SEARCH

SEARCH BY CITATION

Keywords:

  • bone marrow fibrosis;
  • reticulin fibrosis;
  • chronic lymphocytic leukemia;
  • prognosis;
  • overall survival;
  • bone marrow biopsy;
  • myelofibrosis

Abstract

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. FUNDING SOURCES
  8. REFERENCES

BACKGROUND:

Bone marrow (BM) biopsies from patients with chronic lymphocytic leukemia (CLL) may show reticulin fibrosis at diagnosis, but its significance remains unclear. This study sought to assess the prognostic impact of BM reticulin fibrosis in patients with previously untreated CLL.

METHODS:

Data was reviewed from untreated CLL patients in the national Israel CLL database, followed during 1987 to 2012. All bone marrow biopsies were graded for reticulin fibrosis using a modified scoring system containing 4 grades (0-3), based on the European consensus report. Grade of reticulin fibrosis was correlated with overall survival (OS), outcome, and a number of well-recognized prognostic factors for CLL.

RESULTS:

The final cohort included 176 patients (122 males and 51 females). Median age was 63 years (range, 32-86 years) and the 5-year OS was 77.1%. Grade of BM reticulin fibrosis correlated with OS (P < .0001) and mortality (P = .001), and separated patients into 2 groups with different survival curves. Advanced reticulin fibrosis (grades 2-3) was associated with thrombocytopenia (platelet counts of < 100,000/mm3) (P = .025), anemia (P = .018), elevated β2-microglobulin < 4000 μg/mL (P = .048), and the presence of 11q deletion (P = .0015).

CONCLUSIONS:

There was a significant correlation between poor survival and grade of BM reticulin fibrosis. This staining procedure is easy to perform and can readily be added routinely when examining BM biopsies in CLL, because the findings do have prognostic implications. Cancer 2013. © 2013 American Cancer Society.


INTRODUCTION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. FUNDING SOURCES
  8. REFERENCES

A wide variety of benign and malignant disorders are associated with an increase of reticulin fibers in the bone marrow (BM).1 The pattern of the reticulin network can easily be evaluated using light microscopy, and different fiber grading scoring models were used in the past without uniformity.2-4 In principle, grading employs a scale of 4 to 6 grades, where the most severe form is collagen fibrosis.2 Increased reticulin fibrosis is generally a progressive process, sometimes reversible spontaneously or following successful therapy for the underlying disease.1, 5 It accompanies myeloid and lymphoproliferative disorders and other hematologic malignancies, such as hairy cell leukemia,6 myeloproliferative neoplasms (MPNs), particularly primary idiopathic myelofibrosis,7 and may be a prominent feature.6, 8

Although BM reticulin fibrosis was described in 1950,9 there has not been much progress in understanding its pathogenesis. In most cases, it is regarded as a “reactive” process, because the fibroblasts are polyclonal and functionally similar to normal fibroblasts.10, 11 This reactive process appears to be mediated by cytokines secreted by proliferating cellular components of the primary disease, and is an integral part of the BM microenvironmental changes in these disorders.12

In 1981, Rozman et al. reported the prognostic value of the histopathological pattern of leukemic involvement of the BM in chronic lymphocytic leukemia (CLL).13, 14 They described 4 patterns—interstitial, nodular, mixed, and diffuse—and correlated them with patient outcome. The diffuse pattern had negative prognostic impact compared with all other types of nondiffuse involvement. Based on this, BM biopsy (BMB) became part of the work-up in CLL until quite recently. However, according to the revised International Workshop on Chronic Lymphocytic Leukemia (IWCLL)/National Institutes of Health guidelines for diagnosis and treatment of CLL,15 BMB is no longer mandatory for all patients. Currently, there are novel biological prognostic factors used to stratify prognosis in CLL, which have ousted BMB from common practice.

BMBs may also reveal different degrees and patterns of reticulin fibrosis in CLL. However, their incidence and significance have not been extensively studied, and most data are derived from single case reports.16, 17 Today, BMB is not routinely performed, and planning a prospective study to determine prognostic impact of fibrosis in CLL will be difficult.

