Introduction: Diffuse large B-cell lymphoma (DLBCL) is a heterogeneous entity. Lately, several algorithms achieving therapeutically and prognostically relevant DLBCL subclassification have been published. Methods: A cohort of 74 routine DLBCL cases was broadly characterized by immunohistochemistry (IHC), fluorescence in situ hybridization (FISH) of the BCL2, BCL6, and MYC loci, and comprehensive high-throughput sequencing (HTS). Based on the genetic alterations found, cases were reclassified using two probabilistic tools – LymphGen and Two-step classifier, allowing for comparison of the two models. Results: Hans and Tally’s overall IHC-based subclassification success rate was 96% and 82%, respectively. HTS and FISH data allowed the LymphGen algorithm to successfully classify 11/55 cases (1 – BN2, 7 – EZB, 1 – MCD, and 2 – genetically composite EZB/N1). The total subclassification rate was 20%. On the other hand, the Two-step classifier categorized 36/55 cases, with 65.5% success (9 – BN2, 12 – EZB, 9 – MCD, 2 – N1, and 4 – ST2). Clinical correlations highlighted MCD as an aggressive subtype associated with higher relapse and mortality. Conclusions: The Two-step algorithm has a better success rate at subclassifying DLBCL cases based on genetic differences. Further improvement of the classifiers is required to increase the number of classifiable cases and thus prove their applicability in routine diagnostics.

Diffuse large B-cell lymphoma (DLBCL) is the most common lymphoma, affecting diverse molecular pathways and displaying various treatment outcomes [1‒3]. In clinical practice, the most commonly used method – also recommended by the World Health Organization (WHO) [4] and the International Consensus Classification (ICC) [5] – to subclassify DLBCLs is the cell of origin (COO) classification, distinguishing between the germinal center B-cell (GCB) and the non-GCB DLBCL type [6]. In routine diagnostics, this is most commonly achieved by using the immunohistochemical approaches proposed by Hans et al. [7] and the Tally algorithm by Meyer et al. [8]. The resulting phenotypic distinction by COO is linked to better responses in GCB-DLBCL, for which the standard therapy is R-CHOP (rituximab, cyclophosphamide, doxorubicin hydrochloride, oncovin, and prednisone) [9]. COO is also useful in understanding the varied sensitivities to targeted therapies like Bruton tyrosine kinase (Btk) inhibitors [10] or immunomodulatory drugs such as lenalidomide [11]. However, the COO by no means fully accounts for the great clinical heterogeneity of DLBCL.

The COO classification was deduced from a study in the early 2000s by Alizadeh and colleagues [12] based on gene expression profiling (GEP). In more recent years, Schmitz et al. [13] showed a differential mutational distribution depending on COO. Simultaneously, Chapuy et al. [14] have demonstrated that DLBCL is a genetically heterogeneous disorder with characteristic patterns of multiple low-frequency mutations, somatic copy number alterations (SCNAs), and structural variants (SVs) with 5 genetic subgroups of DLBCL (C1 – C5). These groups were more or less equivalent to the ones described by Schmitz et al. [13], therefore C1 = BN2, C2 = A53, C3 = EZB, C4 = ST2, C5 = MCD. Based on these findings in 2020, Wright et al. established the LymphGen classifier, which succeeded to separate DLBCL tumors into 7 subgroups based on defining genetic features, adding the N1 class, which is defined by NOTCH1 mutations. Additionally, the EZB subtype was divided further into MYC+ and MYC- clusters. Crucially, the importance of having a genetically composite group, in which cases are core members of more than one subtype, appreciating the genetic heterogeneity of DLBCL was acknowledged. Furthermore, a year later, Pedrosa et al. [15] came up with a two-step algorithm called the Two-step classifier, which incorporated key statistical features mainly from LymphGen and work by Lacy et al. [16]. However, the Two-step classifier omits the A53 subtype since it is heavily based on SCNA data input, which data are not commonly available in practice.

In this study, we performed a comprehensive characterization of 74 real-life cases of DLBCL by performing IHC, fluorescence in situ hybridization (FISH), along with genomic DNA high-throughput sequencing (HTS). We also aimed to test the applicability of LymphGen and the Two-step algorithms in order to draw conclusions on their prognostic relevance.

Patient Selection

The Institute of Medical Genetics and Pathology of the University Hospital Basel’s archives were searched for patients with DLBCL during the period 2010–2020. Seventy-four cases met the inclusion criteria for tissue microarray (TMA) construction (shown as a whole slide imaging scan, accessible via http://meqvslidewp01.usb.ch/OlyViaWeb). Clinical data included age, sex, Ann Arbor stage, international prognostic index (IPI), sites of involvement (nodal vs. extranodal), administered therapy, date of diagnosis, and date of last follow-up.

