Circulating Tumor DNA as a Biomarker of Treatment Response and Minimal Residual Disease in Diffuse Large B-Cell Lymphoma: A Literature Review

1. Introduction

Diffuse large B-cell lymphoma (DLBCL) is a clinically and biologically heterogeneous group of aggressive non-Hodgkin lymphomas [1]. Although standard immunochemotherapy cures more than 60% of patients, a substantial proportion experience primary refractory disease or relapse [2]. One of the key causes of treatment failure is the complex molecular landscape of DLBCL, which includes genomic instability, a diversity of driver alterations, and clonal evolution under the selective pressure of therapy [3,4,5].

Conventional excisional lymph node biopsy, while remaining the «gold standard» for diagnosis, provides primarily a static immunomorphological and molecular characterization of the tumor at a fixed point in time [6,7]. However, this approach may be insufficiently representative in the setting of spatial heterogeneity caused by genetically discordant subclones in different anatomical sites. Moreover, after treatment, tumor substrate available for biopsy may be absent, and the invasiveness of the procedure itself substantially limits opportunities for serial molecular monitoring [8]. At the same time, positron emission tomography combined with computed tomography (PET/CT), the most widely used imaging modality for assessing changes in the extent of tumor involvement, reflects tumor metabolic activity but does not provide direct information about the molecular mechanisms of resistance or clonal dynamics [9,10].

In this context, the introduction of the «liquid biopsy» concept into clinical practice has created new opportunities for precision oncohematology [9]. Circulating tumor DNA (ctDNA) - a fraction of plasma cell-free DNA released by tumor cells during apoptosis, necrosis, and active secretion, is considered a systemic molecular marker of the tumor process [9]. ctDNA analysis potentially enables minimally invasive detection of tumor-specific genetic alterations and in vivo tracking of clonal disease dynamics, opening perspectives for molecular typing, monitoring of minimal residual disease, and early detection of molecular relapse [11,12].

The aim of this review is to systematize current data on the biological determinants of ctDNA in DLBCL and to evaluate the evolution of analytical platforms, from targeted sequencing approaches to ultrasensitive methods based on detection of phased variants. Particular attention is given to the translational potential of ctDNA as a tool for overcoming the limitations of spatial heterogeneity and for implementing personalized, response-adapted therapeutic strategies.

Although recent reviews, both in the broader context of B-cell lymphomas and specifically in DLBCL, have summarized evidence supporting the prognostic significance of ctDNA, several practice-oriented issues remain insufficiently standardized. In particular, concordance between the ctDNA mutational profile and the tumor tissue profile is not always considered as a separate topic, while limitations to routine implementation are often discussed without integrating biological, analytical, and clinical prerequisites. Accordingly, the present review offers a clinically oriented synthesis of evidence, from baseline risk stratification and early assessment of molecular eradication of the tumor clone to evaluation of MRD after completion of therapy and monitoring during remission, and separately summarizes data on concordance between ctDNA and tumor tissue. Special attention is paid to false-positive results related to clonal hematopoiesis of indeterminate potential and to the distinction between analytical validity and clinical utility, requiring confirmation in prospective studies, including interventional trials.

2. Materials and Methods

Literature Search and Selection

For this review, a literature search was performed in PubMed/MEDLINE, Scopus, and Web of Science, as well as in the ClinicalTrials.gov registry. Study selection was performed by consensus among the authors, and disagreements were resolved by discussion. The search covered English-language publications predominantly from 2012 to March 2026, with emphasis on relevant studies from the last five years. The following keywords and combinations were used: «diffuse large B-cell lymphoma», «DLBCL», «circulating tumor DNA», «ctDNA», «liquid biopsy», «minimal/measurable residual disease», and «targeted sequencing». Original prospective and retrospective clinical studies, meta-analyses, and translational studies dedicated to molecular detection methods were included in the review. Single case reports and non-peer-reviewed preprints were excluded from the main analysis.

3. Results

3.1. Biological Characteristics of Circulating Tumor DNA

Circulating tumor DNA (ctDNA) is a fraction of cell-free DNA carrying genetic and epigenetic alterations characteristic of malignant neoplasms. Cell-free DNA (cfDNA) consists of short double-stranded DNA fragments circulating in plasma and generated predominantly through apoptosis and necrosis [9]. In healthy individuals, the cfDNA pool is formed mainly by hematopoietic cells [9]. A diagnostically critical point is that, in patients with cancer, cfDNA represents a mixture of normal and tumor-derived DNA. Physiological factors such as inflammation, physical activity, and trauma may substantially increase the level of normal cfDNA, thereby reducing the ctDNA fraction and complicating its detection [9].

As a minor fraction of total cfDNA, ctDNA differs from DNA of normal cells by the presence of tumor-specific genetic aberrations, including somatic mutations and unique immunoglobulin gene rearrangements [13]. cfDNA levels in patients with cancer are several-fold higher than normal because apoptotic and necrotic remnants within the tumor mass are inefficiently cleared by phagocytes, which promotes accumulation and release of cellular debris into the systemic circulation [9]. It should be emphasized that blood ctDNA content is characterized by extreme interindividual variability. The ctDNA fraction within total cfDNA ranges from 0.003% to 95%, underscoring that its level depends not only on tumor burden but also on individual patient biology and disease characteristics [14].

One of the key features of cfDNA is its fragmented nature: the main pool of fragments is approximately 167 base pairs in length, corresponding to the size of nucleosome-associated DNA and reflecting apoptotic fragmentation of chromatin in internucleosomal regions [9,15]. A characteristic feature of the tumor fraction is a shift of the fragmentation profile toward shorter molecules, with a peak at approximately 145 bp compared with approximately 167 bp for hematopoietic cell-derived cfDNA. This phenomenon is linked to aberrant hypomethylation and altered chromatin structure in tumor cells, which increase their susceptibility to nuclease degradation [14,16]. An additional factor is transcription-dependent chromatin decondensation, which increases exposure of internucleosomal linker regions to enzymatic degradation [16].