In this study, we quantitated the degree of BM reticulin fibrosis in patients with CLL at diagnosis, and determined possible prognostic significance. We also correlated degree of BM reticulin fibrosis with well-recognized clinical and laboratory prognostic parameters in CLL, overall survival (OS), and outcome in these patients.

MATERIALS AND METHODS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. FUNDING SOURCES
  8. REFERENCES

Patients

This retrospective study was performed on behalf of the Israeli CLL Study Group. The retrieved data from 9 centers in Israel provided information on a total of 195 patients during 1987 to 2012. Treatment-naive patients were included in this study if BMB had been performed as part of their initial investigation at diagnosis or before therapy. Diagnosis of CLL was according to the 1996 IWCLL criteria.15 Data was collected from medical records after approval by all the individual institutes' Helsinki ethics committees. Of the 195 patients included, 19 were excluded because of missing data.

Methods

BMB Analysis

All biopsies examined were routinely stained with hematoxylin and eosin. Gomori or Gordon-Sweet staining methods were used for silver impregnation. Reticulin staining of the BM is not routinely performed in Israel as part of the evaluation of BMBs in patients with CLL. Accordingly, for the purposes of this study, staining for fibrosis was performed by each participating center, and all biopsies were evaluated for pattern of reticulin fibrosis. All pathologists were blinded regarding clinical variables and eventual patient outcome. We preferred to use the more simple and easily applied Thiele semiquantitative scoring system, which includes only 4 grades of reticulin fibrosis (0-3), based on the European consensus2 reported in 2005. We chose not to use the Bauermeister grading model, which includes 6 grades of severity, or the Manoharan model, which includes 5 grades, because they are less simple to apply uniformly and could potentially generate more problems and greater differences in interpretation among the reviewing pathologists, leading to more bias among the 9 centers. Scores were defined as: grade 0, scattered linear reticulin with no intersections; grade 1, loose network of reticulin; and grade 2 and 3, diffuse/dense increase in reticulin with extensive intersections. For the sake of uniformity, we used 2 main categories of reticulin fibrosis: mild/early (grades 0-1) and advanced (grades 2-3). There was no central review panel, and hematopathologists from participating centers classified their final grade as mild or advanced, independently using the above criteria.

We correlated the grade of reticulin fibrosis (mild, advanced) with established prognostic factors for CLL: age, sex, Binet and Rai stages, spleen size, pattern of BM infiltration by CLL, β2-microglobulin levels, and percentage of CD38 and CD23 positivity and ZAP70 (70-kD zeta-chain–associated protein) expression, determined by flow cytometry according to standard techniques. Fluorescence in situ hybridization (FISH) analysis, OS, and eventual outcome were also correlated with the grade of reticulin fibrosis.

The pattern of BM infiltration by CLL was classified into 2 main categories: diffuse and nondiffuse, according to Rozman et al.13

Statistical Analysis

Overall survival was estimated using the Kaplan-Meier method and 2-tailed log-rank test. OS was calculated from date of diagnosis to date of death or last follow-up. A multivariate Cox proportional hazards model was used to evaluate fibrosis grading as predictive for OS, adjusting for other known prognostic variables. All 2-sided P values > .05 were determined to be statistically significant without adjustment for multiple comparisons. Categorical and continuous variables were compared using the chi-square and analysis of variance tests (Kruskal-Wallis test).

RESULTS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. FUNDING SOURCES
  8. REFERENCES

Demographic Characteristics

These characteristics are summarized in Table 1. The median age was 63 years (range, 32-86 years); 122 patients (70.5%) were males and 51 (29.5%) were females.

Table 1. Essential Clinical, Laboratory, and Histopathological Data; Outcome; and Survival According to Severity of Marrow Reticulin Fibrosis
CharacteristicReticulinFibrosis GradingTotal 176 (%)P
Grade 0-1 N = 118 (67%)Grade 2-3 N = 58 (33%)
  1. Abbreviations: BMB, bone marrow biopsy; PB, peripheral blood; SD, standard deviation.