Protein Expression by Immunohistochemistry

Immunohistochemistry was performed following ISO15189-accredited standard diagnostic operating procedures of our institute on an automated stainer (Benchmark Ultra from Roche/Ventana Medical Systems, Tucson, AZ, USA). All further details on staining and evaluation techniques are listed in the supplementary file provided (online suppl. Table 1; for all online suppl. material, see https://doi.org/10.1159/000535938).

Fluorescence in situ Hybridization

FISH was performed with locus-specific identifier dual-color, break-apart probes for BCL2 (ZytoLight SPEC BCL2 Z-2192 from Zytomed Systems, Berlin, Germany), BCL6 (01N23-020 from Abbott/Vysis, Green Oaks, IL, USA), and MYC (ZytoLight SPEC MYC Z-2090-200 from Zytomed Systems).

DNA Extraction

The PrepSEQ™ 1-2-3 Nucleic Acid Extraction Kit (#4452222 Thermo Fisher Scientific, Waltham, MA, USA) was used for DNA extraction from 10 μm-thick untreated formaldehyde-fixed paraffin-embedded (FFPE) tissue samples according to the manufacturer’s protocol. Only cases with a tumor cell content (TCC) >20% were considered eligible for HTS.

HTS and Data Analysis

Analysis for mutations by HTS (online suppl. Table 2) was performed using IonAmpliSeq™ customized and validated lymphoma panel comprising 1,428 amplicons (size range: 125–275) of 146 genes (Thermo Fisher Scientific, Waltham, MA, USA) [17].

LymphGen Classifier

The LymphGen tool is an open-access online platform by the National Cancer Institute (https://llmpp.nih.gov/lymphgen/index.php). It uses a supervised algorithm to assign samples based on probability and recognizes that not every DLBCL sample contains sufficient subtype characteristics.

Two-step Classifier

This classification method is based mainly on Lacy et al. [16] which includes Bernoulli mixture-model clustering and subtype analysis in relation to clinical characteristics and outcomes, and the LymphGen subtype characteristics [18]. Therefore, the two-step classifier integrates the most significant features from both studies and classifies cases in two steps.

Statistics

The R package “complex heatmap” [19] was used to create an oncoplot of the mutated genes with hierarchal clustering based on genetic subtype. Kaplan-Meier survival analysis was performed using GraphPad Prism 10 (GraphPad Software Inc., La Jolla, CA, USA). More information can be found in the online supplementary methods provided.

Clinical Characteristics

Forty-one patients were male, and 33 were female. The majority presented nodal involvement (n = 48), whereas 26 had primarily extranodal manifestations (Fig. 1a). Patients’ mean age was 75, with a median of 78 and a range of 38–101. Most were treated with R-CHOP-based treatments (n = 38); all other therapies and information on staging are specified in Table 1. The average follow-up was 50.9 months, the median was 48 months, and the range was 1–264 months. Thirteen patients experienced relapses, and 10 patients died of their lymphoma. Clinical data for 7 patients were unavailable due to loss of follow-up.

Fig. 1.

a A representative image of a diffuse large B-cell lymphoma (DLBCL) case included in our cohort; ×400. b Expression of CD10; immunoperoxidase staining; ×100. c Expression of FOXP1; immunoperoxidase staining; ×100. d Fluorescence in situ hybridization (FISH) analysis applying a break-apart probe for the BCL6 locus showing clearly separated green and red signals (rearranged allele) and fused signs (wild-type allele) in lymphoma cells; ×600.

Fig. 1.

a A representative image of a diffuse large B-cell lymphoma (DLBCL) case included in our cohort; ×400. b Expression of CD10; immunoperoxidase staining; ×100. c Expression of FOXP1; immunoperoxidase staining; ×100. d Fluorescence in situ hybridization (FISH) analysis applying a break-apart probe for the BCL6 locus showing clearly separated green and red signals (rearranged allele) and fused signs (wild-type allele) in lymphoma cells; ×600.

Close modal
Table 1.

Information on therapy and stage

Tretment and disease stagePatients, n
Therapy 
 R-CHOP 38 
 (DA)-R-EPOCH 
 R-Benda 
 Other 8* 
 No therapy 
 Deceased prior to therapy 
 NA 15 
Ann Arbor stage 
 I 
 II 10 
 III 15 
 IV 14 
 NA 26 
Tretment and disease stagePatients, n
Therapy 
 R-CHOP 38 
 (DA)-R-EPOCH 
 R-Benda 
 Other 8* 
 No therapy 
 Deceased prior to therapy 
 NA 15 
Ann Arbor stage 
 I 
 II 10 
 III 15 
 IV 14 
 NA 26 

*These comprised mini R-POCH (n = 1), HOPE (n = 1), HDT-aHCT (allogenic hematopoietic cell transplant) (n = 1), palliation (n = 1), rituximab (n = 1), R + ribomustin (n = 1), R + RTX (n = 1), surgery (n = 1).

