Next-generation DNA methylation sequencing: loci and regions differentially methylated in adolescents with chronic postsurgical pain

Abstract

DNA methylation (DNAm) may link environmental exposures to pain susceptibility. We examined preoperative DNAm patterns (using next-generation enzymatic-methylation sequencing) associated with chronic postsurgical pain (CPSP), a major biopsychosocial problem in adolescents undergoing musculoskeletal surgeries, recruited across six sites (n = 166; 15.5 (IQR 14.3–16.8) years; 51% male). Pain and functional measures were followed longitudinally. CPSP was defined as pain>3/10 beyond two months post-surgery. We identified 289 differentially methylated loci (DML; HB p < 0.1) that satisfied robustness criteria after bootstrapping, and 20 regions (DMR) (FDR p < 0.05) associated with CPSP. Using CPSP-specific background, DMLs, annotated to 57 genes, were enriched for biological processes (Cellular response to cAMP (p = 0.001), neuronal action potential (p = 0.01), sodium ion transmembrane transport (p = 0.03), potassium ion transmembrane transport (P = 0.035) and regulation of postsynaptic membrane potential (P = 0.041)), molecular functions (Intracellular camp-activated cation channel activity (p = 0.001), cAMP binding (p = 0.003), voltage-gated sodium channel activity (p = 0.01), and voltage-gated potassium channel activity (p = 0.02)), and cellular components (HCN channel complex (p = 0.002), presynaptic active zone membrane (p = 0.006), and postsynaptic membrane (p = 0.049)). DMRs did not enrich gene pathways but showed transcription factor motif enrichment for stress response (TGF-β/SMAD), neuronal differentiation/plasticity (NR4A/TEAD), inflammation/immunity (STATs), and circadian signaling (RORA/RORB/NPAS2). Study findings suggest potential DNAm signatures associated with CPSP and generate hypotheses regarding their possible role in CPSP development.

Highlights

• •CPSP-associated DNAm signatures were identified in adolescents undergoing major surgery
• •289 loci associated with CPSP enriched ion-channel and neuronal pathways

1. Introduction

In the United States, nearly 5% of children undergo surgery annually, often experiencing postoperative pain (Rabbitts and Groenewald, 2020; Groenewald et al., 2022). While short-term postoperative pain is expected, pain lasting beyond the healing period (≥2–3 months) is pathological and considered chronic postsurgical pain (CPSP) (Schug et al., 2019). CPSP affects 11–60% of pediatric surgical patients, depending on surgery type (Dugan et al., 2022; Rosenbloom and Katz, 2022). Beyond its substantial economic burden CPSP can adversely affect the emotional, physical, and social well-being of both children and caregivers (Rosenbloom and Katz, 2022; Ellyson et al., 2022; Lo et al., 2022). Given its widespread impact, there is a clear need to develop personalized interventions. To enable this, there is a critical need to identify predictive biomarkers and the mechanisms underlying development of CPSP.

Although CPSP is genetically regulated (James, 2017; Young et al., 2012; Hocking et al., 2012), genetic association studies suffer from a lack of replicability (Kim et al., 2009) exposing critical gaps in our understanding of susceptibility (Katz, 2012). Bridging these gaps necessitates consideration of the role of environmental factors (such as parental care, socioeconomic status (SES) and psychosocial factors) on genetics, which can prime pain responses after surgery and the development of CPSP (Campbell et al., 2018; Dourson et al., 2022). Environmental factors can modify gene expression via epigenetic mechanisms such as DNA methylation (DNAm). DNAm, the addition of a methyl group to the 5′ position of cytosine at CpG sites, (a sequence of a cytosine (C) followed immediately by a guanine (G) nucleotide in a DNA strand), can decrease transcriptional activity and alter expression of nociceptive and immune genes (Polli et al., 2019), influencing pain responses to surgical stress and the transition from acute to chronic pain (Buchheit et al., 2012). Although methylation patterns between brain and blood cells differ, both may be influenced similarly by nutrients, stress, parenting behavior, etc (Weaver et al., 2005). Animal studies have shown that DNAm assessed in the prefrontal cortex and T cells at 9 months in post nerve-injury and sham rats are well correlated (72%) (Massart et al., 2016). Also, cross-tissue comparison of blood and brain methylation patterns in humans demonstrate the utility of using blood-based epigenome-wide association studies to identify biomarkers of neurological and brain related disease phenotypes (Davies et al., 2012; Hannon et al., 2015). Thus, blood DNAm has emerged as a promising biomarker in prognostic medicine (Heyn and Esteller, 2012; Thompson et al., 2022) owing to its stable methylation patterns that are faithfully retained during long-term storage (Groen et al., 2018), and its ability to yield reliable results across most laboratories (consortium B, 2016).