Mechanisms of ctDNA release include both passive processes, such as apoptosis and necrosis, and active processes, including secretion within extracellular vesicles. The main source is apoptosis, in which caspase-dependent activation of endonucleases leads to internucleosomal DNA fragmentation and the formation of distinctive fragments approximately 166 base pairs in length [14,15]. In lymphomas with high proliferative activity and pronounced spontaneous tumor cell lysis, the apoptotic mechanism plays a key role in generating the baseline plasma ctDNA level. Antitumor therapy, including chemoimmunotargeted agents, enhances apoptotic cell death, which may be accompanied by a transient increase in plasma ctDNA levels. Such dynamics are considered a potential early pharmacodynamic marker of treatment response. Unlike apoptosis, necrosis is a less regulated process, often driven by adverse conditions within tumor tissue such as hypoxia and nutrient deprivation. It leads to the release of longer DNA fragments, up to thousands of base pairs, which are subsequently enzymatically cleaved by macrophages and released in shortened form [14]. In addition, ctDNA can be actively secreted via extracellular vesicles such as exosomes (30-150 nm), which originate from living cells and may contribute to tumor invasiveness, progression, and treatment resistance [14,17]. Study results demonstrate that rates of apoptosis and necrosis do not always correlate with cfDNA release, while ctDNA concentrations may depend on the cell-cycle phase, highlighting the role of active release [17]. It should be noted that the quantitative contribution of active secretion to the plasma ctDNA pool in lymphomas, including DLBCL, remains insufficiently defined. Since a substantial proportion of mechanistic data was obtained in solid tumor models and cell lines [17], direct extrapolation of these mechanisms to DLBCL should be approached with caution [18]. The authors indicate that the physicobiological characteristics of ctDNA depend on the mechanism of its release into the bloodstream. Analysis of these parameters may potentially allow indirect assessment of biological processes occurring in tumor tissue. A number of studies suggest that extracellular vesicle-associated DNA may participate in intercellular communication, influencing cells of the tumor microenvironment and possibly cells in distant tissues [14,19].

After entering the bloodstream, ctDNA undergoes rapid degradation, with a half-life ranging from 16 minutes to 2.5 hours, due to the action of blood DNases, active nucleosome clearance, and filtration in organs such as the kidneys and lymph nodes [20]. Alternative estimates indicate a half-life of 30 to 120 minutes, making the plasma ctDNA level a reflection of current tumor dynamics rather than accumulated changes [16]. Studies in patients with metastatic melanoma have shown a strong correlation between tumor PET activity and ctDNA abundance [21]. These data suggest that metabolically active and rapidly growing tumors may release ctDNA more intensively per unit of tumor burden. At the same time, in lymphoproliferative diseases, including DLBCL, ctDNA release is determined by different biological prerequisites and often results in a higher tumor DNA fraction in plasma compared with many solid tumors. This is particularly typical of aggressive lymphomas, in which high tumor burden is associated with elevated ctDNA levels. This feature makes DLBCL one of the most informative models for clinical application of «liquid biopsy»; however, patterns established in solid tumors cannot be extrapolated to lymphomas without validation [22]. Overall, these biological characteristics emphasize the potential of ctDNA as a dynamic biomarker reflecting tumor processes in real time [23].

According to the literature, ctDNA contains distinct epigenetic features, including changes in DNA methylation, histone modifications, and nucleosome positioning, which arise in early carcinogenesis and affect gene expression without altering the DNA sequence [24,31]. Tumor cells exhibit global genomic hypomethylation, leading to chromosomal instability and oncogene activation, alongside hypermethylation of tumor suppressor gene promoters, which suppresses their expression [24]. This phenomenon also explains the shortening of ctDNA fragments, because hypomethylated DNA becomes more accessible to nucleases during apoptosis [14]. Analysis of methylation patterns allows accurate identification of tumor tissue of origin and assessment of treatment pharmacodynamics, supporting its consideration as one of the most promising epigenetic biomarkers for screening [24].

In addition to methylation, epigenetic features of ctDNA include histone modifications such as acetylation, methylation, and phosphorylation, which affect chromatin accessibility and therefore DNA fragmentation [25]. ctDNA fragmentation patterns closely correlate with gene expression: actively transcribed regions with open chromatin produce shorter fragments [25]. Nucleosome footprinting uses this relationship to determine the epigenetic status and tissue of origin of circulating DNA, because nucleosome positioning has tissue-specific patterns that change during oncogenesis [14]. In tumor cells, these patterns reflect dysregulated gene expression and epigenetic reprogramming, enabling differentiation between malignant and normal DNA fragments and, in the context of lymphomas, distinction between subtypes based on their unique gene-expression programs and chromatin states, as well as tracking of changes in tumor biology during therapy and clonal evolution [25].

In lymphomas, including DLBCL, methylation profiling has been studied less extensively than in solid tumors. Nevertheless, this approach has identified distinctive epigenetic features associated with different subtypes that may be used for molecular classification and prognosis [24]. Specific methylation markers correlate with clinical outcomes in DLBCL, and their integration with mutational profiling improves molecular subtyping and provides additional prognostic information. Thus, analysis of epigenetic ctDNA features is a promising approach for precision oncology, complementing genetic biomarkers in the assessment of treatment response and detection of relapse in patients with DLBCL [26].

3.2. ctDNA Detection

The spectrum of analytical methods for ctDNA detection in DLBCL ranges from single-locus polymerase chain reaction (PCR)-based approaches to multiplex next-generation sequencing (NGS) panels. Selection of a specific technological platform is determined by the clinical or research objective, from monitoring previously identified somatic mutations to comprehensive genetic profiling and assessment of minimal residual disease in the setting of pronounced clonal tumor heterogeneity [27]. Strict standardization of the preanalytical stage is a fundamental requirement for molecular profiling. Plasma is the preferred biomaterial, because serum use carries a high risk of genomic contamination due to leukocyte lysis during blood clotting. Blood is collected in K2EDTA tubes and centrifuged no later than 6 hours after collection. Use of stabilizing tubes (Streck BCT) allows sample preservation for up to 14 days, which is particularly important during transportation [28].

