Risk Factors of Follicular Lymphoma

Abstract

Introduction

Non-Hodgkin Lymphoma (NHL) is a heterogeneous group of malignancies with over thirty different subtypes. Follicular lymphoma (FL) is the most common form of indolent NHL and the second most common form of NHL overall. It has morphologic, immunophenotypic and clinical features significantly different from other subtypes. Considerable effort has been devoted to the identification of risk factors for etiology and prognosis of FL. These risk factors may advance our understanding of the biology of FL and have an impact on clinical practice.

Areas covered

The epidemiology of NHL and FL is briefly reviewed. For FL etiology and prognosis separately, we review clinical, environmental and molecular (including genetic, genomic, epigenetic and others) risk factors suggested in the literature.

Expert opinion

A large number of potential risk factors have been suggested in recent studies. However, there is a lack of consensus, and many of the suggested risk factors have not been rigorously validated in independent studies. There is a need for large-scale, prospective studies to consolidate existing findings and discover new risk factors. Some of the identified risk factors are successful at the population level. More effective individual-level risk factors and models remain to be identified.

1. Introduction

Non-Hodgkin Lymphoma (NHL) is a heterogeneous group of malignancies of lymphocyte origin. It usually arises or is present in lymphoid tissues, such as lymph nodes, spleen and bone marrow. During the past three decades, there have been consistent reports of increase in the incidence of NHL worldwide [1]. NHL is the fifth most commonly diagnosed malignancy in the US among both men and women. There are over thirty NHL subtypes, with the two most common subtypes – diffuse large B-cell lymphoma (DLBCL) and follicular lymphoma (FL) – accounting for about 30% and 20% of all NHL cases, respectively [2].

FL is defined as a lymphoma of follicle center B-cells, which has at least a partially follicular pattern [2]. It usually begins in the lymph nodes, is most often indolent, and grows more slowly compared with some other forms of NHL for example DLBCL. Despite the fact that most FLs are advanced at the time of diagnosis, about 50% of people survive at least eight to ten years after diagnosis, and many patients may not require treatment for a long time. Recent clinical studies have suggested that the survival for patients with FL has improved over the last few years, although more research is needed to confirm this finding. There is no known cure, and patients with FL may be treated with a combination of chemotherapy, monoclonal antibodies, and/or radiation therapy, or they may be followed closely with watchful waiting. Despite the indolent characteristics, the over survival is not affected by classic treatment, and a significant number of FL patients die within two years. Treatment of FL is challenging because of frequent recurrence.

FL is positive for the B-cell markers CD10, CD19, CD20 and CD22, but almost always negative for CD5. According to the WHO criteria [3], FL can be morphologically graded into grade 1 (<5 centroblasts per high-power field (hpf)), grade 2 (6–15 centroblasts/hpf) and grade 3 (>15 centrobalsts/hpf). Grade 3 can be further subdivided into grade 3A (centrocytes still present) and grade 3B (the follicles consist almost entirely of centroblasts). Grades 1, 2, and 3A are considered to be indolent and incurable, whereas grade 3B is considered an aggressive but curable disease similar to DLBCL. It has been noted that although this grading system is valuable from a pathological perspective, its clinical relevance is still debatable. In addition, the reproducibility of the grading system is still being studied [4]. Over time, histologic transformation of FL from an indolent disease to a DLBCL may occur in 10–70% of patients, with an estimated risk of 3% per year, and is associated with rapid progression of lymphadenopathy, extranodal disease, B symptoms and elevated serum LDH [5]. FL may also transform to Burkitt lymphoma, although much less commonly [6].

Considerable effort has been devoted to the identification of NHL risk factors [8]. In this article, we focus on FL and refer to other publications for discussions on other types of NHL. “Classic” research on FL has been focused on clinical measurements and environmental exposures. More recently, with the fast development of high-throughput profiling techniques, molecular-level analysis has been extensively conducting using various platforms including gene expression profiling, array comparative genomic hybridization (aCGH), single nucleotide polymorphisms (SNP) arrays and several other novel technologies that measure methylation status and epigenome. In the following sections, we review risk factors identified for FL etiology and prognosis separately. For each aspect, clinical measurements, environmental exposures, and molecular (which includes genetic, genomic, epigenetic and others) risk factors are reviewed. In addition, we also provide brief discussions on several pitfalls in the pursuit of FL risk factors. In this article, we adopt a loose definition for “risk factors”. What we review include factors that have been confirmed or suggested to have an impact on risk or survival as well as those that are involved in the pathogenesis of FL and hence may be associated with factors that directly contribute to risk or survival. Such a strategy has been motivated by the lack of consensus and limited knowledge of risk factors. We acknowledge that the identification of FL risk factors is an extremely complex process, involving a large number of steps including study design, execution, analysis, validation and others. Due to the limited scope of this article, inevitably, some important aspects will be missed. We refer to [7,8,9,10] and references therein for related discussions.