COO-Classification by Immunohistochemical Algorithms

IHC algorithms were performed for all cases (n = 74). The Hans algorithm had a 96% success rate when compared to Tally with 82% (Fig. 1b, c). The Hans algorithm classified 40 cases as GCB and 31 as non-GCB, whereas Tally classified 30 as GCB and 31 as non-GCB. Furthermore, 3 and 13 cases were not evaluated by Hans and Tally, respectively, due to missing marker data on the TMA and a lack of material for additional staining. From the 13 non-evaluable cases by Tally, 5 were classified as GCB and 8 as non-GCB by Hans. The 3 cases that were not evaluable by Hans were all classified as non-GCB by Tally. Discrepancies between the Hans and the Tally algorithms were seen in 5 cases. Four of the cases were GCB by Hans and non-GCB by the Tally, and 1 case was non-GCB by Hans and GCB by the Tally algorithm.

FISH Analysis of MYC, BCL2, and BCL6 Rearrangements

FISH analysis was carried out for cases with sufficient material available (BCL2: 56 cases; BCL6: 57 cases; MYC: 54 cases). There were 13 DLBCLs with BCL6 translocations (Fig. 1d), 12 with BCL2 translocations, and 4 with MYC translocations. Of these cases, 4 had both BCL2 and BCL6 rearrangements, and 1 case presented with BCL2 and MYC rearrangements, which was reclassified to high-grade B-cell lymphoma, double-hit.

High-Throughput Sequencing

Sixty-two DLBCLs had enough DNA available for HTS. Of these 62 cases, 55 passed quality control (QC) and exhibited 205 pathogenic mutations in 53 genes. The most commonly mutated genes were TP53 (n = 17), KMT2D (n = 13), SOCS1 (n = 11), CREBBP (n = 9), and TNFAIP3 (n = 9). Additionally, 4 cases were found to have an MYD88L265P mutation (Fig. 2). The range of mutations per case was between 1 and 20; the median was 3 mutations/case.

Fig. 2.

Oncoplot showing mutated genes/case (n = 55) with its variant type, clustered by genetic subtype (by the two-step classifier). Additionally, mortality, relapse, cell of origin (COO) subtype: germinal center B-cell-like (GCB) and non-GCB by immunohistochemistry (IHC), and therapy are presented. NA was given when data was unavailable, except for the genetic subtype, where NA is when a case lacks defining genetic features and cannot be categorized.

Fig. 2.

Oncoplot showing mutated genes/case (n = 55) with its variant type, clustered by genetic subtype (by the two-step classifier). Additionally, mortality, relapse, cell of origin (COO) subtype: germinal center B-cell-like (GCB) and non-GCB by immunohistochemistry (IHC), and therapy are presented. NA was given when data was unavailable, except for the genetic subtype, where NA is when a case lacks defining genetic features and cannot be categorized.

Close modal

Comparison between the Two-step Classifier and LymphGen for Subclassifying DLBCL

The Two-step classifier categorized 36/55 submitted cases, with a 65.5% success rate. Nine cases were BN2, 12 – EZB, 9 – MCD, 2 – N1, and 4 – ST2. From the 19 cases that were not categorized, 7 were genetically composite (1 was BN2/ST2, 2 were MCD/BN2, 3 were EZB/ST2, and 1 was BN2/EZB), and 12 lacked defining genetic features. Since the Two-step classifier does not automatically identify cases that are genetically composite, manual evaluation was conducted by scrutinizing the scores for the various genetic subtypes (online suppl. Table 3). As expected, most MCD cases were of non-GCB and most EZB cases were of GCB origin. The N1 cases were GCB and non-GCB (Fig. 2).

