Samples were analyzed using antibodies to PD-L1 (28–2, CELL MARQUE, USA), HLA-DR (TAL.1B5, Dako, Denmark), PD-1 (NAT.105, CELL MARQUE, USA), CTLA-4 (CAL49, Abcam, UK), MUM1 (EAU32, Leica), and Ki-67 (K2, Leica). IHC procedures were performed on tumor biopsies: mediastinal masses (n = 190, 90%), lymph nodes (n = 12, 6%), and extranodal sites of involvement (n = 10, 5%). For each sample, 4 μm slices were stained using a standard immunohistochemistry protocol for formalin-fixed paraffin-embedded (FFPE) tissues, with a ready-to-use detection system that provides high signal amplification without biotin, utilizing the Leica Bond-MAX immunostainer. Surgipath Sub-X Leica medium was used as the final mounting medium. Antibody dilutions and the type of buffer for epitope retrieval (ER)—ER1 (pH = 6) or ER2 (pH = 9)—were optimized experimentally in advance. The expression of PD-L1, HLA-DR, MUM1, and Ki-67 was assessed in CD20+ tumor cells. The reaction was considered reliable in the presence of a positive control—small T-cells and macrophages. The threshold value was set at 50% positive large tumor cells, while PD-1 and CTLA-4 expression were evaluated in CD3+ T-cells within the tumor microenvironment.

Fluorescence in situ hybridization (FISH) was performed on 31 tumor biopsy samples to detect chromosomal aberrations involving MYC/8q24 (n = 31), BCL2/18q21 (n = 31), BCL6/3q27 (n = 31), and del17p13 (n = 16) loci. Standard protocols were used, and preparations were analyzed using an Axio Imager Z2 (Carl Zeiss, Germany) fluorescence microscope with result documentation performed using the ISIS imaging system (MetaSystems, Germany). At least 200 interphase nuclei with high-quality signals were assessed for each sample.

Conventional cytogenetic analysis (CCA) involved short-term culture of homogenized cell suspensions from 31 biopsy material samples. Karyotypes were analyzed using a Zeiss Axioscope microscope equipped with the IKAROS imaging system. A minimum of 20 metaphase spreads were analyzed per sample to detect chromosomal abnormalities.

The CytoScan™ HT-chromosome microarray analysis (CMA) 96F array SNP-oligonucleotide microarray was used for the analysis, which was performed by the Genomed Laboratory of Molecular Pathology in Moscow, Russia. The samples were DNA isolated from 15 mediastinal tumor biopsy samples, with a quantity ranging from 100 ng to 200 ng, and an A260/A280 ratio of at least 1.8. The results were processed using the Multi Sample Viewer Software (v.1.1.0.11) and Chromosome Analysis Suite (ChAS 4.3.0.71) (Thermo Fisher Scientific, USA). The cutoff for a CNA size was set at ≥ 5 Mb, following the guidelines of Schoumans et al. [24].

Mutations in the CD58 and B2M genes were evaluated in a subset of 48 patients. DNA was extracted from tumor biopsies by standard proteinase K-SDS digestion and phenol-chloroform extraction [25]. We analyzed functionally important regions of the CD58 and B2M genes, i.e. the promoter region, exons 1–2 of the B2M gene, and exons 1–3 of the CD58 gene and exon-intron junctions, by Sanger sequencing. For the amplification of target fragments, we used primers designed in the laboratory of genetic engineering at our center (Table 1).

Polymerase chain reaction (PCR) was carried out using PCR Master Mix (Thermo Fisher Scientific, USA) following the manufacturer’s protocol. PCR products were analyzed by electrophoresis in 1.5% agarose gel and then purified using Wizard PCR Preps DNA purification system (Promega, USA). To determine the nucleotide sequence of genes, Sanger sequencing was performed using the ABI PRISM^® BigDyeTM Terminator v.3.1 kit (Thermo Fisher Scientific, USA) on a Nanofor-05 automatic genetic analyzer (Syntol, Russian Federation) according to the manufacturer’s protocols. The obtained nucleotide sequences were compared with the corresponding reference sequences from the NCBI database (B2M: NM_004048.4; NG_012920.2; CD58: NM_001779.3) [26]. The results of Sanger sequencing were described in accordance with the recommendations of the Human Genome Variation Society (HGVS) [27]. The pathogenicity of variants was assessed using the following prediction tools [SIFT v.6.2.1 (J. Craig Venter Institute) [28], PROVEAN v.1.1.5 (J. Craig Venter Institute) [29], PolyPhen-2 v.2.2.2 (Harvard Medical School) [30], and MutationTaster (Charité – Universitätsmedizin Berlin) [31]] and ClinVar (NCBI, U.S. National Library of Medicine) [32].

