Integrating genomic features for prognosis in Chinese patients with diffuse large B-cell lymphoma following R-CHOP therapy

Abstract

Diffuse large B-cell lymphoma (DLBCL) exhibits 30–40% rates of relapse or refractory disease after R-CHOP. Identifying patients at risk for treatment failure is essential for planning subsequent therapeutic strategies, including stem cell mobilization or cellular therapy. To identify genomic predictors in Chinese patients, we performed targeted sequencing of tumor and matched oral DNA from 147 cases. Mutations in TP53 (hazard ratio (HR) = 2.78; P = 0.002), SRP72 (HR = 2.79; P = 0.010), MYC (HR = 1.97; P = 0.033), BCL2 (HR = 2.54; P = 0.021), and ASXL2 (HR = 2.20; P = 0.048) were significantly associated with poor PFS. A five-gene mutation risk score combined with baseline lactate dehydrogenase (LDH) and bone marrow tumor cell status improved PFS prediction over the International Prognostic Index (IPI). In addition, TP53 (HR = 4.45; P = 0.003), SRP72 (HR = 4.16; P = 0.014), and MYC (HR = 2.64; P = 0.047) mutations were strongly associated with increased risk of relapse. Integrating the three-gene mutation risk score with baseline LDH showed superior ability to predict relapse compared with the IPI. Combining mutation-derived scores with clinical factors enhanced prognostic discrimination, aiding early identification of patients for stem cell or cellular therapy.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00277-026-07085-y.

Introduction

Diffuse large B-cell lymphoma (DLBCL) is the most common non-Hodgkin lymphoma subtype in adults. The R-CHOP regimen achieves long-term remission in 60–70% of patients [1]. However, approximately 30–40% of patients exhibit primary resistance or relapse [2], and the underlying molecular mechanisms remain poorly understood. Accurate risk stratification at diagnosis is beneficial for targeted treatment. Recent next-generation sequencing (NGS) studies have revealed high-frequency gene mutations in DLBCL, including tumor protein p53 (TP53), B-cell leukemia/lymphoma 2 (BCL2), and myeloid differentiation primary response 88 (MYD88) [3, 4]. Nevertheless, systematic analyses focusing on Chinese populations are still lacking, especially those integrating tissue-based targeted sequencing with long-term follow-up data.

Traditional prognostic tools, such as the International Prognostic Index (IPI) and molecular classification (germinal center B-cell-like (GCB) and activated B-cell-like (ABC) subtypes), have limitations in guiding personalized therapy [5, 6]. The IPI is based on five clinical factors [7] but cannot capture genetic mutations. The GCB and ABC classification established by gene expression profiling (GEP) [6] is the gold standard but is limited by high cost and technical complexity. Immunohistochemistry (IHC) is a more accessible surrogate [8, 9] but shows only partial concordance with GEP [4, 10].

Prior studies predominantly relied on Western cohorts or small-sample sequencing [3], whereas Chinese DLBCL patients may have distinct mutational features. Moreover, many studies lacked match normal tissues, potentially confounding results with germline variants. In this study, we performed exome sequencing of 521 clinical hotspot genes in tumor tissues and paired normal controls, combined with longitudinal follow-up of overall survival (OS) and progression-free survival (PFS). Unlike ctDNA or whole-genome studies [11], our tumor-normal paired sequencing enhanced low-frequency mutation detection and eliminated clonal hematopoiesis interference. Comprehensive follow-up data provided robust genotype-outcome associations. Identifying high-risk patients at diagnosis may inform early stem cell mobilization or cellular therapy.

Materials and methods

Patient selection and sample collection

Inclusion criteria were as follows: (1) histologically confirmed DLBCL, NOS (confirmed by independent pathology review); (2) received front-line R-CHOP-21 regimen (at least 4 cycles, up to 6–8 cycles); (3) no prior history of malignancy or prior chemotherapy/radiotherapy; (4) available baseline formalin-fixed paraffin-embedded (FFPE) tumor tissue for DNA extraction; and (5) complete clinical and follow-up data.

This retrospective study included consecutive Chinese DLBCL patients treated at our center (May 2016–July 2024). A total of 147 patients were enrolled. All cases were confirmed as DLBCL, NOS by two expert hematopathologists; high-grade BCL, Burkitt lymphoma, and transformed FL were excluded. Tumor tissue samples and matched buccal cells were collected from all patients prior to R-CHOP therapy for DNA extraction and next-generation sequencing (NGS). A panel-based NGS approach targeting 521 clinically relevant hotspot genes was used to characterize the DLBCL mutation landscape (Table S1).

