For this proof-of-concept study, a virtual MTB approach was conducted in a retrospectively assembled cohort of prPBL aiming to address an urgent unmet medical need in an extremely rare and aggressive subtype of B-cell non-Hodgkin lymphoma of post-germinal origin. As previously reported, screening of our institutional database revealed 76 PBL cases whose biopsy specimens were transferred to the reference center for Haematopathology, University Hospital Schleswig Holstein Campus Lübeck and Haematopathology Lübeck for centralized histopathologic expert review between January 1998 and December 2020. Due to the rarity of this entity, the sample size was not statistically predetermined. The number of cases included in the study corresponds to all PBL cases that have been referred to the reference center of Hematopathology within two decades. Investigations were not randomized, and investigators were not blinded. After excluding PBL cases with insufficient or unrepresentative tissue samples, the molecular landscape of PBL was characterized based on WES and WTS from 33 and 20 PBL cases, respectively (20). From this cohort, 14 PBL cases presented with the primary refractory disease were selected, deducing potential advantages from applying a certified MTB-pipeline approach in the era of precision oncology. No data were excluded from analyses.

Sample preparation, whole exome, and RNA-sequencing from formalin-fixed, paraffin-embedded (FFPE) tissue sections, as well as the process of raw data preparation, filtering, and the detection of single nucleotide variants (SNVs), short insertions and deletions (indels), somatic copy number aberrations (SCNAs) and fusion genes, were performed as previously described by Witte et al. (20) and Künstner et al. (22). MSI sensor was applied for the detection of microsatellite instability (MSI). For gene expression analysis from RNA-seq data, STAR ALIGNER (version 2.7.2b) and MIXNORM were used. The hg19 genome served as the reference genome. Several steps of bioinformatic analysis are integrated into the institutional MTB pipeline at University Cancer Center Schleswig-Holstein (UCCSH), which is certified for routine clinical diagnostics ( Figure 1 ).

Standardized sheets (genomic reports) that were constructed for MTB database research upon Medical Informatics for Research and Care in University Medicine (MIRACUM) pipeline analysis list the tumor mutational burden (TMB) and the percentual content of tumor cells (23). Apart from microsatellite (MSI) status, we calculated the BRCAness score (cut-off ≥ 20%), and variant allele frequency (VAF) for each mutation and provided information regarding tumor heterogeneity as we reported on tumor subclones. Genomic reports were provided by the Medical Systems Biology Group (University of Lübeck). An individual genomic report was prepared for each case.

ANNOVAR was used for the annotation of filtered variants. Coverage for reference and alternative alleles for each variant was extracted using VCF- QUERY (VCFTOOLS version 0.1.13). Somatic variants were filtered as follows: at least 8 reads per base, minimum VAF of 5%, and only variants with a frequency below 0.001 in 1000 genomes, gnomAD, or ExAC, were considered for subsequent downstream analysis. Serving as a component for treatment prioritization, the effect of strong deleterious effects (CADD phred score > 20) was assessed per sample, and a gene set variation analysis was performed for WES and RNA-seq data against HALLMARK gene sets. More details on bioinformatics are provided in the Supplementary Material .

In the first step, the recurrence of a genomic alteration was checked to employ the databases COSMIC, OncoKB (prognostic & diagnostic levels), ClinVar (clinically relevant variation), and cBioPortal. Afterward the functional relevance of an alteration was verified with ProteinPaint. Once an alteration was found to be recurrent as well as functionally relevant, its therapeutic vulnerability was annotated with CIVIC, OncoKB (therapeutic & FDA levels), Cancer-Genome-Interpreter (CGI), and the Drug-Gene-Interaction database (DGIdb) in a third step of database research ( Figure 1 ). Apart from genomic alterations, we included immunohistochemical findings identified upon histopathologic diagnostics (CD19, CD30, CD38, and CD79B) to potentially serve as a relevant therapeutic target ( Supplemental Table 1 ). A rationale for an immunotherapeutic strategy was recommended if a high TMB status was detected (≥10 mut/Mb) or in samples with MSI high status.

Finally, the annotation ended up concluding research of ongoing studies on https://clinicaltrials.gov and preclinical data on https://pubmed.ncbi.nlm.nih.gov. Only resources open for academic research purposes were considered for MTB recommendations. Evidence levels for MTB recommendations were defined in accordance with the NCT/DKTK MASTER program ( Figure 1 ) and with the European Society for Medical Oncology Scale for Actionability of Molecular Targets (ESCAT) (4, 24). Additionally, genetic alterations were classified by biomarker/treatment baskets proposed by Horak et al. ( Figure 1 ) (4).

