Cell‐of‐origin determined by both gene expression profiling and immunohistochemistry is the strongest predictor of survival in patients with diffuse large B‐cell lymphoma

Abstract The tumor cells in diffuse large B‐cell lymphomas (DLBCL) are considered to originate from germinal center derived B‐cells (GCB) or activated B‐cells (ABC). Gene expression profiling (GEP) is preferably used to determine the cell of origin (COO). However, GEP is not widely applied in clinical practice and consequently, several algorithms based on immunohistochemistry (IHC) have been developed. Our aim was to evaluate the concordance of COO assignment between the Lymph2Cx GEP assay and the IHC‐based Hans algorithm, to decide which model is the best survival predictor. Both GEP and IHC were performed in 359 homogenously treated Swedish and Danish DLBCL patients, in a retrospective multicenter cohort. The overall concordance between GEP and IHC algorithm was 72%; GEP classified 85% of cases assigned as GCB by IHC, as GCB, while 58% classified as non‐GCB by IHC, were categorized as ABC by GEP. There were significant survival differences (overall survival and progression‐free survival) if cases were classified by GEP, whereas if cases were categorized by IHC only progression‐free survival differed significantly. Importantly, patients assigned as non‐GCB/ABC both by IHC and GEP had the worst prognosis, which was also significant in multivariate analyses. Double expression of MYC and BCL2 was more common in ABC cases and was associated with a dismal outcome. In conclusion, to determine COO both by IHC and GEP is the strongest outcome predictor to identify DLBCL patients with the worst outcome.

non-GCB by IHC, were categorized as ABC by GEP. There were significant survival differences (overall survival and progression-free survival) if cases were classified by GEP, whereas if cases were categorized by IHC only progression-free survival differed significantly. Importantly, patients assigned as non-GCB/ABC both by IHC and GEP had the worst prognosis, which was also significant in multivariate analyses. Double expression of MYC and BCL2 was more common in ABC cases and was associated with a dismal outcome. In conclusion, to determine COO both by IHC and GEP is the strongest outcome predictor to identify DLBCL patients with the worst outcome.  3,4 Several studies have shown a survival benefit for DLBCL patients with a GCB phenotype compared to an ABC phenotype, 5,6 whereas other have not. [7][8][9] In addition, a third group with unclassified cases (UC) was reported, and proposed to have an inferior outcome similar to ABC-DLBCL. 4,10 In the updated World Health Organization (WHO) classification of Tumors of Hematopoietic and Lymphoid Tissues, 2 information on the cell-of-origin (COO), either by immunohistochemical (IHC) stainings 6,[11][12][13] or by GEP, is required for a definite DLBCL diagnosis.
In clinical practice, however, the use of GEP has not been widely adopted. Most GEP technologies require fresh-frozen tumor tissue.
But in the daily, clinical diagnostic work-up, formalin fixed, paraffinembedded (FFPE) tissue is the primary source, and fresh-frozen material is not routinely collected. Therefore, IHC algorithms have been developed as substitutes and applied with varying concordance to GEP. The most commonly used classification is the Hans algorithm 11 based on the IHC staining results of three proteins: CD10, BCL6 and MUM1, although other systems have also been proposed. 6,12,13 However, these IHC algorithms will only identify two groups; GCB or non-GCB, since they cannot identify cases classified as UC by GEP.
In recent years, the NanoString technology Lymph2Cx assay was developed based on GEP, which shows a strong concordance to the original COO model and can be applied on FFPE tissue. 10,[14][15][16] This assay uses a limited set of 15 pre-specified genes and five housekeeping genes, and has the potential to identify all three subgroups of DLBCL. 17,18 In this study, our aim was to investigate the concordance between the Lymph2Cx assay and the IHC algorithm by Hans et al. 11   Patients with a known previous history of a low-grade lymphoma were excluded. Clinical information was collected from patient records.
Patients were followed-up with clinical examinations and radiologic examinations were used when relapse or progressive disease was suspected. Age-adjusted International Prognostic Index (aaIPI) was used (one point for each: (a) Ann Arbor stage III-IV; (b) elevated serum lactate dehydrogenase (LDH); (c) and ECOG performance status 2-3). Here 0-1 is considered to be low risk and 2-3 is considered to be high risk, in accordance with national guidelines in Sweden and Denmark.