We previously identified 637 differentially methylated positions associated with CPSP using Illumina microarrays (Chidambaran et al., 2019) in pediatric patients undergoing spine fusion. We also reported two CpG sites in the mu-1 opioid receptor gene (OPRM1) promoter associated with CPSP using targeted pyrosequencing (Chidambaran et al., 2017). However, given the vast number of CpG sites in the human genome (approximately 28 million), the possibility of missing additional CPSP-associated loci with the use of arrays and targeted sequencing exists, particularly those in intergenic regions (Pidsley et al., 2016).

In this study, we performed whole-genome Next-generation enzymatic methylation sequencing (NG EM-seq) in a larger adolescent multisite cohort undergoing diverse musculoskeletal surgical procedures, to examine preoperative DNAm patterns (differentially methylated sites and regions) associated with CPSP. Compared with array-based platforms, NG EM-seq approaches provide broader and more uniform methylome coverage, including intergenic and non-promoter regions, making it less likely that CPSP associated loci are missed. We also hypothesized that DNAm will have a regulatory role in pain persistence and impact preoperative pain to CPSP transition.

2. Results

2.1. Recruitment and subject characteristics

A total of 343 participants were recruited preoperatively from six clinical sites. Whole blood samples were collected prior to surgery for analysis. Of these, 135 were excluded because their DNA had been allocated to other analyses, and 39 were lost to follow-up resulting in 168 participants with available samples for methylation sequencing (Fig. 1). This final cohort included male (52%) and female (48%) participants who were predominantly White (82%), with a median age of 15.5 years (IQR 14.3–16.8). Based on longitudinal assessments, 108 (64%) met criteria for CPSP, which was defined as any pain score > 3/10 on PainDETECT questionnaires beyond 2 months after surgery. Subjects with CPSP reported greater functional disability as measured by the Functional Disability Inventory (FDI (IQR 6 [3–11.25] compared to non-CPSP (FDI 2 [0–5.5], p < 0.001)) and lower quality of life (PedsQL) at 2–6 months and 10–12 months after surgery, indicating greater and sustained functional impairment throughout recovery. Demographic, pain, psychosocial, surgical and immune cell composition stratified by CPSP outcome are summarized in Table 1. The CPSP and non-CPSP groups were comparable in age, sex, and surgical type, but differed in preoperative and immediate postoperative pain and preoperative psychosocial measures (p < 0.1).

Fig. 1.

Overview of study and analytical pipeline. A. Consort style flow diagram showing recruitment, exclusions, and samples included in the study. B. Workflow for DNA methylation analysis for Chronic postsurgical pain associations.

Table 1.

Univariable comparisons of demographic, clinical and immune variables between participants with and without chronic postsurgical pain (CPSP).