The relevance of genomic alterations was considered based on VAF and CADD score. Variants of unknown significance were excluded. Each recommended drug was either approved by the Food and Drug Administration (FDA) and/or European Medicines Agency (EMA) or at least designated for FDA fast-track development as a minimum requirement for consideration within the scope of this study (cutoff date 31^st October 2022).

Treatment was recommended if at least one agent targeted at least one genomic alteration, one protein with relevant expression levels or if the TMB status was high. As data on the efficacy of agents targeting a non-mutated pathway component up- or downstream in an altered pathway detected upon gene set enrichment analysis remains insufficient, such agents were not considered for MTB recommendations. We excluded immunotherapeutic rationales based on the mutational signature SBS26, which was found in 6 cases as this rationale represents a biological rationale, so far lacking clinical validation (25).

Considering the matching score (MS) calculation by Sicklick et al., we adapted the calculation to our retrospective virtual MTB setting and provided a modified matching score (mMS) for each PBL case (26). The calculation was performed by dividing the number of alterations serving as a potential target for recommended drugs by the total number of characteristic alterations after excluding variants of unknown significance and a VAF lower than 5%. Apart from genomic alterations, immunohistochemical targets and immunotherapeutic rationales (TMB > 10mut/Mb, MSI status), as well as positive BRCAness scoring pleading for PARP-inhibition, were considered equally for mMS calculation. In the present calculation, a synergistic and well-established drug combination targeting the same aberration (such as dabrafenib plus trametinib for BRAF mutations) the impact was counted as one. Results ranged from 0% to 100%. Higher scores represented better matches.

x = number of targetable vulnerabilities

(genomic alterations + IHC + MSI - high status + TMB - high status + BRCAness score)

y = number of characteristic and significant alterations

(targetable + un - targetable with a VAF ≥ 5%)

Combinations of drugs were considered and recommended according to the I-PREDICT study (9). Congruent to the approach of Sicklick et al. (9), the participating pharmacist screened each potential combination for feasibility in the light of drug interactions.

Each mutation was proved for its biological relevance and conclusiveness based on VAF, gene set enrichment analysis, and RNA-sequencing data, if available.

In summary, the prioritization of MTB recommendations was a multifactorial process simultaneously considering several considerations ( Figure 1 , Supplementary Table 2 ).

The detailed MTB workflow is visualized in Figure 1 . In total, we performed three rounds of virtual multicentre MTB conferences in accordance with the institutional standards of UCCSH, retrospectively discussing the 14 prPBL cases (1^st round: 4 cases; 2^nd round: 5 cases; 3^rd round: 5 cases). A board-certified hematologist presented the case. The conference included at least a molecular oncologist, a bioinformatician, a pathologist, a pharmacist, a radiologist, and a medical geneticist. Centralized documentation of MTB recommendations was conducted in each case ( Supplementary Material ).

Data was taken from accession number EGAD00001006795 (European genome-phenome archive (EGA)).

This retrospective study was approved by the ethics committee of the University of Lübeck (reference no. 18-311), conducted in accordance with the Declaration of Helsinki, and patients have provided written informed consent regarding routine diagnostic and academic assessment, including genomic studies of their biopsy specimen in addition to the transfer of their clinical data.

Here, we report on the potential of molecularly stratified treatment options in 14 cases presenting with prPBL (median age 76.5 years, range 56 - 91). Additional PBL cases responding to initial cytoreductive treatment served as a comparison cohort (n = 19) (20). The majority of patients were male (11/14; 79%) and presented with advanced-stage disease (10/14; 71%) as well as an adverse prognostic constellation (11/14; 79% had an NCCN-IPI scoring ≥4). All patients had elevated lactate dehydrogenase (LDH) serum levels and the frequency of reduced performance status, according to the Eastern Cooperative Oncology Group (ECOG), demonstrated the frailty of patients included in the present cohort. Half of the current cohort was HIV positive (7/14 cases) and had underlying EBV (7/14 cases) infections at initial diagnosis. In 6 cases, we detected both HIV and EBV infections (43%). Cytoreductive treatment was applied in 13/14 cases (92.8%). More than half of the patients (8/14; 57%) received a CHOP-based (cyclophosphamide, hydroxydaunorubicin, vincristine, and prednisolone) treatment in 1^st line setting. In relapsed or refractory settings, 10/14 (71.4%) patients were eligible for 2^nd line cytoreductive treatment. Across any line of treatment, only five PBL patients responded to therapy (partial remission; PR), and another five PBL patients had stable disease (SD) as the best response. Three PBLs were completely refractory (progressive disease; PD) towards any treatment approach, and one PBL died shortly after initial clinical presentation. This underlines the urgent unmet clinical need for individualized treatment options among rare entities in hematology, such as PBL. Tumor cell content, immunohistochemical findings, and the contribution of MYC alterations were comparable between primary refractory cases and the comparison cohort. All baseline clinicopathological characteristics are summarized in Table 1 .

