Longitudinal Changes in Immune Activation Serum Biomarkers Prior to Diagnosis and Risk of B-cell NHL Subtypes

This article is featured in Highlights of This Issue, p. 155

Errors in DNA modifying activities that are associated with the activation of B lymphocytes, including immunoglobulin heavy chain gene (IgH) class-switch recombination (CSR) and somatic hypermutation (SHM) of Ig genes, can contribute to the development of mature B-cell malignancies, including several histologic subtypes of non–Hodgkin lymphoma (NHL) of B-cell origin (1, 2). In fact, there is ample evidence that chronic stimulation of B cells by infection [e.g., human immunodeficiency virus (HIV), H. pylori, hepatitis C virus], or by forms of autoimmunity that involve autoantibody production (e.g., Sjögren's; refs. 3–7), is associated with an increased risk of B-cell NHL. Therefore, factors that promote B-cell activation, including cytokines and other immune stimulatory molecules, would be expected to be associated with an increased risk for the development of B-cell NHL (8). The overall goal of this investigation is to determine whether an immune system environment that promotes chronic B-cell stimulation precedes the development of NHL of B-cell origin, by determining if the rate of change in serum levels of B cell–stimulatory cytokines and molecules that are associated with B-cell activation precede the development of follicular lymphoma (FL) and/or diffuse large B-cell lymphoma (DLBCL).

There are several reports in the literature, in study populations worldwide, which have examined the association between pre-diagnosis B-cell activation markers and NHL risk by quantifying serum/plasma levels of various cytokines and other immune molecules (9–27); levels of circulating (serum/plasma) sCD30, CXCL13, and IL10 have been associated with NHL risk in several studies (12, 14–23, 25–27). In most of these studies, however, single serum samples were available for analysis at varying time points prior to NHL diagnosis, resulting in cross-sectional examinations of serum cytokine levels in relation to NHL risk. Three of these cohort studies measured a limited number of cytokines in two serum samples per participant (25–27). In all three studies, there was limited ability to examine the changes of these immune markers over time in FL and DLBCL subtypes because of small sample sizes.

Larger studies of NHL using several serial samples within individuals to assess a panel of B-cell activation molecules would allow for a more robust evaluation of immune marker trajectories prior to diagnosis, especially by NHL subtypes given the heterogeneity in B-cell NHL and the likelihood that the trajectories of immune activation molecules differ by histologic subtypes. For these reasons, we conducted a prospective study, with independent review of pathology results, in an ethnically diverse, apparently immunocompetent population without HIV infection to determine whether the risk of B-cell NHL subtypes is associated with pre-diagnosis serum levels of 15 cytokines, chemokines, and other immune-related biomarkers: IL1β, IL2, IL4, IL5, IL6, IL10, IL12, CXCL13, IL8, TNFα, IFNγ, GM-CSF, VEGF, sCD30, IgE. To better characterize the underlying cytokine immune milieu prior to diagnosis, we constructed several traditional statistical models such as linear mixed equations and Generalized Estimating Equations (GEE) models as well as a machine learning algorithm based on random forests to simultaneously assess cytokine levels variable importance (28). We also used more traditional regression methods to assess the relationship of several variables to diagnosis.

As we were able to retrieve up to three serial serum samples collected over several years prior to B-cell diagnosis in this study cohort, we also evaluated the longitudinal changes in levels of these molecules and the subsequent development of FL and DLBCL, two common histologic subtypes of NHL that emerge from germinal center B cells.

Incident cases of NHL of the FL or DLBCL histologic subtypes, identified from 1992 through 2008 among relatively young, ethnically diverse active-duty military personnel with an archived specimen in the DoD Serum Repository (DoDSR) drawn prior to the diagnosis date, were eligible for inclusion in the study. Cases were identified from the Armed Forces Institute of Pathology National Pathology Repository and the military Automated Centralized Tumor Registry. Pathology reports were obtained and reviewed by study pathologist for over 90% of FL and DLBCL cases, to confirm the diagnosis and date of diagnosis. Histologic determination was based on the WHO 2008 Classification (29). Each case was matched whenever possible with two controls on age (±1 year), gender, race, branch of service, and dates of blood collection (±90 days), via the Defense Medical Surveillance System (DMSS), the capabilities of which include linking individuals identified from various demographic and medical databases to serum specimens archived in the DoDSR (30). All study subjects were HIV seronegative, and had no previous cancer diagnosis, other than nonmelanoma skin cancer.

