristics of these patients are summarized in Table 1.26, 27 Total RNA extracted from the tumor tissues was profiled using our recently established real-time PCR-based assays for 96 cancer-related miRNAs (listed in a supplementary table by Wang28). These cancer-related miRNAs were chosen based on literature mining, which indicates the dysregulation of these miRNAs in various human cancers. In addition, PCR-based profiling assays for all these miRNAs had been designed and validated experimentally in our recent study.28

The miRNA profiling data were analyzed to identify individual miRNAs that were correlated with disease outcome using univariate Cox proportional hazards regression models. Among all the miRNAs included in the profiling study, 13 were found to be significantly correlated with OS (Table 2). The independent prognostic values of these candidate miRNAs were evaluated further by controlling for treatment protocols (chemotherapy and radiotherapy status) and disease stage with multivariate Cox regression analysis. Six miRNAs, including miR-142-3p, miR-31, miR-146a, miR-26b, miR-24, and miR-193b, retained their prognostic significance in this multivariate analysis and thus were selected for further model development.

Table 2. miRNAs Correlated With Overall Survivala
miRNA NamePFold Change
  • Abbreviation: miRNA, microRNA.

  • a

    The P values were calculated using the Wald test in univariate Cox regression analysis, and were adjusted further with permutation tests. The fold change values were log2 transformed, representing the average expression difference of the miRNAs in 2 patient groups (deceased vs alive).

miR-142-3p.00079−0.54
miR-31.00151.14
miR-146a.0050−0.34
miR-26b.0081−0.25
miR-24.00910.15
miR-29b.0098−0.30
miR-203.0120.61
miR-155.014−0.41
miR-193b.0150.48
miR-26a.019−0.29
miR-215.0200.34
miR-101.029−0.26
let-7i.044−0.11

To more closely examine the expression profiles of these selected miRNAs in relation to patient survival, the 101 patients were stratified into 2 groups based on survival outcome (deceased vs alive). The expression profiles of these 6 miRNAs in the 2 patient groups are shown in Figure 1.28, 31 It is interesting to note that 3 miRNAs (miR-142-3p, miR-146a, and miR-26b) were preferentially overexpressed in the surviving patients, whereas the remaining miRNAs (miR-31, miR-24, and miR-193b) were overexpressed in the patients who died.

thumbnail image

Figure 1. The expression profiles of 6 prognostic microRNAs (miRNAs) are shown in 101 patients with oropharyngeal squamous cell carcinomas. The patients were stratified into 2 groups according to their survival status (deceased or alive). The expression profiles of individual miRNAs were determined in each group, and were normalized using a quantile-based scaling method as described previously.28 The normalized expression data were observed by further scaling to the value range of 0 to 1 and grouped into multiple bins with Weka (Waikato Environment for Knowledge Analysis) software.31 The P values were calculated using the Wald test in univariate Cox regression analysis, and further adjusted with permutation tests.

Download figure to PowerPoint

A miRNA Signature to Predict Cancer Survival

Given the significant correlation between miRNA expression and patient survival, we hypothesized that a combined expression signature of multiple miRNAs could be used for outcome prediction. Thus, the 6 miRNAs selected from the profiling studies were used to build a prognostic model to predict OS as follows:

  • equation image

in which S represents the risk score for each patient and E represents the normalized expression level of individual miRNAs from each tumor. The coefficient for each miRNA in this equation is the Z score from the Cox regression analysis, defined as Z = (Cox regression coefficient)/(standard error of the coefficient).

In this prediction model, a high-risk score predicts poor survival for the patient. By the median risk score, the 101 patients were divided into 2 cohorts of similar size. In this way, 51 patients were predicted to be of high risk (with ≥ the median score) and 50 to be of low risk (with < the median score). Kaplan-Meier survival analysis indicated that the 2 cohorts exhibited distinct risks for death (P = 4.8E-05) (Fig. 2A). Similarly, the miRNA risk score was also found to be significantly prognostic of DSS (P = 7.7E-05) (Fig. 2B).

thumbnail image

Figure 2. Kaplan-Meier survival analysis was used to evaluate the microRNA (miRNA) signature using the training cohort. A risk score was calculated based on the final miRNA prediction model and assigned to each patient. Based on the risk score, the patients were stratified into either the low-risk group or the high-risk group. The prognostic value of the miRNA risk scores was evaluated with regard to (A) overall survival and (B) disease-specific survival. The P values were calculated using the log-rank test. (C and D) Leave-one-out cross-validation was used to evaluate the miRNA modeling strategy. The cross-validated results from all 101 rounds were combined for prognostic evaluation using (C) overall survival and (D) disease-specific survival.