The course of the disease, information on clinical characteristics, and treatment sequences for each prPBL case is visualized in Figure 2A . Survival analysis revealed significantly inferior PFS (p< 0.0001) and OS (p = 0.002) in prPBL compared to those cases responding to first-line treatment ( Figure 2B ).

Since the genomic and transcriptomic landscape of PBL recently has been described comprehensively (20, 27, 28), we carved out genomic features of primary refractory cases to detect potential therapeutic vulnerabilities in this difficult-to-treat hematologic malignancy and verified their biological significance based on transcriptomic data.

Our WES analysis revealed 3,955 SNVs and indels involving 2,700 genes after variant filtering. Investigations regarding the functionality of such mutations revealed that 17.1% were functionally relevant, whereas 26.5% were functionally neutral, and 56.4% were found to be functionally inconclusive. Across the detected mutations, the most frequent alterations were missense mutations (74.2%), followed by frameshift mutations (16.1%; indels) and nonsense mutations (9.7%). Further investigations revealed loss of function (LOF) in 48.4% and gain of function (GOF) in 51.6% among the spectrum of mutations ( Figure 2C ). The median TMB was slightly higher in primary refractory cases (4.06 mut/Mb in primary refractory PBL vs. 3.09 mut/Mb in the comparison cohort; Figure 2D ; Supplemental Table 3 ). Given the limited sample size, this result was statistically insignificant (p = 0.337). TMB values in PBL displayed an overall low to intermediate TMB (29, 30). As previously reported, no evidence for MSI-based hypermutations in PBL, and matched germline DNA was not available for comparative analysis (20). After the identification of mutational subtypes and their functionality, the MIRACUM pipeline analysis identified 26 relevant genes carrying driver mutations ( Figure 3A ).

In total, MIRACUM pipeline analysis identified 47 variants potentially driving cancer growth across 26 genes in the cohort of 14 prPBL. Among these variants, individualized database research assigned potential treatment recommendations for 15 variants involving 9 genes. For targets amenable to multiple agents, we favored agents with the highest level of evidence according to the NCT/DKTK classification (reference). The heterogeneous spectrum of treatment recommendations comprised well-known agents such as idelalisib (targeting PI3KCD) ( 31) or everolimus (targeting mTOR) ( 32) and novel agents such as eprenetapopt (targeting TP53) ( 33), napabucasin (targeting STAT3) ( 34) or bemarituzumab (targeting FGFR2) ( 35). Moreover, recurrent LOF variants in TP53 were previously related to treatment recommendations with CDK4/6 inhibitors (e.g., palbociclib/abemaciclib) (4). However, novel data suggest the inefficacy of CDK4/6 inhibitors in TP53-mutated malignancies as such mutations promote resistance to this class of drugs prompting us to exclude this recommendation (36). Results from the conducted MIRACUM pipeline analysis are outlined in Figure 3A .

Second, treatment recommendations based on immunohistochemical investigations were found in each case. Among the four immunohistochemical targets, a rationale for anti-CD38 (e.g., daratumumab) (37), anti-CD19 (e.g., tafasitamab) (38), anti-CD30 (brentuximab vedotin) (39) and anti-CD79B (e.g., polatuzumab vedotin) (40) was found in 14/14 (100%), 6/14 (43%), 2/14 (14%) and 2/14 (14%) cases, respectively ( Figure 3A ).

Third, exhaustive database research and annotation for recurrent genomic alterations beyond the MIRACUM pipeline analysis revealed four additional targetable mutations potentially acting as relevant drivers (ERBB2, KIT, IDH2, and TET2) ( Figure 3B ). In the era of precision oncology, alterations in such genes represent trailblazers for molecularly stratified treatment strategies. Further characteristics of annotated mutations are summarized in Supplemental Table 4 . Particular attention was paid to the functional alignment concerning the biological significance and conclusiveness of each annotated mutation.

For potential treatment recommendations emerging from the annotation process, FDA approval was available for 14 agents, EMA was approved for 12 agents, and three agents were designated for FDA fast-track development ( Figure 3C ).