2. Etiology

2.1 Clinical and environmental risk factors

The etiology of FL is still poorly understood [8]. It has been suggested that age, gender and ethnicity may affect a person’s likelihood of developing FL. The incidence of FL increases with age. Although in principal FL may occur at any age, it is extremely rare in children. The median age at diagnosis is 60–65 years. Women have a slightly higher risk of developing FL than men. The incidence of FL is low among Chinese and Japanese. People of Jewish ancestry have a higher incidence. In the US, the incidence of FL is higher in Caucasian than in African-American.

Risk factors that may increase the risk of NHL overall also include medications that suppress the immune system (for example if a person has just had an organ transplant, he/she is more susceptible because immunosuppressive therapy has reduced the body’s ability to fight off new illnesses). We note that immune suppression is more tightly connected to aggressive lymphomas, while the evidence for FL has been less definitive. Infection with certain viruses and bacteria may also increase risk. Viruses that have been implicated in the development of FL include the Epstein-Barr virus (EBV), human T-cell lymphotropic virus (HTLV) type I, and the herpesvirus associated with Kaposi sarcoma (i.e., human herpesvirus HHV-8) [11]. It is worth pointing out that although they have been implicated in the development of FL, these viruses are linked mostly with diffuse or high-grade lymphomas. Congenital immunodeficiencies have been associated with lymphoma. Acquired immunodeficiencies may include infection with the human immunodeficiency virus (HIV). Note that most lymphomas associated with HIV are intermediate-grade or high-grade lymphomas.

Certain chemicals, such as those used to kill insects and weeds, may increase the risk of developing NHL. Possible mechanical link between pesticides and the development of FL is discussed in [12]. However, significantly more research is needed on this specific topic. An increased risk of FL was found among women who started using hair dyes before 1980, and cannot be excluded for women who started using in 1980 and after [13].

Lifestyle factors may also contribute to the risk of FL. A case-control study conducted in Italy linked tobacco use to the development of FL [14]. The researchers reported a 50% increased risk for 16–33 pack-years to an 80% increased risk with 34 pack years or greater. There was no increased risk for other NHL subtypes. In an independent study, analysis of data on over 6,500 NHL cases also suggested a positive link between smoking and incidence of FL, but not other NHL subtypes. United States researchers from Yale University reported that long-term cigarette smoking in women increased the risk of developing FL [17]. Analysis of the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial, which had 1,264 histologically confirmed NHL cases, suggested that smoking was not associated with NHL overall but was inversely associated with FL (ever smoking vs never: hazard ratio(HR)=0.62, 95% confidence interval [0.45, 0.85]) [16]. Such conflicting results, although not ideal, are not uncommon in epidemiologic studies. Published studies have noted the discrepancy on smoking effect, however there is no satisfactory explanation. Alcohol intake may directly affect immune function, which is an important etiologic factor for lymphoid malignancies in general. Morton and others [17] conducted a pooled analysis of nine case-control studies of NHL and showed that ever drinkers, compared with never drinkers, had a 17% lower risk of NHL (no detailed investigation on FL was conducted). This finding can be potentially explained by a beneficial effect of moderate alcohol consumption on immune system. However, no significant dose-response relationship was observed, which could be used to argue against a possible biological mechanism for the observed inverse association. Analysis of the PLCO data suggested that alcohol consumption was unrelated to NHL (drinks/week: p-value for trend=0.187) [16]. In six large prospective cohort studies of alcohol and NHL, researchers found that moderate alcohol intake was associated with a 41% reduction in risk among Iowa women, and heavier alcohol intake was associated with a 33% risk reduction among United Kingdom women, a 23% risk reduction among retired US men and women, and a 40% reduction among Japanese men. However, alcohol intake was not associated with NHL risk among Finnish male smokers or US men and women in a cancer screening trial. Chang and others [18] investigated whether the history of alcohol drinking affected the risk of NHL using the California Teachers Study cohort, a prospective cohort with 496 women diagnosed with B-cell NHL. It was found that women who were former alcohol drinkers at baseline were at an elevated risk of overall B-cell NHL and FL (rate ratio=1.81, 95% CI [1.00, 3.28]). The researchers argued that it was important to identify both current and past alcohol consumption status. In most of the aforementioned studies, subtype specific analysis was not carefully conducted. Thus, alcohol drinking may be suggested as a potential risk factor for FL, but more analyses/studies focusing on FL are needed to confirm its role. In the Netherlands Cohort Study, it was found that the rate ratio of lymphatic malignancies per 4-unit increase in BMI (body mass index) at 20 years of age was 1.13 (95% CI [1.01, 1.25]). The overall rate ratio of lymphatic malignancies per 5-cm increase in height was 1.08 (95% CI [1.02, 1.15]). However, the rate ratio for FL alone was not significant and at 1.15 (95% CI [0.95, 1.40]) [19]. In a prospective cohort study with 37,931 Iowa women among whom there were 261 cases of NHL and 58 cases of FL, Cerhan and others [20] observed no overall association between anthropometric characteristics (including BMI) and risk of overall NHL or FL. In contrast, in a population-based case-control study, Skibola and others [21] found that the risks of NHL and FL were positively associated with an overweight or obese status compared to a normal-weight status. In a Scandinavian study with 3,055 NHL cases and 3,187 population-based controls, Cheng and others did not find any association between BMI and risk of FL [22].