Age (mean ± SD)62.1 ± 10.764.8 ± 10.463.0 ± 10.7.1009
Sex
 Female39 (33.3)12 (21.4)51 (29.5).1081
 Male78 (66.7)44 (78.6)122 (70.5)
White blood cells47,31749,05247,868.0864
Hemoglobin12.7 ± 2.311.7 ± 2.512.4 ± 2.4.0183
Platelets > 100,000/mm3102 (86.4)38 (67.9)140 (80.5).0039
β2-microglobulin, μg/mL > 400018 (22.2)13 (40.6)31 (27.4).0482
Positive for CD5/CD19, % in PB71.1 ± 19.667.8 ± 26.170.0 ± 21.9.9544
BMB% infiltration65.9 ± 27.857.3 ± 34.162.9 ± 30.2.2903
Pattern of infiltrate
 Nondiffuse48 (45.3)25 (48.1)73(46.2).9013
 Diffuse58 (54.7)27 (51.9)85 (53.8)
Positive for CD23, % in PB68.6 ± 29.962.3 ± 32.566.6 ± 30.8.2636
ZAP70-positive, %26 (50.0)19 (67.9)45 (56.3).1246
CD38-positive, %39 (41.9)23 (54.8)62 (45.9).1662
Spleen size >15 cm25 (24.5)14 (30.4)39 (26.4).4489
Death (%)33 (28.0)31 (53.4)64 (36.4).0010
5-y mortality12 (10.2)23 (39.7)35 (19.9)<.0001

Approximately half of the patients were diagnosed in early stage; 86 patients (48.9%) had Binet stage A, 56 patients (31.8%) had Binet stage B, and 34 patients (19.3%) had Binet stage C (Table 2). A similar distribution of stages was seen with the Rai classification (Table 2). Peripheral blood analysis showed an average white blood cell count of 20,100/mm3 (range, 1500-533,000/mm3), average hemoglobin of 12.4 g/dL (range, 4.2-17 g/dL), average platelet counts of 168,000/mm3 (range, 11,000-434,000/mm3), and absolute lymphocyte count ranged between 5280 and 510,614/mm.3 The median follow-up for the entire cohort after diagnosis was 66 months (range, 2-244 months). At the time of writing, 63.6% of patients were alive.

Table 2. Binet/Rai Staging and Grade of Bone Marrow Reticulin Fibrosis
 ReticulinFibrosis Grading  
 Grade 0-1 N = 118 (67%)Grade 2-3 N = 58 (33%)Total 176 (%)P
Binet stage
 A64 (54.2)22 (37.9)86 (48.9).0016
 B40 (33.9)16 (27.6)56 (31.8)
 C14 (11.9)20 (34.5)34 (19.3)
Rai stage
 043 (37.7)12 (21.8)55 (32.5).0125
 126 (22.8)12 (21.8)38 (22.5)
 232 (28.1)14 (25.5)46 (27.2)
 35 (4.4)3 (5.5)8 (4.7)
 48 (7.0)14 (25.5)22 (13.0) 

BM Histopathology

A pattern of CLL infiltration was reported in 158 biopsies, of which 85 (54%) displayed a diffuse pattern and 73 (46%) a nondiffuse pattern. The extent of BMB involvement was calculated by evaluating the percentage of leukemic infiltrate within the whole section, ranging from 5% to 100%. There were no differences seen in the estimated percentage of CLL cells present when patient groups with mild or advanced fibrosis were compared (65.9 ± 27.8 and 57.3 ± 34.1, respectively; P = .29).

Grade of Reticulin Fibrosis

A total of 22 patients (12.5%) had reticulin fibrosis grade 0, 96 patients (55%) had grade 1, 33 patients (20%) had grade 2, and 22 patients (12.5%) had grade 3. Because of the relatively small cohort size, and for the sake of uniformity, we merged grades 0 and 1 as one group (mild reticulin fibrosis) and grades 2-3 as another (advanced reticulin fibrosis), thereby creating 2 convenient and larger subgroups for comparison of prognostic parameters, clinical risk factors, and OS.

