Circulating miR-148a-3p Correlates with Inadequate Induction Response in Pediatric Hodgkin Lymphoma

Introduction

Pediatric classical Hodgkin’s lymphoma (cHL) is highly curable, with an estimated five-year survival rate exceeding 90%.^1,2 Despite these favorable outcomes, approximately one-third of patients in the EuroNet-PHL trials failed to achieve an adequate metabolic response, measured by quantitative 18F-Fluorodeoxyglucose-Positron Emission Tomography (18F-FDG-PET), after two cycles of induction chemotherapy with vincristine, etoposide, prednisone, and doxorubicin (OEPA), necessitating intensified consolidation chemotherapy and radiotherapy to ensure durable remission.^1,2 However, the biological mechanisms driving these distinct responses to induction therapy are not well defined.

Circulating microRNAs (miRNAs) — around 22-nucleotide noncoding RNAs — have emerged as promising, minimally invasive biomarkers for cancer diagnosis, prognosis, and therapy monitoring³ due to their high stability in blood and regulatory role in gene expression.⁴ In classical Hodgkin’s lymphoma (HL), studies have identified aberrant miRNA expression in both tissue and cell-free samples that contributes to pathogenesis by influencing immune evasion, loss of B-cell phenotype, drug sensitivity and resistance, and activation of growth-promoting pathways via the regulation of target genes.^5,6 Based on this evidence, we hypothesized that specific circulating miRNA signatures may reflect tumor biology and can predict early therapeutic outcomes in pediatric patients.

Therefore, this study investigates whether serum miRNA expression at diagnosis correlates with the early metabolic response to induction chemotherapy in pediatric cHL.

Materials and Methods

To test this hypothesis, serum miRNA profiles were analyzed in a cohort of 35 newly diagnosed pediatric HL patients in the EuroNet-PHL-C2 trial (EudraCT: 2012–004053-88). Biobank serum samples were collected across multiple German centers with written informed consent from participants or legal guardians, processed within 24 h, and stored at −80 °C. Only samples without visible hemolysis were included. The cohort (ages 3.4–18 years; male:female ≈1.5:1) predominantly presented with stage II disease (60%), with stages III and IV each comprising 20%. Elevated ESR was observed in 57%, B‑symptoms in 40%, bulky disease in 43%, and extranodal involvement in 31% (Table  S1). These baseline characteristics were comparable to those observed in clinical studies.

All patients underwent OEPA chemotherapy followed by 18F-FDG-PET–based early response assessment (ERA).¹ Metabolic response was quantified using the lesion‑to‑liver activity ratio (qPET), applying a threshold of 1.3 to distinguish adequate from inadequate metabolic response.⁷ Patients with qPET <1.3 were classified as ERAqPET‑negative or adequate responders (AR) (n=13), and those with qPET ≥1.3 as ERAqPET‑positive or inadequate responders (IR) (n=22) (Table  S1).

Peripheral blood was collected at diagnosis (day 0) and post‑induction (day 57). Serum was isolated by double centrifugation to remove debris and platelets and stored at −80 °C. Total RNA, including miRNAs, was extracted using the miRNeasy Serum/Plasma Kit (Qiagen). Minimal hemolysis was confirmed (ΔCt miR‑23a–miR‑451a <5),⁸ and stable endogenous miRNA recovery was verified across samples (Figure  S1A–C).

Small RNA libraries were prepared using the QIAseq‑miRNA Library Kit (Qiagen), quantified by qRT‑PCR, and sequenced (single‑end 75 bp reads) on an Illumina NextSeq‑500 (Illumina, Inc). Reads were trimmed, aligned to GRCh37, and annotated using miRBase 20.⁹ On average, 488 miRNAs per sample were detected at ≥1 TPM (Figure  S2A), consistent with previous serum miRNA studies.¹⁰ Data were normalized using the trimmed mean of M‑values (TMM), and differential expression analysis was performed using edgeR (Bioconductor).¹¹ Principal component analysis (PCA) identified three technical outliers (using PCA score-plot: Hotelling T² test), which were excluded from downstream analyses (Figure  S2B).

