Diffuse large B-cell lymphoma (DLBCL), the most common histologic type of non-Hodgkin lymphoma (NHL) in both immunocompetent and immunosuppressed individuals, is a clinically and molecularly heterogeneous entity. In the general population, 5-year relative survival is 59% [1], though the clinical course is highly variable. Gene expression profiling of DLBCLs has revealed distinct disease subtypes that derive from B-cells at varying stages of differentiation, specifically germinal center B-cell (GCB)-DLBCL, post-germinal center or activated B-cell (ABC)-DLBCL, and the rare primary mediastinal large B-cell lymphoma [2,3]. People with ABC-DLBCL have a substantially worse prognosis than people with GCB-DLBCL or primary mediastinal large B-cell lymphoma, and certain molecular characteristics appear distinctly related to the cell of origin, such as the association of BCL2and MYCtranslocations with GCB-DLBCL and constitutive activation of the nuclear factor- κB (NF-κB) pathway in ABC-DLBCL [4–7].

Human immunodeficiency virus (HIV) causes extensive immune dysfunction, including loss of T-cell function, altered cell signaling and chronic B-cell activation, with accompanying increases in B-cell proliferation, differentiation and apoptosis [8,9]. People with advanced HIV infection have 30-100-fold increased risk for DLBCL [10] and tend to present with aggressive disease that is more frequently of extranodal origin than in the general population [8]. Epstein- Barr virus (EBV), uncommon in DLBCLs in HIV-uninfected individuals, is detectable in over 80% of DLBCLs in HIV-infected individuals [11].

Previous studies of the cell of origin of DLBCLs in HIV-infected individuals have yielded varying results. Several studies have suggested the predominance of post-germinal center DLBCL based on gene expression profiling [12], immunohistochemistry [13] or demonstration of immunoglobulin gene somatic hypermutation [14], but other studies have not supported those findings [15–22]. Several gene expression [12,23] and immunohistochemistry-based [13,17,18,24] studies have reported an intermediate phenotype with increased co-expression of GCB and ABC markers, which complicates classification by cell of origin. Finally, studies of the prognostic significance of molecular characteristics have been conflicting [17–19], and few studies have related cell of origin to other clinicopathologic characteristics [13,17,20].

To advance understanding of the pathogenetic contribution of HIV to lymphomagenesis, we classified DLBCLs by cell of origin in archived specimens from HIV-infected and HIV-uninfected cases using the most current immunohistochemistry-based methods [25], evaluated the prognostic significance of this cell of origin classification, and compared the frequency of key molecular and viral characteristics.

All HIV-infected cases were male, with a median age at DLBCL diagnosis of 37 years (range 25–58 years) and median year of diagnosis of 1992 (range 1987–1998; largely prior to the introduction of highly active antiretroviral therapy [HAART] to treat HIV and rituximab to treat DLBCL) (Table I). In contrast, HIV-uninfected cases were evenly divided by sex, tended to be older (median age: 65 years, range 16–90) and were diagnosed over a wider range of time (median year: 1992, range 1977–2003; also largely in the pre-rituximab era). HIV-infected individuals were predominantly diagnosed with DLBCL of primary extranodal origin (61%), with 20% occurring in the brain or central nervous system, 14% in the lower gastrointestinal tract and 8% in the stomach, whereas only 39% of HIV-uninfected individuals were diagnosed with DLBCL of primary extranodal origin, with no particularly common sites. All HIV-infected and 99 (85%) HIV-uninfected cases were deceased as of 2011.

We compared the expression frequency of individual markers of B-cell differentiation by HIV status (Table II). Compared with HIV-uninfected cases, DLBCLs in HIV-infected cases generally were less likely to express germinal center markers, particularly BCL6 (28% vs. 46%, p= 0.039) and GCET1 (12% vs. 27%, p= 0.042) and more likely to express activated B-cell markers, particularly MUM1 (69% vs. 48%, p= 0.018).

Using the Tally algorithm to classify DLBCLs according to cell of origin [25], HIV-infected cases were more likely to have ABC-DLBCL than HIV-uninfected cases (HIV-infected: 83%, HIV-uninfected: 54%, p< 0.001). Among HIV-infected individuals, only 31% of the DLBCLs had one of the three translocations we evaluated, most commonly MYC/IgH, and the presence of translocations was not related to cell of origin (p= 0.648; Table III). In contrast, 47% of the DLBCLs among HIV-uninfected individuals had at least one of the three translocations we evaluated, and the presence of translocations was strongly related to cell of origin (p= 0.009). Specifically, among HIV-uninfected cases, the MYC/IgH[t(8;14)(q24;q32)] and IgH/BCL2[t(14;18)(q32;q21)] translocations were significantly more common in GCB- than ABC-DLBCL (p= 0.010 and < 0.001, respectively), and the BCL6/IgH[t(3;14)(q27;q32)] translocation was significantly more common in ABC-DLBCL (p= 0.039). EBV was detected in 63% of DLBCLs in HIV-infected cases and occurred almost exclusively in the ABC type (p= 0.002), whereas EBV was rarely detected in DLBCLs of HIV-uninfected cases (3%). In all cases, regardless of HIV status, ABC-DLBCL was somewhat more likely to be of primary extranodal origin than GCB-DLBCL, but these differences were not statistically significant. Patient characteristics such as sex and age were not significantly related to cell of origin among HIV-infected and HIV-uninfected cases.