| RNA extraction
Extraction of RNA from FFPE tissue was done according to the AllPrep DNA/RNA Mini Kit for FFPE protocol (Qiagen, Hilden, Germany). That protocol allows for the simultaneous purification of genomic DNA and total RNA from the same biological sample. Purification of RNA was done with the AllPrep column flow-through, using an RNeasy Mini spin column.

| NanoString assay
Samples were analyzed with the Lymph2CX assay on a NanoString instrument according to the manufacture's instructions. The dataset was analyzed using the research use only (RUO) version of the NanoString Lymphoma Subtyping Test (LST), which is based on the Lymph2Cx assay, to determine the COO molecular subtype of each sample. 19 The LST algorithm measures the geometric mean of five housekeeping genes (HK geomean), to ensure RNA quality based on a pre-defined clinical QC threshold of 128. An HK geomean value below 64 was deemed as insufficient RNA quality to provide a subtyping result. A value between 64 and 128 was considered to be borderline quality since it meets previously published thresholds for RNA quality within clinical research studies, 18

| Immunohistochemical stainings
The IHC stainings for CD10, BCL2, BCL6, MUM1 and MYC were performed at the different sites according to routine procedures in each diagnostic laboratory. The stainings were re-evaluated semi-quantitatively by each site's hematopathologists (authors MH, MF, MA, SBE, HMP). The Hans algorithm was applied to classify tumors as GCB or non-GCB by IHC, and included CD10, BCL6 and MUM1 stainings with a cut-off of 30% positive tumor cells. For MYC, a cut-off of 40% was applied and for BCL2 50%. Since insufficient material was a problem in a majority of the cases, FISH analyses for BCL2 and MYC were not performed.

| Cell-of-origin groups
The following subgroups were defined according to GEP or IHC: 1. ABC = ABC type defined by GEP and classified with the Lymph2Cx assay. Additional COO groups were studied and are presented in the supplementary material (Supplementary methods Tables S1 and S2).

| Ethics
The study was conducted in accordance with the Declaration of

| Statistical analyses
Tabulated values were compared using the chi-square or the Fisher's exact test. Student's ttest was used to compare means between groups.
Pearson's test was applied to determine correlative associations between parameters. Overall survival (OS) was calculated from the date of diagnosis to the date of death of any cause. Progression-free survival (PFS) was calculated from the date of diagnosis to the date of lymphoma progression or death due to any cause. Survival curves and univariate analyses were performed using the Kaplan-Meier method, and the log-rank test and Cox proportional hazards regression were used to compare differences between groups. Cases with missing information on clinical or pathological variables were not included in the survival analyses. Multivariate Cox proportional hazards regression models included prognostic variables of at least borderline significance (P < .10). Cases with one or more missing variables were omitted from the multivariate analysis. The proportional hazards assumption was tested and was not violated. A P value <.05 was considered to be statistically significant. Statistical analyses were performed using RStudio 1.1.383 (www.r-project.org).

| Univariate survival analysis
Patients classified as ABC according to the Lymph2Cx assay had significantly inferior five-year survival rates at 58% for OS and 56% for PFS. This is compared with 71% for OS and 69% for PFS in the GCB-GEP group, and 82% for OS and 78% for PFS in the UC-group ( Figure 1A,D). Patients categorized as non-GCB by the Hans algorithm showed inferior five-year survival rates at 65% for OS and 62% for PFS, compared with 72% for OS and 71% for PFS in the GCB-IHC group ( Figure 1B,E). Patients grouped as ABC according to the Lymph2Cx assay and non-GCB by the Hans algorithm demonstrated inferior five-year survival rates at 53% for OS and 51% for PFS. This is compared with 74% for OS and 72% for PFS in cases that were not ABC and non-GCB combined ( Figure 1C,F). There were other variables associated with inferior OS (Table 2) and PFS (

| UC cases
Of 42 cases categorized as UC according to the Lymph2Cx assay, 33 (79%) were non-GCB and 9 (21%) were GCB-IHC according to the Hans algorithm (Table S3) In conclusion, GEP combined with IHC to classify cases as ABC/non-GCB is the best predictor of inferior survival, in both uni-and multivariate analyses, probably by identifying cases at the extreme ends of the GCB and ABC spectrum. We also found that cases classified by IHC as non-GCB, were more often GCB-GEP or UC than vice versa for the GCB-IHC Lastly, we want to thank Uppsala-Umeå Comprehensive Cancer Consortium and grants provided through regional agreement between Umeå University and Västerbotten County Council on cooperation in the field of Medicine, Odontology and Health.