Expression of miR‑148a‑3p was validated using qRT‑PCR (Qiagen). RNA was reverse‑transcribed and amplified according to manufacturer instructions, and expression was calculated using the 2⁻ΔΔCt method (Ct < 38; GeneGlobe, Qiagen). Values were normalized to the global Ct mean and compared with ERAqPET‑negative controls. Diagnostic performance was assessed using receiver operating characteristic (ROC) curve analysis, with area under ROC curve (AUROC) used to determine sensitivity and specificity. Statistical analyses were performed using Mann–Whitney U-tests, linear regression and Spearmann correlation, with data visualization conducted in R and GraphPad Prism 10 (San Diego, CA, USA). Statistical significance was defined as p < 0.05.

Results

Pre-induction serum sequencing identified 173 miRNAs with a TMM ≥50, of which 119 were reported in adult HL studies (Table S2). A comparison of these 119 miRNAs in the ERAqPET subgroups revealed 24 HL-associated miRNAs with divergent baseline expressions (Table 1). Among these, ten miRNAs—including miR-223-3p, miR-143-3p, miR-345-5p, miR-24-3p, miR-629-5p, miR-140-5p, miR-148a-3p, miR-190b, miR-181a-5p, and miR-27b-3p—were elevated at diagnosis in inadequate responders, whereas fourteen—miR-4732-3p, miR-144-3p, let-7e-5p, miR-103a-3p, let-7f-5p, miR-126-5p, miR-374a-5p, miR-26b-5p, miR-98-5p, miR-125a-5p, miR-142-3p, miR-26a-5p, miR-15a-5p, and miR-20a-5p—were downregulated in the ERAqPET-positive cohort. Unsupervised hierarchical clustering and PCA of these miRNAs distinctly separated adequate and inadequate responders (Figure 1A and B). Additional PCA projections using higher-order principal components (PC3 vs. PC4) demonstrated consistent group separation (Figure S3).

Table 1 Expression of Differentially Expressed miRNAs Between ERAqPET Negative and ERAqPET Positive Patients’ Samples Before Start of Induction Therapy

Figure 1 Baseline serum-miRNA profiling distinguishes ERAqPET-positive and ‑negative Hodgkin lymphoma patients and predicts post-Induction PET response via miR-148a-3p. (A) Heat map and hierarchical clustering by samples and miRNAs were performed on the listed samples (bottom) using the 24 differentially expressed miRNAs (Left) based on normalized TMM counts for each miRNA for each patient analyzed from qPET <1.3 (Red) and qPET ≥1.3 (Blue) groups. The colour scale shown (in Log fold) illustrates the relative expression level of the indicated miRNA across all samples: Blue indicates low expression of miRNA, and red indicates high expression of miRNA. (B) Principal component analysis plot for 24 significantly differentially expressed miRNAs in adequate responder (AR) and inadequate responder (IR) patients. Patient’s samples from ERAqPET negative <1.3 (Red dots) and ERAqPET positive ≥1.3 (Blue triangles) groups cluster separately, with a closer relationship of patients with each other. (C) Baseline miR-148a-3p expression at diagnosis (Day 0). Box plots display normalized read counts (TMM values) for miR-148a-3p. Boxes represent interquartile ranges (Tukey method), horizontal lines indicate medians, and whiskers show data ranges. Significant differential expression between groups is indicated (p < 0.05). (D) Linear regression shows association between pre-treatment miR-148a-3p expression and post-OEPA ERAqPET response in HL patients. Scatter plot show normalized expression (Log2-transformed TMM values) of miR-148a-3p (y-axis) versus ERAqPET scores (x-axis). Spearman’s rank correlation test was performed for statistical significance (p <0.05).