Historically, allele-specific PCR based on analysis of individual clonal immunoglobulin gene rearrangements was considered the principal method for detecting minimal residual disease. Introduction of droplet digital PCR increased the analytical sensitivity of the method to 0.01% when monitoring recurrent somatic mutations, such as MYD88 L265P. However, the clinical applicability of PCR-based approaches in DLBCL is limited by the absence of universal molecular targets, as well as by pronounced clonal heterogeneity and tumor evolution, which require the development of individualized patient-specific primers for each patient [9,27]. To overcome these limitations, NGS technologies were introduced. Immunoglobulin gene sequencing (Ig-HTS, for example clonoSEQ) is based on identification of a unique clonal V(D)J rearrangement that serves as a specific molecular marker of tumor cells. This platform has been approved by the FDA for MRD monitoring in chronic lymphocytic leukemia and multiple myeloma. However, the clinical informativeness of this method in DLBCL is limited because the high frequency of somatic hypermutation in primer-binding regions prevents identification of the tumor clonotype in approximately 20% of patients [9,27,28].

Limitations of amplicon-based approaches were overcome with the introduction of targeted hybrid-capture enrichment methods, an example of which is CAPP-Seq (Cancer Personalized Profiling by deep Sequencing). The use of disease-specific probes enables selective enrichment of genomic regions with recurrent mutations and simultaneous analysis of a broad spectrum of genes, thereby increasing analytical sensitivity through multiplex accounting of mutational markers [27,28]. Owing to error-correction technologies such as integrated digital error suppression (iDES), CAPP-Seq achieves analytical sensitivity of approximately 0.002%, allowing detection of individual mutant molecules against a substantial excess of normal DNA [27,28,29]. An alternative approach, cSMART (circulating single-molecule amplification and re-sequencing technology), used by Zhang et al., is characterized by high analytical accuracy through unique identification of each molecule. The method demonstrates high concordance with tissue biopsy results and can identify additional mutational variants, reducing the impact of spatial tumor heterogeneity [30].

Further technological development is aimed at increasing analytical sensitivity required for detecting minimal residual disease at different stages of therapy. Duplex sequencing provides sensitivity down to 0.0002%; however, its application is limited by low efficiency of duplex molecule formation [27]. One of the most promising methods is PhasED-Seq (Phased Variant Enrichment and Detection Sequencing). This approach is based on detection of phased variants, defined as two or more somatic mutations located in cis on the same DNA molecule, which is characteristic of B-cell lymphomas due to aberrant activity of activation-induced cytidine deaminase. This method reduces both technical and biological background noise and can achieve a limit of detection of up to 1 mutant molecule per 2 million molecules (0.00005%) while preserving high molecular yield, opening new perspectives for relapse-risk stratification [27,28,31].

Thus, broad targeted sequencing panels (CAPP-Seq, cSMART) are appropriate for initial genotyping and molecular classification of DLBCL subtypes. In contrast, ultrasensitive MRD monitoring favors technologies with minimal background signal that use multiple phased variants to increase detection specificity at extremely low ctDNA concentrations (PhasED-Seq) [27,31].

In the context of clinical application, ctDNA analysis methods are classified into two conceptually distinct approaches depending on whether prior tumor material analysis is required: tumor-informed and tumor-agnostic approaches (Table 1) [32,33,34]. This distinction determines not only analytical characteristics but also the scalability of the method in routine practice.

Table 1. Comparative characteristics of tumor-informed and tumor-agnostic approaches to ctDNA analysis.

Characteristic	Tumor-informed approach	Tumor-agnostic approach
Principle	Personalized panel based on variants identified by sequencing tumor tissue	Fixed gene panel covering genes recurrently mutated in the given nosology
Example platforms	Signatera(TM), PhasED-Seq	CAPP-Seq, cSMART, targeted NGS panels
Need for baseline tumor material	Required (prior identification of tumor-specific variants is necessary)	No
Analytical sensitivity	High (usually <=10-4; in selected technologies, down to 10-6)	Variable (usually 10-3-10-4; with error-corrected panels, down to 10-5)
Possibility of de novo genotyping	No	Yes
Optimal application	MRD monitoring; early relapse detection	Initial genotyping; dynamic monitoring
Scalability	Limited	High
Turnaround time	Longer	Standard
Key limitations	Unavailable when archival material is lacking; does not detect variants arising de novo during clonal evolution	High background noise; risk of false-positive results due to clonal hematopoiesis of indeterminate potential

Table 1. Comparative characteristics of tumor-informed and tumor-agnostic approaches to ctDNA analysis.

Characteristic	Tumor-informed approach	Tumor-agnostic approach
Principle	Personalized panel based on variants identified by sequencing tumor tissue	Fixed gene panel covering genes recurrently mutated in the given nosology
Example platforms	Signatera(TM), PhasED-Seq	CAPP-Seq, cSMART, targeted NGS panels
Need for baseline tumor material	Required (prior identification of tumor-specific variants is necessary)	No
Analytical sensitivity	High (usually <=10-4; in selected technologies, down to 10-6)	Variable (usually 10-3-10-4; with error-corrected panels, down to 10-5)
Possibility of de novo genotyping	No	Yes
Optimal application	MRD monitoring; early relapse detection	Initial genotyping; dynamic monitoring
Scalability	Limited	High
Turnaround time	Longer	Standard
Key limitations	Unavailable when archival material is lacking; does not detect variants arising de novo during clonal evolution	High background noise; risk of false-positive results due to clonal hematopoiesis of indeterminate potential

Table 2. Comparative characteristics of ctDNA detection methods in DLBCL.