On the other hand, LymphGen version 1.0 classified 11/55 cases, with a total subclassification rate of 20%. One case was BN2, 7 – EZB, 1 – MCD, and 2 – genetically composite (EZB/N1). From the 9 BN2 cases classified by the Two-step classifier, only 1 was classified as BN2 by the LymphGen, and the rest of the cases were unclassified. From the 12 EZB cases classified by the Two-step classifier, 6 were also categorized as EZB by the LymphGen, 1 case was genetically composite (EZB/N1), and the other 5 cases were unclassified. Also, from the 9 cases classified as MCD by the Two-step classifier, only 1 was classified as MCD by LymphGen, and the other 8 were not classified. From the 2 N1 cases classified by the Two-step classifier, only one was categorized as genetically composite (EZB/N1), and the second case was unclassified. All 4 ST2 cases classified by the Two-step classifier were not classified by LymphGen. The 19 cases that were left unclassified by the Two-step classifier were also unclassified by LymphGen except for 1 case that was categorized as EZB and was genetically composite by the Two-step classifier. Furthermore, when LymphGen version 2.0 was applied to our incomplete input data, we only obtained an 18% success rate. Ten cases were classified as follows: 1 – BN2, 6 – EZB, 1 – MCD, and 2 genetically composite cases (EZB/N1) (online suppl. Table 3). The success rates of the three algorithms can be seen in Table 2. Figure 3 summarizes the results obtained by the various investigations of our cohort.

Table 2.

Rates of genetic subtypes, classified by the different algorithms

Rates, %MCDBN2EZBST2N1A53Other
MYC+MYCNAcompositeunclassified
Expected 8.7 13.3 5.9 17.6 6.4 1.7 5.8 40.6 
Two-step classifier output 16.4 16.4 1.8 14.5 5.5 7.3 3.6 Not applicable 12.7 21.8 
LymphGen V1.0 output 1.8 1.8 1.8 9.1 1.8 3.6 80 
LymphGen V2.0 output 1.8 1.8 1.8 7.3 1.8 3.6 81.8 
Rates, %MCDBN2EZBST2N1A53Other
MYC+MYCNAcompositeunclassified
Expected 8.7 13.3 5.9 17.6 6.4 1.7 5.8 40.6 
Two-step classifier output 16.4 16.4 1.8 14.5 5.5 7.3 3.6 Not applicable 12.7 21.8 
LymphGen V1.0 output 1.8 1.8 1.8 9.1 1.8 3.6 80 
LymphGen V2.0 output 1.8 1.8 1.8 7.3 1.8 3.6 81.8 

The output from the 3 classifiers was compared to the expected rates (by LymphGen V1.0).

Our cohort contained 17 cases (30.9%) with TP53 mutations that did not fulfill the assignment criteria of other categories and might have fit the A53 subtype provided their data on somatic copy-number alterations (SCNAs) was present and they had an altered copy-number.

Fig. 3.

Flow chart summarizing the results obtained by immunohistochemistry (IHC), fluorescence in situ hybridization (FISH), high-throughput sequencing (HTS), and the various classifiers applied.

Fig. 3.

Flow chart summarizing the results obtained by immunohistochemistry (IHC), fluorescence in situ hybridization (FISH), high-throughput sequencing (HTS), and the various classifiers applied.

Close modal

Reclassification of Discrepant Cases by IHC-Based COO Classification

As mentioned above, there were 5 cases that were differentially categorized by the two IHC algorithms. Therefore, in order to refine the COO, we looked at their mutational profiles (Table 3).

Table 3.

Discordant cases

CaseHans algorithmTally algorithmMutated genesGenomic profileDetermined COO
Non-GCB (CD10−/BCL6+/MUM1+) GCB (CD10−/GCET+/FOXP1−/MUM1+/LMO2+) BCL10, KMT2D, BCL2 EZB GCB 
GCB (CD10−/BCL6+/MUM1−) Non-GCB (CD10−/GCET−/FOXP1+/MUM1−/LMO2−) EBF1 BN2 Non-GCB 
GCB (CD10−/BCL6+/MUM1−) Non-GCB (CD10−/GCETdim+/FOXP1+/MUM1−/LMO2−) XPO1, PIM1 (2 variants), CD79B MCD Non-GCB 
GCB (CD10−/BCL6+/MUM1−) Non-GCB (CD10−/GCET−/FOXP1+/MUM1−/LMO2−) TNFAIP3, TP53 BN2 Non-GCB 
GCB (CD10dim+/BCL6+/MUM1+) Non-GCB (CD10dim+/GCET−/FOXP1+/MUM1+/LMO2−) NA NA GCB 
CaseHans algorithmTally algorithmMutated genesGenomic profileDetermined COO
Non-GCB (CD10−/BCL6+/MUM1+) GCB (CD10−/GCET+/FOXP1−/MUM1+/LMO2+) BCL10, KMT2D, BCL2 EZB GCB 
GCB (CD10−/BCL6+/MUM1−) Non-GCB (CD10−/GCET−/FOXP1+/MUM1−/LMO2−) EBF1 BN2 Non-GCB 
GCB (CD10−/BCL6+/MUM1−) Non-GCB (CD10−/GCETdim+/FOXP1+/MUM1−/LMO2−) XPO1, PIM1 (2 variants), CD79B MCD Non-GCB 
GCB (CD10−/BCL6+/MUM1−) Non-GCB (CD10−/GCET−/FOXP1+/MUM1−/LMO2−) TNFAIP3, TP53 BN2 Non-GCB 
GCB (CD10dim+/BCL6+/MUM1+) Non-GCB (CD10dim+/GCET−/FOXP1+/MUM1+/LMO2−) NA NA GCB 

