Educational theme: 5hmC profiles of cfDNA predict R-CHOP treatment response in patients with DLBCL

This article is part of our educational series on the role of circulating tumor DNA (ctDNA) in the management of patients with lymphoma.

Diffuse large B-cell lymphoma (DLBCL) accounts for ~30% of all non-Hodgkin lymphomas. The standard treatment regimen usually consists of rituximab, cyclophosphamide, doxorubicin, vincristine, and prednisone (R-CHOP), however 30–50% of patients do not respond to this type of treatment. Currently the International Prognostic Index (IPI) is used to predict high-risk patients, however, it cannot accurately predict the therapeutic effect of R-CHOP in patients with DLBCL. More recently, analysis of cell-free DNA (cfDNA), which is found in the peripheral blood and emerges from the original tumor as it rapidly divides, has emerged as a novel, non-invasive tool for diagnosis and prognosis of cancers.¹

Epigenetic modification of DNA, especially methylation at the fifth carbon of cytosine (5-methylcyctosine; 5mC) can influence gene expression and cancer development. 5mC in cfDNA is dynamic, reversible, and can be oxidized into 5-hydroxymethylcytosine (5hmC). 5hmC has recently been shown to be associated with DLBCL prognosis. Chen et al. have investigated whether genome-wide analyses of 5hmC profiles could be used to predict treatment response to R-CHOP in patients with DLBCL. The results of this study were published in the journal Clinical Epigenetics and are summarized below.¹      

Study design and patient characteristics¹

5hmC profiles were analyzed by genome-wide analysis (5hmC-Seal technology) of plasma cfDNA from 86 patients with DLBCL across multicenter studies in China, before R-CHOP was administered. Baseline characteristics can be found in Table 1.

• 52.3% of patients had intermediate-/high-risk DLBCL (IPI score ≥3).

Table 1. Baseline characteristics*

β2M, beta 2 macroglobulin; GCB, germinal center B cell; IPI, International Prognostic Index; LDH, lactate dehydrogenase; SD, standard deviation. *Data from Chen et al, 20211
Characteristic	N = 86
Mean age, years (SD)	54.59 (15.56)
Male/female, %	53.5/46.5
Ann Arbor stage, %               I               II               III               IV               Unknown	7.0 20.9 8.1 55.8 8.1
IPI, %               0               1               2               3               4               5               Unknown	9.3 14.0 20.9 32.6 17.4 2.3 3.5
Cell of origin, %               GCB               Non-GCB               Unknown	26.7 70.9 2.3
Mean LDH level, U/L	364.33
Mean β2M level, mg/L	2.84

Patients were randomly divided into the training cohort (n = 56) or the validation cohort (n = 30).

The efficacy of R-CHOP was evaluated in all patients after four treatment cycles using the Lugano 2014 criteria.

Results

Training cohort¹

• 35 patients were identified as responders (those with a complete response or partial response) and 21 patients as non-responders (those with stable or progressive disease).
• Differential analysis of the 5hmC regions between responders and non-responders revealed differences (fold change >0.5) in 205 regions.
  • This resulted in upregulation of 124 genes and down regulation of 81 genes, in responders compared to non-responders.
  • 5hmCs occurred in the intronic, intergenic, and promoter regions.
• The genes most enriched for 5hmCs were for signaling pathways involved in alpha–beta T-cell differentiation, protein-lysine N-methyltransferase activity, and histone H3-K9 modification.

Validation cohort¹

• 22 patients were identified as responders and eight patients as non-responders.
• By using a logistics regression model with an elimination algorithm, 13 of the 205 5hmCs were found to distinguish responders from non-responders in both the training and validation cohorts (Table 2). 

Table 2. Logistic regression coefficients of the 13 5hmC markers that predict R-CHOP response*

SE, standard errors of coefficients; Z value, Wald z-statistic value. *Data from Chen et al, 20211
Gene (intercept)	Coefficient	SE	Z value	P value
THAP3 (chr1_6721489_6721898)	0.7712	0.145	1.865	< 0.05
SMYD3 (chr1_246290825_246291238)	0.39	0.149	1.955	< 0.05
OR2G2 (chr1_247755954_247756505)	3.1779	0.108	1.344	< 0.05
ALKBH3 (chr11_43905400_43905804)	3.3423	0.128	1.306	< 0.05
RNASEH2C (chr11_65511519_65512429)	1.5211	0.072	2.061	< 0.05
ARHGEF12 (chr11_120211662_120212234)	−3.8797	0.115	−3.225	< 0.001
ZNF280D (chr15_56982146_56982638)	−1.2266	0.177	−3.250	< 0.001
SLC5A11 (chr16_24916341_24916920)	0.6683	0.076	0.149	< 0.05
CTDP1 (chr18_77500908_77501376)	−2.573	0.103	−3.182	< 0.001
GPR15 (chr3_98270705_98271079)	0.1052	0.167	2.348	< 0.05
GOLGB1 (chr3_121430838_121431239)	0.8526	0.178	2.982	< 0.01
FBXL4 (chr6_99461404_99461922)	1.7188	0.101	2.165	< 0.05
LMBR1 (chr7_156700537_156701031)	1.0942	0.078	0.579	< 0.05

• The 13 markers could effectively predict response to R-CHOP treatment (area under curve [AUC] = 0.78) with a sensitivity of 0.82 and a specificity of 0.75.
  • ARHGEF12 and ZNF280D showed the best predictive performance (AUC = 0.76 in the validation cohort).
• Although lactate dehydrogenase (LDH) level and stage of disease could also predict response to R-CHOP, the AUCs were smaller than that of the 13 5hmC markers (0.646 and 0.658, respectively), even when LDH and stage were combined (AUC = 0.669).

The Cancer Genome Atlas (TCGA)-DLBCL dataset¹

To further understand the potential associations between the 13 5hmC marker genes and R-CHOP treatment response, mRNA expression profiles of the 13 5hmC marker genes were compared to that of B-lymphocyte antigen CD20 (MS4A1), a rituximab target gene, in 48 patients with DLBCL from the TCGA dataset.

• mRNA expression of MS4A1 was positively correlated with that of ARHGEF12 (rho = 0.385), FBXL4 (rho = 0.376), GOLGB1 (rho = 0.434), and LMBR1 (rho = 0.45).
• The gene expression profiles that were highly positively associated with that of ARHGEF12 included RHOA (rho = 0.667), RHOB (rho = 0.604), CDC42 (rho = 0.676), ROCK1 (rho = 0.832), GNA12 (rho = 0.721), and GNA13 (rho = 0.784).
• Patients with high expression of ARHGEF12 and CDC42 had significantly lower overall survival rates than those with low expression (p = 0.079 and p = 0.0033, respectively).

Conclusion

There were 13 5hmC markers derived from cfDNA that effectively distinguished responders and non-responders to R-CHOP therapy. This was superior to existing clinical indicators, such as LDH levels and staging of disease, and therefore these 5hmCs may serve as effective biomarkers of response. 5hmC of ARHGEF12 had the best prognostic score, and its expression was positively associated with genes involved in the Rho signaling pathway. This pathway is known to influence cancer initiation, proliferation, metastasis, and drug resistance.

Limitations to the study include that only Chinese patients were included, and the sample sizes were small. Therefore, the results may not fully represent all patients with DLBCL and so further studies in larger, more diverse cohorts are required.  

For other articles in this theme click on the links below:

The role of ctDNA in lymphoma management

Monitoring treatment response in lymphoma using cfDNA

CSF analysis of ctDNA for lymphoma with CNS involvement