Cell-of-origin classification using the Hans and Lymph2Cx algorithms in primary cutaneous large B-cell lymphomas

Primary cutaneous diffuse large B-cell lymphoma, leg type (PCDLBCL-LT) and primary cutaneous follicle center lymphoma with a diffuse population of large cells (PCFCL-LC) are both primary cutaneous B-cell lymphomas with large-cell morphology (CLBCL) but with different clinical characteristics and behavior. In systemic diffuse large B-cell lymphoma, not otherwise specified (DLBCL-NOS), gene-expression profiling (GEP) revealed two molecular subgroups based on their cell-of-origin (COO) with prognostic significance: the germinal center B-cell-like (GCB) subtype and the activated B-cell-like (ABC) subtype. This study investigated whether COO classification is a useful tool for classification of CLBCL. For this retrospective study, 51 patients with PCDLBCL-LT and 15 patients with PCFCL-LC were analyzed for their COO according to the immunohistochemistry-based Hans algorithm and the NanoString GEP-based Lymph2Cx algorithm. In PCFCL-LC, all cases (100%) classified as GCB by both Hans and Lymph2Cx. In contrast, COO classification in PCDLBCL-LT was heterogeneous. Using Hans, 75% of the PCDLBCL-LT patients classified as non-GCB and 25% as GCB, while Lymph2Cx classified only 18% as ABC, 43% as unclassified/intermediate, and 39% as GCB. These COO subgroups did not differ in the expression of BCL2 and IgM, mutations in MYD88 and/or CD79B, loss of CDKN2A, or survival. In conclusion, PCFCL-LC uniformly classified as GCB, while PCDLBCL-LT classified along the COO spectrum of DLBCL-NOS using the Hans and Lymph2Cx algorithms. In contrast to DLBCL-NOS, the clinical relevance of COO classification in CLBCL using these algorithms has limitations and cannot be used as an alternative for the current multiparameter approach in differentiation of PCDLBCL-LT and PCFCL-LC. Supplementary Information The online version contains supplementary material available at 10.1007/s00428-021-03265-5.


Introduction
Primary cutaneous diffuse large B-cell lymphoma, leg type (PCDLBCL-LT) and primary cutaneous follicle center lymphoma with a diffuse population of large cells (PCFCL-LC) are both primary cutaneous B-cell lymphomas with largecell morphology (CLBCL), but with different clinical characteristics and behavior [1]. PCDLBCL-LT is an aggressive lymphoma with a 5-year disease-specific survival (DSS) of 56% despite standard immunochemotherapy (R-CHOP). PCFCL-LC usually has an indolent behavior with a 5-year DSS of 95% and is preferably treated with local radiotherapy [1,2]. PCDLBCL-LT was initially recognized as a subgroup of PCFCL-LC with a somewhat different morphology, typical presentation with skin lesions on the leg(s), and a more

Case selection
Skin biopsies of 66 patients with a CLBCL, including 51 patients with PCDLBCL-LT and 15 patients with PCFCL-LC, were selected from the archives of the Departments of Pathology of the Leiden University Medical Center (LUMC), The Netherlands (n = 60), and the University Hospitals Leuven (UZL), Belgium (n = 6). Only cases with diffuse infiltrates that were predominantly composed of large B cells were included. Clinical information and follow-up data were collected from the Dutch Registry of Cutaneous Lymphoma and/or from medical records. Thirty-five patients with PCDLBCL-LT and 11 patients with PCFCL-LT were reported in previous studies analyzing immunohistochemical and genetic markers [8,17]. FFPE skin biopsies were centrally reviewed by AS, PJ, and RW and all patients were diagnosed according to the criteria of the 2016 revision of the WHO classification and the 2018 update of the WHO-EORTC classification [1,18]. At time of diagnosis, presence of extracutaneous disease was excluded by standard staging procedures, consisting of a fluorodeoxyglucose positron emission tomography-computed tomography (FDG PET-CT) scan or a CT scan in combination with a bone marrow biopsy.
In addition to GEP, the general molecular profile was determined via targeted next-generation sequencing (tNGS) with the LYMFv1 panel including 52 B-cell lymphoma-related genes, according to the methods described in a previous study [20]. The complete molecular profile of PCDLBCL-LT and PCFCL-LC was included in Supplemental Fig. 1. For the purpose of the current study, only the mutational status of MYD88 and CD79B and loss of CDKN2A were reported.