Download figure to PowerPoint

One common issue in computational modeling is the risk of overtraining (ie, the model may work well with the training data, but not with independent testing data). To evaluate the potential risk of overtraining from our modeling strategy, leave-1-out cross-validation was performed. In this cross-validation analysis, for each round, miRNA profiles from 100 training tumors were used to build a prediction model and the 1 remaining tumor was reserved for independent model testing. The process was rotated 101 rounds until all the tumors had been used for model testing. For each validation round, candidate miRNAs were selected based on their prognostic significance for OS of the training tumors as well as independence from treatment protocols and disease stage, using the same miRNA selection strategy as described earlier. Similarly, a prediction model was built in each validation round by combining the selected miRNAs, using the Z score for each miRNA as the coefficient in the model (details described earlier). The median score from each model was used to classify the reserved independent testing case. In this way, 101 slightly different models (from the final model) were developed and tested, and the cross-validated results from all 101 rounds were combined for performance evaluation. The cross-validated data were prognostic of both OS (Fig. 2C) and DSS (Fig. 2D), demonstrating the robustness of our computational modeling strategy.

Correlation of miRNA Expression With HPV Expression

Unlike most other types of head and neck cancer, previous studies have indicated that oropharyngeal cancer is strongly correlated with HPV infection.3-7 To comprehensively analyze the expression profile of HPV in oropharyngeal cancer, we developed a new real-time RT-PCR method to quantify the transcriptional activity of E6 and E7 genes from 13 oncogenic HPV types (discussed earlier). With this new method, HPV E6 and E7 transcripts were detected in 82 of the 101 cancer patients, including 75 cases of HPV-16, 6 cases of HPV-33, and 1 case of HPV-35.

The prognostic significance of HPV detection in patients with oropharyngeal SCC was evaluated using univariate Cox regression analysis. We evaluated both the status of the HPV detection (present or absent) and the expression level of HPV (averaged E6 and E7 expression). Both HPV status and HPV expression level were found to be significantly correlated with OS (P = .017 and P = .034, respectively). Thus, consistent with previous studies, HPV infection was a prognostic marker for oropharyngeal SCC. To determine whether the inclusion of HPV could further enhance the prognostic performance of the miRNA-based prediction model, the miRNA signature and HPV infection were evaluated using multivariate Cox analysis. As a result, HPV status was no longer found to be significant with the incorporation of the miRNA risk score (P = .49). In contrast, the significance of the miRNA signature was retained (P < .001). Thus, the prognostic value of the HPV status was already reflected in the miRNA signature, and adding this HPV feature is not likely to further enhance the model performance.

To examine the impact of HPV infection on miRNA model performance further, patients were stratified into 2 groups according to their HPV status (HPV positive vs HPV negative). The miRNA model was applied to the prognosis of these 2 patient groups separately. Survival analysis indicated that the miRNA signature was prognostic for both groups, with P values of .0063 and .0037 for HPV-positive and HPV-negative patients, respectively (Fig. 3). Thus, the prognostic performance of the miRNA signature was independent of the HPV status.

thumbnail image

Figure 3. Kaplan-Meier survival analysis was used to evaluate the independence of the microRNA (miRNA) signature from human papillomavirus (HPV) infection. The patients were stratified into 2 groups based on their HPV status. The miRNA model was applied to (A) patients who were HPV positive (HPV+) or (B) patients who were HPV negative (HPV-) separately.

Download figure to PowerPoint

Individual miRNA expression profiles were also found to be correlated with HPV status. Student t tests were performed to evaluate miRNA expression differences between the HPV-negative and HPV-positive patients. Among all the miRNAs analyzed, 5 were found to be significantly correlated with HPV status (miR-9, miR-223, miR-31, miR-18a, and miR-155) (Table 3).32 It is interesting to note that miR-31 was also included in the miRNA signature, which could help explain why HPV status was not found to be independent of the miRNA signature in our multivariate Cox analysis.

Table 3. miRNAs Correlated With HPV Statusa
miRNA NameFold ChangeP
  • Abbreviation: miRNA, microRNA.

  • a

    The fold change values were log2 transformed. The P values determined by the Student t test were adjusted further for multiple testing using the false discovery rate approach.32

miR-91.981.7E-05
miR-31−1.731.7E-03
miR-223−0.988.6E-04
miR-1550.803.8E-02
miR-18a−0.523.0E-02

The miRNA Signature Was Independent of Clinicopathologic Features

We assessed whether the new miRNA signature had independent prognostic value within the context of commonly used clinical parameters, including age at diagnosis, sex, race, smoking history, histologic type, stage, radiotherapy status, and chemotherapy status. The prognostic significance of the miRNA signature was evaluated for OS after controlling for the clinical features in a multivariate Cox regression model. The miRNA signature was still found to be statistically significant on this multivariate survival analysis, with a hazard ratio of 3.22 and P = .0022 (Table 4). Thus, the prognostic significance of the miRNA signature was independent of the clinical features.

Table 4. Cox Regression Analysis to Evaluate the Independent Prognostic Value of the MiRNA Signature and Clinical Parameters
ParameterHRPa
  • Abbreviations: HR, hazards ratio; K, keratinizing; miRNA, microRNA; NK, nonkeratinizing.