Case 1 was categorized as EZB (with mutations in BCL10, KMT2D, BCL2) by the Two-step classifier and therefore rather matches the Tally algorithm as being of probable GCB origin.

Case 2 was BN2 (with an EBF1 mutation) and therefore it could be of either COO subtype. When re-evaluating the IHC findings, we had more arguments for the non-GCB than the GCB subtype (expression of FOXP1 and lack of LMO2).

Case 3 was MCD (with mutations in XPO1, PIM1, CD79B) and therefore it was more likely to be non-GCB, correlating to the Tally algorithm.

Case 4 was BN2 (with mutations in TNFAIP3, TP53) and thus it could be of either COO subtype. Re-evaluating the IHC findings, we had more arguments for the non-GCB than the GCB subtype (expression of FOXP1 and lack of LMO2).

Case 5, which was GCB by Hans and non-GCB by the Tally algorithm, was not sequenced due to a lack of material available, and therefore its mutational profile could not be analyzed. Due to its weak CD10 expression along with BCL6, it was considered to belong to the GCB-type.

Correlation of Molecular Data and Clinical Outcome

Finally, we attempted to correlate clinical outcomes, i.e., relapses and mortality, with genetic subclassification. Most patients who relapsed and died of lymphoma were MCD, followed by patients who were not genetically subclassified. One BN2 case relapsed, and another died of lymphoma. One EZB, one ST2, and one genetically unclassified case each relapsed and died of lymphoma. One N1 and one composite (MCD/BN2) patient experienced a relapse only. Finally, 1 genetically unclassified patient had a mortality event only (online suppl. Fig. 1; Table 4).

It is largely acknowledged that DLBCL is a heterogeneous disease with wide variations regarding clinical presentation, outcome, and underlying lymphomagenesis. Thus, it is vital to achieve an improved subclassification of this disease to provide better, more tailored treatments and to identify those with a higher risk for progression and relapse. This study attempts to assess whether the revolutionary improvement in our understanding and subcategorization of DLBCLs achieved in the last decade is already applicable to daily routine practice and whether it results in a better correlation regarding patients’ outcomes.

Seventy-one cases were evaluated by the Hans and 61 by the Tally algorithms, respectively. The Hans algorithm classified more cases as GCB than Tally, which may be because BCL6 weighs more in this algorithm and is not strictly exclusive to GCB cells [20]. We had 5 cases that were differentially classified by the two IHC algorithms; however, by considering their mutational profiles, i.e., genetic subclassification, we could conclude their potential COO in 2/5 cases. For the remaining three cases, we looked in detail at the staining intensities from both the Hans and Tally algorithms for the COO confirmation. Based on our evaluation, Tally COO seems to be closer to the biological reality considering the better correlation with the genomic classifiers.

Our FISH analysis established that BCL6 was the most commonly rearranged gene (n = 13), followed by BCL2 and MYC, which is in accordance with previously published studies and further confirms the comparability of our cohort to the general aspects of DLBCLs [21, 22]. In efforts to genetically subclassify our cases, we used the LymphGen algorithm [18] and the Two-step classifier [15]. We originally attempted to categorize our cases using LymphGen version 1.0, which resulted in a 20% success rate. By applying version 2.0 [23], which did not require SCNA data, we only obtained an 18% success rate. Therefore, there are no clear discrepancies between the two versions’ rigor. Four cases had MYD88L265P mutations, which is a dominant genetic feature of the MCD genetic subtype. However, both LymphGen versions were unable to classify 3 of these cases and 1 case was categorized as genetically composite (EZB/N1) (online suppl. Table 3).

The Two-step classifier had a clearly higher success rate in classifying our cases (65.5%). This highlights its better applicability for routine cases as no SCNA data are needed. Furthermore, the Two-step classifier algorithm seems more robust than those of both LymphGen variants. However, it omits the A53 subtype, which is important for inclusion considering that the most frequently mutated gene was TP53 in our cohort and it is very frequently mutated in DLBCL in general [24]. It is worth appreciating the benefit of using a two-step algorithm while also acknowledging the existence of composite cases that reflect the genetic heterogeneity of DLBCL.