There exists a rationale for immune checkpoint blockade in tumors with high TMB status and DNA MMR deficiency (41–44). We used a cut-off for TMB-high status of ≥10 mut/Mb (43). Contrary to expectations, all PBL cases presented with a TMB<10 mut/Mb. In six cases, we found a predominant DNA MMR deficiency signature (single base substitution signature SBS26) (25, 45). For this constellation, the rationale for immunotherapy based on a mutational signature associated with DNA MMR deficiency corresponds to a molecular evidence level no higher than m3/m4 (in vitro data/biologic rationale). Consequently, there was no TMB- or DNA-MMR-deficiency-related recommendation for immunotherapy in the present prPBL cohort. However, immune checkpoint blockade was recommended in two cases (X%) harboring ROS1 alterations associated with resistance towards crizotinib and other targeted agents by propagating an immune escape mechanism (46). Moreover, we applied the calculation of the UCCSH MTB-pipeline BRCAness score (based on SBS6 signature) to predict the responsiveness towards PARP inhibitors (47). The BRCAness score incorporates mutations coming along with homologous recombination deficiency (HRD), such as BRCA1 or BRCA2 losses or alterations mimicking these losses (ATM, CHEK2, RAD51) (48). In the present cohort, BRCAness scoring revealed no evidence for PARP inhibition.

This proof-of-concept approach evaluated treatment options associated with NCT-DKTK molecular levels of evidence of at least m2B or higher for each prPBL case beyond anti-CD38 antibodies (daratumumab: 14x m1C rationale, ESCAT tier IIA; 32%). In the light of therapeutically addressable genomic alterations, 19 potential therapeutic vulnerabilities were assigned to m2A (n = 10; 23%) or m2B (n = 9; 20%) rationales. As already stated, immunohistochemical targets displayed a promising therapeutic option in primary refractory PBL. Assignment to molecular evidence levels revealed two m1C (brentuximab vedotin; 5%) and nine m2A (tafasitamab and polatuzumab vedotin; 20%) rationales.

Considering ESMO Scale for Clinical Actionability of molecular Targets (ESCAT) (reference), it has to be mentioned that recommendations based on higher ESCAT levels can hardly be reached in rare entities, especially among the spectrum of hematologic malignancies, due to the lack of prospective (randomized) clinical trials and their underrepresentation in MTB settings. Consequently, the experience in the molecularly stratified treatment of rare hematologic malignancies lags far behind recent developments in the field of solid tumors. However, the assignment of ESCAT evidence levels to therapeutic vulnerabilities found in this difficult-to-treat entity allocated 8 recommendations to ESCAT tier IC (18%), 14 recommendations to ESCAT tier IIA (32%), two recommendations to ESCAT tier IIB (5%) and 20 recommendations were allocated to ESCAT tier IIIA (45%). More preclinical options (NCT DKTK evidence level m3/m4 or ESCAT tier IV/V) were not considered in the present cohort, characterized by an urgent need for treatment recommendations in primary refractory setting due to the aggressive biologic features in PBL.

The calculation of the mMS ranged from 21% to 100% (median: 50%). In eight cases, a mMS of ≥ 50% was calculated (57%). The prognostic impact of the mMS remains speculative at this point, as no recommended treatment was administered in this virtual setting. However, the higher the mMS, the more alterations and pathways relevant to each case can be addressed upon recommended treatment strategies. Therefore, it can be expected that the mMS will be associated with overall response rates (ORR) and patient outcomes referring to the MS calculation reported in the I-PREDICT study (9).

The assignment to treatment baskets was made under the consideration of the drug’s mechanism of action rather than its functionality regarding the targeted alteration. According to baskets, we identified tyrosine kinases (TK; n=5), PI3K-MTOR-AKT pathway (PAM; n=3), cell cycle alterations (CC; n=2), RAF-MEK-ERK cascade (RME; n=2), immune evasion (IE; n=2) and others (OTH; n=4) for targeted treatment recommendations. An additional treatment basket based on B-cell-specific immunohistochemical markers has been added: B-cell targets (BCT; n=25) representing the elementary therapeutic basket in the present study ( Figure 3D ). Our analysis revealed 3.0 MTB recommendations in the median across the entire cohort of prPBL (range 2 – 5) ( Figure 4A ). All potential MTB treatment recommendations are summarized in Table 2 .