2.2 Molecular risk factors

As with other types of cancers and other subtypes of NHL, there is increasing evidence that molecular (including genetic, genomic, epigenetic and others) risk factors may have independent contributions to the risk of FL beyond clinical risk factors and environmental exposures [8,23,24,25].

The most common acquired nonrandom chromosomal translocation in FL patients is the t(14; 18) translocation, which is found in more than 80% of all cases. This translocation, generated through the BCL-2/IGH rearrangement, results in the overexpression of the BCL-2 gene, which encodes apoptosis regulator proteins and has been implicated in a number of cancers including melanoma, breast, prostate and lung carcinomas. In fact, the detection of t(14; 18) product by FISH or polymerase chain reaction (PCR) has been frequently used in the diagnosis and follow-up of patients with FL. In some cases, the presence of BCL-2 staining in biopsies may be significant for the patient’s prognosis or likelihood of relapse. The interaction between t(14; 18) translocation and pesticides was investigated in [12], which showed that occupational exposure to pesticides would increase BCL-2/IGH prevalence together with the frequency of BCL-2/IGH-bearing cells especially during the high pesticides use period. However, it should be noted that this translocation is not unique to FL. It has been detected in healthy patients as well as patients with other types of tumors. A small fraction of FL (~5%) does not exhibit the classic t(14;18) but instead contains alterations affecting BCL-6 at 3q27, including t(3;14)(q27;q32) [26]. This leads to the deregulated expression of transcriptional repressor BCL-6, which is normally required for germinal centers formation. Protein encoded by gene BCL-6 acts as a sequence-specific repressor of transcription and has been shown to modulate the STAT-dependent Interleukin 4 (IL-4) responses of B cells.

Microarray studies have been conducted, searching for genes whose expressions are associated with the etiology of FL [27]. Gene expression profiling of normal germinal center B (GCB) cells has been shown to be unchanged in FL, supporting the perspective that FL arises from this stage of B-cell differentiation. In a cDNA microarray study with 588 genes, Husson and others [28] identified 28 genes that were down-regulated and 37 genes that were up-regulated in FL cells compared with normal GCB cells. The expression level of each differentially expressed gene was then verified by quantitative PCR, resulting in 24 up-regulated genes and 8 down-regulated genes (with p-value<0.10). Up-regulated genes in FL included 2 cell cycle regulator proteins that are involved in G1 arrest, p21 and p16, which is consistent with the low proliferative nature of FL cells. The identified up-regulated genes also included cell cycle regulator proteins (CDK10, p120, p21, and p16), genes that are involved in cell-cell interactions (TNF, IL2RG, and IL4RA), and transcription factors PAX5 and Id-2 that are involved in norm B-cell development. Down-regulated genes in FL included MRP8 and MRP14, which are involved in adhesion.

Conde and others [29] conducted a three-stage genome-wide association study and identified two variants associated with FL at 6q21.32 (rs10484561 and rs7755224). The combined sample contained 1,465 FL cases and 6,011 controls. For rs10484561, the combined OR (odds ratio) was 1.95 (95% CI [1.72, 2.22]) and the combined p-value= 1.12 × 10⁻²⁹. This marker is in the human leukocyte antigen (HLA) class II region, which is essential for immune function and has been shown to be associated with multiple cancers. Smedby and others [30] conducted a three-stage genome-wide association study, starting with a genome-wide scan of 379 FL cases and 791 controls followed by validation using 1,049 cases and 5,790 controls, and identified a second independent FL-associated locus on 6q21.32, rs2647012 (combined OR=0.64, combined p-value=2 × 10⁻²¹), located 962 base pairs away from rs10484561 (r² < 0.1 in controls). After mutual adjustment, the associations at the two SNPs remained genome-wide significant (rs2647012: OR=0.7, adjusted p-value=4 × 10⁻¹²; rs10484561, adjusted OR=1.64, adjusted p-value=5 × 10⁻¹⁵). Haplotype and coalescene analyses indicated that rs2647012 arose on an evolutionarily distinct haplotype from that of rs10484561 and tagged a novel allele with an opposite (protective) effect on FL risk.