Method	Analytical sensitivity	Analytical specificity	Advantages	Limitations	Level of clinical validation	Source
ddPCR	10-4	High	Absolute quantitative assessment, cost-effectiveness; monitoring of known recurrent mutations (MYD88 L265P, etc.).	Low multiplexing capacity (single loci); inability to perform comprehensive genotyping; high sensitivity to contamination	Limited clinical validation in DLBCL; used exclusively for targeted monitoring of previously identified molecular targets	[35,36]
Ig-HTS	~10-6	High	Standardized monitoring of V(D)J rearrangements.	Requires baseline tumor material to identify a clonal marker; the clonal reporter is not identified in all patients; somatic hypermutation and low cfDNA input may reduce analytical efficiency	Moderate clinical validation in DLBCL without broad implementation in practice	[10,37,38]
CAPP-Seq	up to 10-6	High	Multiplex analysis; simultaneous molecular profiling and assessment of ctDNA/MRD kinetics. Integration of iDES minimizes sequencing background noise	Laboratory workflow and bioinformatic analysis are labor-intensive. When ultra-low VAFs are detected, preanalytical quality and complex filtering of biological background noise are critical.	High validation for prognosis assessment and dynamic monitoring in DLBCL, but not yet integrated into routine clinical standards	[29,39]
cSMART	~10-6	High	Single-molecule barcoding based on NGS; well suited for non-invasive plasma genotyping (validated mainly in solid tumor models)	Limited clinical validation and absence of standardized protocols for use in DLBCL	Limited clinical validation in DLBCL; used mainly within research protocols	[40,41]
PhasED-Seq	~10-6 to ~5 x 10-7	High	Detection of linked (phased) variants providing ultrahigh sensitivity. Clinical significance of ctDNA-MRD monitoring in large B-cell lymphomas has been demonstrated in prospective studies.	Requires panel customization and baseline tumor material to identify phased variants. Bioinformatic analysis and laboratory logistics are highly complex.	High clinical validity in DLBCL with inclusion in NCCN recommendations	[23,31,42,43]

Table 2. Comparative characteristics of ctDNA detection methods in DLBCL.

Method	Analytical sensitivity	Analytical specificity	Advantages	Limitations	Level of clinical validation	Source
ddPCR	10-4	High	Absolute quantitative assessment, cost-effectiveness; monitoring of known recurrent mutations (MYD88 L265P, etc.).	Low multiplexing capacity (single loci); inability to perform comprehensive genotyping; high sensitivity to contamination	Limited clinical validation in DLBCL; used exclusively for targeted monitoring of previously identified molecular targets	[35,36]
Ig-HTS	~10-6	High	Standardized monitoring of V(D)J rearrangements.	Requires baseline tumor material to identify a clonal marker; the clonal reporter is not identified in all patients; somatic hypermutation and low cfDNA input may reduce analytical efficiency	Moderate clinical validation in DLBCL without broad implementation in practice	[10,37,38]
CAPP-Seq	up to 10-6	High	Multiplex analysis; simultaneous molecular profiling and assessment of ctDNA/MRD kinetics. Integration of iDES minimizes sequencing background noise	Laboratory workflow and bioinformatic analysis are labor-intensive. When ultra-low VAFs are detected, preanalytical quality and complex filtering of biological background noise are critical.	High validation for prognosis assessment and dynamic monitoring in DLBCL, but not yet integrated into routine clinical standards	[29,39]
cSMART	~10-6	High	Single-molecule barcoding based on NGS; well suited for non-invasive plasma genotyping (validated mainly in solid tumor models)	Limited clinical validation and absence of standardized protocols for use in DLBCL	Limited clinical validation in DLBCL; used mainly within research protocols	[40,41]
PhasED-Seq	~10-6 to ~5 x 10-7	High	Detection of linked (phased) variants providing ultrahigh sensitivity. Clinical significance of ctDNA-MRD monitoring in large B-cell lymphomas has been demonstrated in prospective studies.	Requires panel customization and baseline tumor material to identify phased variants. Bioinformatic analysis and laboratory logistics are highly complex.	High clinical validity in DLBCL with inclusion in NCCN recommendations	[23,31,42,43]

3.3. Concordance of ctDNA and Tumor Biopsy Mutational Profiles

Detection of ctDNA in the plasma of patients with DLBCL provided the basis for studying the concordance between ctDNA mutations and tumor biopsy findings in DLBCL.

Studies evaluating concordance between ctDNA and tumor tissue mutational profiles show that liquid biopsy can reproduce key genetic characteristics of DLBCL while simultaneously detecting additional aberrations inaccessible through analysis of tissue biopsy. In a prospective single-center study led by Song G-Y, paired formalin-fixed paraffin-embedded tumor tissue samples and plasma ctDNA were analyzed in 52 patients with newly diagnosed DLBCL using an NGS panel comprising 112 genes. Mutation-level concordance was 74.7% (R2=0.276), while ctDNA analysis identified molecular aberrations in 98.1% of cases and detected key driver mutations at a higher frequency. In addition, the proportion of unique mutations detected exclusively in plasma was 60%, confirming pronounced molecular genetic heterogeneity [44].

Recent studies using the LymphGen genomic classification confirm high concordance between ctDNA and tumor tissue in defining molecular DLBCL subtypes and underscore the role of ctDNA in overcoming unclassifiable cases and spatial heterogeneity. In a recently published study by Moia et al. that included 166 patients with newly diagnosed DLBCL, concordance between NGS results from ctDNA and tumor material for verification of molecular subtype according to the LymphGen genomic classification was 95.8% [45]. In a large study including 443 patients with DLBCL, the mutational landscape of ctDNA and tumor substrate was compared, with a median concordance of 82%. Notably, in 45.9% of cases, ctDNA mutational profiling identified a LymphGen DLBCL subtype that was unclassifiable based on tumor biopsy sequencing [46]. In 2023, Yi Xia et al. performed targeted sequencing of ctDNA and tumor genomic DNA in 104 patients with DLBCL followed by LymphGen molecular classification. The results demonstrated the added diagnostic value of liquid biopsy relative to tissue analysis: in the paired cohort (n=62), 43.5% of patients (27/62) had molecular genetic variants detected exclusively in ctDNA and absent on sequencing of the tumor biopsy. In particular, the MCD subtype was identified solely by plasma analysis in 35.3% of cases (6/17). The authors attributed this phenomenon to spatial tumor heterogeneity and regarded molecular ctDNA profiling as complementary to conventional tissue analysis [47]. Multiple studies confirm that, in the setting of spatial heterogeneity, ctDNA analysis is a more informative tool for assessing the molecular disease profile than analysis of material obtained by biopsy from a single site [9,11,23,48,49].

In summary, concordance between ctDNA and tumor tissue mutational profiles varies substantially depending on the analytical platform used and the composition of the targeted panel, which must be considered when interpreting results. At the same time, the available evidence supports ctDNA as a clinically meaningful alternative to tumor biopsy for molecular profiling of DLBCL, with added value in the setting of pronounced spatial tumor heterogeneity. On this basis, integration of ctDNA mutational analysis into molecular profiling and risk-stratification algorithms in patients with DLBCL appears scientifically justified, but requires further standardization and prospective clinical validation.