RNA-sequencing data was available in 10 cases (71.4%). A gene set enrichment analysis and an integrated analysis of WES and RNA-seq data were performed for these cases. RNA expression patterns were used to underline or refute the biological significance of an annotated and potentially addressable mutation ( Figure 4B ). Based on this integrated analysis, one potential candidate driver mutation (KIT) was excluded as we evaluated inconclusive results for mRNA expression. KIT represents an oncogene, and consecutively we expected an overexpression of KIT mutation-related transcripts. However, the present analysis revealed an inconclusive under-expression. Others (n = 18) were found to be conclusive. We assumed driver mutations to be recurrent based on database research, functionally relevant based on RNA expression patterns, and represent an essential mutation within an altered pathway enriched in a sample (49). Following these criteria, we identified seven conclusive driver mutations (one in a tumor suppressor gene and six in oncogenes) and nine conclusive mutations in oncogenes, as well as one conclusive mutation in a tumor suppressor gene ( Figure 4B ). However, the significance of specific variants (driver or not) remained unresolved due to the lack of RNA-seq data in four cases. In such cases, a variant was categorized as a provisional mutation but not as a driver mutation.

Apart from this, RNA-seq did not reveal a distinct transcriptomic signature of prPBL compared to the comparison cohort ( Supplementary Figure 1 ). However, we found exclusive expression patterns of TAF9, CCDC125, ALMS1and ZNF462 in prPBL but not in the comparison cohort. RNA expression of such genes did not contribute novel insights into the pathogenesis of these difficult-to-treat cases.

Congruent to previous results published by our work group, the current re-evaluation of RNA-seq data did not reveal any novel, recurrent, and therapeutically relevant genomic fusion beyond those affecting MYC ( Supplemental Table 5 ).

MTB recommendations for drug combinations considered insights from previous studies such as the I-PREDICT (NCT02534675) or the TOP-ART trials (NCT03127215) and other studies investigating the efficacy of drug combinations within the context of targeted therapies (9). Previous studies highlighted the advantages of drug combinations in molecularly stratified treatment settings (9). Preferably, drug combinations were chosen based on available datasets demonstrating their feasibility and efficacy in distinct entities or basket trials (11/14 cases; 79%). Novel drug combinations were selected considering the potential of overlapping drug toxicities, the molecular evidence levels for involved single agents, and the availability of such agents. Due to a distinct toxicity profile, immune checkpoint inhibition was a promising component for several drug combinations (4/14 cases; 29%). Moreover, several therapeutic options identified by immunohistochemical assessment harbored the potential for various additional drug combinations in all cases ( Supplemental Table 6 ). There is growing evidence for using immunotherapeutic agents and/or agents targeting immunohistochemical assessable structures in the context of cytoreductive drug combinations in MTB settings as those combinations represent the standard of care among a variety of both solid and hematologic malignancies (40, 50, 51). However, there is still significant room for improvement in determining the combination of targeted therapeutics in the era of precision oncology.

The median turnaround time from DNA/RNA isolation to virtual MTB recommendations was 28 days. In this virtual MTB approach, solely recommendations on treatment but not additional diagnostics were enunciated. After the exclusion of one inconclusive mutation upon integrated WES and RNA-seq analysis, the standardized institutional UCCSH MTB pipeline application revealed 43 treatment recommendations across the 14 cases of prPBL. Because PBL represents an extremely rare and aggressive hematologic malignancy, most recommendations were based on evidence deduced from evidence generated in different entities (m2A-B; 28/43 recommendations; 65%). Subsequently, class m1 evidence (NCT-DKTK) was gained from case reports (m1C; 16/43 recommendations; 37%) as prospective clinical trials are in short supply (37, 39, 52).

As outlined in the methods section, treatment prioritization reflected a multifactorial process that incorporated patient-related clinical features (such as ECOG-PS), drug availability, molecular evidence levels, the calculation of the mMS, the modality of a therapeutic target, the biologic significance of an alteration (based on RNA-seq data, VAF and gene set enrichment analysis) and the feasibility of drug combinations. Most recommendations were based on single agents (38/43; 88%). The spectrum of MTB recommendations expanded when potential drug combinations were considered (28 additional recommendations, Supplementary Table 6 ), and/or alternative agents to preferred recommendations were considered as well (10 additional recommendations; Supplementary Table 7 ). The decision towards the preference for a specific agent over another addressing the same target was made based on available molecular evidence levels (e.g., daratumumab = m1C versus isatuximab = m2A).

In summary, whole exome and partially whole transcriptomic sequencing data of 14 primary refractory PBL cases were processed through the UCCSH MTB pipeline. They revealed a total of 43 treatment recommendations in this aggressive and chemo-refractory hematologic B-cell malignancy. Treatment recommendations comprised molecular evidence levels from m2B to m1C rationales. The heterogeneous distribution of treatment basket allocations underlines the diversity of potential treatment strategies in a virtual second-line setting ( Figure 4C ), and demonstrates the relevance of molecular diagnostics in rare and aggressive B-cell malignancies such as PBL. The interdependence between case-related targets, treatment recommendations, and evidence levels is visualized in Figure 4D .