The work-up of cases by IHC and FISH analysis is a routine standard diagnostic procedure, as recommended by the WHO and the ICC. However, genomic sequencing is not universally performed in all cases due to its considerably high price and, so far, little clinical relevance. Moreover, the type and utilization of custom panels differ across different institutions, contingent on their budget constraints. The cost for sequencing a patient with the panel that we have used for this research evaluation is ∼CHF 2,000. The usage of the algorithms is currently free of charge, and they can be used to evaluate not only whole cohorts but also 1 case at a time. Yet, advanced statistical knowledge is beneficial to comprehend the algorithms and a certain level of proficiency for working with RStudio is required when using the Two-step algorithm. An advantage of the LymphGen algorithm is that it considers mutations/fusions/translocations/amplifications/gains/homozygous and heterozygous deletions in 114 genes, whereas the Two-step classifier considers 40 genes only, which is quite limiting. Additionally, these 40 genes must be deliberated in the input file, even if they are not present in the custom panel that has been used for the sequencing. Even though both genetic algorithms have their advantages and disadvantages, currently the Two-step classifier performs better in subclassifying day-to-day cases in our experience and requires only one input file, whereas LymphGen entails 3 files, making the data preparation process more laborious.

Finally, we correlated all our molecular data to each other and to the available clinical information. The majority of our non-GCB cases were enriched for MCD. GCB cases were enriched for EZB, concordant with previous studies [13, 18]. In concordance with these studies, MCD-defined patients exhibited the highest rates of relapse and mortality, confirming MCD to be a biological risk feature for aggressive behavior [25]. More specifically, MYD88L265P mutation leads to increased nuclear factor κB (NF-κB) activity as the MYD88 protein serves as an adapter, connecting the interleukin-1 (IL-1) receptors and Toll-like receptors (TLRs) to activate NF-κB, which plays a role in immune responses to invading pathogens [26]. Additionally, pointing towards the central importance of unopposed NF-κB signaling in MCD DLBCL, MYD88L265P mutations co-occur with mutations in the B-cell receptor (BCR) subunit CD79B, activating mutations in CARD11 coiled-coil (CC) domain and loss of TNFAIP3, which all drive NF-κB [27]. Importantly, such mutations were observed in our MCD cases, and it is crucial for these patients to be recognized since they can benefit from treatment options with lenalidomide [28] or Btk inhibitors [29].

We recognize that a notable limitation of our study is the relatively small sample size, considering the prevalence of DLBCL. Nevertheless, our aim was to mirror a real-world diagnostic scenario; thus, we consciously refrained from gathering a large cohort, reflecting our routine approach of evaluating cases promptly rather than anticipating to accumulate them. Additionally, we opted not to carry out gene expression profiling (GEP) to define the COO of any of the cases, even the 5 discordant ones that had a less clearly defined COO, as we acknowledged that GEP is not usually performed as a routine diagnostic tool due to financial restraints.

Our real-life cohort demonstrates that by using high-level state-of-the-art molecular techniques, DLBCL cases can be classified and stratified according to current algorithms. The Two-step classifier shows superiority to LymphGen based on our available data. By applying these algorithms, we could confirm that patients can be stratified to identify more aggressive subtypes, such as MCD, and hence potentially implement targeted treatments. However, as both algorithms had a high rate of “unclassifiable” cases, the robustness of the algorithms needs to be improved to ensure their applicability in daily practice.

All patients have given general written informed consent for their archival tissue to be used for scientific research, and in accordance with the Swiss Federal Act on Research involving Human Beings, Article 38, small quantities of (such archived) bodily substances from generally consented patients are allowed to be used for anonymized research purposes. Thus, the study has been conducted in accordance with the Declaration of Helsinki and the current local laws. This study protocol was reviewed and approved by the Ethics Committee for Northwestern and Central Switzerland, Hebelstrasse 53, 4056 Basel, Switzerland, with approval number [EKNZ 2014-252].

The authors have no conflicts of interest to declare.

This work was supported by Krebsliga Schweiz [KFS-5228-02-2021].

V.-S.I. provided methodology and investigation and wrote the original draft. V.V. provided DNA extraction and initial genomic sequencing. A.T. and T.M. provided patient samples, clinical data, supervision, and conceptualization. M.D., F.S., and J.K. provided patients’ clinical data. S.D., A.T., and T.M. performed review and editing. S.D. and A.T. provided funding acquisition. All authors read and approved the final manuscript.