The DoDSR, which is managed by the Armed Forces Health Surveillance Center, has archived over 40 million serum specimens from active-duty and reserve personnel with more than 2 million new samples added yearly. Residual serum is stored from routine HIV type I (HIV-I) screening and from HIV-I testing of pre- and post-deployment specimens. During handling outside the freezers, the specimens are kept on beds of dry ice to preclude freeze–thaw cycles and then are stored at −30°C.

Serum levels of IL1β, IL2, IL4, IL5, IL6, IL10, IL12, IL8, TNFα, IFNγ, GM-CSF, VEGF were quantified by multiplexed immunometric assay (Luminex platform) 12-plex bead array immunoassay (High Sensitivity Human Cytokine Panel, R&D Systems). Briefly, Luminex microparticles pre-coated with analyte-specific antibodies were incubated with diluted serum samples, followed by a biotin–antibody and by a streptavidin–phycoerythrin conjugate. The fluorescence intensity of each analyte's microparticles was quantified using a Bioplex 200 (Luminex) System Analyzer (Bio-Rad), and the data analyzed using BioPlex Manager (v 4.1.1) software, using a five parameter logistic curve fit, with assay results expressed in pg/mL. Serum levels of sCD30 (Bender MedSystems), CXCL13 (R&D Systems), and total IgE were determined by ELISA and also expressed as pg/mL. Total serum IgE was determined using the CIA-7.12 and CIA-4.15 monoclonal antibodies as previously described (31, 32).

In many subjects, serum levels of IL2, IL4, and IL12 were undetectable. For these markers as well as other markers with undetectable results, we first assigned the value of zero to the undetectable level and then we added the value of 1.0 for log transformation.

Laboratory personnel were blinded to samples’ case–control status. We randomly ordered case and control samples within matched sets and tested all samples of a given set on the same Luminex or ELISA plate. We interspersed blinded quality control aliquots of pooled sera from the control population and observed good to excellent within-plate and between-plate assay coefficients of variation (CV). Intra-plate CVs for IL8, IL10, TNFα, VEGF, sCD30, IgE, and CXCL13 were 6.2%, 18.4%, 8.8%, 9.3%, 10.8%, 10.7%, and 9.5%, respectively. Inter-plate CVs for IL8, IL10, TNFα, VEGF, sCD30, IgE, and CXCL13 were 8.5%, 25.1%, 11.1%, 11.6%, 11.6%, 17.0%, and 11.4%, respectively. Serum levels of several of the cytokines measured (IL1β, IL2, IL4, IL5, IL6, IL12, GM-CSF, IFNγ) were either below or near the level of detection in most determinations, precluding the calculation of meaningful CV estimates for those molecules.

Prior to analysis, data quality was explored by assessing outliers and skewness. Groups were compared using the t test or Wilcoxon Rank Sum test as appropriate. All of the immune marker variables had non-normal distributions with right-skewing, thus immune-biomarker results were natural log-transformed to meet model assumptions. The immune markers were analyzed as continuous variables as well as categorical variables with cut points for each assay set at the quartiles of the control distributions. For immune markers with very limited data above the level of detection, data were analyzed as dichotomous variables, comparing the fraction of detectable versus undetectable between cases and controls.

Longitudinal data were analyzed by using linear mixed modeling with both fixed and random effects and odds ratios (OR) and 95% confidence intervals (CI) between cases and controls were estimated. This allowed each individual to deviate randomly from the population probability distribution. To assess nonlinearity, a spline-based lowess regression model also was fitted. An interaction term was included to assess the relation between time to diagnosis and FL or DLBCL. GEE analyses were also conducted to allow for the examination of repeated measures data and to determine the trajectory of the immune marker measurements as time to diagnosis approached. All P values were two-tailed. Data were analyzed using SAS version 9.2.