Treatment and response assessment

All patients received standard R-CHOP-21 regimen. PET/CT was performed to evaluate treatment response according to the Lugano criteria. Patients were classified into the complete response (CR) group (n = 102) and the non-CR group (n = 45) based on their best response to R-CHOP therapy.

DNA extraction and sequencing

Briefly, DNA was extracted from FFPE tumor sections and buccal cells using QIAamp kits (Qiagen). Libraries were prepared using hybridization capture targeting 521 genes and sequenced on Illumina platforms (PE150). See Supplemental Methods for detailed procedures.

Data processing and analysis

Raw reads were aligned to hg38 using BWA [12]. SAMtools [13] was used for variant calling. The filter 2 parameters of SNP and InDel are as follows: QUAL ≥ 20, DV ≥ 4, and MQ ≥ 30. CoNIFER [14] was used for CNV detection in this study. ANNOVAR [15] was used for annotation. Detailed parameters are in Supplementary Methods.

Statistical analysis

Primary endpoints were PFS and OS. PFS was defined as time from treatment initiation to progression or death; OS as time to death. Log-rank test compared survival curves. Univariate and multivariate Cox regression identified prognostic factors. Model performance was assessed using C-index and time-dependent AUC (12,24,36 months) with DeLong’s test. All analyses used R 4.5.2; P < 0.05 was significant. See Supplemental Methods for detailed procedures.

Results

Landscape of DNA mutations in 147 samples

We sequenced 147 tumor samples and matched buccal cells from newly diagnosed DLBCL patients. Patient characteristics are in Table 1. All 147 patients received the standard R-CHOP-21 regimen with a fixed 21-day cycle interval. The median number of cycles administered was 6 (range: 3–8). More than 90% of patients received ≥ 85% of the planned dose for all chemotherapy agents. Dose delays (> 7 days) occurred in 14 patients (9.5%), primarily due to grade 3–4 neutropenia or infection. No patient received a dose-dense schedule (e.g., R-CHOP-14). The cohort included 102 CR and 45 non-CR patients.

Table 1.

Patient information^a)

Variables	Total (n = 147)	CR (n = 102)	Non–CR (n = 45)
Sex, n (%)
Female	68 (46.3)	49 (58.0)	19 (42.2)
Male	79 (53.7)	53 (52.0)	26 (57.8)
Age, median (Q1, Q3)	63 (52.3, 69)	62.5 (52, 69.25)	64 (53.5,70)
ECOG, n (%)
≤ 2	130 (88.4)	92 (90.2)	38 (84.4)
>3	17 (11.6)	10 (9.8)	7 (15.6)
Ki67, median (Q1, Q3)	80 (70, 90)	80 (70, 90)	80 (70, 90)
MYC fusion gene, n (%)	12 (8.2)	9 (8.8)	3 (6.7)
BCL2 fusion gene, n (%)	16 (10.9)	10 (9.8)	6 (13.3)
BCL6 fusion gene, n (%)	37 (25.2)	29 (28.4)	8 (17.8)
Central nervous system involvement at initial diagnosis.”	6 (4.1)	6 (5.9)	0 (0)
Ann Arbor staging system, n (%)
I/II	36 (24.5)	30 (29.4)	6 (13.3)
III/IV	111 (75.5)	72 (70.6)	39 (86.7)
IPI at apheresis, n (%)
0–3	105(71.4)	78 (76.5)	27 (60)
4–5	42 (28.6)	24 (23.5)	18 (40)
BM tumoral cell (%), median (Q1, Q3)	0.0 (0.0, 0.0)	0.0 (0.0, 0.0)	0.0 (0.0, 0.0)
Number of extranodal tumor involvement sites, median (Q1, Q3)	4 (2,6)	3 (2,5)	5(2,7)

a) * P < 0.05, statistical significance

ECOG Eastern Cooperative Oncology Group, IPI international prognostic index BM Bone marrow

As shown in Fig. 1A-D, median number of DNA mutations detected in the 147 samples was 19. Missense mutations accounted for 61.1% (Fig. 1A-B). The 10 most frequently mutated genes (Fig. 1F) were: PIM1 (47%), KMT2D (37%), MYC (31%), BTG1 (29%), H1-4 (28%), MYD88 (27%), DUSP2 (25%), FAT4 (24%), DTX1 (22%), and TP53 (21%) (Fig. S1A). Overall, 4,416 mutations were detected, with 51 genes mutated in ≥ 10% of patients (Fig. 1G). Additional recurrent mutations (BCL2, BCL6, CREBBP, SOCS1, ACTB, SRP72) were detected in 10–20% of patients.