Epigenetic events in general have important implications in cancer. Recent molecular studies have established that activation of various oncogenes and silencing of tumor suppressor genes are required for FL development and progression [31]. Commonly methylated genes in FL include the androgen receptor gene, which encodes a member of the receptor group that binds and mediates the actions of androgens. DNA promoter hypermethylation of the androgen receptor gene is common in FL. McDonald and others [32] reported 16/19 samples of T and B lymphomas positive for methylated androgen receptor genes. A similar analysis of grade 1 and 2 FL found that 25/26 samples tested were positive for methylation at the androgen receptor promoter region [33]. SHP1 is a phosphotyrosine phosphatase with important roles in the regulation of immune system cell differentiation and activation. Methylation of SHP1 promoter region appears common across multiple subtypes of lymphomas (including DLBCL, MALT lymphoma, plasmacytomas and mantel cell lymphoma). In [34], 32/33 FL samples assayed were methylated at the SHP1 promoter region. In contrast, no promoter methylation of SHP1 was observed in 20 normal reactive lymph node samples. The almost uniform methylation of SHP1 in FL samples and the high rates across several lymphoma subtypes suggest that the down-regulation of SHP1 due to promoter hypermethylation may play an important role in lymphomagenesis. DAPK (dead-associated protein kinase) is a calcium-calmodulin-dependent serine/threonine kinase that is involved in apoptosis. Methylation of DAPK may be common in FL. In two studies [35,36], 25/29 FL samples were positive for aberrant DAPK methylation. The finding is consistent with the hypothesis that methylation of DAPK is a common early epigenetic phenomenon of FL, which allows cells to escape the normal apoptosis process. p16 (cyclin-dependent kinase inhibitor 2A) is located on chromosome 9q21 and the most commonly altered gene in human malignancies. Homozygous deletion of p16 is common in multiple cancers, however, not in patients with newly diagnosed FL. Two studies [37,38] have investigated p16 deletions in FL. A combined analysis of published data showed aberrant p16 promoter methylation in 5/16 FL cases. p15 is also a cyclin-dependent kinase inhibitor and located 25 kilobases from the p16 gene. Its protein shares significant areas of homologous amino acid sequences with p16. Methylation of the p15 promoter region has been reported in acute myeloid and lymphoblastic leukemia, multiple myeloma and lymphomas. However, few studies have specifically detailed the histologic subtype of the lymphomas analyzed, making it impossible to analyze FL. Of those that have, 10/27 FL samples were found to have detectable p15 methylation [39,40]. The p57 gene (cyclin-dependent kinase inhibitor 1C) is located on chromosome 11p15.5. The expression of p57 gene is absent in various hematological cell lines. In [41], 8/18 FL samples were observed to have methylation of the p57 promoter region. p14 protein is also a cyclin-dependent kinase inhibitor known to induce cell cycle arrest at G1 and G2. Mice with homozygous deletion of p14 frequently develop numerous tumors including lymphomas. Despite interesting findings in animal studies, Baur and others [39] failed to find methylation of the p14 promoter region in FL samples. Hypermethylation of p14 does not appear to be a common epigenetic event in FL. As suggested in [31], genes whose methylation may be involved in FL etiology also include GSTP1, IL-12 receptor beta-2, Snk/plk2 and GADD45-gamma. More research is needed to draw definitive conclusions on these genes.

3. Prognosis

3.1 FLIPI and FLIPI2

Prognosis of FL has attracted extensive attention. For a brief review of the history on FL prognosis research, we refer to [42] and references therein.

In 2004, the Follicular Lymphoma International Prognostic Index (FLIPI) was built from a retrospective analysis of more than 4,000 patients with FL treated between 1985 and 1992 [42]. The endpoint was OS (overall survival). In a multivariate analysis, five parameters were used to build this index: age (>60 versus ≤60), serum LDH level (>upper limit of normal (UPLN) versus ≤UPLN), number of nodal areas (>4 versus ≤4), hemoglobin level (<120 versus ≥120g/L) and Ann Arbor state (III-IV versus I-II). Three risk groups, low, intermediate and high, were distinguished. The low risk group was defined as “number of risk factors=0–1”. It accounted for 36% of the patients, with five-year OS 90.6% and ten-year OS 70.7%. The intermediate group was defined as “number of risk factors=2” and accounted for 37% of the patients, with five-year OS 77.6% and ten-year OS 50.9%. The high risk group contained the rest 27% of the patients, with five-year OS 52.5% and ten-year OS 35.5%. Using independent cohorts of patients, several retrospective studies have confirmed the effectiveness of FLIPI. Particularly, the distributions of patients and differences in hazard ratios for death between different prognostic subgroups were found to be very similar to those in [42]. The FLIPI has thus been considered as a standard prognostic index for FL. In addition, although the FLIPI was originally constructed for OS, in several recent studies, it has been used as a prognostic index for PFS (progression-free survival). FLIPI may have several limitations. For example, even though it has been used in some PFS studies, the FLIPI was not originally designed for PFS. In addition, the samples were collected between 1985 and 1992, during which period some important clinical and biological parameters were not collected and thus not used to construct FLIPI. In addition, patients in the FLIPI cohort only received conventional chemotherapy or no treatment, making them significantly different from those in more recent studies.