Variable	Non-CPSP (N = 58)		CPSP (N = 108)		P Value
Patient demographics
Age (years)1	15.6	(14.4, 16.9)	15.4	(14.3, 16.7)	0.91
Weight1	55.1	(51.4, 64.1)	55.3	(49.2, 62.3)	0.63
Sex2 Female	25	(45%)	55	(51%)	0.34
Race2					0.77
White	47	(34%)	90	(66%)
Black	5	(33%)	10	(67%)
Asian	3	(60%)	2	(40%)
Native Hawaiian or Pacific Islander	0	(0)	1	(100%)
Others	1	(25%)	3	(75)
Unknown	2	(50%)	2	(50%)
Deprivation Index1	0.3	(0.3, 0.4)	0.3	(0.2, 0.4)	0.98
Preoperative pain, psychosocial and functional characteristics1
Pediatric Pain Screening Tool	1.0	(0.0, 2.0)	1.0	(0.0, 4.0)	0.08
PainDetect Q2. Strongest pain	2.0	(0.0, 4.0)	4.0	(2.0, 7.0)	<0.001
PainDetect Q3. average pain	1.0	(0.0, 3.0)	2.0	(0.0, 4.25)	0.011
Functional Disability Inventory score	3.0	(1.0, 7.0)	6.0	(1.0, 12.0)	0.09
Pediatric Quality of Life score	84.2	(71.7,94.8)	78.2	(65.8,91.8)	0.14
Postoperative pain and psychosocial characteristics1
2–6 month follow up
PainDetect Q2. Strongest pain	2.0	(1.0,3.0)	5.0	(4.0,7.0)	<0.001
PainDetect Q3. average pain	1.0	(0.0, 1.0)	2.0	(1.0,4.0)	<0.001
Functional Disability Inventory score	2.0	(0.0, 5.5)	6.0	(3, 11.25)	<0.001
Pediatric Quality of Life score	89.1	(80.4,96.5)	80.4	(66.3,88.0)	<0.001
10–12 month follow up
PainDetect Q2. Strongest pain	0.0	(0.0, 2.0)	4.0	(1.0, 5.0)	<0.001
PainDetect Q3. average pain	0.0	(0.0, 1.0)	1.0	(0.75, 3.0)	<0.001
Functional Disability Inventory score	0.0	(0, 2.0)	2.0	(0,5.0)	0.001
Pediatric Quality of Life score	94.6	(87.0,99.0)	87.5	(76.1,95.7)	0.004
Surgical factors
Surgical duration (minutes)1	116.0	(93.5, 239.0)	189.5	(95.75, 233.0)	0.37
Surgery type2					0.63
Pectus / Nuss	32	(55%)	55	(51%)
Spine correction	26	(45%)	53	(49%)
Perioperative analgesia
Regional Anesthesia2					0.19
Epidural	13	(22%)	16	(15%)
Erector Spinae-catheter	19	(33%)	33	(30%)
Cryoablation	0	(0%)	6	(6%)
None/unknown	26	(45%)	53	(49%)
Analgesics2
Dexmedetomidine – Yes⁎	12	(21%)	30	(29%)	0.34
Ketamine – Yes	8	(14%)	15	(14%)	0.99
Ketorolac – Yes	31	(53%)	52	(48%)	0.63
Acetaminophen- Yes	52	(90%)	98	(91%)	0.79
Acute postoperative pain and opioid use
Area under Curve for postoperative pain POD0–1	174.0	(126.0, 254.1)	219.9	(153.8, 275.6)	0.018
Morphine equivalents/kg POD0–1	0.8	(0.6, 1.2)	0.9	(0.7, 1.3)	0.27
DNAm sequencing
Batch2					<0.001
1	52	(90%)	66	(61%)
2	6	(10%)	42	(39%)
Immune cell types1
Monocytes	0.1	(0.0, 0.1)	0.1	(0.0, 0.1)	0.29
Natural killer cells	0.0	(0.0, 0.1)	0.0	(0.0, 0.1)	0.95
Helper T cells (naïve + memory + T regulatory)	0.2	(0.1, 0.2)	0.2	(0.1, 0.2)	0.97
Cytotoxic T cells (naïve + memory)	0.1	(0.0, 0.1)	0.1	(0.0, 0.1)	0.98
Granulocytes	0.6	(0.5, 0.6)	0.6	(0.5, 0.7)	0.49
B-cells	0.1	(0.0, 0.1)	0.1	(0.0, 0.1)	1.00

Abbreviation: AUC: area under the curve for pain scores over postoperative days 0–2. CPSP: if pain scores beyond 2-month post-surgery>3/10, IQR, interquartile range; POD, postoperative day.