Fig. 1.

Landscape of DNA mutations in 147 samples. (A) Bar chart showing the number of six major mutations in the 147 samples, with the x-axis representing the logarithmic value of the mutation count; (B) Bar chart illustrating the distribution of three types of mutations; (C) Bar chart displaying the frequency spectrum of six types of single nucleotide variants (SNVs); (D) Bar chart depicting the number of DNA mutations in each sample, with the y-axis representing the logarithmic value of the mutation count; (E) Box plot presenting the number of six major mutations; (F) Mutation frequency of the top 10 genes with the highest mutation rates; (G) The DNA mutation landscape at baseline, with genes sorted by their mutation frequency

The impact of prognostic-related genes on overall survival

By performing Kaplan–Meier survival analysis on genes with a mutation frequency greater than 10% (Table S2), we identified that the TP53 mutation was significantly associated with poor OS in patients with DLBCL (Fig. S2A). Patients with TP53 mutations had median OS of 13 vs. 16 months in wild-type (HR = 6.76, P = 0.0026).

The cumulative hazard function further supports the adverse prognostic impact of TP53 mutations. Patients with TP53 mutations exhibited a consistently higher cumulative hazard of death throughout the follow-up period than those with TP53 wild-type (HR = 12.98, P = 0.005), suggesting a persistently elevated risk of mortality associated with TP53 alterations (Fig. S2B). The non-CR group showed higher TP53 mutation frequency than CR group (31.11% vs. 16.83%, P = 0.044) (Fig. S2C).

To further validate these findings and adjust for potential confounding clinical factors, we performed a multivariate Cox proportional hazards regression analysis incorporating both clinical variables and mutation data. Specifically, clinical covariates included age, Eastern Cooperative Oncology Group (ECOG) performance status, Ann Arbor staging (> 2 vs. ≤2), baseline LDH level (> 207 U/L vs. ≤207 U/L), bone marrow tumoral cell infiltration (%), plasma albumin concentration, and the IPI score prior to treatment. In addition, we included the mutation status of genes identified as significant (P < 0.05) in the univariate Cox regression analysis, such as TP53. This integrative approach enabled us to assess the independent prognostic value of genetic alterations in the context of established clinical risk factors. TP53 mutation remained an independent adverse prognostic factor for OS (HR = 14.02, 95% confidence interval [CI]: 2.42‒81.28, P = 0.003), whereas other mutations were not statistically significant after adjustment (Table S3). Moreover, a higher percentage of bone marrow (BM) tumor cells (%) was identified as an independent predictor of poor OS (HR = 1.08, 95% CI: 1.04‒1.13, P = 0.0001; Table S3).

Given the strong prognostic impact of TP53 mutations, we investigated whether co-mutations with other frequently mutated genes could refine the risk stratification. Gene pairs predictive of adverse prognosis were identified using the top 20 most frequently mutated genes (Table S4). Notably, patients harboring co-mutations in MYC and TP53, KMT2D and TP53, CREBBP and TP53, or H1-4 and TP53 had significantly worse OS than patients carrying single mutations or wild-type alleles, suggesting a synergistic detrimental effect of these co-occurring alterations on survival outcomes.

The impact of prognostic-related genes and clinical factors on progression-free survival

Kaplan–Meier (KM) survival analyses were performed to identify candidate factors associated with PFS (Fig. 2A-F). Patients harboring TP53 mutations had a significantly shorter PFS than wild-type patients (median PFS: 9 vs. 14.5 months; HR = 2.78; P = 0.0013) (Fig. 2A). Similarly, SRP72 (median PFS: 9 vs. 13.5 months; HR = 2.79; P = 0.0076) (Fig. 2B), MYC (median PFS: 11 vs. 14 months; HR = 1.92; P = 0.0293) (Fig. 2C), ASXL2 (median PFS: 10 vs. 13 months; HR = 2.2; P = 0.0419) (Fig. 2D) and BCL2 (median PFS: 9.5 vs. 14 months; HR = 2.54; P = 0.0166) (Fig. 2E) mutations were associated with worse PFS. In contrast, ACTB mutations conferred a protective effect (median PFS, 13 vs. 13 months; HR, 0.244; P = 0.0347).

Fig. 2.