The FLIPI2 was built to solve the problems encountered by FLIPI [7,43,44]. It has PFS as the endpoint, as opposed to OS with FLIPI, and integrates all the modern parameters prospectively collected. It has five prognostic parameters: longest diameter of the largest tumor mass >6 versus ≤6cm, serum β2 microglobulin level >UPLN versus ≤UPLN, bone marrow involved or not, hemoglobin level ≤120 versus >120 g/L, and age >60 versus ≤60 years. In the analysis of 812 patients (88% of whom were treated with rituximab), FLIPI2 classified patients into three risk groups. The low risk group was defined as having none of the risk factors and accounted for 20% of the patients. The 3-year PFS was 91%, and the 5-year PFS was 79.5%. The intermediate group had 1–2 of the risk factors and accounted for 53% of the patients. The 3-year PFS was 69%, and the 5-year PFS was 51%. The high risk group had 3-year PFS 51% and 5-year PFS 19%. Arcaini and others [45] conducted retrospective analysis of 280 patients (among whom 262 were diagnosed after 1995 and 190 were treated with rituximab) and confirmed the accuracy of FLIPI2 for PFS. Compared with FLIPI, FLIPI2 needs to be confirmed in more independent studies. Note that FLIPI2 has an endpoint different from that of FLIPI and is not intended to replace FLIPI. Instead, they may complement each other.

3.2 Other clinical and environmental risk factors

Lifestyle factors also have an impact on prognosis. Geyer and others [46] evaluated the association between pre-diagnosis cigarette smoking, alcohol use and BMI with overall survival in 1,286 patients enrolled through population-based registries in the US from 1998 through 2000. It is noted that the study patients included multiple subtypes. It was found that compared with never smokers, former smokers (HR=1.59, 95% CI [1.12, 2.26]) and current smokers (HR=1.50, 95% CI [0.97, 2.29]) had poorer survival, and poorer survival was found to be positively associated with smoking duration, number of cigarette smoked per day, pack-years of smoking and shorter time since quitting (all p-values<0.01). Alcohol use was associated with poorer survival (p=0.03) compared with nonusers. Those who drank >43.1 g/week (median intake among drinkers) had poorer survival (HR=1.55, 95% CI [1.06, 2.27]), whereas those drinkers consuming less than this amount demonstrated no significantly different survival (HR=1.13, 95% CI [0.75, 1.71]). Greater BMI was associated with poorer survival (p=0.046), but this survival disadvantage was only observed for obese individuals (HR=1.32 for BMI ≥30 versus BMI 20–24.9; 95% CI [1.02, 1.70]). These results held for lymphoma-specific survival and were broadly similar for DLBCL and FL. Han and others [47] were among the first to test the hypothesis that a higher intake of fruits and vegetables was associated with better NHL survival [48]. Using a population-based cohort of 568 women with newly diagnosed NHL followed for a median of 7.7 years, the researchers found 32% better overall survival for NHL patients who had higher pre-diagnosis intake of vegetables and fruits (HR=0.68, 95% CI [0.49, 0.95]) after adjustment for demographic and clinical variables. Total vegetables (HR=0.58, 95% CI [0.38, 0.89]), green leafy vegetables (HR=0.71, 95% CI [0.51, 0.98]) and citrus fruits (HR=0.73, 95% CI [0.54, 0.99]) showed the strongest associations. Similar results held for both DLBCL and FL.

Janikova and others [49] analyzed a total of 181 PET (positron emission tomography) scans in 117 patients with FL in a retrospective study. It was found that PET at staging was able to substantially change treatment strategy in a significant proportion of FL patients. Persisting PET positivity after treatment was predictive for a high risk of an early relapse and might identify patients with poor prognosis. A study led by Dana-Faber Cancer Institute researchers analyzed the National Comprehensive Cancer Network’s Non-Hodgkin’s Lymphoma Outcome database, which included 953 eligible patients, and showed that PET can upstage 18% to 45% of FL patients and lead to changes in their management.