Binet and Rai Staging Classification Correlates With Grade of Reticulin Fibrosis

Both classifications stratified patients according to outcome, and there was a correlation between advanced stage of disease and the presence of advanced reticulin fibrosis. Binet stage C patients presented more frequently with grade 2-3 reticulin fibrosis (34.5%) compared to grade 0-1 (11.9%). Differences were statistically significant (P = .0016). A similar trend was evident in Rai stage 4, where only 7% had early grade fibrosis: 0-1 compared to 25.5% with advanced grade 2-3 (P = .0125) (Table 2). In univariate analysis (Table 3), there was correlation between inferior overall survival and both Binet stage C (hazard ratio [HR] = 5.755) and advanced grade 2-3 reticulin fibrosis (HR = 10.877). When bivariate Cox proportional hazard model was applied for grade of reticulin fibrosis as well as Binet stage, both parameters were found to be significant for adverse prognosis (advanced fibrosis > grade 2 with an HR of 4.506, and Binet stage C with HR = 2.085) (Table 4).

Table 3. Univariate Analysis of Survival Based on Reticulin Fibrosis Grading or Binet Stage
ParameterHazard Ratio95% Hazard Ratio Confidence Limits
  1. Abbreviation: ref, reference value.

Fibrosis grade 01 (ref) 
Fibrosis grade 12.312(0.298-17.912)
Fibrosis grade 2+10.877(1.468-80.589)
Binet stage A1 (ref) 
Binet stage B1.974(0.853-4.571)
Binet stage C5.755(2.494-13.276)
Table 4. Bivariate Cox Proportional Hazard Model of Fibrosis and Binet Stage
ParameterPr > Chi-SquareHazard Ratio95% Hazard Ratio Confidence Limits
  1. Abbreviation: Pr, P-value; ref, reference value.

Fibrosis grade 0-1 1 (ref) 
Fibrosis grade 2+<.00014.506(2.211-9.183)
Binet stage C.00082.085(1.360-3.196)

Prognostic Parameters and Grade of Reticulin Fibrosis

Bivariate analysis was performed comparing mild (grade 0-1) and advanced (grade 2-3) reticulin fibrosis to other clinical and laboratory prognostic parameters in CLL. Advanced reticulin fibrosis was more frequent in patients with thrombocytopenia < 100,000/mm3 (P = .0039), and hemoglobin levels and advanced fibrosis was also correlated. Patients with grade 2-3 fibrosis had a mean hemoglobin of 11.7 g/dL compared with 12.7 g/dL for grades 0-1 (P = .0183). No differences were evident regarding white blood cell counts (Table 1).

Comparing flow cytometer immunophenotyping results for expression of CD23, CD38, ZAP70, and coexpression of CD5/CD19, no differences were seen between the 2 groups (Table 1). Furthermore, there were no differences between the various histopathological patterns of CLL infiltration (diffuse pattern versus all other types) and grade of reticulin fibrosis. Patients with advanced fibrosis more frequently had higher serum β2-microglobulin levels (s > 4000 μg/mL or 1.5× normal values, P = .048). FISH analysis for 17p deletion, trisomy 12, 11q deletion, and 13q deletion was available for only some of the patients (85, 67, 75, and 63, respectively). Correlation with advanced grades of fibrosis were statistically significant for 11q deletion (P = .0015). Of the 85 patients examined for 17p deletion, 7 were positive. A trend for correlation with advanced fibrosis and 17p deletion was shown (Fig. 1, upper left panel), but sample size is too small to draw definitive conclusions (Fig. 1).

thumbnail image

Figure 1. Kaplan-Meier curves for overall survival for patients tested with fluorescent in situ hybridization (FISH). Correlation of grade of fibrosis and FISH positivity are shown for 17p deletion, 11q deletion, trisomy 12, and 13q deletion.

Download figure to PowerPoint

Survival Analysis, Outcome, and Grade of Reticulin Fibrosis

Five-year OS was 77.1% for the entire cohort; 85.7%, for Binet stage A, 74.15% for Binet stage B, and 46.1% for Binet stage C. There was a statistically significant correlation between grade of reticulin fibrosis and OS and outcome (Fig. 2). The 5-year OS for grade 0-1 was 86.92% compared with 51.9% for patients with grade 2-3 (Fig. 2). After a median follow-up of 66 months, the probability of being alive with reticulin fibrosis grade 0-1 was 72% compared with 46.6% for grade 2-3 (P = .0010; Table 1). The 5-year mortality for early fibrosis was 10.2% compared with 39.7 % for patients who had more advanced fibrosis (P = .001; Table 1).

thumbnail image

Figure 2. Kaplan-Meier curves are shown for overall survival of all patients according to grade of fibrosis.