3.4. Clinical Significance of ctDNA in DLBCL

Before ctDNA can be effectively used as an indicator of therapeutic efficacy, a prognostic factor, and a tool for assessing minimal residual disease, its baseline quantitative characteristics must be determined before initiation of antitumor therapy.

At present, tumor burden is commonly classified as high or low according to ctDNA concentration. In the foundational study by Kurtz D.M. et al., a risk-stratification method based on baseline ctDNA concentration was developed. Through statistical analysis, the investigators identified an optimal cutoff of 2.5 log haploid genome equivalents per milliliter of plasma (2.5 log hGE/mL). The authors found that exceeding this baseline ctDNA level was significantly associated with inferior event-free survival both during first-line therapy (HR 2.6; p=0.007) and during treatment of relapsed disease (HR 2.9; p=0.01), independently of the International Prognostic Index, DLBCL molecular subtype, and total metabolic tumor volume in multivariable analysis [50]. Similar results were obtained in the study by Li M. et al. With a ctDNA detection rate of 98% (51/52), statistically significant correlations were shown with disease stage (p=0.0107), baseline LDH activity (p=0.0044), International Prognostic Index (p=0.0053), and B symptoms (p=0.0007), supporting ctDNA as an alternative indicator of tumor burden. Further analysis demonstrated that patients with high ctDNA levels had significantly inferior progression-free survival (PFS) (median survival 9.5 months versus not reached; HR 2.298; 95% confidence interval [CI] 1.014-5.207; p=0.04) and overall survival (OS) (median survival 36.8 months versus not reached; HR 3.474; 95% CI 1.119-10.79; p=0.022). In multivariable analysis in that study, baseline ctDNA level was the only independent prognostic factor for both PFS (HR 2.47; 95% CI 1.35-4.5; p=0.004) and OS (HR 2.49; 95% CI 1.238-5.0; p=0.011). It should be emphasized that detailed prognostic stratification is critically necessary for patients with advanced-stage disease because of the high heterogeneity of clinical outcomes in this category. Thus, in the analyzed study, patients with advanced-stage disease and high baseline ctDNA levels had significantly inferior PFS (median survival 5.1 months versus 18.7 months; HR 2.296; 95% CI 1.014-5.199; p=0.04) and OS (median survival 26.0 months versus not reached; HR 2.971; 95% CI 1.012-8.723; p=0.038) compared with patients with low baseline ctDNA levels. The authors concluded that initial ctDNA concentration is a convincing independent biomarker for risk stratification and prediction of clinical outcomes in patients with DLBCL [51]. This pattern was also observed in another study led by Narkhede M., in which a personalized approach yielded a ctDNA detection rate of 95% (39/41). In addition, further analysis showed that median ctDNA concentrations differed fundamentally depending on disease stage, with very high levels in advanced forms compared with localized disease. High baseline ctDNA levels correlated with advanced disease stage (I-II versus III-IV; p=0.0007) and adverse R-IPI group (0-2 versus 3-5; p=0.0008). However, no statistically significant association with event-free or overall survival was observed in that study, most likely because of the limited sample size [52]. A 2025 systematic review and meta-analysis including 53 studies found a clear prognostic association between elevated baseline ctDNA and a high risk of DLBCL progression (HR 2.50; 95% CI 2.15-2.90) [13].

Thus, the literature analysis convincingly demonstrates that baseline ctDNA level is a reliable biomarker of tumor burden. High baseline ctDNA concentration is associated with adverse clinical and immunomorphological parameters of DLBCL and enables individualized prediction of early progression and reduced OS. This supports inclusion of quantitative ctDNA assessment in initial risk-stratification algorithms for patients with DLBCL. At the same time, it should be recognized that threshold values defining high ctDNA tumor burden are not fully standardized and depend primarily on the platform used, the quantitative reporting format, and the characteristics of the included cohort, limiting direct comparability of results across studies.

Quantitative monitoring of ctDNA dynamics during immunochemotherapy represents a modern basis for implementing response-adapted therapy in DLBCL, with interim kinetics of tumor clone eradication being a key component. A pioneering work in this field demonstrating the potential of ctDNA-guided treatment within a response-adapted DLBCL concept was the study by Kurtz D.M. [50]. Using CAPP-Seq, the authors showed that a 2-log reduction in ctDNA level after the first treatment cycle, defined as early molecular response (EMR), and a 2.5-log reduction after the second cycle, defined as major molecular response (MMR), stratified clinical outcomes. Patients who achieved EMR or MMR had significantly higher 2-year event-free survival. In the EMR group, EFS was 83%, compared with 50% among non-responders (p=0.0015). In the group with documented MMR, EFS was 82% versus 46%, respectively (p<0.001). The prognostic value of EMR and MMR was convincingly confirmed in a second independent validation cohort. Thus, Kurtz D.M. was the first to demonstrate the prognostic potential of ctDNA detection in DLBCL.

These findings were confirmed in subsequent studies. In the same study by Li M. et al., patients with a documented molecular response had more favorable PFS compared with the group that did not achieve molecular response. Median PFS was not reached in patients with molecular response, whereas it was 6.25 months in non-responders (HR 5.348; p=0.0015). Similar results were observed for OS (median survival not reached versus 25.87 months; HR 4.0; p=0.028). In addition, assessment of ctDNA elimination at an interim treatment stage may represent a potential non-invasive alternative to PET/CT for prognostic evaluation of PFS (HR 3.65; p=0.0033) and OS (HR 3.536; p=0.016) [51].