The sequencing datasets generated and analyzed during the current study have been deposited in the European Nucleotide Archive (ENA) at EMBL-EBI under accession number PRJEB65320. Further inquiries can be directed to the corresponding author.

1.
Campo
E
,
Swerdlow
SH
,
Harris
NL
,
Pileri
S
,
Stein
H
,
Jaffe
ES
.
The 2008 WHO classification of lymphoid neoplasms and beyond: evolving concepts and practical applications
.
Blood
.
2011
;
117
(
19
):
5019
32
.
2.
Smith
A
,
Crouch
S
,
Lax
S
,
Li
J
,
Painter
D
,
Howell
D
.
Lymphoma incidence, survival and prevalence 2004-2014: sub-type analyses from the UK’s Haematological Malignancy Research Network
.
Br J Cancer
.
2015
;
112
(
9
):
1575
84
.
3.
Teras
LR
,
DeSantis
CE
,
Cerhan
JR
,
Morton
LM
,
Jemal
A
,
Flowers
CR
.
2016 US lymphoid malignancy statistics by World Health Organization subtypes
.
CA Cancer J Clin
.
2016
;
66
(
6
):
443
59
.
4.
Alaggio
R
,
Amador
C
,
Anagnostopoulos
I
,
Attygalle
AD
,
Araujo
IBO
,
Berti
E
.
The 5th edition of the world Health organization classification of haematolymphoid tumours: lymphoid neoplasms
.
Leukemia
.
2022
;
36
(
7
):
1720
48
.
5.
Campo
E
,
Jaffe
ES
,
Cook
JR
,
Quintanilla-Martinez
L
,
Swerdlow
SH
,
Anderson
KC
.
The international Consensus classification of mature lymphoid neoplasms: a report from the clinical advisory committee
.
Blood
.
2022
;
140
(
11
):
1229
53
.
6.
Hwang
HS
,
Yoon
DH
,
Suh
C
,
Park
CS
,
Huh
J
.
Prognostic value of immunohistochemical algorithms in gastrointestinal diffuse large B-cell lymphoma
.
Blood Res
.
2013
;
48
(
4
):
266
73
.
7.
Hans
CP
,
Weisenburger
DD
,
Greiner
TC
,
Gascoyne
RD
,
Delabie
J
,
Ott
G
.
Confirmation of the molecular classification of diffuse large B-cell lymphoma by immunohistochemistry using a tissue microarray
.
Blood
.
2004
;
103
(
1
):
275
82
.
8.
Meyer
PN
,
Fu
K
,
Greiner
TC
,
Smith
LM
,
Delabie
J
,
Gascoyne
RD
.
Immunohistochemical methods for predicting cell of origin and survival in patients with diffuse large B-cell lymphoma treated with rituximab
.
J Clin Oncol
.
2011
;
29
(
2
):
200
7
.
9.
Lenz
G
,
Wright
G
,
Dave
SS
,
Xiao
W
,
Powell
J
,
Zhao
H
.
Stromal gene signatures in large-B-cell lymphomas
.
N Engl J Med
.
2008
;
359
(
22
):
2313
23
.
10.
Wilson
WH
,
Young
RM
,
Schmitz
R
,
Yang
Y
,
Pittaluga
S
,
Wright
G
.
Targeting B cell receptor signaling with ibrutinib in diffuse large B cell lymphoma
.
Nat Med
.
2015
;
21
(
8
):
922
6
.
11.
Goy
A
,
Ramchandren
R
,
Ghosh
N
,
Munoz
J
,
Morgan
DS
,
Dang
NH
.
Ibrutinib plus lenalidomide and rituximab has promising activity in relapsed/refractory non-germinal center B-cell-like DLBCL
.
Blood
.
2019
;
134
(
13
):
1024
36
.
12.
Alizadeh
AA
,
Eisen
MB
,
Davis
RE
,
Ma
C
,
Lossos
IS
,
Rosenwald
A
.
Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling
.
Nature
.
2000
;
403
(
6769
):
503
11
.
13.
Schmitz
R
,
Wright
GW
,
Huang
DW
,
Johnson
CA
,
Phelan
JD
,
Wang
JQ
.
Genetics and pathogenesis of diffuse large B-cell lymphoma
.
N Engl J Med
.
2018
;
378
(
15
):
1396
407
.
14.
Chapuy
B
,
Stewart
C
,
Dunford
AJ
,
Kim
J
,
Kamburov
A
,
Redd
RA
.
Molecular subtypes of diffuse large B cell lymphoma are associated with distinct pathogenic mechanisms and outcomes
.
Nat Med
.
2018
;
24
(