To prioritize candidates, we focused on miR-148a-3p, which exhibited the greatest expression difference between ERAqPET groups (Bonferroni-corrected p = 5.5×10⁻⁵) (Figure 1C). Linear regression revealed a significant positive correlation between baseline miR-148a-3p expression and qPET metabolic parameters (Spearman p = 1.43×10⁻⁵) (Figure 1D). After induction therapy, miR-148a-3p expression did not differ between the ERAqPET groups (Figure S4A). However, the longitudinal analysis showed significant shifts in change of expression between both the groups (Figure S4B), predominantly underscoring its predictive value at baseline.

qRT-PCR validation confirmed elevated baseline miR-148a-3p levels in inadequate responders. However, post-treatment differences were abrogated (Figure 2A), reinforcing its potential as a baseline biomarker. High ERAqPET values at diagnosis were associated with B symptoms (p = 0.017), extranodal lesions (p = 0.0009), elevated ESR ≥30 mm/h (p = 0.019), and bulky disease (p = 0.026); Figure 2B). qRT-PCR data for miR-148a-3p confirmed associations with B symptoms (p = 0.03) and extranodal lesions (p = 0.004), although ESR (p = 0.239) and bulky disease (p = 0.073) did not reach significance (Figure 2B).

Figure 2 qRT-PCR–validated serum miR-148a-3p and early PET response differentiate responders from non-responders to OEPA induction-therapy in Hodgkin lymphoma. (A) miR‑148a‑3p expression dynamics during induction therapy. Relative miR‑148a‑3p expression levels at diagnosis (Day 0) and after induction therapy (Day 57), normalized to the mean of the ERAqPET‑negative (AR) group. Violin plots show the distribution of values; violins indicate interquartile ranges (Tukey method), solid horizontal lines mark medians, and whiskers denote data ranges (dashed for violins). Statistically significant differences between groups (p < 0.05) are annotated. (B) Comparison analysis: Early response qPET and baseline miR‑148a‑3p expression across clinical subgroups. Upper panels: early response qPET values. Lower panels: baseline miR‑148a‑3p expression in the same subgroups, normalized to the mean of the ERAqPET‑negative (AR) group. Violin plots display relative expression changes. Boxes and violins represent interquartile ranges (Tukey method), solid horizontal lines indicate medians, and whiskers show data ranges (dashed for violins). Significant differences between groups (p < 0.05) are indicated. (C) Correlation analysis: Upper panels show association between treatment response (qPET) on the x-axis and clinical parameters on the y-axis and lower panels show association between baseline miR-148a-3p expression and corresponding clinical parameters. Correlation plots show relative miRNA expression (2^(-ΔΔCt) values) on the x-axis versus corresponding clinical values on the y-axis. Statistical significance was assessed using Spearman’s rank correlation test (p < 0.05). (D) Receiver operating characteristic (ROC) curve for miR-148a-3p levels. ROC analysis was performed using qRT-PCR data to evaluate the diagnostic performance of miR-148a-3p. The black solid line represents the experimental ROC curve; the red dashed line represents the identity line (AUC = 0.5). Area under the ROC curve (AUC), p-value, and sensitivity/specificity percentages at the selected threshold are indicated.

Linear regression analysis revealed significant positive associations between baseline miR‑148a‑3p expression and several clinical markers; these relationships were corroborated by Spearman’s rank correlation. Baseline miR‑148a‑3p correlated with serum CCL17 (thymus‑ and activation‑regulated chemokine, TARC) (p = 0.009), white blood cell (WBC) count (p = 0.0044), immunoglobulin A (p = 0.035), and immunoglobulin G (p = 0.006). No significant correlations were observed with C‑reactive protein (CRP) (p = 0.06) or immunoglobulin M (p = 0.055) (Figure 2C), highlighting its association with some selective inflammatory and immunological disease features. Box plot analysis confirmed that miR-148a-3p levels remained stable across age and sex. This indicates that the miR-148a-3p associated with ERAqPET status is independent of these demographic variables (Figure S5).

ROC analysis was performed to assess whether miR‑148a‑3p expression is associated with metabolic response following OEPA induction therapy. The curve demonstrated separation between groups (Figure 2D), with an AUROC of 0.823 (95% CI 0.677–0.967), a sensitivity of 0.57, and a specificity of 1.00 for distinguishing adequate responders from inadequate responders. The association was statistically significant (p = 0.0031), indicating that miR‑148a‑3p may serve to distinguish responders within this cohort.