Alcoceba M. et al. evaluated the prognostic role of ctDNA dynamics. According to study criteria, 38 patients achieved a major molecular response, demonstrating a 2.5-log reduction in ctDNA after two induction cycles of antitumor therapy, whereas 6 patients did not achieve such response. Comparative PFS analysis showed a statistically significant difference: 2-year PFS was 76% among patients achieving major molecular response versus 0% among non-responders (p<0.001). Additional inclusion of structural variants to assess qualitative molecular response allowed the authors to expand the cohort by 6 cases. With this parameter considered, patients were reclassified. Achievement of qualitative molecular response was shown to be significantly associated with favorable prognosis: 2-year PFS was 80% versus 10% in the non-response group (p<0.001) [53]. The largest confirmation of the clinical significance of ctDNA monitoring was presented by Roschewski M. et al., who performed an integrative analysis of data from 5 prospective first-line studies and evaluated MRD kinetics by ctDNA level in 137 patients with large B-cell lymphomas. Undetectable ctDNA was achieved in 55% of patients after 2 induction cycles and in 78% by the end of systemic therapy. Two-year PFS was 96% in patients with undetectable ctDNA after 2 cycles versus 67% in those with persistent ctDNA detection (HR 6.9; p=0.0025). After completion of therapy, 2-year PFS was 29% in patients with ctDNA-MRD-positive disease versus 97% in those with MRD-negative remission (HR 28.7; p<0.0001) [54].

Soscia et al. studied the prognostic significance of MRD monitoring through analysis of immunoglobulin gene rearrangements in ctDNA in 73 patients with DLBCL. Interim assessment of molecular response enabled stratification of patients by risk of progression. At a median follow-up of 40 months, PFS was 78.1% in the group that achieved molecular response compared with 30.8% in the group with detectable ctDNA (p<0.0001). The ability of the method to refine prognosis in patients with partial response according to imaging deserves special attention: ctDNA status significantly differentiated this heterogeneous group into favorable- and unfavorable-prognosis subgroups (p=0.018) [55].

The available data indicate that MRD status at the end of therapy is decisive for verification of sustained deep remission. ctDNA positivity after completion of therapy is considered one of the most important molecular markers of residual disease persistence and, in several studies, demonstrates greater prognostic accuracy than standard imaging methods. Support for this comes from a systematic review and meta-analysis by Majid et al., which included data from 490 patients. The meta-analysis demonstrated a statistically significant association between ctDNA persistence after therapy and a substantially higher risk of disease progression. This association was statistically robust despite pronounced heterogeneity among the included studies (I2=75%) and, importantly, was evident even among patients who achieved complete metabolic response on PET/CT [56]. The clinical superiority of molecular over metabolic stratification was shown in the prospective study by Sworder et al. Use of the PhasED-Seq platform revealed a marked separation, in which the prognostic value of PET/CT was attenuated in the presence of ctDNA status, underscoring the high sensitivity of ctDNA for detecting persistent disease in patients with complete metabolic response [57]. Integration of ctDNA-MRD assessment with PET/CT findings after completion of therapy allows four clinical scenarios to be defined, each implying a differentiated management strategy [58]. It should be emphasized that the proposed algorithm is conceptual and requires validation in prospective interventional studies (Table 3).

Table 3. Clinical scenarios and management strategies for patients with DLBCL based on integration of PET/CT results and ctDNA molecular response after completion of therapy.

PET/CT	Molecular response by ctDNA	Interpretation	Management strategy	Literature source
-	MRD -	1. Deep molecular-metabolic remission 2. Low risk of relapse	1. Standard dynamic follow-up 2. De-escalation of PET/CT monitoring frequency may be considered	[54,65]
+	MRD -	1. Likely false-positive metabolic activity 2. Relatively favorable prognosis	1. Repeat PET/CT scan 2. Dynamic ctDNA monitoring	[54,66]
-	MRD +	1. Minimal residual disease in metabolic remission 2. High risk of early relapse	1. Intensive ctDNA monitoring every 3 months 2. Consider enrollment in a clinical trial of early intervention	[54]
+	MRD +	Unfavorable prognosis	1. Immunomorphological verification 2. Planning of salvage therapy (HDCT with auto-HSCT, CAR T-cell therapy, or bispecific antibodies)	[54,67]

Table 3. Clinical scenarios and management strategies for patients with DLBCL based on integration of PET/CT results and ctDNA molecular response after completion of therapy.

PET/CT	Molecular response by ctDNA	Interpretation	Management strategy	Literature source
-	MRD -	1. Deep molecular-metabolic remission 2. Low risk of relapse	1. Standard dynamic follow-up 2. De-escalation of PET/CT monitoring frequency may be considered	[54,65]
+	MRD -	1. Likely false-positive metabolic activity 2. Relatively favorable prognosis	1. Repeat PET/CT scan 2. Dynamic ctDNA monitoring	[54,66]
-	MRD +	1. Minimal residual disease in metabolic remission 2. High risk of early relapse	1. Intensive ctDNA monitoring every 3 months 2. Consider enrollment in a clinical trial of early intervention	[54]
+	MRD +	Unfavorable prognosis	1. Immunomorphological verification 2. Planning of salvage therapy (HDCT with auto-HSCT, CAR T-cell therapy, or bispecific antibodies)	[54,67]

In addition to assessing MRD status immediately after therapy completion, dynamic molecular monitoring during remission is a promising direction aimed at detecting relapse at the molecular level. The seminal work in this field was a retrospective study of prospectively collected samples by Roschewski M. et al., in which serial ctDNA analysis by Ig-HTS in serum from patients with DLBCL in remission demonstrated the ability to detect molecular relapse a median of 3.5 months before the emergence of clinical and radiological signs of disease progression. Detection of ctDNA during follow-up was associated with an extremely high risk of subsequent clinical relapse (HR 228; positive predictive value 88%; negative predictive value 98%) [59]. Additional value of serial molecular monitoring was demonstrated by Scherer F. et al., who showed that ctDNA analysis by CAPP-Seq can identify patterns of clonal evolution and the emergence of new genomic aberrations that may determine resistance mechanisms and justify personalized selection of salvage therapy [11]. In summary, dynamic ctDNA-based molecular monitoring provides a significant temporal advantage over standard imaging methods in relapse detection. This conceptually supports a transition toward early pre-emptive therapy strategies at the preclinical stage of progression. However, the optimal frequency and duration of molecular monitoring, as well as the clinical appropriateness of pre-emptive therapy based on isolated molecular relapse, remain subjects of ongoing research.