HIV-infected cases had worse OS compared with HIV-uninfected cases (hazard ratio [HR] = 3.82, 95% confidence interval [CI] 2.57-5.67). Among both HIV-infected and HIV-uninfected cases, ABC-DLBCL was associated with adverse OS compared with GCB-DLBCL (HIV-uninfected: HR = 1.64, 95% CI 1.03-2.62, p= 0.038; HIV-infected: HR = 2.38, 95% CI 0.95–5.99,p= 0.066) (Figure 1). Patterns were similar in analyses stratified by calendar year of DLBCL diagnosis (≤ 1990, 1991-1995, ≥ 1996), and for analyses of HIV-uninfected individuals restricted to males aged < 60 years.

To address the potential for increased frequency of an intermediate phenotype (co-expression of GCB and ABC markers) in DLBCLs in HIV-infected individuals, we compared our classification of cell of origin to the immunostain algorithms reported in previous studies. Most patients with the phenotype BCL6–/MUM1+/CD138 – described by Carbone et al.[13] were classified as ABC-DLBCL using the Tally algorithm (among HIV-infected: 22 [43%] cases were BCL6-/MUM1+/CD138–, of which 21 were classified as ABC-DLBCL; among HIV-uninfected: 32 [28%] were BCL6-/MUM1+/CD138–, of which 20 were classified as ABC-DLBCL). The pattern was similar for the phenotype described by Hoffmann et al.[18](CD10 + and/or BCL6 + and MUM1 + and/or CD138+; nine [18%] HIV-infected, all classified as ABC-DLBCL; 24 [21%] HIV-uninfected, of which 18 were classified as ABC-DLBCL). In contrast, few cases were identified as CD10+/BCL6+/MUM1 + based on Chadburn et al.[17](one [2%] HIV-infected, five [4%] HIV-uninfected) in our study, and we could not directly compare our results to Madan et al.[24] because we did not include all necessary immunostains.

In a direct comparison of DLBCLs in HIV-infected and HIV-uninfected individuals, we observed striking differences in the distributions of certain molecular characteristics, supporting alternative pathways of lymphomagenesis in disease subgroups defined by both cell of origin and HIV status. Specifically, DLBCLs in HIV-infected cases were predominantly of ABC origin, with EBV-positivity almost completely concordant with the ABC phenotype. DLBCLs in HIV-infected cases tended to lack IgH/BCL2and BCL6/IgHtranslocations, regardless of the cell of origin. In contrast, DLBCLs in HIV-uninfected cases were approximately evenly divided by cell of origin, with the BCL6/IgHtranslocation more common in ABC-DLBCL and the IgH/BCL2translocation slightly more common in GCB-DLBCL. The MYC/IgHwas more common in GCB-DLBCL, regardless of HIV status. These results provide further insight into the pathogenetic mechanisms of lymphomagenesis in HIV-infected DLBCLs occurring during the pre-HAART and pre-rituximab era. The first distinct pattern of lymphomagenesis in HIV-infected cases, constituting the majority of DLBCLs, is driven by EBV, lacks typical lymphomagenic lesions and is of the ABC-type, with the central nervous system being the most frequent extranodal site. In contrast, the second pattern, similar to HIV-uninfected cases, is independent of EBV, includes typical genetic alterations, and is evenly distributed between ABC- and GCB-types.

The predominance of the ABC phenotype that we observed in HIV-infected cases is consistent with some [12–14], but not all [15-22], previous studies of DLBCL by HIV status, though direct comparison of our results with some studies is hindered by inclusion of histologies other than DLBCL, evaluation of relatively few cases or, for immunohistochemistry-based studies, use of various antibodies (alone or in combination, with a range of cut-offs to determine positivity) to identify the cell of origin. Although we identified some DLBCLs with the “intermediate” phenotypes described previously, most of these patients were classified as ABC-DLBCL by the Tally algorithm in our study. Thus, our data do not clearly support previous studies that have suggested a particular propensity of DLBCLs in HIV-infected individuals to express an intermediate phenotype with increased co-expression of GCB and ABC markers [12,13,17,18,23,24]. However, DLBCLs that cannot be classified by cell of origin also have been reported in the general population [7], suggesting that further gene expression profiling studies with the ability to evaluate an intermediate phenotype with sufficient numbers of HIV-infected and HIV-uninfected cases are warranted.