  • a

    P values were calculated using the Wald test.

miRNA signature (high- vs low-risk score)3.22.0022
Age at diagnosis1.00.94
Sex (male vs female)1.18.77
Race (white vs others)1.44.46
Smoking status (yes vs no)1.50.39
Histologic type (NK vs K/hybrid)1.01.98
Stage (I/II/III vs IV)1.15.77
Radiotherapy status (definitive vs postoperative)1.70.19
Chemotherapy status (yes vs no)1.31.47

Validation of the miRNA Signature With an Independent Cohort

The miRNA-based prognostic model was validated further with 49 independent cases in a validation cohort (Table 1).26, 27 First, miRNA expression profiles in the 49 patients were determined by real-time RT-PCR and applied to the miRNA prognostic model. As a result, a risk score was calculated by the prediction model and assigned to each of the 49 patients in the validation cohort. In this way, the patients in the validation cohort were stratified into either the high-risk group (20 patients) or the low-risk group (29 patients) based on the same threshold risk score determined by the training cohort. Kaplan-Meier survival analysis indicated the 2 patient groups had significantly different OS rates (P = .018) (Fig. 4A). Similarly, these 2 patient groups had significantly different DSS rates (P = 0.017) (Fig. 4B). Thus, the new miRNA model retained its prognostic significance when applied to an independent patient cohort.

thumbnail image

Figure 4. Kaplan-Meier survival analysis was used to evaluate the microRNA signature using the validation cohort for (A) overall survival and (B) disease-specific survival.

Download figure to PowerPoint

DISCUSSION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Note Added in Proof
  8. REFERENCES

One major challenge in miRNA biomarker studies is the lack of high-quality tumor tissues for expression analysis. Typically, miRNA expression profiling is performed using freshly frozen tumors for the extraction of high-quality RNA. However, the majority of tumors are routinely preserved with FFPE, leading to highly degraded RNA in the archived tumors. Given our exclusive focus on oropharyngeal SCC, it was challenging to collect a large number of frozen tumors for miRNA profiling analysis. Instead, we relied on using archived FFPE tumor tissues that were collected following standard clinical practice. To address the challenge of analyzing low-quality RNA, we developed a new PCR-based profiling method for miRNA expression analysis that has robust performance in detection sensitivity and specificity, even when applied to FFPE tumors.28 In the current study, this new method was applied to the profiling of 150 archived oropharyngeal tumors, resulting in an experimentally validated miRNA signature for the prognosis of oropharyngeal SCC.

As the first step, we analyzed 101 tumors for the development of a miRNA-based prognostic model. The robustness of the modeling strategy has been demonstrated by cross-validation using the same training tumors as well as in a multivariate analysis to control for clinical features. More importantly, the model also has been validated in a separate experiment using a patient cohort comprised of 49 independent cases. The miRNA model was demonstrated to also be useful for prognosticating the validation cohort, which indicated the general applicability of the model. The robust performance of the miRNA model reflected the independent prognostic value of the miRNAs included in the model because these miRNAs were selected based on their prognostic significance as well as independence from common clinical features. Thus, the miRNA signature represents a new biomarker-based strategy for the identification of patients with high-risk oropharyngeal SCC who could benefit from improved treatment strategies.

Among the miRNAs in the prognostic model, miR-31 has been shown to be significantly upregulated in patients with head and neck cancer and to play an oncogenic role in oral carcinogenesis.33, 34 miR-24 also is upregulated in patients with head and neck cancer, leading to enhanced cancer cell proliferation and reduced apoptosis.35 Consistent with these studies, we demonstrated that both miR-31 and miR-24 were associated with poor prognosis in patients with oropharyngeal SCC. Conversely, miR-146a functions as a tumor suppressor is many cancer types.36-39 Consistent with these findings, we demonstrated that miR-146a was associated with a favorable prognosis. Thus, the prognostic value of the miRNAs in the model could result from the functional roles of these miRNAs in oropharyngeal SCC.

Consistent with previous studies, the current profiling analysis also identified HPV as a prognostic marker for oropharyngeal SCC. It is important to note that we further demonstrated that the new miRNA signature was even more significant than HPV at prognosticating oropharyngeal SCC. Given the important role of HPV infection in oropharyngeal tumor initiation, it is interesting to note that we have identified 5 miRNAs that are correlated with HPV transcriptional activity (Table 3).32 In particular, one such HPV-correlated miRNA, miR-31, also was found to be of prognostic value in patients with oropharyngeal SCC. Another miRNA, miR-9, was found to be a prognostic marker for cervical cancer, as revealed in our recent study.40 Because the majority of cervical cancers are caused by HPV infection, the prognostic value of miR-9 in patients with cervical cancer is likely the result of its strong association with HPV infection. These HPV-related miRNAs may help to elucidate the molecular mechanisms as well as identify novel therapeutic targets for HPV-induced cancers, including both oropharyngeal and cervical cancers.

Note Added in Proof

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Note Added in Proof
  8. REFERENCES

FUNDING SUPPORT

Supported by the National Institutes of Health (grant R01GM089784), the Foundation for Barnes-Jewish Hospital Cancer Frontier Fund, and the Washington University Diversity Scholars Program.

CONFLICT OF INTEREST DISCLOSURES

The authors made no disclosures.

REFERENCES

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Note Added in Proof
  8. REFERENCES