Impact of different DNA mutations before treatment on progression-free survival. (A) Progression-free survival curves for different TP53 before treatment. (B) Progression-free survival curves for different SRP72 status before treatment. (C) Progression-free survival curves for different MYC status before treatment. (D) PFS curves for different ASXL2 status before treatment. (E) Progression-free survival curves for different BCL2 status before treatment. (F) Comparison of gene mutation frequencies between the non-relapsed and relapse groups in all patients. Bar plots display the mutation frequencies of the five genes (TP53, SRP72, BCL2, MYC, and ASXL2) in both groups

The results of the univariate Cox regression analysis were consistent with those of the KM survival analysis (Table 2). Mutations in TP53 (HR = 2.78, P = 0.002), SRP72 (HR = 2.79, P = 0.010), BCL2 (HR = 2.45, P = 0.021), MYC (HR = 1.97, P = 0.033), and ASXL2 (HR = 2.20, P = 0.048) were significantly associated with poorer PFS.

Table 2.

Multivariable analysis of progression-free survival of patients with DLBCL treated with R-CHOP therapy^a)

Characteristics	Untivariable COX		Multivariable COX
	HR (95%CI)	P	HR (95%CI)	P
Age	0.99 (0.97–1.02)	0.612	-	-
ECOG	0.99 (0.66–1.52)	0.989	-	-
Baseline LDH (> 207 U/L vs. ≤207 U/L)	3.73 (1.45–9.58)	0.006	2.62(1.05–6.55)	0.039
Ann Arbor staging(> 2 vs. ≤2)	2.93(1.04–8.25)	0.042	-	-
IPI	1.11(0.87–1.42)	0.404	-	-
BM tumoral cell (%)	1.03(1.01–1.05)	0.024	1.03(1.01–1.06)	0.031
Plasma Albumin Concentration	0.94(0.89–0.99)	0.024	-	-
Risk score (>1.61 vs≤ 1.61)b)	4.34( 2.27–8.31)	9.19e-06	3.92( 2.01–7.66)	6.45e-05

a)* P < 0.05, statistical significance

b)Risk score: Risk score was constructed from five genes (TP53, SRP72, BCL2, MYC, and ASXL2)

ECOG Eastern Cooperative Oncology Group, LDH lactate dehydrogenase, IPI international prognostic index, BM Bone marrow

Weights were assigned as the natural logarithm of the hazard ratio (ln[HR]) from univariate Cox regression (TP53 = 1.02, SRP72 = 1.03, BCL2 = 0.93, MYC = 0.68, ASXL2 = 0.79)

The patient score was calculated as the weighted sum of mutation status (0 = wildtype; 1 = mutated), and patients were stratified into Low- and High-risk groups using a cutoff of 1.61

To evaluate the cumulative impact of multiple gene mutations on patient outcomes, we developed a weighted mutation-based risk score incorporating five genes of interest (TP53, SRP72, BCL2, MYC, and ASXL2) as follows: the weight assigned to each gene was derived from the natural logarithm of the HR (ln[HR]) obtained from univariate Cox regression analyses for PFS. The resulting weights were: TP53 = 1.02, SRP72 = 1.03, BCL2 = 0.93, MYC = 0.68, and ASXL2 = 0.79. For each patient, the risk score was calculated as the weighted sum of the binary mutation status (0 = wild-type, 1 = mutated) across five genes. Patients were then stratified into low- and high-risk groups using a cutoff value of 1.61.

To further validate these findings and adjust for potential confounders, multivariate Cox proportional hazards regression analyses were performed, incorporating the clinical and genetic variables (Table 2). The final model identified three independent PFS predictors. The high-risk group, as defined by the mutation-weighted risk score, exhibited a significantly increased hazard of progression or relapse (HR = 3.92, 95% CI: 2.01–7.66, P = 6.45e-05) (Table 2). In addition, higher baseline LDH levels (> 207 U/L vs. ≤207 U/L) (HR = 2.62, 95% CI: 1.05–6.55, P = 0.039) and increased BM tumor cell percentage (HR = 1.03, 95% CI: 1.01–1.06, P = 0.031) were associated with a greater risk of disease progression.

To further explore the relationship between gene mutations and disease recurrence, we compared the mutation frequencies of five genes (TP53, SRP72, BCL2, MYC, and ASXL2) between the relapsed and non-relapsed patient groups using co-bar plot analysis (Fig. 2F). The relapse group showed notably higher mutation frequencies in all five genes. These observations suggest the potential enrichment of TP53, SRP72, BCL2, MYC, and ASXL2 mutations in patients who experience relapse, supporting their putative roles in disease progression and therapeutic resistance.