Download figure to PowerPoint

DISCUSSION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. FUNDING SOURCES
  8. REFERENCES

The current guidelines for the diagnosis and treatment of CLL were updated in 2008 and require persistent B-cell lymphocytosis above 5000 × 109 cells/L in the peripheral blood,15 whereas BM biopsy with more than 30% lymphocytes is no longer needed for diagnosis. Although BMBs are now not performed routinely in CLL, they still may provide useful and practical data on the disease, particularly in relation to causes of cytopenias and response to therapy. BMB involvement by CLL includes nodular, interstitial, mixed, and diffuse patterns,18, 19 where the former 3 are usually accompanied by better preservation of normal architecture showing hematopoiesis, but the bone marrow is replaced by CLL cells in the diffuse pattern. It is well recognized that patients with diffuse involvement have a shorter survival than those with nondiffuse patterns, and that this finding has prognostic value.13 In addition, earlier reports from the 1980s have demonstrated the value of sequential biopsy during long-term follow-up of patients with CLL.18 A more recent study reported in 2006 by Schade et al.20 provided further support for the role of BMBs in CLL and showed an association between the pattern of infiltration (diffuse versus nondiffuse), ZAP70 expression, and immunoglobulin variable heavy chain mutational status. They also noted that combining the findings from BM biopsies with other prognostic parameters resulted in a better clinical stratification of patients with CLL.20

Fibrosis of the BM accompanies a number of benign and malignant diseases.1, 5 In most cases, fibrosis is composed of reticulin fibers, but niches of the BM microenvironment also contain collagen, laminin, and fibronectin. In contrast to BM collagen fibrosis, which has adverse clinical significance and is usually irreversible, the significance and clinical relevance of potentially reversible reticulin fibrosis of the BM varies in different diseases. Mild degrees of reticulin fibrosis can be identified in healthy individuals, and its presence is not always indicative of disease.21 In the past, the need for a uniform terminology and grading system for fibrosis resulted in the proposal of several scoring systems,3, 4 the most recent of which is the European consensus published in 2005 by Thiele et al.2 Although the scoring systems proposed, including the European consensus criteria, were intended for use in MPNs, no others models are available for “non-MPN” disorders with fibrosis. Because of this, the same grading systems have also been applied to evaluate BM fibrosis independently of its cause. We chose the Thiele scoring method2 for our study, because it simplifies earlier descriptions, reducing the categories to 4 grades which also includes the presence of some reticulin in normal BMs, defined as grade 0. In our study, combining and distinguishing very early grades of reticulin fibrosis (grade 0-1) and more advanced fibrosis (grade 2-3) has given us more confidence in the uniformity of our evaluations in a national multicenter setting.

The clinical significance of reticulin fibrosis is well established in primary idiopathic myelofibrosis,7 and in recent years has also been evaluated in other myeloproliferative neoplasms such as polycythemia vera, chronic myeloid leukemia,22 and essential thrombocytosis.23 Its clinical and prognostic significance has also been evaluated in other hematologic malignancies such as myelodysplastic syndromes8 and acute lymphoblastic leukemia,24 and in nonmalignant hematologic diseases such as immune thrombocytopenic purpura.25 In contrast, the significance of reticulin fibrosis has rarely been evaluated in patients with CLL, and most of the relevant data are based on case reports describing either rare associations or coexistence of 2 hematologic malignancies, eg, CLL and myelofibrosis.12, 13, 23-26 We feel confident that none of our cases with advanced fibrosis were due to coexistence of MPN, first, because all biopsies were performed routinely before treatment and as evaluation of progressive disease. Furthermore, none showed advanced collagen fibrosis, nor was there any correlation with spleen size, which is usually a hallmark of primary myelofibrosis.