3.3 Molecular risk factors

Dave and others [50] conducted microarray gene expression profiling of 191 FL patients, searching for genes associated with prognosis. The researchers defined two expression profiles, referred to as the immune-response-1 and -2 signatures, associated with long and short survival, respectively. The immune-response-1 signature consisted of T-cell-specific genes CD7, CD8B1, LEF1, ITK and STAT4 and macrophage lineage genes ACTN1 and TNFSF13B (BAFF). The immune-response-2 signature included genes expressed by macrophages and/or dendritic cells, such as TLR5, FCGR1A, SEPT10, LGMN and C3AR1 (complement 3a receptor 1). Cell sorting experiments confirmed that the immune-response signatures mainly reflected expression levels of the various, non-neoplastic CD19- cell populations. It is worth noting that, Tibshirani [51] analyzed the same data using various standard statistical techniques and argued that “… our analysis sheds serious doubt on the reproducibility of the authors’ (Dave et al.) biologic findings”, highlighting the analytic challenges faced by high-throughput gene expression studies [8]. Several research groups compared gene expression profiles of low-grade FLs (histological grades 1–2) with those of grade 3 FLs or FLs that had transformed into DLBCLs. Glas and others [52] conducted supervised classification on a training set of paired samples from patients who experienced either an indolent or aggressive disease course, and established a gene expression profile with 81 genes. It was shown that in the training set, this gene signature had 100% accuracy distinguishing low-grade from high-grade diseases. The classification accuracy was 93% in an independent validation set. In a third set of FL samples where histologic grading was ambiguous, this gene signature showed a classification accuracy of 94%. Genes that were significantly up-regulated in the aggressive phase of the disease included those involved in cell cycle (such as genes CCNE2, CCNA2, CDK2, CHEK1 and MCM7) and DNA synthesis (including genes TOP2A, POLD3A, HMGA1, POLE2, GMPS and CTPS) as well as those reflecting increased metabolism (including genes FRSB, RARS, HK2 and LDHA) and activation of several signaling pathways (including genes FRZB, HCFCR1, PIK4CA, and MAPK1). Genes that are derived from the reactive infiltrate of T cells and macrophages (CD3C, CXCL12 and TM4SF2) were up-regulated in the indolent phase of the disease. Lossos and others [53] profiled 12 FLs with transformation and identified a set of 671 genes that exhibited at least a three-fold variation in the biopsy pairs of three or more patients. The researchers identified two distinct gene profiles possibly associated with FL prognosis. Five out of the 12 cases displayed enhanced expression of C-MYC and its target genes, whereas in four cases a decreased expression of C-MYC and its target genes was observed. De Vos and others [54] profiled four FLs with documented progressions. Among the top 36 up-regulated and 36 down-regulated genes, seven genes were also identified by [53]. Genes CDA and GAPD, two genes reflecting levels of metabolism, were among the overlapped and up-regulated. Genes IRF8 and of PTPRC were also identified in the two studies and down-regulated. The researchers also noted down-regulation of different T-cell markers upon transformation such as CD7, FYB (Fyn binding protein) and SEMA4D (CD100). Elenitoba-Johnson and others [55] studied 11 FLs that transformed into DLBCL. The findings included 67 significantly up-regulated and 46 down-regulated genes in DLBCL. Interesting up-regulated genes included the growth factor/cytokine receptors MET (the hepatocyte growth factor receptor), FGFR3 (fibroblast growth factor receptor 3), LTBR (lymphotoxin b receptor) and PDGFRB (platelet-derived growth factor receptor b). In addition, gene p38BMAPK was also found up-regulated in DLBCL. This finding was confirmed in follow-up mechanical study. Janikova and others [56] conducted gene expression profiling of 31 non-selected patients with FL, 12 of whom were in relapse and the remaining 19 newly diagnosed. The researchers employed template matching and defined two gene sets composed of genes differentially expressed among samples. These gene sets shared an over-representation of genes with similar biological functions and were termed T-CELL and PROLIFERATION profiles. The poor profile was defined by a high PROLIFERATION score and/or low T-CELL score. The poor profile cohort contained a significantly higher proportion of relapsed cases. In addition, a comparison of samples from initial diagnosis and from relapse showed significant differences mainly in the T-CELL profile.

With the analysis of 278 patients, Cerhan and others [57] found that SNPs in genes IL8 (rs4073; HR_TT=2.14, 95% CI [1.26, 3.63]), IL2 (rs2069762, HR_GG/TT=1.80, 95% CI [1.06, 3.05]), IL12B (rs3212227; HR_AC/CC=1.83, 95% CI [1.06, 3.06]) and IL1RN (rs454078; HR_AA=1.93, 95% CI [1.11, 3.34]) were the most significant predictors for survival. A summary score using the number of deleterious genotypes from these genes was shown to be significantly associated with survival (p=0.001). A risk score that combined the four SNPs with clinical and environmental risk factors was even more strongly associated with survival (p-value<0.001). The 5-year survival estimates were 96%, 72% and 58% for the low, intermediate and high risk groups, respectively. Wrench and others [58] confirmed the association of SNP rs10484561 and rs6457327 with risk of FL and demonstrated that SNP rs6457327 predicted both time to and risk of FL transformation independently of clinical and environmental risk factors, particularly including FLIPI. Han and others [59] investigated overall survival of 101 FL patients. This study took a candidate-gene approach. A total of 1,229 SNPs representing 122 KEGG pathways were analyzed. The researchers identified two pathways as having independent predictive power for prognosis beyond clinical and environmental risk factors. The two pathways were Endometrial cancer, which included ten SNPs from genes CASP9, CCND1, MLH1, MYC, TP53 and CTNNB1, and Melanogenesis, which included five SNPs from genes MC1R and CTNNB1. It was noted that the identified pathways differed significantly from those for DLBCL and NHL overall. With the same data, Ma and others [60] adopted a data mining approach using thresholding and identified 131 genes (187 SNPs) as associated with prognosis of FL. The identified genes/SNPs represented the following KEGG pathways: apoptosis (including genes CASP9, IL1A, IL3, IRAK2, IRAK3), Cytokine-cytokine receptor interaction (gene IFNGR2, IL10, IL15, IL1A, IL3, IL4R), focal adhesion (RAC1, RAC2), Leukocyte transendothelial migration, and multiple signaling pathways (including B cell receptor signaling pathway, calcium signaling pathway, Fc epsilon RI signaling pathway, Insulin signaling pathway, Jak-STAT signaling pathway, MAPK signaling pathway, p53 signaling pathway, Toll-like receptor signaling pathway, VEGF signaling pathway, and Wnt signaling pathway).