Constitutive activation of the NF-κB pathway is one of the most important features of ABC-DLBCL, based on increased expression of a range of NF-κB target genes in microarray analyses and dependence of ABC-DLBCL cell lines on NF-κB activity [7]. EBV is known to be a key driver of lymphomagenesis in immunosuppressed individuals [11], and we demonstrated a nearly exclusive association between EBV and ABC-DLBCL in the HIV-infected cases in our study. Our results are consistent with several previous studies that also reported higher EBV prevalence in ABC- than GCB-DLBCL [13,17,20,22], and may support the importance of the transforming EBV protein LMP1 in activating the NF-κB pathway [11], preferentially leading to ABC-DLBCL, although data on NF-κB were not available. EBV also may preferentially lead to ABC-DLBCL by down-regulating BCL6 and up-regulating BCL2 [29-32]. In HIV-uninfected individuals, we observed a higher frequency of the BCL6/IgHin ABC-DLBCL than in GCB-DLBCL, which also is consistent with previous research [33]. Interestingly, although the BCL6/IgHoccurred in DLBCLs of only two HIV-infected cases, both were EBV- and had the ABC phenotype.

GCB-DLBCL is also associated with a range of genetic alterations, including translocations involving the BCL2and MYConcogenes [6,7]. In our study, the MYC/IgHwas indeed more common in GCB-DLBCL, regardless of HIV status. In contrast, we observed a higher frequency of the IgH/BCL2translocation in GCB-DLBCL only in HIV-uninfected cases.

The ABC phenotype was associated with adverse prognosis compared with GCB-DLBCL in our study, regardless of HIV status, although the associations were of borderline statistical significance. Although the prognostic significance of the cell of origin classification independent of the International Prognostic Index is well-established in HIV-uninfected cases [25], some studies [18,19,22], but not all [17], have reported a similar association in HIV-infected cases. These inconsistent results may be related to the use of different classification approaches to identify the cell of origin. Additionally, the role of underlying differences in patients' immune status may explain some inconsistencies, because the risk of DLBCL in HIV-infected individuals is related to immune suppression [34], and DLBCL risks have declined since the introduction of HAART [10]. Almost all cases in our study were diagnosed before 1996, so further studies in the HAART era that can directly evaluate the relationship of molecular characteristics to the degree of immunosuppression are warranted, and would be hypothesized to observe a higher proportion of GCB-DLBCL in HIV-related cases due to improved immune function.

The cell of origin may have been misclassified for some cases in our study because we used immunohistochemistry rather than gene expression profiling [35]. However, the Tally algorithm has the best concordance with gene expression profiling studies to date [25], and does not rely on BCL6, the germinal center marker associated with the most rapid antigen degradation [36]. In addition, we included blinded duplicates to ensure the reproducibility of our immunostain interpretation. Some misclassification also could have occurred because we used different cut-offs to determine positivity for FOXP1 (80%), MUM1 (80%) and GCET1 (60%) since the Tally algorithm (30% cut-off for all immunostains) was published after we completed immunostain interpretation. The difference in the cut-offs would bias away from classification of ABC-DLBCL because we used more stringent criteria for the two ABC markers, and thus our identification of a high proportion of ABC-DLBCL in HIV-infected cases is likely conservative. Our analyses of OS should be interpreted cautiously because we lacked detailed clinical information (e.g. International Prognostic Index parameters, DLBCL treatment, CD4 count, HIV treatment) and because nearly all HIV-infected cases were diagnosed before the introduction of HAART, and nearly all patients were diagnosed prior to the introduction of rituximab as standard therapy for DLBCL. Finally, because cases identified for this study had tissue specimens that otherwise would have been discarded by local area pathology departments (e.g. for logistical or storage constraints), our population was not necessarily representative of all DLBCL cases diagnosed in the LA SEER registry.

In summary, our results provide strong evidence for both similarities and differences in the molecular characteristics of DLBCLs in HIV-infected and HIV-uninfected individuals. Specifically, among HIV-infected cases, DLBCLs were predominantly EBV+, with nearly all EBV+ DLBCLs of the ABC-type and lacking typical lymphomagenic translocations. In contrast, a small proportion of DLBCLs in HIV- infected cases and nearly all DLBCLs in HIV-uninfected cases were EBV–, with near even distribution of ABC and GCB phenotypes, and increased frequency of the MYC/IgHand IgH/BCL2translocations in GCB-DLBCL. Future research evaluating the distribution and prognostic significance of molecular characteristics of DLBCLs in HIV-infected individuals should include patients treated with HAART and be able to measure the relationship of molecular characteristics to the degree of immunosuppression. In addition, analyses that directly compare a wider range of genetic alterations in DLBCLs by HIV status would provide further insight into the pathogenetic mechanisms of lymphomagenesis.