Missing data were imputed using rfimpute in the randomForest package in R. We also used Fuzzy Forests, a machine learning algorithm that extends Random Forests in the case of correlation (28). Fuzzy Forests works in stages, first with an unsupervised learning part that creates weighted correlation networks on all potential predictor variables. The second stage consists of a supervised learning part that ranks the variable importance in terms of prediction. Fuzzy Forests is freely available from CRAN and was implemented using Fuzzy Forest package using R Studio version 1.1.463.

The institutional review boards of the Walter Reed Army Institute of Research and UCLA approved the research protocol.

The datasets generated during the current study are available from the corresponding author on reasonable request.

The study population was comprised of 500 NHL cases and 985 controls, all of whom were on active duty on the case diagnosis date (Table 1). Comparison of the distribution among study subjects according to age, gender, race, or branch of service did not reveal any notable differences between the case and control groups as expected, as the controls were matched to cases on these factors (Table 1). The median age of cases was 36 years, 92% of cases were male, and 67% were white. Pathology review of all cases determined that 211 were DLBCL and 142 were of the FL subtype (Table 2).

Two or more serial pre-NHL diagnosis serum samples were available from the DoDSR for 86.2% of NHL cases and 86.0% of controls. Three serial pre-NHL diagnosis serum samples were available for 71% of NHL cases and 70.7% of controls. The majority of the specimens tested came from the time period most proximal to NHL diagnosis, with 61% of the serum specimens tested obtained during the time window 0 to 5 years preceding NHL diagnosis, and 11% of specimens collected >10 years pre-diagnosis.

We conducted the analyses separately for DLBCL and FL. We modeled serum levels of sCD30, CXCL13, IL10, IL5, IFNγ, TNFα, IL6, GM-CSF, IgE, IL8, IL1β, and VEGF as continuous variables and estimated an OR per unit increase in the immune marker on the natural log scale with adjustment for the matching factors (Table 3). When all of the biomarkers were entered into the same model, only sCD30, CXCL13, IL10, and VEGF remained significant among DLBCL cases compared with controls. When the B-cell activation molecules were mutually adjusted for the other biomarkers, only sCD30 and CXCL13 remained significant for FL.

Changes in mean levels of sCD30, CXCL13, IL10, TNFα, and INFγ, from the earliest to most recent measurement before diagnosis in DLBCL cases compared with controls, are shown in Table 4; Supplementary Fig. S1 for DLBCL, and Table 5; Supplementary Fig. S2 for FL. P values were calculated for the delta between the earliest and the newest mean of the serum sample for each of these markers using the t test (Table 4 for DLBCL and Table 5 for FL, respectively). Longitudinal changes in biomarkers prior to diagnosis showed increases in serum levels of sCD30, CXCL13, IL10, TNFα, and IFNγ in cases compared with controls in DLBCL. For FL, longitudinal changes prior to diagnoses showed increases in serum levels of sCD30, CXCL13, and IL10. In both DLBCL and FL subtypes, a strong, statistically significant increase in serum levels from the earliest to the most recent specimens was noted for sCD30 and CXCL13. In contrast, a significant rate of change in IL10 serum levels in FL and DLBCL subtypes was observed in specimens collected closer to diagnosis. Serum levels of TNFα and IFNγ were elevated only in specimens closest to diagnosis, comparing cases with controls in DLBCL subtypes.

The results of the lowess analysis for DLBCL and FL for sCD30, CXCL13, and IL10 are shown in Fig. 1, with individual biomarker levels plotted on the y axis versus years to diagnosis on the x-axis. sCD30 and CXCL13 levels appeared to be elevated as early as 5 years prior to diagnosis for FL cases compared with controls. CXCL13 levels were seen to be higher in cases than in controls even earlier in DLBCL subtype. IL10 trajectories were noted closer to diagnosis for both FL and DLBCL.