Statistical analysis
Comparison of the characteristics between the COO subgroups was performed with the χ 2 test for categorical variables and the Kruskal-Wallis test for continuous variables. The overall survival (OS) and DSS were estimated with the cumulative proportion surviving at the end of 5 years of follow-up duration. In time-to-event analysis, the starting point was the date of the histological diagnosis. An event was defined as death by any cause for OS and death by lymphoma/lymphoma-related death for DSS. Patients without an event at the last time of follow-up were censored. Risk groups were compared with the log-rank test. Median follow-up duration was calculated with the reverse Kaplan-Meier method [21]. Statistical analyses were performed with IBM SPSS Statistics 25. A p-value of < 0.05 was considered statistically significant.

Results
The clinical characteristics and follow-up data, immunohistochemistry, genetic alterations, and the results of COO classification of the 51 PCDLBCL-LT and 15 PCFCL-LC are presented in Table 1

Clinical characteristics
Patients with PCDLBCL-LT included 26 females and 25 males. The median age at diagnosis was 78 (range, 47 to 92) years. Forty-three of 51 (84%) patients presented with skin lesions on one or both legs. In the remaining patients, disease was located on the upper lip (n = 1), cheek (n = 1), abdomen (n = 1), arm (n = 3), hip (n = 1), and penis (n = 1). Initial therapy with multiagent chemotherapy (n = 21 with rituximab and n = 5 without rituximab) or radiotherapy (n = 23) resulted in a complete remission in 48 of 49 (98%) cases. In two patients, no treatment was given because of spontaneous remission of the skin lesions (n = 1) or sudden unrelated death (n = 1). After a median follow-up of 11.3 (range, 0.17 to 16.3) years, 20 patients were still alive with (n = 8) or without (n = 12) evidence of disease, 16 patients had died of lymphoma, and 15 patients had died of unrelated disease. At 5 years, the overall survival was 48% and the disease-specific survival was 59%.
Patients with PCFCL-LC included four females and 11 males with a median age at diagnosis of 58 (range, 37 to 67) years. Fourteen of 15 (93%) patients presented with (localized) skin lesions in the head-and-neck region and/or trunk. The remaining patient presented with skin lesions on the right arm and left leg. Initial treatment consisted of radiotherapy and resulted in a complete remission in all patients. After a median follow-up of 7.0 (range, 4.3 to 16.1) years, all 15 patients were alive with (n = 8) or without (n = 7) evidence of disease.

Hans algorithm
In PCDLBCL-LT, the immunohistochemical markers of the Hans algorithm, CD10, BCL6, and MUM1, showed expression in 7 of 51 (14%), 34 of 51 (67%), and 42 of 51 (82%) patients, respectively (Fig. 1). As most cases expressed MUM1 and were negative for CD10, the majority (38/51; 75%) was classified as non-GCB. In the remaining 13 cases (25%), GCB classification was attributed to the expression of CD10 in seven patients and isolated expression of BCL6 in six patients. Comparison between PCDLBCL-LT cases classified as either non-GCB or GCB showed no differences in the expression of BCL2 and IgM as well as in survival.
In the PCFCL-LC group, all patients (15/15; 100%) were classified as GCB subtype. This was attributed to both CD10 and BCL6 expression in eight patients and to isolated BCL6 expression in seven patients. None of the cases expressed MUM1.