Mutations in the TP53 gene were analyzed in a subset of 35 patients. DNA was extracted from tumor biopsies or sections of FFPE tissue blocks [33]. Exons 4 through 10 of the TP53 gene were amplified in five separate PCR reactions. Library preparation was performed using the Nextera XT DNA (Illumina, San Diego, CA, USA), followed by sequencing on a MiSeq genetic analyzer (Illumina, USA). Bioinformatics analysis was conducted using a pipeline that included tools such as Trimmomatic [34], BWA [35], SAMtools [36], VarDict [37], and Annovar [38]. The identified variants were further assessed for potential pathogenicity using the Franklin by Genoox platform [39] and the SESHAT [40] online databases.

The analysis of the E571 mutations (predominantly E571K or E571G) in the XPO1 gene was performed on tumor biopsy DNA samples from 36 patients using allele-specific PCR (AS-PCR) on the CFX96 Touch Real-Time PCR Detection System from Bio-Rad Laboratories, Inc. (USA). The primers and probes used in this study are listed in Table 2.

The PCR conditions for TaqMan real-time AS-PCR included preheating at 94°C for 300 s, followed by 45 cycles of thermal cycling. The denaturation step was at –94°C for 20 s, while the annealing and elongation steps were at –60°C for 50 s. Each primer was used at a concentration of 10 pmole per reaction, and each probe was used at 5 pmole per reaction. The reaction volume was 25 mL and the PCR buffer, magnesium chloride, dNTPs, and Taq polymerase were provided by Syntol LLC (Moscow, Russia). Primers and TaqMan probes were synthesized at Syntol according to the authors’ design.

STR profiles of the tumor cell DNA were analyzed on a cohort of 93 patients. A diagnostic system for the investigation of allelic imbalance (AI) in microsatellite repeats located near the PD-L1/PD-L2 genes (loci 9p24.1) and CIITA genes (loci 16p13.13) using the STR-PCR method has been developed. The methodology has been described in detail previously [41]. The primers for microsatellite repeats located near the HLA (loci 6p21.3) were adopted from the publication by Chambuso et al. [42] (2019).

Tumor DNA was isolated from the biopsy samples taken at diagnosis. Control DNA was isolated from the blood samples collected during CR or from bone marrow without tumor involvement. AI was assessed by comparing heterozygous markers from tumor DNA with those of matched control DNA. Patients exhibiting homozygous inheritance for any of the studied markers were excluded from further analysis due to the inability to evaluate AI.

AI of microsatellite repeats was examined in the regions of HLA [loci 6p21.3, (GT)n and (CA)m], PD-L1/PD-L2 [loci 9p24.1, (GT)n and (TTAT)m] and CIITA [loci 16p13.13, (CA)n and (GT)m] using STR-PCR (Table 3) with fragment analysis. Six separate PCR reactions with specific primers to amplify target loci markers on a DNAEngine thermal cycler (Bio-Rad, USA) were used. PCR products were then subjected to capillary electrophoresis using the Nanophor-05 genetic analyzer (Syntol LLC, Russia), and the data were analyzed using GeneMarker software, version 3.0.1 (SoftGenetics, USA).

Categorical variables were analyzed using Fisher’s exact test for small sample sizes and the χ² test when the minimum expected value for all categories exceeded 5. Continuous variables were assessed using nonparametric methods, including the Mann-Whitney U test for comparisons between two groups and the Kruskal-Wallis H test for comparisons among three groups. Survival analysis was conducted using the Kaplan-Meier method to estimate survival, with group comparisons performed using the log-rank test. Odds ratios (OR) with corresponding 95% confidence intervals (CI) were calculated to evaluate the association between binary categorical variables and outcomes. OR were computed using contingency tables. Fisher’s exact test or χ² test was used to assess the statistical significance of these associations, depending on sample size and cell frequencies. Overall survival (OS) was defined as the time from the date of diagnosis to death from any cause. Event-free survival (EFS) was calculated as the time from the initiation of chemotherapy to the earliest occurrence of relapse, disease progression, and switch to alternative anti-cancer therapy due to refractory disease or PR. All statistical analyses were performed using R version 4.1 (R Core Team, 2017). Statistical significance was defined as P < 0.05.

CCA and FISH analyses of MYC, BCL2, and BCL6 rearrangements were performed in 31 patients with PMBCL, while del17p13 was assessed in 16 patients. The frequency of detected markers is presented in Figure 3. Due to the low mitotic activity of tumor cells, standard cytogenetic analysis was challenging. Adequate mitoses were obtained in only 16 (52%) PMBCL samples, among which a complex karyotype was identified in 8 cases (50%). Translocations involving the BCL6 and MYC loci were each observed once, whereas BCL2 translocations and del17p13 were not detected. Copy gains (trisomy/duplication/amplification) of MYC and BCL6 were identified in 6 cases (19%), while BCL2 gains were observed in 5 cases (16%). The analyzed cytogenetic markers did not demonstrate any impact on the prognosis in PMBCL.