We also correlated grade of reticulin fibrosis with other clinical and laboratory parameters regarded as risk factors in CLL. In recent years, reliable prognostic and predictive factors have been established for OS in CLL, the most important being FISH analysis,26 mutational status, ZAP70,27 and CD38 positivity,28 or serum β2-microglobulin levels.29 However, not all of these are performed routinely worldwide, and some require highly specialized laboratories to provide reliable results. In light of our findings, we suggest performing a simple staining for reticulin, for all BM biopsies, if already done as part of the work-up before starting treatment. This easily performed staining procedure can add information regarding prognosis and survival in CLL. In this regard, it is of interest that there was a correlation between the degree of the fibrosis and cytopenias (anemia and thrombocytopenia), but not with the pattern of involvement by CLL, indicating that reduction of normal hematopoiesis is affected by the presence of advanced fibrosis, independent of the degree of BM infiltration by CLL cells. Advanced reticulin fibrosis (grades 2-3) may indeed be a reflection of more advanced disease and clinical stage, but does not appear to relate to extent of BM involvement by CLL or to bear any relationship to the amount of CLL cells seen in the marrow but is related to stroma-leukemic cell interactions. In this context, Burger et al.30 in their investigation of the relationship between CLL cells and the microenvironment, defined different patterns of crosstalk which were shown to have a varying influence on leukemic cell growth in the different organ niches. In this respect, more advanced BM reticulin fibrosis may be relevant and play a role as part of a “dysfunctional environment” within the BM niche.30 Serum β2-microglobulin levels are also considered as a simple surrogate marker for tumor burden assessment, treatment-free survival, and OS.29 In our study, we identified a correlation between advanced fibrosis and elevated β2-microglobulin levels, and OS. FISH analyses was performed in only some cases, but there was a correlation between high-risk chromosomal deletions (17p and 11q) and advanced fibrosis. It is indeed possible that 17p and 11q deletion are more prevalent in patients with CLL who have more advanced reticulin fibrosis, but our cohort is too small to conclude this definitively. This will need future validation in a larger patient cohort.

The pathogenesis of fibrosis in BM is still poorly understood, but has been studied best in hairy cell leukemia by Cawley et al.,31 as well as in MPNs.12 It is apparent that cytokines such as transforming growth factor-β, interleukin-1, and platelet-derived growth factor play a major role, together with overexpression of adhesion molecules such as CD44.32 However, the exact mechanism involved is still not completely understood, and even less is known about pathogenesis of fibrosis in CLL. Kimura et al.33 correlated interleukin-1 levels with the presence of reticulin fibrosis in a patient with CLL, but no other studies evaluated this association. To our knowledge, only one earlier study dealt indirectly with the topic of BM reticulin pattern in CLL.34 This involved a heterogeneous cohort of 55 patients, some of whom were therapy-naive and others already treated by the time biopsy was done. They reported an inverse correlation between the expression of CD23 and the degree of reticulin fibrosis, but did not relate to fibrosis as a prognostic parameter in CLL. In our study, we could not confirm this observation, and there was no significant correlation between the grade of fibrosis and CD23 expression. Regarding other prognostic parameters, it was evident that patients with more advanced fibrosis had higher serum β2-microglobulin levels, but no correlation was found with CD38 expression and ZAP70 positivity.

Some patients with CLL develop fibrosis during the course of their disease, and it could be multifactorial in origin. This could reflect the natural history of the disease, or be related to therapy given for progressive disease. Because of this and in order to avoid any bias, we decided only to include therapy-naive patients and assess if grade of reticulin fibrosis in untreated patients has prognostic significance. Our results indicate that it does provide additional information for risk stratification in CLL.

In conclusion, we suggest performing this simple histochemical staining procedure when evaluating BM biopsies of patients with CLL. Further studies are obviously required to better understand the pathogenesis of fibrogenesis in CLL.

FUNDING SOURCES

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. FUNDING SOURCES
  8. REFERENCES

No specific funding was disclosed.

CONFLICT OF INTEREST DISCLOSURE

The authors made no disclosure.

REFERENCES

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. FUNDING SOURCES
  8. REFERENCES