In addition to single-gene effects, we explored the prognostic impact of gene-gene interactions by analyzing co-mutations among the top 20 mutated genes (Table S4). Several co-mutated gene pairs, including MYC-TP53, TP53-CREBBP, DUSP2-KMT2C, MYC-TET2, DUSP2-BRCA2, TP53-BTG2, BTG1-BRCA2, MYD88-BRCA2, DTX1-BRCA2, and BTG1-KMT2C were significantly associated with poor PFS. Conversely, co-mutations in PIM1-SOCS1 and SOCS1-ACTB appeared to confer a protective effect and were associated with longer PFS.

Mutation difference between the non-relapse and relapse groups

Among patients achieving CR, univariate Cox regression analysis of PFS identified TP53 (HR = 4.45, P = 0.003), SRP72 (HR = 4.16, P = 0.014), and MYC (HR = 2.64, P = 0.047) as relapse risk factors.

The patients were stratified into mutated and wild-type groups for each gene mutation. The TP53-mutated group exhibited a substantially higher cumulative incidence of relapse than the wild-type group (P = 0.0016; Fig. 3A). Similarly, SRP72 mutations were associated with a significantly increased incidence of relapse (Fig. 3B, P = 0.0078). For MYC, patients with mutations demonstrated a moderately increased cumulative relapse risk compared to wild-type (Fig. 3C, P = 0.04). To evaluate the cumulative impact of multiple gene mutations on patient outcomes, we developed a weighted mutation-based risk score incorporating three genes of interest: TP53, SRP72, and MYC. The weight assigned to each gene was derived from ln[HR] obtained from univariate Cox regression analyses for PFS. The resulting weights were as follows: TP53 = 1.49, SRP72 = 1.43, and MYC = 0.97. For each patient, the risk score was calculated as the weighted sum of the binary mutation status (0 = wild-type, 1 = mutated) across the three genes. Patients were then stratified into low- and high-risk groups using a cutoff value of 2.4. Kaplan–Meier survival analysis demonstrated that patients in the high-risk group had a significantly shorter time between CR and relapse than those in the low-risk group (Fig. 3D, log-rank P = 0.00011), indicating that the composite score effectively predicted adverse outcomes.

Fig. 3.

Cumulative incidence of relapse stratified by gene mutation status. (A) TP53 mutation. (B) SRP72 mutation. (C) MYC mutation. Each panel shows the cumulative incidence of relapse in patients with and without the indicated mutations. In all cases, the mutation-positive groups exhibited a higher relapse incidence than the wild-type counterparts. Log-rank P-values are reported for each comparison. The shaded areas indicate the 95% confidence intervals. The number of patients at risk at each time point is shown below the x-axis. (D) Time between complete remission (CR) and relapse or end of follow-up curves, stratified by genetic risk groups. The risk score is calculated using a weighted sum of mutations in TP53, SRP72, and MYC, with weights derived from their subdistribution hazard ratios. Patients are classified into the low-risk and high-risk groups using a cutoff of 2.4. The high-risk group showed significantly poorer outcomes than the low-risk group. (E) Comparison of gene mutation frequencies between the non-relapsed and relapsed groups in patients with CR. Bar plots display the mutation frequencies of three genes (TP53, SRP72, MYC) in the non-relapse and relapse groups

Multivariate Cox regression models adjusted for clinical stage were used to further validate the prognostic significance. In the interval from CR to relapse or last follow-up, the high-risk group showed a significantly increased hazard of progression or relapse (HR = 5.22, 95% CI: 1.84–14.79, P = 1.89e-03) (Table 3). In addition, higher baseline LDH levels (> 207 U/L vs. ≤207 U/L) (HR = 6.31, 95% CI: 1.38–28.94, P = 0.018) were associated with a greater risk of disease relapse (Table 3). These findings support the clinical utility of mutation-weighted risk scores based on biologically relevant genes and their individual contributions to relapse risk and survival.

Table 3.