At present, the ctDNA-guided treatment paradigm, as a component of response-adapted strategy, is being actively incorporated into the design of prospective studies aimed at evaluating the possibility of modifying treatment intensity under dynamic ctDNA control [60].

Table 4. Key clinical studies evaluating circulating tumor DNA in diffuse large B-cell lymphoma.

Study (year)	Design	N patients	ctDNA detection method	Tumor burden/response assessment	Key results	Practical conclusion
Kurtz DM et al. (2018)	Prospective cohorts (first-line and salvage)	217	CAPP-Seq	High burden: >2.5 log hGE/mL; early molecular response (EMR): >=2-log reduction by cycle 1; major molecular response (MMR): >=2.5-log reduction by cycle 2	High baseline ctDNA level was associated with inferior EFS (HR 2.6; p=0.007 for first-line therapy; HR 2.9; p=0.01 for salvage therapy). Achievement of EMR/MMR stratified outcomes	Baseline concentration is a surrogate marker of tumor burden; early kinetics (EMR/MMR) allow assessment of treatment sensitivity and risk stratification.
Li M et al. (2022)	Real-world cohort (first-line)	52	Targeted NGS	High level: > median (2.44 log hGE/mL); clearance after 2-4 cycles	High baseline level was an independent predictor of inferior PFS (HR 2.47; p=0.004) and OS (HR 2.49; p=0.011). In stage III-IV disease, it identified the poorest-prognosis group (PFS 5.1 vs 18.7 months). Clearance after 4 cycles predicted PFS (HR 3.65; p=0.003) and OS (HR 3.536; p=0.016), complementing PET.	Baseline level is important for stratification, especially in advanced stages; interim monitoring is a non-invasive alternative/complement to PET.
Narkhede M et al. (2024)	Retrospective analysis of a personalized assay	41	Signatera(TM) commercial platform	Pre-treatment detection; clearance during therapy	ctDNA was detected in 95%. Baseline level correlated with stage (p=0.0007) and R-IPI (p=0.0008). Clearance was associated with improved EFS (HR 6.5; p=0.003) and OS (HR 22; p=0.005). Clearance preceded PET by a mean of 97 days.	A tumor-informed approach enables assessment of disease extent; ctDNA clearance is an early marker of therapeutic efficacy.
Alcoceba M et al. (2024)	Single-center prospective first-line study	44	Targeted NGS (single-nucleotide and structural variants)	Major molecular response (>=2.5-log reduction after 2 cycles); qualitative response including structural variants	Major molecular response was achieved in 38 patients and not achieved in 6; 2-year PFS was 76% vs 0% (p<0.001). Qualitative response: 80% vs 10% (p<0.001). In multivariable analysis, major molecular response and PET DeltaSUVmax reduction >66% were independent predictors.	ctDNA kinetics after 2 cycles is a powerful early prognostic marker; combined analysis with PET improves risk-stratification accuracy.
Roschewski M et al. (2025)	Integrative analysis of 5 prospective first-line studies	137	PhasED-Seq (phased variants)	ctDNA clearance after 2 cycles and at end of therapy	ctDNA elimination after 2 cycles: 2-year PFS 96% vs 67% (HR 6.9; p=0.0025). End of therapy: 97% vs 29% (HR 28.7; p<0.0001). ctDNA-MRD assessment outperformed and complemented PET in discordant cases.	ctDNA-MRD status after therapy is the most powerful prognostic factor for confirming deep remission; it helps resolve false-positive PET findings.
Soscia R et al. (2025)	Retrospective observational study	73	Ig-HTS (Ig rearrangements)	MRD at interim time points and after therapy	2-year PFS was 78.1% in MRD-negative patients vs 30.8% in MRD-positive patients (p<0.0001). After therapy, MRD persistence in 21.3% of patients was associated with 100% relapse risk (PPV 100%, NPV 84%). MRD assessment refined prognosis in patients with partial response on CT (p=0.018).	Ig-HTS MRD assessment is effective for risk stratification, especially after completion of therapy, and is valuable when imaging results are indeterminate.
Roschewski M et al. (2015)	Prospective post-therapy monitoring	126	Ig-HTS (serum)	Emergence of ctDNA during follow-up	ctDNA detection was associated with relapse risk (HR 228); PPV 88%, NPV 98%. Molecular relapse was detected a median of 3.5 months before clinical manifestations.	ctDNA enables detection of relapse long before clinical and radiological signs.

Table 4. Key clinical studies evaluating circulating tumor DNA in diffuse large B-cell lymphoma.