Studies have linked 1p36 deletions with worse prognosis in FL [61]. With a cohort consisting of 251 specimens, the researchers identified 46 cases with nonsynonymous mutations affecting TNFRSF14 (tumor necrosis factor receptor superfamily member 14). OS and disease-specific survival (DSS) were associated with the presence of TNFRSF14 mutation in patients whose overall treatment included rituximab. It was further shown that inferior OS and DSS were most pronounced in patients whose lymphomas contained both TNFRSF14 mutations and 1p36 deletions after adjusting for FLIPI.

O’Shea and others [62] studied 185 FL patients to assess the prognostic relevance of aUPD (acquired homozygosity in the form of segmental acquired uniparental disomy). It was found that the number of genetic abnormalities was predictive of outcome. In particular, the presence of more than three abnormalities was associated with inferior OS. Sites of recurrent aUPD were detected on 6p, 16p, 12q, 1p36, 10q and 6q. In multivariate analysis, aUPD on 1p36 was correlated with shorter OS. aUPD on 16p was predictive of transformation and correlated with poorer PFS.

With its important implications in the etiology of FL, methylation was also investigated as potential risk factors for prognosis [63]. It was found that methylation profiles were conserved in sequential FL and transformed-FL biopsies, suggesting that widespread methylation represents an early event in lymphomagenesis and may not contribute substantially to FL transformation. Significant correlation (p-value<0.05) between FL methylation values and reduced gene expression was demonstrated for up to 28% of loci. Such findings suggested that it may not be sensible to attribute FL prognosis to methylation.

Minimal residual disease (MRD) may affect FL prognosis and be the major cause of relapse [64]. Rambaldi and others [65] studied MRD following sequential administration of CHOP and rituximab in previously untreated patients with FL. It was shown that for patients achieving PCR-negative status following CHOP and therefore excluded from rituximab, freedom from recurrence was 52%, and for patients treated with rituximab, a durable PCR-negative status was associated with a better clinical outcome. Zipf and Johnston [66] provided detailed account of using PRC based techniques for detecting MRD.

High tumor burden is also associated with shorter PFS and OS, even in the rituximab era. It is noted that there is no gold standard to define tumor burden, although a number of methods are in clinical use, including for example the Groupe D’Etude des Lymphomes Follicularies (GELF) criteria. Studies have found that patients with follicular histology, high tumor burden and MRD appeared to gain the greatest benefit from maintenance rituximab [67].

4. Discussion

Identification of FL risk factors is of significant interest. In cancer study overall, it is estimated that one-third of cancer occurrences are preventable, and that one-third of people with cancer can recover fully if the cancer is detected at an early stage. Identifying risk factors contributing to FL etiology may help specify high risk population and facilitate preventive actions. FL is an indolent disease. For many patients, a “watch and wait” strategy may be sufficient [68,69]. Identifying patients possibly with an aggressive path can be the first step in formulating effective treatment strategies. Some recent studies such as [8] investigate NHL overall. Different subtypes of NHL, although sharing a certain degree of similarity, differ significantly from each other. From a practical point of view, it is more sensible to study each subtype separately.

For presentational clarity, we provide separate discussions on etiology and prognosis, and on clinical/environmental and molecular risk factors. The development and progression of FL is a continuous process and caused by the combined effects of multiple types of risk factors. Thus, in clinical practice, a large number of risk factors need to be considered simultaneously. In this article, we review a large number of potential risk factors suggested in the literature. We note that most of them have not been extensively validated in independent studies, and there is a lack of consensus. Even the successful ones, such as the FLIPI, may be of limited practical value. A few possible reasons for the lack of success are provided in the “Expert Opinion” section. Research on FL is moving fast. In addition, because of our limited knowledge in this area, our review is far from comprehensive. The present review may need to be updated in the near future.

Identification of risk factors is the first step in clinical practice. Even though a large number of possible risk factors have been suggested, most of them have not been effectively utilized in clinical practice. Due to the limited scope of this article, we refer to [5,7,70,71] for discussions on FL treatment regimens and implications of identified risk factors in initial treatment selection.