Multivariate Cox regression analysis of time between CR to relapse or end of follow-up in patients with DLBCL who achieved CR after treatment with R-CHOP therapy^a)

Characteristics	Untivariable COX		Multivariable COX
	HR (95%CI)	P	HR (95%CI)	P
Age	0.99 (0.96–1.02)	0.652	-	-
ECOG	1.12 (0.61–2.08)	0.713	-	-
Baseline LDH (> 207 U/L vs. ≤207 U/L)	5.88 (1.32- 25)	0.020	6.31(1.38–28.94)	0.018
Ann Arbor staging(> 2 vs. ≤2)	1.91(0.54–6.7)	0.314	-	-
IPI	1.19(0.81–1.75)	0.366	-	-
BM tumoral cell (%)	0.91(0.47–1.77)	0.782	-	-
Plasma Albumin Concentration	0.90(0.83–0.99)	0.025	-	-
Risk scoreb) (>2.4 vs≤ 2.4)	6.09( 2.18–17.00)	5.59e-04	5.22( 1.84–14.79)	1.89e-03

a)* P < 0.05, statistical significance 

b)Risk score: Risk score was constructed from three genes (TP53, SRP72 and MYC)

ECOG Eastern Cooperative Oncology Group, LDH lactate dehydrogenase, IPI international prognostic index, BM Bone marrow

Weights were assigned as the natural logarithm of the hazard ratio (ln[HR]) from univariate Cox regression for PFS (TP53 = 1.49, SRP72 = 1.43, MYC = 0.97)

The patient score was calculated as the weighted sum of mutation status (0 = wildtype; 1 = mutated), and patients were stratified into Low- and High-risk groups using a cutoff of 2.4

To assess the association between gene mutations and disease relapse, we compared the mutation frequencies of three genes of interest (TP53, SRP72, and MYC) between the relapsed and non-relapsed patient groups using a co-bar plot analysis (Fig. 3E). The relapse group showed noticeably higher mutation frequencies in TP53, SRP72, and MYC genes. These results suggest the potential enrichment of TP53, SRP72, and MYC mutations among relapsed patients, supporting their roles in promoting disease progression or treatment resistance. This visual comparison further corroborated the findings of the survival analyses.

Prognostic impact of combined genetic and clinical factors in overall and CR cohorts

To assess the prognostic value of gene mutations beyond traditional clinical features, Cox regression analyses were performed in the overall cohort. One analysis included only the International Prognostic Index (IPI), while a second incorporated the gene mutation risk score (TP53, SRP72, BCL2, MYC, and ASXL2) in addition to clinical variables, including baseline LDH (> 207 U/L vs. ≤207 U/L) and BM tumoral cell (%). The concordance index (C-index) of the IPI for PFS prediction was 0.53, whereas the composite assessment of gene mutation risk score and clinical factors increased the C-index to 0.781, indicating improved discrimination. Time-dependent receiver operating characteristic (ROC) analysis at 1, 2, and 3 years also demonstrated higher AUC values for the clinical + gene mutation risk score combined evaluation than for the IPI (Fig. 4A). For example, at 36 months, the area under the curve (AUC) increased from AUC_clinical = 0.501 to AUC_combined = 0.940. DeLong’s test confirmed that the improvement in predictive accuracy was statistically significant at most time points (p-values: 36 months = 0.000837) (Fig. 4B). Furthermore, risk stratification based on the median predicted risk score from the combined evaluation divided the patients into high- and low-risk groups. Kaplan–Meier analysis showed a significantly inferior PFS in the high-risk group (log-rank P < 0.0001; Fig. 4C), supporting the clinical utility of the combined evaluation.

Fig. 4.

Prognostic improvement performance of composite assessment of mutations and clinical factors in the overall and complete remission (CR) cohorts. (A-C) Prognostic performance of the International Prognostic Index (IPI) versus the clinical + gene combined evaluation in the overall cohort. (A) Time-dependent area under the curve (AUC) curves at 1, 2, and 3 years for the IPI and clinical + gene combined evaluation. (B) Time-dependent receiver operating characteristic (ROC) curves comparing the IPI (blue) and the combined clinical + gene combined evaluation (red) for predicting progression-free survival (PFS) in the overall cohort. The combined evaluation consistently shows higher AUC values at 36 months. (C) Kaplan–Meier curves of PFS stratified by the median risk score from the combined evaluation in the overall cohort. Patients in the high-risk group exhibit significantly worse PFS than those in the low-risk group (log-rank P < 0.0001). (D-E) Prognostic evaluation performance in the R-CHOP CR subgroup. (D) Time-dependent AUC curves show improved discrimination for the clinical + gene combined evaluation. (E) Time-dependent ROC curves comparing the IPI (blue) and clinical + gene (red) combined evaluation in patients who achieve CR after R-CHOP therapy. The clinical + gene combined evaluation provides improved predictive accuracy at all time points. (F) Kaplan–Meier curves of time between CR and relapse or end of follow-up stratified by combined evaluation risk score in the CR cohort. The high-risk group has significantly poorer outcomes (log-rank P = 0.00011)