CMA was performed in 15 patients with PMBCL. Genomic aberrations were detected in all analyzed cases, with a median number of aberrations of 15 (range, 6–25). The most frequent abnormalities were observed on chromosomes 6 and 9, detected in 13 out of 15 patients (87%). The predominant type of alteration was copy number gain, with a total of 96 such events recorded, of which 46% (44/96) were amplifications. Amplifications were most found on chromosome 9, occurring in 73% (11/15) of patients (15 events). In total, 68 deletions and 47 cases of cnLOH were identified. These aberrations most frequently affected chromosome 6: deletions were detected in 73% (11/15), and cnLOH in 60% (9/15) of cases. Biallelic loss was observed in locus 17q24.1. The localization and extent of the detected genomic aberrations are visualized in Figure 4. Focusing on loci of interest (9p24.1, 16p13.13, and 6p21.3), amplification of 9p24.1 was the most frequent aberration, identified in 13 (87%) cases. Quantitative abnormalities of the 6p21.3 locus were observed in 7 (47%) cases, including cnLOH in 5 (33%) cases, deletion in 1 (7%) case, and gain in 1 (7%) case. For the 16p13.13 locus, quantitative abnormalities were found in 6 (40%) cases, including cnLOH in 3 (20%) cases and deletion in 3 (20%) cases.

We have assessed microsatellite repeats flanking key genes (PD-L1/PD-L2, CIITA, and HLA) involved in immune evasion in PMBCL. The STR profile analysis included PMBCL patients with heterozygous inheritance for at least one marker of the pair. AI near the PD-L1/PD-L2 (9p24.1) was detected in 24/73 (33%) and 29/73 (40%) patients, respectively. AI near the CIITA (16p13.13) was observed in 16/70 (23%) patients for the CA marker and in 15/70 (21%) for the GT marker. For the HLA (6p21.3), AI was identified in 16/29 (55%) patients for the CA marker and in 17/29 (59%) patients for the GT marker. The frequency of detected markers is presented in Figure 3.

The next step involved comparing the results obtained from STR-profile analysis with those from the CMA in 15 PMBCL patients. Aberrations larger than 5 Mb, along with microdeletions, were analyzed in accordance with the guidelines for genomic array analysis in acquired hematological neoplastic disorders [23]. In this exploratory study, we evaluated microdeletions, microduplications, and cnLOH sites to assess the inclusiveness of genes potentially involved in the pathogenesis of PMBCL. High-resolution CMA of tumor DNA revealed that amplification, pseudo-hyperdiploidy, and cnLOH manifest as AI. Thus, AI near the regions of interest may reflect underlying genomic instability. Allelic loss can result from either absolute loss of DNA content or copy-neutral loss of a parental allele. However, while STR analysis can identify the involvement of specific loci in pathogenesis, it cannot determine the precise chromosomal event (e.g., deletion or duplication) leading to AI.

Mutations in TP53 gene were detected in 4/35 (11%) patients. In 3 (9%) patients, the identified mutations were classified as pathogenic according to online databases. Notably, one tumor sample exhibited a rare combination of two pathogenic mutations in cis configuration, located in proximity, which may indicate a unique mechanistic feature. In the fourth patient, the mutation identified was classified as a variant of uncertain significance, and its potential impact on disease progression could not be determined. Consequently, this case was excluded from survival analyses. Mutations in the B2M gene were detected in 29 out of 48 patients (60%), representing the most frequently altered gene in this cohort. Mutations in the CD58 gene were observed in 18 out of 48 patients (38%). E571K mutation in XPO1 gene was identified in 7 out of 36 patients (19%). The frequency of detected markers is presented in Figure 3.

IHC analysis of key markers was performed to evaluate their expression in PMBCL tumor samples. The results are presented in Figure 3. Expression of PD-L1 was observed in 41% of samples (n = 46). The PD-1 receptor was detected in 85% of samples (n = 46). Expression of CTLA-4 was identified in 96% of tumor samples (n = 46). HLA-DR positivity was observed in 76% of samples. MUM1 was present in 88% of samples (n = 65). A high proliferation index (Ki-67 > 70%) was detected in 43% of samples (n = 212).

