DLBCL tumor specimens and patients’ clinical data were obtained at the Hematology Unit, “P. Giaccone” Hospital of Palermo. Elderly patients (≥ 65-year-old) have been selected for the study and further classified in two cohorts of responder (n=13) and non-responder (n=6) to first-line R-CHOP therapy (validation cohort). A panel of hematologist and pathologist at the “P. Giaccone” Hospital followed the ESMO Clinical Practice Guidelines and Italian Society of Hematology guidelines for diagnosis, treatment and follow-up of DLBCL patients.

The training cohort (GSE10846, n=414) has been divided in two groups: patients treated with CHOP (n=181) and patients treated with R-CHOP (n=233). Patients’ cohort has been filtered by age (≥ 65-year-old) (n=188) and subsequently divided in responder (n=94) and non-responder (n=94) (7).Differential expressed genes (DEGs) (fold change >1.4, p-value <0.05, κ=387) have been clustered in the responder and non-responder groups and according to LDH levels. The activated B-cell (ABC) and germinal center B-cell (GCB) molecular subtypes have been reported as annotations.

Finally, the top 30 up-regulated genes of responder and non-responder patients have been used to perform enrichment analysis in order to identify the main signatures involved in gene ontology (MSigDB), considering molecular function, biological process and cellular component (p-value<10^-7). The signatures associated with the first 30 upregulated genes were also computed by the QIAGEN Ingenuity Pathway Analysis software.

The association between features and patients’ overall survival was assessed by using Cox proportional- hazards model. Specifically, in the univariate analysis Cell of origin (COO), ECOG performance status, Extranodal Sites, IPI, LDH, sex and stage parameters were dichotomized according to (8, 9). The dichotomization of the identified signature was defined by using the median expression of each gene (MYO6, EBF1 and CALR). In the multivariate analysis, we combined our signature with each previously described feature.

To generate the Kaplan-Meier curves of overall survival by using the GSE10846 dataset, the initial population was filtered by age (≥ 65-year-old). “High” and “Low” groups were defined by using the median expression of each gene (MYO6, EBF1 and CALR) in the patient cohort.

All analyses were performed with R survival, survminer, and coxph libraries. Graphs were created by using the ggplot2 library.

Total RNA from FFPE tumor tissue specimens was isolated by using RNeasy FFPE Kit (Qiagen). 300 ng of total RNA was retro-transcribed with the high-capacity c-DNA reverse transcription kit (Applied Biosystem). In order to perform a four-gene multiplex assay, we used specific Droplet digital PCR (ddPCR- QX200 Droplet Reader) gene expression assays with FAM (n=2) and HEX (n=2) fluorophores. To optimize the multiplex reactions, from 100 to 300nM gene-specific primers have been used in combination with ddPCR supermix for probes (No-dUTP) and 25 ng of cDNA samples. Droplets were generated using the QX200 Droplet Generator (Bio-Rad) and dispensed into a 96 well-PCR plate. PCRs were performed in a ProFlex PCR System (Applied Biosystem) with the following protocol: 1x (95°C for 10 min), 50x (94°C for 30 sec, 56°C for 1 min), 1x (98°C for 10 min). After gene target amplification, samples were analyzed using QX200 Digital Droplet Reader (Bio-Rad). Gene expression analyses (copies/µl) were performed using QX Manager Software (1.2 Standard Edition) and normalized by using GAPDH.

FFPE lymphoma tissue specimens, stratified by age ≥ 65-year-old, were obtained from 11 responder patients and 4 non responder patients treated with R-CHOP.

Antigen retrieval was performed using the PT link system (Dako, Agilent Technologies, Santa Clara, CA, USA). Thereafter, sections were permeabilized with the 0.1% TRITON X-100 PBS for 10 min on ice, followed by 3% H₂0₂ and 10% human serum blocking incubation.

All slides were exposed overnight at 4°C to primary antibodies against Calreticulin (CAL-R) (ab22683; mouse IgG1; Abcam, Cambridge Science Park, UK), Myosin VI (MYO6) (MUD-19; mouse IgG1, Sigma-Aldrich), and EBF-1 (HPA061169; rabbit; Sigma-Aldrich). Staining was revealed using a biotin-streptavidin system (Dako LSAB2 System-HRP) and detected with the DAB substrate chromogen system (Dako). Nuclei were counterstained with Mayer’s Hematoxylin (Lillie’s Modification) Histological Staining Reagent (Dako). Staining was analyzed using an Olympus BX60 microscope. Immunohistochemical analysis were quantified with Image J.

Diffuse large B-cell lymphoma (DLBCL) is a heterogeneous disease, causing high mortality in elderly patients. Despite the adverse effects, CHOP- and R-CHOP-based therapies result effective in the two-thirds of DLBCL patients, of which the rest portion experiences disease recurrence. Being DLBCL elderly patients the more susceptible to standard therapy side effects, in order to identify the genes predictive of therapy response, a gene expression analysis of 188 pretreatment biopsies of patients with an age ≥ 65 was retrieved from a publicly available dataset (GSE10846) ( Figure 1A ) (7).