5. Conclusion

In this study, we provide review of risk factors for the risk and prognosis of FL, the second most common subtype of NHL. Our review suggests that for both etiology and prognosis, a large number of clinical, environmental and molecular risk factors have been suggested. As with other subtypes of NHL and some other types of cancers, many of those risk factors have not been rigorously validated in independent studies. More relevant discussions are provided in the “Expert opinion” section. Much more effort is needed before the reviewed risk factors can be more effectively used in clinical practice.

6. Expert opinion

In the literature, a large number of possible FL risk factors have been suggested, some of which are reviewed in this article. Because of our limited knowledge, inevitably, some important risk factors are omitted. Our limited review suggests that many of the suggested risk factors have low reproducibility and have not been successfully validated in independent studies. Some studies have even reached conflicting results. As discussed in [8,72] and references therein, multiple factors may contribute to the such an observation. In epidemiologic studies, subjects in different studies may have significantly different clinical and demographic characteristics, casting serious doubt on the comparability of those studies. In addition, even with comparable subjects, different studies may have different study objectives, collect different sets of information and adopt different statistical analysis techniques. Patients in different studies may also have different treatment regimens. In prognosis studies, treatment regimen has a tremendous impact on the prognostic path and should be accounted for in prognosis modeling and risk factor identification.

Compared with clinical and environmental data, molecular data are considerably more complicated, and so the identification of molecular risk factors can be much more challenging. There are several pitfalls in molecular studies. First of all, they may face the same subject comparability problem as with epidemiologic studies. In addition, different studies may adopt different technologies (gene expression profiling, IHC, flow cytometry, etc), which do not measure the same features. Gene expression studies make the inaccurate assumption that gene and protein expression are perfectly matched. IHC studies often fail to report on immunoarchitectural features (the distribution of cells in relation to the malignant follicles) that may be more important than the total number of cells. Flow cytometry techniques offer the possibility of more objectively enumerating many thousands of cells, but interpretation of these data is very much operator-dependent and similarly does not address aspects of immunoarchitecture. Moreover, as discussed in a recent study [8], existing analytic techniques for molecular data also have limitations. It is not uncommon that analyzing the same molecular data using different statistical techniques leads to significantly different discoveries. A representative example has been provided in [51]. It is worth noting that, several years after its publication, there is still no agreement on the analysis results for the data presented in [50,51]. In addition, some molecular studies focus on analyzing molecular data only, without properly adjusting for clinical and environmental risk factors.

The development and progression of FL is a complex process and involves the interactions of multiple clinical risk factors, environmental exposures, genetic, genomic, epigenetic events and others. It is unfortunate that most existing studies and analysis approaches have not taken an integrative approach. Rather they only focus on a subset of potential risk factors, which may result in biased estimates and false risk factor identification. For some cancers, there are studies attempting to more systematically investigate and integrate a large number of different types of risk factors. However, to the best of our knowledge, there is still no such study on FL. Studies that can more comprehensively collect and process information are much needed.

In multiple prognosis studies, the FLIPI has been shown to be effective. Similar success has been observed for a few other risk factors/indexes. However, it has been noted that these risk factors are only effective at a population level, predicting “average” risk and progression for a large group of subjects. The prediction accuracy for a single subject has been disappointingly low. In recent cancer research, a lot of attention has been devoted to personalized medicine, with the ultimate goal of predicting cancer development and progression at an individual level. For FL, there has been no risk model that is successful for such a purpose. It should be noted that the risk factors reviewed in this study, because of their commonality, are unlikely to be effective for an individual patient.

Our recommendations for future research include the following. First, existing literature needs to be more carefully and more systematically examined. Risk factors suggested in multiple studies warrant a higher priority in future validation studies. In addition, it is of significant interest to examine the causes of discrepancies in the identified risk factors. The differences between cohorts’ characteristics and technical aspects (such as platforms and profiling protocols in molecular studies) need to be quantified. Such examination may help determine how much discrepancy can be attributed to incomparability among studies (and how much cannot be explained) and improve reproducibility of identified risk factors. Second, it is worthwhile to reanalyze existing studies, particularly molecular studies, using more advanced analytic methods. For example, in most published gene expression and SNP studies, the interplays between genes and between genes and environmental risk factors have not been sufficiently accounted for. There is a need to reanalyze such data using methods that take a system biology point of view. We refer to [8] for more detailed discussions. Third, there is a need for carefully-designed, prospective studies that take an integrative approach and comprehensively collect information on multiple types of risk factors. The International Lymphoma Epidemiology Consortium (InterLymph) may serve as a prototype for such studies.

Article Highlights.

• Follicular lymphoma is the second most common subtype of NHL and the most common indolent one.

• Despite a significant amount of effort, the etiology and prognosis of FL remain poorly understood.

• A large number of potential risk factors have been suggested in the literature. However, there is a lack of consensus, and many suggested risk factors have not been rigorously validated in independent studies.

• More studies are needed to consolidate existing findings and discover more FL risk factors. There is also need for research on personalized risk factors and models.