To explore the applicability of the gene mutation risk score combined evaluation in a more homogeneous clinical setting, the same analyses were repeated for patients who achieved CR after R-CHOP treatment. The combined evaluation (clinical + gene) incorporated the gene mutation risk score (TP53, SRP72, and MYC) in addition to clinical variables, including baseline LDH (> 207 U/L vs. ≤207 U/L). Similarly, the combined evaluation improved performance, with the C-index increasing from 0.593 to 0.914. The time-dependent AUC values were consistently higher for the combined evaluation (clinical + gene) at all time points (Fig. 4D), and DeLong’s test indicated significant differences (36 months, P = 0.000117) (Fig. 4E). Kaplan–Meier survival analysis demonstrated a clear separation between high- and low-risk patients, with the high-risk group experiencing significantly more relapses (log-rank P = 0.00011; Fig. 4F).

Overall, these findings suggest that incorporating gene mutation information alongside key clinical variables provides additional prognostic insight for PFS and relapse risk in both the overall cohort and patients achieving CR.

Pathogenic driver genes and potential novel therapeutic strategies in DLBCL

Using the OncodriveCLUST algorithm, we identified MYD88, USP7, LRRN3, MYC, NOTCH1, CD79B, and RB1 as potential pathogenic driver genes (Fig. S3A). Mutations in MYD88, NOTCH1, and CD79B activate the NF-κB pathway, promoting tumor cell survival and proliferation. MYC mutations may drive proliferation and suppressing apoptosis. USP7 stabilizes oncoproteins including c-MYC, while RB1 mutations lead to uncontrolled proliferation.

We further annotated the functional domains using the Pfam domain function to identify regions enriched in amino acid-altering mutations. The five most frequently mutated Pfam domains were Pkinase, Myc_N, Actin, BTG1, and V-set (Fig. S3B).

Mutation co-occurrence and exclusivity patterns were assessed using the somatic interaction function. Among the top 40 mutated genes at baseline, 142 co-occurring gene pairs and seven mutually exclusive gene pairs were identified (P < 0.05) (Fig. S3C). MYD88 showed significant co-occurrence with PIM1, CD79B, PRDM1, and ETV6 but was mutually exclusive with FOXO1, FAT1, and CCND3. PIM1 frequently co-occurred with BTG1, DUSP2, FAT4, BRCA2, CD79B, MAG, PRDM1, IGLL5, LRRN3, and ETV6. In contrast, TP53 co-occurred with PRDM1 and FAT1 but was mutually exclusive to SOCS1. These findings reveal complex genetic interactions that may influence the biological and therapeutic responses of DLBCL patients.

We performed drug-gene interaction analysis using the DGIdb database via the maftools package. Several mutated genes were identified as potential drug targets (Fig. S3D). Notably, TP53, PIM1, KMT2D, and MYC exhibited clinically actionable interactions with the existing targeted agents. TP53 mutations are associated with sensitivity to kinase inhibitors, and MYC with transcriptional regulation inhibitors. These findings highlight actionable molecular targets and provide insights for personalized treatment in DLBCL.

Discussion

Although R-CHOP remains the cornerstone of first-line therapy for DLBCL, 30–40% of patients relapse, underscoring the need for predictive biomarkers [1, 16]. While clinical indices like IPI have prognostic value, genomic features offer improved risk stratification [3, 4]. In this study, we analyzed pretreatment genomic features of 147 Chinese DLBCL patients treated with R-CHOP.

Our results demonstrate that a combined evaluation (clinical + gene) incorporating gene mutation risk score in addition to clinical variables, including baseline LDH (> 207 U/L vs. ≤207 U) and BM tumoral cell (%), is significantly associated with PFS, establishing a key prognostic indicator for R-CHOP outcomes. Our results align with previous evidence linking ctDNA dynamics to DLBCL outcomes [17]. Furthermore, specific gene mutations with profound clinical implications have been identified. Consistent with landmark DLBCL genomic studies [3, 18], TP53 mutations strongly predicted adverse outcomes (lower CR rate, shorter PFS and OS), indicating TP53 dysfunction as a principal driver of R-CHOP resistance.