¹median (IQR) presented with the interquartile range and compared using Wilcoxon/Kruskal-Walli's test.

^⁎Missing data (n = 5).

²Categorical variables (N (%) presented and compared using Fisher's exact tests/Chi-square.

2.2. Differential methylation analyses

2.2.1. DNA methylation pre-processing

Genomic DNA was successfully extracted from all available preoperative blood samples. Sequencing using the Enzymatic Methyl-seq (EM-seq) approach on the NovaSeq 6000 platform produced high-quality data, with a mean/median Phred score of 36 across samples (Supplementary Fig. 1). A schematic overview of the experimental workflow is provided in Fig. 2. NG-EMseq processing for the 168 participants was done in two batches (batch 1: n = 120; batch 2: n = 48). Two samples failed FASTQ quality control and were excluded, resulting in a final analytic cohort of 166 subjects (Fig. 1). To assess potential retention bias, demographic and clinical characteristics were compared between included participants and those without follow-up data (Supplementary Table 1). We found no difference in median age, weight preop FDI, and strongest preoperative pain scores between those who completed follow-up and those who did not, but a difference in sex/race was noted (p = 0.003) with the follow-up group including more males (51%) while the lost ton follow-up group had more females (74%) and non-white patients (p = 0.001). In addition, the lost to follow-up group included mostly spine surgery patients (97%) vs. 49% in follow-up group resulting in higher surgical duration (266 min [IQR 227.5–334.5] vs. 180 min [IQR 96–236], p < 0.001).

Fig. 2.

Study Design. Of 343 subjects recruited across six pediatric institutions in the US, 166 subjects had evaluable DNAm and outcomes, and were included in analyses. Pain, functional, and psychological surveys (see Methods and Supplementary Information) were administered preoperatively and postoperatively at 2–6 and 10–12 months. Pain scores and analgesic use were noted for immediate postoperative period (postoperative days 0–1). DNA was extracted from preoperative blood samples, which were converted by enzymatic methylation next-generation sequencing (Em-seq) and read by Novaseq 6000, followed by differential methylation analysis.

2.2.2. Adjustment for cell composition

To ensure observed methylation differences reflected true biological variation, we estimated and adjusted for individual cell-type fractions prior to differential methylation analysis (Qi and Teschendorff, 2022). For all the NG Em-seq samples (n = 166), in silico estimations of the different cell-type fractions was performed using the reference library from Salas, Zhang (Salas et al., 2022) in EpiDISH (Qi and Teschendorff, 2022). These cell-type proportions were included as covariates in the methylation analysis to account for inter-individual differences in immune cell composition. The in silico estimates were compared with empirical data from flow cytometry and clinical complete blood counts (CBCs) for available subjects to validate the accuracy. Flow cytometry data (methods available in previous publication) (Vijayarajan et al., 2026) were available for 97 participants, while 14 subjects had CBC data, drawn preoperatively for clinical purposes. All but one cell type showed moderate correlations between in silico and measured fractions (Supplementary Table 2), supporting the reliability of the EpiDISH-based deconvolution.

2.2.3. CpG methylation profiling and quality control

Technical replicate analysis confirmed high reproducibility across methylation calls. Replicate concordance increased with sequencing depth and reached optimal correlation at a 7× minimum coverage threshold (mean R² for replicates =0.9155; Supplementary Fig. 2). Multiple filtering steps were performed to remove potential confounding or unreliable sites. Specifically, 241,621 CpG sites overlapping known SNPs were excluded to avoid allelic bias, and 1,416,267 CpG sites located on sex chromosomes were removed to minimize sex-related methylation variability. Sites with coverage <7× (∼14 million CpGs) were masked to prevent inclusion of low-confidence calls, and 27,215,822 rare CpG sites not present in at least 90% of samples were excluded. After all quality control and filtering steps, 44,255 CpG sites remained and were used for downstream differential methylation analysis (Supplementary file 1).