Discussion

The current management of pediatric high-risk cHL requires precise approaches that integrate polychemotherapy and radiotherapy de-escalation^1,2 with novel immunotherapy agents to optimize efficacy and minimize long-term toxicity.¹² Although quantitative 18F-FDG-PET serves as the gold standard for response assessment,⁷ serum-based miRNA profiling has emerged as a promising adjunct for monitoring treatment responses in various malignancies.¹³

Our pilot study demonstrated that pre-treatment circulating miRNA signatures can discriminate between adequate and inadequate responders after two OEPA-cycles, effectively “rediscovering” several circulating and tissue-specific miRNAs previously implicated in adult cHL.^5,6 These results suggest that pretreatment miRNA profiles could complement early risk stratification.

Among the differentially expressed miRNAs, miR-148a-3p is of particular interest as it functions as either a tumor suppressor or an oncogene in a context-dependent manner.¹⁴ In various cancers, including cHL, miR-148a-3p is frequently downregulated by DNA methylation–mediated epigenetic silencing compared to non-malignant cells, which affects pathways involved in cell proliferation, migration, and invasion.^14,15

Paradoxically, our pediatric cohort exhibited elevated baseline serum miR-148a-3p levels, specifically among the inadequate responders. This may reflect the heterogeneous methylation patterns previously reported in cHL: Paczkowska et al documented methylation-induced silencing of miR-148a-3p in four of seven cHL cell lines and two of six primary Hodgkin and Reed/Sternberg (HRS) cell samples compared with non-malignant cells.¹⁵ Alternatively, high serum levels may originate from non-malignant sources, highlighting its potential role as a biomarker for aggressive, pro-tumorigenic disease. Although often downregulated in most cancers, miR-148a-3p is upregulated in glioma, osteosarcoma, and prostate cancer, where higher expression correlates with advanced stage and poorer prognosis.^16–18

Mechanistically, miR-148a-3p directly targets critical tumor suppressors, such as DNA methyltransferase 1 (DNMT1) and phosphatase and tensin homolog (PTEN), binding to their 3′-untranslated regions (UTRs) to inhibit translation.¹⁹ By suppressing DNMT1, miR-148a-3p induces global DNA hypomethylation, a recognized epigenetic hallmark of tumorigenesis.^20,21 Beyond epigenetic regulation, miR-148a-3p overexpression may facilitate HRS cell immune escape by downregulating β-microglobulin and impairing HLA class I expression, as demonstrated in colorectal tumors.²² Additionally, it regulates B cell differentiation via targets such as BACH2, MITF, BIM, and PTEN,²³ potentially linking it to the aberrant B cell transcriptional program characteristic of cHL.

Elevated baseline miR-148a-3p levels in our cohort were associated with higher CCL17/TARC levels, an established pediatric-cHL diagnostic marker,²⁴ as well as increased Immunoglobulin G and Immunoglobulin A titers, both of which are recognized features of Hodgkin lymphoma.^25,26 The association with B‑symptoms and extranodal involvement further supports its link to aggressive disease features. Importantly, age and sex did not appear to influence baseline miR-148a-3p expression, suggesting that the observed associations with metabolic response are unlikely to be driven by demographic confounding.

Furthermore, diagnostic performance analyses indicate an association between miR-148a-3p expression and metabolic response. While these results suggest that miR-148a-3p may help to distinguish responders from non-responders within this cohort, validation in larger, independent patient populations will be required before clinical applicability can be inferred. Future studies may explore whether integrating miR-148a-3p into broader panels of soluble biomarkers could enhance multi-parametric strategies for early risk assessment in pediatric cHL.

Conclusion

Our preliminary findings indicate that circulating miR-148a-3p levels at diagnosis are associated with metabolic tumor activity, inflammatory markers, and adverse clinical features in pediatric cHL. These observations suggest that miR-148a-3p may contribute to multi-parametric biomarker approaches aimed at the early identification of patients at risk for delayed metabolic remission. The decline of elevated pre-treatment levels following induction therapy is consistent with an association with disease burden, although its predictive relevance requires confirmation. Given the exploratory nature of this pilot study, validation in larger, independent cohorts will be necessary to define its biological and potential clinical utility.