Fuzzy forest analysis confirmed sCD30, CXCL13 and IL10 as the most important variables in predicting cases from controls for both FL and DLBCL. In fact, the two most predictive markers were CXCL13 at time point 3 (closest to diagnosis) and sCD30 at time point 3, and clustered together in the weighted correlation network in the forest part of the algorithm. It is important to point out that the module formation is an unsupervised learning technique where the outcomes are not known to the algorithm. It is also interesting to note that it grouped the time point 3 biomarkers separately from time points 1 and 2 for the most cases. This is most likely because there is a dramatic increase in these activation markers shortly before diagnosis.

The availability of serum samples collected longitudinally in the DoDSR allowed the prospective assessment of the association between circulating serum levels of 15 cytokines, chemokines and other immune-related biomarkers and the risk of FL and DLBCL, two common histologic subtypes of NHL, over a period of nearly 20 years preceding the diagnosis of these B-cell NHL. These results indicate that sCD30, CXCL13, and IL10 levels are elevated with increasing serum concentrations over time preceding the clinical recognition of DLBCL and FL, in presumably immunocompetent persons, and suggest that B cell–stimulatory cytokines may contribute to the etiology of these NHL subtypes. In addition, these three biomarkers have the potential to serve as a biomarker for NHL. These findings are in general agreement with the findings of several prior studies examining prediagnostic B cell–activation associated molecules in single samples per participant, both in immunocompetent persons (12, 14–23), and in the setting of HIV infection (33–35), and further characterize the role of immune markers in lymphomagenesis by examining immune marker trajectories. In addition, several studies have noted that polymorphisms in the IL10 gene, as well as in the gene encoding the receptor for CXCL13, are associated with risk for the development of FL and/or DLBCL (36–38).

CD30 is a member of the TNF receptor superfamily and the receptor for CD30 ligand (CD153 or TNFSF8), a TNF-like immune stimulatory molecule (39). CD30 is expressed on Hodgkin lymphoma (HL) Reed-Sternberg (HRS) cells (40, 41), where it has the potential to serve as a receptor for signals that may enhance their proliferation. The results presented here indicate that sCD30 levels in circulation are elevated for several years preceding NHL diagnosis in FL and DLBCL subtypes and are markedly increased as time to diagnosis approaches.

Elevated serum levels of sCD30 have been reported to precede the diagnosis of acquired immune deficiency syndrome (AIDS)-associated B-cell NHL (31, 34) and B-cell NHL in immunocompetent subjects (12, 14, 15, 18, 19, 22, 23), neither of which typically express CD30, as well as HL in immunocompetent persons (32). In a nested case–control study of a subset of participants enrolled in two cohorts, CLUE-1 and CLUE-11, sCD30 levels were measured in two serum samples (26); however, this study did not contain sufficient numbers of FL and DLBCL cases with repeated samples to allow for analyses by subtype. Spath and colleagues (25) examined the trajectory of sCD30 and CXCL13 in two serum samples from 30 FL and 42 DLBCL cases that had specimens stored in The Northern Sweden Health and Disease Study. Consistent with our findings, increasing levels of sCD30 over time towards diagnosis was associated with FL risk. Purdue and colleagues (27), however, reported elevated but stable sCD30 levels, with no evidence of an increasing trajectory, in two samples from the Janus Serum Bank in Norway for 39 cases of FL and 66 cases of DLBCL. It is noteworthy that we observed a significant trajectory for sCD30 levels and risk of DLBCL. Neither Spath and colleagues (25) nor Purdue and colleagues (27) reported similar findings. One possible explanation is that our sample size was much larger with 211 DLBCL cases matched 1:2 to controls. We also had access to three specimens per individual for 71% of our study population, which provided a more stable estimate of serial immune marker levels. The median time from the earliest specimen collection to diagnosis was shorter, however, compared with the prior two studies (25, 27), which may be the reason for the differences in the findings. As DLBCL is considered an aggressive NHL subtype, with a relatively short latency, these findings suggest that the observed sCD30 trajectory is likely indicative of the presence of an emerging DLBCL.