Lymph2Cx algorithm
In contrast to the Hans algorithm, COO classification of the PCDLBCL-LT patients with the Lymph2Cx panel identified only nine of 51 (18%) patients as ABC subtypes (Fig. 1). The other cases were classified as GCB (20 of 51; 39%) or as UI (22 of 51; 43%). The Lymph2Cx prediction scores ranged between 0.00 and 1.00. Regarding the patient characteristics, immunohistochemistry, MYD88 and CD79B mutations, loss of CDKN2A, and survival of the PCDLBCL-LT patients, there were no statistically significant differences between the three COO subgroups. All nine patients in the ABC subgroup were also classified as non-GCB by the Hans algorithm (100% concordance), but in the GCB category, classification of only six of 20 (30%) patients was concordant. The UI category consisted of a mixture of cases classified as non-GCB (15/22; 68%) and GCB (7/22; 32%) by Hans. In the PCFCL-LC group, COO classification with the Lymph2Cx algorithm was fully concordant with the Hans algorithm and all 15 patients were classified as GCB subtypes. Their prediction scores were all determined at 0.00.

Genetic profile
The results of mutational analysis with our 52-gene panel are presented in Supplementary Fig. 1 for the PCDLBCL-LT and Supplementary Fig. 2 for the PCFCL-LC. Thirtysix of 47 (77%) PCDLBCL-LT patients harbored a MYD88 mutation, including the hotspot L265P in 30 (83%) of these patients, and 22 of 47 (47%) patients had a CD79B mutation, with the hotspot Y196 mutation in 13 (59%) of these cases.