Following unsupervised hierarchical clustering, the dichotomization of training cohort patients in responder and non-responder to CHOP and R-CHOP therapy allowed the identification of differentially expressed genes (DEGs) associated to a poor outcome ( Figure 1B ). The median age of the responder and non-responder cohorts of patients, to standard therapy, was comparable (74.04 versus 75.12-year-old) thus allowing the exclusion of age-related deaths. Of note, responder patients were mainly characterized by the GCB-like (62,5%) molecular subtype associated with a favorable outcome ( Figure 1B ). In accordance with well-established negative prognostic LDH parameter, responder DLBCL patients harbored lower LDH levels (1,0887 versus 1,8112) with respect to non-responder patients ( Figure 1B ; Supplementary Figures S1A, B ). Analysis of gene expression profile, including 387 genes, of responder versus non-responder patients showed ten most differentially expressed genes (p-value ≤ 0.001) ( Figure 1C ; Supplementary Table 1 ). Specifically, ATAD3A, CALR, CWF19L1, GALT, MAGEA9, MAPKI8IP3, PSLNR, SEPTIN7P13, SLC19A1 and SLC38A5 resulted up-regulated in non-responder patients, while high expression levels of EBF1, EDNRA, CCDC18, CCDC186, FYB, MALAT1, MIS18BP1, MYO6, THRAP3 and TOP1, characterized responder DLBCL patients ( Figure 1C ).

The enrichment analysis of top 30 upregulated genes computed with Molecular Signatures Database (MSigDB), revealed six signatures associated with cell cycle, cell division and cytoskeleton organization, which are related to B cell malignant neoplasia ( Figures 1D, E ; Supplementary Table 2 ). Together these data provide evidence that these gene signatures may select responder from non-responder patients, identifying patients with a better life expectancy.

In order to validate the expression levels of previously identified genes in dictating the dichotomization in life expectancy, we analyzed a cohort of naïve DLBCL patients, diagnosed and in follow-up at the Hematology/Oncology Unit of the “P. Giaccone” Hospital in Palermo, treated with R-CHOP as first -line therapy (validation cohort), selecting the frail cohort of DLBCL patients (≥ 65-year-old) ( Table 1 ).

To overcome the low abundance and integrity of RNA content on formalin-fixed paraffin embedded (FFPE) biopsy samples, we adopted an implemented quantitative multiplex droplet digital PCR-based assay ( Figure 1C ; Figure 2A ), from which emerged that two out ten DEGs, previously identified, EBF1 and MYO6 resulted up-regulated, at both mRNA and protein levels, in responder DLBCL patients ( Figures 2B–D ; Supplementary Figures 2A, B ; Supplementary Table 3 ). Furthermore, the gene expression analysis of the ten upregulated genes, arisen from the non-responder included in the dataset (GSE10846), displayed an increasing trend of CALR mRNA levels, although not reaching statistical significance, which was paralleled by high protein expression levels ( Figures 2B–D ; Supplementary Table 4 ). These data indicate that EBF1, MYO6 and CALR could predict DLBCL patients’ response to R-CHOP therapy and aid in the stratification of responder versus non-responder patients.

To investigate the clinical significance of the identified signature, the magnitude of EBF1, MYO6, and CARL expression levels has been correlated to the survival data of ≥ 65-year-old DLBCL patients of the training cohort.

Transcriptome microarray analysis of a cohort of 154 DLBCL revealed a significant negative correlation between the signature EBF1^low, MYO6^low and CALR^high expression and survival probability of patients, assuming a more pronounced significance than single gene expression ( Figure 3A ). Univariate analysis denoted that EBF1^high, MYO6^high and CALR^low signature expression is an independent positive prognostic factor of overall survival showing a higher statistical significance over several important clinical parameters, such as ECOG performance status, extranodal sites and stage ( Supplementary Figure 3A ). Importantly, EBF1^high, MYO6^high and CALR^low signature expression significantly increased the prognostic value of ECOG performance status, extranodal sites, IPI, LDH and stage ( Supplementary Figure 3B ).

In line with the signature prognostic value, from the STRING network analysis emerged an implication of EBF1, MYO6, and CALR in the regulation of tumor progression together with the control of B lymphocyte gene transcription, intracellular vesicle transport and protein folding ( Figure 3B ; Supplementary Figure S3C ). Being restricted the signature expression to DLBCLs as compared to non-tumoral lymphoid cells, a specific targeting of EBF1, MYO6 and CALR could be exploited for therapeutic intervention ( Supplementary Figure S4A ).

To investigate whether the EBF1, MYO6 and CALR differential expression could be an age-independent signature to predict R-CHOP response, we assessed their expression levels in a heterogeneous age group, which revealed a comparable expression level in adults as well as elderly DLBCL patients ( Figure 4A ). Of note, beside in the over 65-year-old, EBF1^low, MYO6^low and CALR^high signature expression is able to dichotomize the response to R-CHOP treatment in under 65-year-old DLBCL patients ( Figure 4B ; Supplementary Figure S4B ). Although EBF1^high, MYO6^high and CALR^low signature expression was an independent positive prognostic factor in under 65-year-old DLBCL patients, it did not increase the prognostic value of the selected clinical features in a multivariate analysis ( Supplementary Figures 4C, D ).

Our study unveils a robust predictive and prognostic signature able to determine the response to R-CHOP treatment in both under and over 65-year-old DLBCL patients. Notably, EBF1^high, MYO6^high and CALR^low signature expression ameliorates the prognostic power of the most important clinical parameters and, in particular, of the IPI clinical risk scoring system in the elderly patients.