A comprehensive analysis of clinical, IHC, and molecular markers was conducted. The initial step involved frequency analysis to evaluate potential predictors (Figure 5). Given the low number of fatal outcomes, OS analysis did not yield conclusive predictors. The focus shifted to patients failing to achieve CR following induction therapy. Early events in this subgroup warranted a 12-month time point for analysis. A total of 223 patients were included, with censoring at 12 months. The cohort encompassed patients with progression, relapse, or PR requiring second-line therapy and auto-HSCT within 12 months. Patients under observation without reaching the 12-month mark were excluded.

Patients with extramediastinal involvement had higher risk of poor outcomes compared to those without involvement (OR = 2.26, 95% CI: 1.08–4.73, P = 0.025). The presence of bulky tumors (n = 212) was associated with a significantly higher risk of poor outcomes compared to the absence of such tumors (n = 11) (P = 0.018). Soft tissue or breast tissue involvement (n = 51) was associated with a trend toward poorer outcomes compared to absence (n = 172), approaching statistical significance (OR = 1.92, 95% CI: 1.01–3.68, P = 0.038). AI near the PD-L1 (9p24.1) was significantly associated with a higher risk of poor outcomes (OR = 3.39, 95% CI: 1.18–9.76, P = 0.020). LOH at any of the three loci—9p24.1 (PD-L1/PD-L2), 16p13.13 (CIITA), or 6p21.3 (HLA)—was significantly associated with poorer outcomes (OR = 4.56, 95% CI: 1.36–15.29, P = 0.009). Among all markers analyzed, AI at these loci demonstrated the highest risk for adverse outcomes following induction therapy.

Subsequent analysis assessed EFS among patients treated with R-DA-EPOCH or RmNHL-BFM-90 regimens, stratified by STR profiles at 9p24.1, 16p13.13, or 6p21.3 loci. In cases with AI at any locus, EFS decreased to 50% (95% CI: 39–65) compared to 81% (95% CI: 66–100) in patients with stable STR profiles (P = 0.028; Figure 6A). In a subgroup of 12 patients treated with nivolumab and R-DA-EPOCH, all exhibited AI in at least one locus, yet no adverse events were observed, yielding an EFS of 100%. This suggests that nivolumab may mitigate the negative impact of AI (P = 0.004; Figure 6B).

We analyzed the OS outcomes in a cohort of 33 patients with R/R PMBCL treated with different regimens. Specifically, we compared OS between patients receiving nivolumab in combination with chemotherapy (n = 8) and those treated with chemotherapy alone (n = 25). The addition of nivolumab demonstrated a clear trend toward improved OS at 36 months, with survival rates of 86% (95% CI: 63–100) in the nivolumab group compared to 44% (95% CI: 28–68) in the chemotherapy-only group (Figure 7). Despite this clinically meaningful difference, statistical significance was not achieved (P = 0.083), likely due to the small sample size. The choice of chemotherapy in R/R cases was determined by the patient’s performance status. For most patients, platinum-containing regimens were selected, including R-DHAP (n = 17, 52%) and rituximab, ifosfamide, carboplatin, etoposide (R-ICE) (n = 6, 18%). Additionally, seven (21%) patients received rituximab, dexamethasone, carmustine, etoposide, cytarabine, melphalan (R-DEXA-BEAM). Two (6%) patients were treated with R-CHALD (rituximab, chlorambucil, etoposide, methotrexate, dexamethasone) and one (3%) with R-GIDOX (rituximab, gemcitabine, ifosfamide, dexamethasone, oxaliplatin).

The present study has several limitations that must be acknowledged. The small sample size constrained the statistical power of some analyses, particularly in subgroup comparisons, calling for cautious interpretation of the results. This was especially relevant for the immunotherapy cohort, where the limited number of patients and observed events restricted the robustness and generalizability of the findings. Future research should focus on long-term follow-up to evaluate the durability of responses and survival outcomes in patients receiving ICIs. Additionally, exploring combination strategies, such as integrating immunotherapy with novel targeted agents, holds promise for improving outcomes in high-risk PMBCL subsets. To validate the current findings and further refine treatment approaches, larger, prospective studies are needed to provide more conclusive evidence and support personalized, risk-adapted strategies in PMBCL management.

The findings emphasize the importance of integrating ICIs, such as nivolumab, into first-line treatment for PMBCL patients, particularly those with high-risk clinical features. Furthermore, the strong prognostic role of AI at key loci (9p24.1, 16p13.13, 6p21.3) underscores the need for routine molecular profiling to guide risk-adapted treatment strategies. To validate these findings, larger, prospective studies with extended follow-up are required. Additionally, further exploration of ICIs in both first-line and R/R settings, combined with molecular characterization, will refine personalized treatment approaches and improve outcomes for patients with PMBCL.