In addition to TP53, BCL2, and MYC, we identified previously unreported associations between SRP72 and ASXL2 mutations and poor PFS, highlighting their potential as novel DLBCL biomarkers. Conversely, mutations in ACTB (β-actin), a cytoskeletal protein, were unexpectedly correlated with improved PFS. Although the mechanistic roles of SRP72, ASXL2, and ACTB mutations in DLBCL pathogenesis and treatment response warrant further investigation, their identification highlights novel potential prognostic genes for R-CHOP efficacy.

Our study had some limitations. First, the findings derive from a single Chinese patient cohort, requiring validation in diverse populations. Nevertheless, all patients underwent baseline PET/CT for accurate staging, which is essential for appropriate treatment allocation and prognostic assessment [19]. Regarding follow-up, all patients in our cohort underwent regular surveillance including physical examination, blood tests, and imaging (ultrasonography or CT) every 3–6 months, consistent with strategies shown to be effective for relapse detection in lymphoma survivors [20]. Second, our analysis focused on pretreatment samples; however, dynamic monitoring of DNA during and after R-CHOP therapy may offer more sensitive prognostic information and insights into emerging resistance mechanisms, as demonstrated in other lymphoma treatment contexts [21, 22]. Third, although we identified significant associations, the functional mechanisms linking specific genes, such as SRP72, ASXL2, and ACTB, to the R-CHOP response remain to be elucidated. Future studies integrating transcriptomic and functional data are needed.

Our prognostic model was developed in R-CHOP-treated patients, but frontline therapy for DLBCL is evolving. Picardi et al. reported favorable outcomes with liposomal doxorubicin-containing regimens [23], and the POLARIX trial established Pola-R-CHP as a new standard [24]. Although our model requires validation in these novel regimens, the identified mutations (TP53, MYC, BCL2) are biologically linked to chemoresistance and may remain relevant for risk stratification.

In conclusion, this study defined the key genomic features predictive of R-CHOP treatment outcomes in a large Chinese DLBCL cohort. We established that specific mutations, most notably in TP53, SRP72, BCL2, MYC, and ASXL2, were robust indicators of poor prognosis, whereas ACTB mutations denoted a more favorable subgroup. Pretreatment TP53 mutation status is a particularly powerful predictor of CR, PFS, and OS. These genomic markers provide clinicians with valuable tools for the early identification of patients at high risk of R-CHOP failure. Identifying patients harboring high-risk mutations prior to therapy could inform early consideration of autologous stem cell collection or CAR-T therapy, potentially improving outcomes for those at elevated risk of R-CHOP failure.

Supplementary Information

Below is the link to the supplementary material.

Author contributions

SH contributed to the study’s conception and design, conducted the hospital chart review, interpreted data, drafted the work and wrote the final report. LHZ and HQZ assisted with the statistical analysis and wrote the manuscript. KG and JY conducted the hospital chart review. YXH also helped author the manuscript and contributed to the study’s conception and design. YZ and ZC and JSH and XJY and WJW and JS and WYZ and GQW and LY and LXW and MMZ assisted with pathological diagnosis. YXH and HH designed the research study and participated in critical revision of manuscript. All authors reviewed and approved the final manuscript.

Funding

This work was sponsored by grants from the High Technology Research and Development Center of the National Natural Science Foundation of China (2025YFF0500705), NSFC International (Regional) Cooperation and Exchange Program (82561160165), the National Natural Science Foundation of China (Grant No. 82341206, 82270234), the National Science Fund for Distinguished Young Scholars (Grant No. 82425002) and Sanming Project of Medicine in Shenzhen (SZSM202111004). It was also supported by the Zhejiang Provincial Department of Science and Technology “Jianbing Lingyan + X” Research and Development Project Plan (Grant No. 2025C02075, 2024C03156).

Data availability

The Exome sequencing data reported in this paper have been deposited at Genome Sequence Archive in National Genomics Data Center (https://ngdc.cncb.ac.cn/gsa) and publicly available as of the date of publication. Accession numbers are HRA012974. Source data are included with this paper. Any other potential type of data used to interpret the finding can be provided upon request to corresponding author.

Declarations

Ethics approval and consent to participate

This study was approved by the Clinical Research Ethics Committee of the FirstAffiliated Hospital, Zhejiang University School of Medicine (Approval No. 2025B-0998). The informed consent was waived because this study was a retrospective study with review of related data through the electronic medical record. This research adheres to the ethical principles outlined in the Declaration of Helsinki (1975 revision).

Competing interests

The authors declare no competing interests.