IL10 is a B cell–stimulatory cytokine that also can suppress T-cell and monocyte responses, including antiviral cell-mediated responses (42). IL10 can enhance BCL2 expression, thereby promoting B-cell viability (43). IL10 has the potential to contribute to the pathogenesis of NHL in several ways; IL10 can directly stimulate B cells and also can inhibit antiviral T cell–mediated responses, such as those involved in the immunoregulation of Epstein-Barr virus (EBV) infection, so may also promote B-cell activation driven by EBV. IL10 is produced by various immune cell subsets, including regulatory T and B cells and monocytes (44, 45). DLBCL tumor cells also can produce IL10, allowing cell-autonomous growth and immune escape by these tumor cells (46). Elevated levels of IL10 in circulation have been seen to be associated with an increased risk for NHL, both in HIV-infected (33) and in presumably immunocompetent persons (16, 20, 21). In addition, elevated levels of regulatory B cells (IL10-producing cells) in circulation were seen to precede the diagnosis of AIDS-associated NHL (47). Genetic polymorphisms in the IL10 gene also have been seen to be associated with risk for NHL in several studies [reviewed in (37, 38)]. To our knowledge, the current study is the first to report on the trajectory of IL10 serum levels prior to FL and DLBCL diagnosis, finding that the rate of change in IL10 levels increased closer to diagnosis compared with results for sCD30 and CXCL13.

CXCL13 is a chemokine produced by follicular dendritic cells and helper T cells in secondary lymphoid organs (48). The receptor for CXCL13 is CXCR5 (BLR1). CXCL13 plays a central role in the trafficking of B cells into and within lymphoid follicles, that is necessary for antigen exposure and activation within germinal centers, leading to the differentiation of B cells into antibody-producing plasma cells (49–51). Aberrant CXCL13 and CXCR5 participate in the pathogenesis of multiple subtypes of lymphomas (52–55), and elevated levels of human CXCL13 were seen in a NOD-SCID xenograft model of AIDS-associated Burkitt lymphoma (56). Elevated levels of circulating CXCL13 have been seen to be associated with an increased risk for NHL in both HIV-infected persons (35, 57) and in presumably immunocompetent persons (17, 19, 23). In addition, the gene encoding the receptor for CXCL15, CXCR5/BLR1, was seen to be associated with risk for FL in a large GWAS study carried out in the Interlymph consortium (36), suggesting an important role for CXCL13 and its receptor in the pathobiology of NHL. Spath and colleagues (25) also examined changes in CXCL13 serum levels in repeated samples. A significant slope was found for DLBCL, comparing cases with controls, but no trajectory was found for FL. In contrast, we observed a significant trajectory of CXCL13 levels with both FL and DLBCL subtypes.

Our study had some limitations. The study population was composed mainly of young men, with a median age of diagnosis of 36 years, which is younger than the age distribution of the cohorts in other similar investigations (25–27). Our results, therefore, may not apply to women or be extrapolated to all cases of FL or DLBCL. In addition, the median follow-up time from the earliest serum sample to incident disease was 5.5 years, which is a relatively short window compared with other studies, and hampered our ability to interpret whether the observed immune markers trajectories may reflect early disease processes. The study also had considerable strengths. The availability of three serum samples for 71% of the study population, collected longitudinally over several years before the onset of B-cell NHL, allowed the examination of the temporal dynamics of these markers. Pathology reports were retrieved for over 90% of the cases and each report was reviewed by a study pathologist to determine NHL histologic subtypes. We controlled for potential confounding by matching on known demographic risk factors for B-cell NHL. The current study also improves on evidence from previous similar investigations as the sample size was relatively large allowing a more robust assessment of the rate of change prior to diagnosis for a wide range of B-cell activation molecules

The use of machine learning techniques in immunological studies has been limited until recently. In this study, machine learning indicated that three biomarkers, namely sCD30, CXCL13, and IL10, are important markers for predicting lymphoma. While machine learning identified the same three markers that have already been identified using classical modeling approaches, and three markers that are already well established in the literature, obtaining the same finding using an alternative, unrelated analytical approach enhances the strengths of this finding, and points to common biological pathways leading to NHL.