Study (year)	Design	N patients	ctDNA detection method	Tumor burden/response assessment	Key results	Practical conclusion
Kurtz DM et al. (2018)	Prospective cohorts (first-line and salvage)	217	CAPP-Seq	High burden: >2.5 log hGE/mL; early molecular response (EMR): >=2-log reduction by cycle 1; major molecular response (MMR): >=2.5-log reduction by cycle 2	High baseline ctDNA level was associated with inferior EFS (HR 2.6; p=0.007 for first-line therapy; HR 2.9; p=0.01 for salvage therapy). Achievement of EMR/MMR stratified outcomes	Baseline concentration is a surrogate marker of tumor burden; early kinetics (EMR/MMR) allow assessment of treatment sensitivity and risk stratification.
Li M et al. (2022)	Real-world cohort (first-line)	52	Targeted NGS	High level: > median (2.44 log hGE/mL); clearance after 2-4 cycles	High baseline level was an independent predictor of inferior PFS (HR 2.47; p=0.004) and OS (HR 2.49; p=0.011). In stage III-IV disease, it identified the poorest-prognosis group (PFS 5.1 vs 18.7 months). Clearance after 4 cycles predicted PFS (HR 3.65; p=0.003) and OS (HR 3.536; p=0.016), complementing PET.	Baseline level is important for stratification, especially in advanced stages; interim monitoring is a non-invasive alternative/complement to PET.
Narkhede M et al. (2024)	Retrospective analysis of a personalized assay	41	Signatera(TM) commercial platform	Pre-treatment detection; clearance during therapy	ctDNA was detected in 95%. Baseline level correlated with stage (p=0.0007) and R-IPI (p=0.0008). Clearance was associated with improved EFS (HR 6.5; p=0.003) and OS (HR 22; p=0.005). Clearance preceded PET by a mean of 97 days.	A tumor-informed approach enables assessment of disease extent; ctDNA clearance is an early marker of therapeutic efficacy.
Alcoceba M et al. (2024)	Single-center prospective first-line study	44	Targeted NGS (single-nucleotide and structural variants)	Major molecular response (>=2.5-log reduction after 2 cycles); qualitative response including structural variants	Major molecular response was achieved in 38 patients and not achieved in 6; 2-year PFS was 76% vs 0% (p<0.001). Qualitative response: 80% vs 10% (p<0.001). In multivariable analysis, major molecular response and PET DeltaSUVmax reduction >66% were independent predictors.	ctDNA kinetics after 2 cycles is a powerful early prognostic marker; combined analysis with PET improves risk-stratification accuracy.
Roschewski M et al. (2025)	Integrative analysis of 5 prospective first-line studies	137	PhasED-Seq (phased variants)	ctDNA clearance after 2 cycles and at end of therapy	ctDNA elimination after 2 cycles: 2-year PFS 96% vs 67% (HR 6.9; p=0.0025). End of therapy: 97% vs 29% (HR 28.7; p<0.0001). ctDNA-MRD assessment outperformed and complemented PET in discordant cases.	ctDNA-MRD status after therapy is the most powerful prognostic factor for confirming deep remission; it helps resolve false-positive PET findings.
Soscia R et al. (2025)	Retrospective observational study	73	Ig-HTS (Ig rearrangements)	MRD at interim time points and after therapy	2-year PFS was 78.1% in MRD-negative patients vs 30.8% in MRD-positive patients (p<0.0001). After therapy, MRD persistence in 21.3% of patients was associated with 100% relapse risk (PPV 100%, NPV 84%). MRD assessment refined prognosis in patients with partial response on CT (p=0.018).	Ig-HTS MRD assessment is effective for risk stratification, especially after completion of therapy, and is valuable when imaging results are indeterminate.
Roschewski M et al. (2015)	Prospective post-therapy monitoring	126	Ig-HTS (serum)	Emergence of ctDNA during follow-up	ctDNA detection was associated with relapse risk (HR 228); PPV 88%, NPV 98%. Molecular relapse was detected a median of 3.5 months before clinical manifestations.	ctDNA enables detection of relapse long before clinical and radiological signs.

4. Limitations and Future Perspectives

Despite rapid advances in circulating tumor DNA analysis technologies, clinical integration of ctDNA in DLBCL is still accompanied by several unresolved issues. First, a substantial proportion of available data has been obtained in observational, often single-center studies, whereas completed prospective interventional trials demonstrating improved clinical outcomes when therapy is modified according to molecular status are currently lacking. Thus, at the current stage, ctDNA is primarily a highly sensitive prognostic biomarker; however, its predictive value as a tool for therapeutic decision-making requires further confirmation. High analytical sensitivity and prognostic accuracy of the method are not, by themselves, equivalent to proven clinical utility.

Insufficient standardization of the analytical workflow also remains a major limitation. Study results depend on preanalytical conditions, including choice of biomaterial, tube type, time to centrifugation, and sample-storage characteristics. Additional variability is introduced by differences between analytical platforms, targeted panel design, sequencing depth, error-correction algorithms, and bioinformatic filtering. Consequently, direct comparison of data obtained in different studies should be performed cautiously, and universal thresholds for defining high tumor burden and MRD positivity have not yet been fully standardized.

Clonal hematopoiesis of indeterminate potential represents a separate problem and is the main source of biological background noise in mutational ctDNA analysis [34,61,62]. Somatic mutations associated with clonal hematopoiesis, particularly in DNMT3A, TET2, ASXL1, and TP53, may be mistakenly interpreted as tumor-specific, leading to false-positive results [34,62,63]. This issue has particular clinical importance in DLBCL, given the median age of patients at diagnosis and the increasing prevalence of clonal hematopoiesis of indeterminate potential with age. Current approaches to minimizing the impact of clonal hematopoiesis include parallel sequencing of leukocyte DNA to filter germline variants and variants associated with clonal hematopoiesis of indeterminate potential, as well as analytical strategies focused on DLBCL-specific markers, such as immunoglobulin gene rearrangements and phased variants, which by definition are not characteristic of clonal hematopoiesis [10,31,64].

Additional barriers to broad implementation include the cost of highly sensitive NGS platforms, the need for advanced laboratory infrastructure, and the high complexity of bioinformatic interpretation. For a number of methods, including PhasED-Seq and other tumor-informed approaches, dependence on baseline tumor material and the need for individualized panel customization remain, limiting scalability in routine practice.

Future development in this field is likely to be linked to several key directions. The primary task is unification of preanalytical and analytical protocols with establishment of agreed criteria for result interpretation. Interventional studies evaluating the possibility of treatment de-escalation in patients with deep molecular response and, conversely, early intensification in patients with persistent ctDNA are of no less practical interest. Finally, further integration of mutational profiling with epigenetic features, fragmentomics, and imaging data may improve the accuracy of risk stratification and bring truly personalized management algorithms for patients with DLBCL closer to implementation.

Thus, ctDNA is currently one of the most promising tools for molecular monitoring in DLBCL; however, its transition from the research and prognostic domain to a full component of routine clinical decision-making requires further standardization, validation, and proof of practical utility.

5. Conclusions

In recent years, circulating tumor DNA has established a new paradigm of molecular monitoring in DLBCL, integrating opportunities for non-invasive genotyping, dynamic assessment of tumor burden, and highly sensitive detection of minimal residual disease. Modern analytical platforms, from targeted methods to tumor-agnostic approaches, have substantially expanded the potential of liquid biopsy and demonstrated consistent prognostic significance at key treatment stages: before therapy, during early monitoring, and after completion of treatment. The clinical value of baseline ctDNA level, early molecular elimination of the tumor clone, and MRD status has been confirmed in multiple cohort studies, and post-therapy ctDNA assessment has already been reflected in NCCN recommendations. At the same time, the transition of ctDNA from a prognostic domain to a tool that directly determines therapeutic strategy remains the major unresolved issue in this field and requires results from ongoing interventional studies.