Discussion
Diagnosis of CLBCL is primarily made on a combination of the clinical presentation and the morphology and immunophenotype of the tumor cells (supported by specific molecular markers). In order to determine the discriminative performance of COO classification in CLBCL, this study describes the results of the IHC-based Hans algorithm and the GEP-based Lymph2Cx algorithm in 66 patients with a CLBCL, including 51 patients with PCDLBCL-LT and 15 patients with PCFCL-LC. All cases of PCFCL-LC were uniformly classified as GCB by both the Hans algorithm and the Lymph2Cx algorithm. These results correspond to literature with GCB classification of PCFCL-LC in 23 of 25 (92%) cases using Hans and in all ten (100%) PCFCL cases with spindle-cell morphology using NanoString technology [22,23].
In contrast to the homogeneous results in PCFCL-LC, the results of COO classification in PCDLBCL-LT patients were heterogeneous. Using Hans, 75% of the PCDLBCL-LT patients classified as non-GCB and 25% as GCB. However, using the Lymph2Cx algorithm, only 18% of the cases was classified as ABC, while 39% was determined as GCB and 43% as UI.
Regarding the Hans algorithm, our results correspond roughly to a study reporting non-GCB classification in 14 of 23 (61%) PCDLBCL-LT patients [24]. In another study, however, all 32 PCDLBCL-LT cases (100%) had a non-GCB phenotype [22]. In this latter study, CD10 expression, a key marker of the Hans algorithm, was used as an exclusion criterion for diagnosis of PCDLBCL-LT. Although CD10 expression is uncommon in PCDLBCL-LT, it has never been used as an exclusion criterion for diagnosis of PCLDBCL-LT [18,25]. In our cohort of PCDLBCL-LT, all CD10-positive cases showed the characteristic features of PCDLBCL-LT, including presentation on the leg(s), roundcell morphology, lack of (residual) follicular dendritic cellnetworks, expression of BCL2 and MUM1 in all but one case, IgM expression in all cases, and MYD88 mutations in all cases. In addition, there was no difference in survival between the CD10-positive and CD10-negative PCDLBCL-LT cases (OS: 57% vs. 46%, log-rank 0.69; DSS: 69% vs. 57%, log-rank p-value 0.67). Therefore, there is no reason to use CD10 expression for exclusion as PCDLBCL-LT.
Regarding the Lymph2Cx algorithm, no studies specifically using this technique in PCDLBCL-LT patients have been published so far. Interestingly, the percentage of ABC-classified cases (18%) of PCDLBCL-LT in our cohort is much lower than expected based on the similarities in GEP of PCDLBCL-LT with ABC-DLBCL and PCFCL-LC with GCB-DLBCL, as described by Hoefnagel et al. [16]. Also, the immune phenotype of PCDLBCL-LT is characterized by protein expression of the activated B-cell markers BCL2, MUM1, and IgM in the vast majority of the patients. Accordingly, the genetic profile is dominated by NFκB-activating mutations, predominantly in MYD88 and CD79B, which are enriched in ABC-DLBCL and, especially, primary central nervous system lymphoma (PCNSL), primary testicular lymphoma (PTL), and intravascular large B-cell lymphoma [26][27][28]. In contrast, the GCB marker BCL6 is commonly expressed [2] and a significant subgroup of PCDLBCL-LT harbors MYC rearrangements [17,29], generally seen as an early GC event [30]. In PCNSL and PTL, 60 to 96% is classified as non-GCB by the Hans algorithm, corresponding to our results (75%) in PCDLBCL-LT [31][32][33][34]. However, Bödör et al. [31] identified a much higher percentage of ABC cases defined by Lymph2Cx in 77 PCNSL cases (81%) compared with PCDLBCL-LT in our study (18%). On the other hand, Montesinos et al. [35] also demonstrated heterogeneity in GEP profiles in 21 PCNSL cases, as 24% classified as Lymph2Cx algorithm (3 groups), respectively, and log-rank for survival analysis. Bold values are statistically significant b This patient presented with skin lesions on the right arm and left leg c (Immuno)chemotherapy consisted of R-CHOP, CHOP, or (R)CHOP-like regimens, such as R-CEOP. From the patients who received (immuno) chemotherapy, 21 (81%) patients received regimens with rituximab and five (19%) patients, all diagnosed between 2001 and 2005, received regimens without rituximab d No therapy was given because of spontaneous remission of the skin lesions (n = 1) and sudden unrelated death (n = 1) e Data is missing in one case f Data is missing in two cases g Data is missing in three cases h Data is missing in four cases ABC, 33% as GCB, and 43% as UI using the COO algorithm developed by Wright et al. [36]. In addition, a study using microarray-based GEP in 23 PCNSL cases demonstrated overlapping expression of germinal-center and activated B-cell genes [37]. In line with these findings, recent studies in DLBCL-NOS also demonstrated that binary classification into ABC or GCB does not correspond to the more complex biology as identified by transcriptomics and that recurrent genetic alterations are identified beyond the established COO subtypes [38][39][40].
The discrepancy in results of COO classification in PCDLBCL-LT may partly be clarified by the significantly different techniques and diverse selection of genes used for classification. The analysis of Hoefnagel et al. [16] was based on 43 discriminating genes, of which only 3 genes (IRF4, ITPKB, and PIM2) overlapped with the genes in the Lymph2Cx panel. In addition, their study did not address the UI category, as was determined in 43% of our PCDLBCL-LT cases. Nevertheless, the poor correspondence and high percentage (43%) of UI may suggest a different biology of PCDLBCL-LT compared with DLBCL-NOS for which these algorithms were originally developed. Future studies need to investigate this observation with more comprehensive transcriptomic analysis, such as the EcoTyper, as recently described by Steen et al. [40].
In our cohort of PCDLBCL-LT, MYD88 and/or CD79B mutations as well as survival did not statistically differ between the COO subtypes defined by both algorithms. This corresponds to PCNSL, as Bödör et al. [31] also reported equal distribution of MYD88 and/or CD79B mutations and no differences in survival between the COO subgroups.
While our results showed that classification as non-GCB with Hans or ABC/UI with Lymph2Cx strongly favors a diagnosis of PCDLCBL-LT, a GCB outcome does not differentiate between PCFCL-LC and PCDLBCL-LT. Use of COO algorithms for this purpose should, therefore, be interpreted with caution, as incorrect interpretation may have far stretching clinical implications. For instance, PCDLBCL-LT as defined by the 2016 revised WHO classification or the 2018 update of the WHO-EORTC classification but categorized as GCB by either the Hans or Lymph2Cx algorithm may be misdiagnosed as PCFCL-LC and, subsequently, treated incorrectly as a low-grade CLBCL with radiotherapy instead of immuno-polychemotherapy.
In conclusion, our data confirm the GCB-subtype of PCFCL-LC but show that PCDLBCL-LT cases classify along the whole COO spectrum of DLBCL-NOS using the Hans and Lymph2Cx algorithms. In contrast to systemic DLBCL-NOS, the clinical relevance of COO classification in CLBCL using these algorithms is limited and cannot be used as an alternative for the current multiparameter approach in differentiation of PCDLBCL-LT and PCFCL-LC.