Ongoing stimulation by B-cell activation molecules, including sCD30, CXCL13, and IL10, may lead to B-cell activation, promote the viability of these cells, and drive B cells toward a germinal center, where they undergo interactions with other immune cells that lead to IgH CSR and SHM, DNA-modifying activities that are mediated by activation-induced cytidine deaminase (AICDA). This can place B cells at higher risk for the development of the many forms of NHL that emerge from germinal center B cells, as errors that occur during IgH CSR and SHM can result in lymphomagenic lesions, such as oncogene translocation and/or mutation. As the serum levels of sCD30, CXCL13, and IL10 increased progressively over time to NHL diagnosis, the markedly elevated levels closer to diagnosis likely reflect the presence of the nascent lymphoma, systemic inflammation associated with the presence of NHL, and/or the infiltration of reactive lymphocytes into the affected lymph nodes. The stable elevation in B-cell NHL risk with elevated levels of sCD30, CXCL13, and IL10 across several years prior to diagnosis may also indicate the presence of a B cell–stimulatory environment that promotes the genesis of these cancers.

C.M. Ramirez reports grants from NIH during the conduct of the study. E.C. Breen reports grants from NIH during the conduct of the study. O. Martínez-Maza reports grants from NIH during the conduct of the study. No disclosures were reported by the other authors.

The opinions or assertions contained herein are the private views of the authors, and are not to be construed as official, or as reflecting true views of the National Institutes of Health, the Department of the Army or the Department of Defense, Department of Energy or ORAU/ORISE. The investigators have adhered to the policies for protection of human subjects as prescribed in AR 70-25.

L.I. Levin: Conceptualization, supervision, funding acquisition, methodology, writing–original draft, project administration, writing–review and editing. C.M. Ramirez: Conceptualization, resources, data curation, software, formal analysis, investigation, methodology, writing–original draft, writing–review and editing, statistical analysis. E.L. Liao: Data curation, software, formal analysis, investigation, methodology, writing–original draft, statistical analysis. H. Guo: Conceptualization, data curation, software, formal analysis, investigation. B.K. Kim: Data curation, formal analysis, investigation, methodology, pathology review. A.J. Marrogi: Conceptualization, data curation, formal analysis, investigation, methodology, pathology review. L.I. Magpantay: Conceptualization, data curation, software, investigation, methodology. E.C. Breen: Conceptualization, data curation, formal analysis, investigation, methodology, writing–original draft, writing–review and editing. O. Martínez-Maza: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, investigation, methodology, writing–original draft, project administration, writing–review and editing.

We would like to thank Guadalupe Peña and José León Merino for their technical assistance and Dr. Angela Cost and staff at the Armed Forces Health Surveillance Branch for their assistance in the identification of control subjects and retrieval of serum specimens from the Department of Defense Serum Repository.

Data from the DMSS, The Armed Forces Health Surveillance Branch, Defense Health Agency, U.S. Department of Defense, Silver Spring, Maryland were obtained for study subjects from 1990 to 1999. Serum specimens from the Department of Defense Serum Repository, The Armed Forces Health Surveillance Branch, Defense Health Agency, U.S. Department of Defense, Silver Spring, Maryland were retrieved from study subjects from 1988 to 1999.

This work was supported by the NCI at the NIH (grant number R01-CA121195, C.M. Ramirez, E.L. Liao, L.I. Magpantay, E.C.Breen, and O. Martínez-Maza). The UCLA Center for AIDS Research was supported by a grant from the NIH (P30-AI028697) and by the Pendleton Charitable Trust and the McCarthy Family Foundation. The Jonsson Comprehensive Cancer Center is supported by a grant from the NIH (P30-CA016042). This research also was supported in part by an appointment (Hongyu Guo) to the Department of Defense (DOD) Research Participation Program administered by the Oak Ridge Institute for Science and Education (ORISE) through an interagency agreement between the U.S. Department of Energy (DOE) and the DOD. ORISE is managed by ORAU under DOE contract number DE-SC0014664.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.