5
):
679
90
.
15.
Pedrosa
L
,
Fernandez-Miranda
I
,
Perez-Callejo
D
,
Quero
C
,
Rodriguez
M
,
Martin-Acosta
P
.
Proposal and validation of a method to classify genetic subtypes of diffuse large B cell lymphoma
.
Sci Rep
.
2021
;
11
(
1
):
1886
.
16.
Lacy
SE
,
Barrans
SL
,
Beer
PA
,
Painter
D
,
Smith
AG
,
Roman
E
.
Targeted sequencing in DLBCL, molecular subtypes, and outcomes: a Haematological Malignancy Research Network report
.
Blood
.
2020
;
135
(
20
):
1759
71
.
17.
Pillonel
V
,
Juskevicius
D
,
Ng
CKY
,
Bodmer
A
,
Zettl
A
,
Jucker
D
.
High-throughput sequencing of nodal marginal zone lymphomas identifies recurrent BRAF mutations
.
Leukemia
.
2018
;
32
(
11
):
2412
26
.
18.
Wright
GW
,
Huang
DW
,
Phelan
JD
,
Coulibaly
ZA
,
Roulland
S
,
Young
RM
.
A probabilistic classification tool for genetic subtypes of diffuse large B cell lymphoma with therapeutic implications
.
Cancer Cell
.
2020
;
37
(
4
):
551
68.e14
.
19.
Gu
Z
,
Eils
R
,
Schlesner
M
.
Complex heatmaps reveal patterns and correlations in multidimensional genomic data
.
Bioinformatics
.
2016
;
32
(
18
):
2847
9
.
20.
Crotty
S
.
Revealing T follicular helper cells with BCL6
.
Nat Rev Immunol
.
2021
;
21
(
10
):
616
7
.
21.
Tzankov
A
,
Xu-Monette
ZY
,
Gerhard
M
,
Visco
C
,
Dirnhofer
S
,
Gisin
N
.
Rearrangements of MYC gene facilitate risk stratification in diffuse large B-cell lymphoma patients treated with rituximab-CHOP
.
Mod Pathol
.
2014
;
27
(
7
):
958
71
.
22.
Visco
C
,
Tzankov
A
,
Xu-Monette
ZY
,
Miranda
RN
,
Tai
YC
,
Li
Y
.
Patients with diffuse large B-cell lymphoma of germinal center origin with BCL2 translocations have poor outcome, irrespective of MYC status: a report from an International DLBCL rituximab-CHOP Consortium Program Study
.
Haematologica
.
2013
;
98
(
2
):
255
63
.
23.
Wright
GW
Improvements in LymphGen 2.0 NIH
National Cancer Institute
2023
. Available from: https://llmpp.nih.gov/lymphgen/index.php#showver.
24.
Xu-Monette
ZY
,
Wu
L
,
Visco
C
,
Tai
YC
,
Tzankov
A
,
Liu
WM
.
Mutational profile and prognostic significance of TP53 in diffuse large B-cell lymphoma patients treated with R-CHOP: report from an International DLBCL Rituximab-CHOP Consortium Program Study
.
Blood
.
2012
;
120
(
19
):
3986
96
.
25.
Chen
R
,
Zhou
D
,
Wang
L
,
Zhu
L
,
Ye
X
.
MYD88(L265P) and CD79B double mutations type (MCD type) of diffuse large B-cell lymphoma: mechanism, clinical characteristics, and targeted therapy
.
Ther Adv Hematol
.
2022
;
13
:
20406207211072839
.
26.
Iwasaki
A
,
Medzhitov
R
.
Regulation of adaptive immunity by the innate immune system
.
Science
.
2010
;
327
(
5963
):
291
5
.
27.
Shaffer
AL
3rd
,
Young
RM
,
Staudt
LM
.
Pathogenesis of human B cell lymphomas
.
Annu Rev Immunol
.
2012
;
30
:
565
610
.
28.
Hernandez-Ilizaliturri
FJ
,
Deeb
G
,
Zinzani
PL
,
Pileri
SA
,
Malik
F
,
Macon
WR
.
Higher response to lenalidomide in relapsed/refractory diffuse large B-cell lymphoma in nongerminal center B-cell-like than in germinal center B-cell-like phenotype
.
Cancer
.
2011
;
117
(
22
):
5058
66
.
29.
Rui
L
,
Drennan
AC
,
Ceribelli
M
,
Zhu
F
,
Wright
GW
,
Huang
DW
.
Epigenetic gene regulation by Janus kinase 1 in diffuse large B-cell lymphoma
.
Proc Natl Acad Sci U S A
.
2016
113
46
E7260
7
.