2.3. Differentially methylated loci (DMLs) associated with CPSP

Because CPSP is shaped by both biological and clinical influences, we used a two-stage analytic framework to balance broad discovery with control for confounding variables. In the first stage served as a screen, narrowing the genome-wide set of CpG sites to a smaller candidate set with differential DNAm associated with CPSP, and a nominal p < 0.05 threshold was selected to support preliminary discovery and reduce the likelihood of missing potentially relevant CpG sites. The second stage was confirmatory, testing whether these candidate sites remained associated with CPSP after adjustment for relevant clinical covariates, and findings were evaluated using a Benjamini Hochberg false discovery rate threshold of 0.10 to control the expected proportion of false positives across tests. Significant CpG sites identified in the second-stage models were then internally validated using nonparametric bootstrapping with 1000 replications to filter CpG sites that provide the strongest evidence of association. The two stages differ in which variable is modeled as the outcome. Screening was performed using LIMMA (RN Beads) (Muller et al., 2019), a linear modeling framework in which methylation must be the dependent variable and a binary outcome such as CPSP cannot be fitted; this framework was chosen because it estimates variance by pooling information across all CpG sites, which gives more stable results in a sample of our size than testing each site independently (S, 2004). Once the candidate set was reduced, CPSP was modeled directly as the outcome, so that the associations we report are estimated on the clinical scale with covariates adjusted in the same model.

2.3.1. CpG sites mostly located in intergenic/intronic regions differ in methylation between CPSP and non-CPSP group

First, we conducted linear regression with DNAm as dependent variable and CPSP as independent variable, adjusted for age, race, sex and batch. This identified 2210 CpG sites associated with CPSP (p < 0.05; Supplementary file 2). Annotation of these 2210 CpGs with the Annotatr package (version:1.34.0, hg38 build) produced 4675 feature-level annotations, as single CpG sites can overlap multiple genomic features. Most annotated sites occurred in intergenic (37%) and intronic (30%) regions, with smaller fractions in 1–5 kb upstream, lncRNA, exonic, and promoter regions. Only a few CpGs are mapped to untranslated regions (Fig. 3A). Directionally, most differentially methylated CpG sites (DMLs) were hypomethylated (≈ 69%) (mean beta difference at hypomethylated DML = − 0.026, SD = 0.016) while hypermethylated DMLs were less common (≈ 30%) (mean beta difference = 0.022, SD = 0.018). When CpGs were categorized by CpG-island context, most were in inter-CpG regions (≈92%), with relatively few within CpG islands, shores, or shelves (Fig. 3B). Together, these findings indicate that methylation differences linked to CPSP were primarily characterized by hypomethylation within noncoding and regulatory genomic regions.

Fig. 3.

Genomic annotation of differentially methylated CpG sites. (A) Distribution of hypomethylated (red) and hypermethylated (blue) CpG sites across annotated gene regions among the 2210 significant sites identified in LIMMA (linear regression using DNAm as dependent variable and Chronic postsurgical pain as independent variable, adjusted for age, sex, race and batch). A total of 4675 annotations were generated because individual CpGs can overlap multiple genomic features. (B) Proportion of CpG sites across CpG-island features (hg38 reference genome), showing the majority in inter-CpG regions (≈ 92%), with smaller fractions in CpG islands, shores, and shelves. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

2.3.2. Differentially methylated CpG sites associated with CPSP (after adjusting for covariates) were mostly hypomethylated

In the next stage, the 2210 DM CpG sites identified were re-evaluated using logistic regression models, with CPSP status as the dependent variable and methylation values (β values) as independent variables. Although beta-values are bounded and can yield large coefficient magnitudes, they provide direct biological interpretability as percentage methylation. To improve interpretability, effect sizes can also be rescaled to represent a 10% increase in methylation (Antoun et al., 2022). Also, to adjust for clinical covariates impacting CPSP, we tested all the perioperative factors that showed differences between CPSP and non-CPSP groups (p < 0.1), namely, preoperative average and strongest pain over the last four weeks, preoperative PainDETECT total score, preoperative functional disability, PPST, post operative pain AUC and batch of DNAm (Table 1) for association with CPSP. After stepwise regression, batch (β = 1.82, p = 0.001) and preoperative pain (strongest pain over last 4 weeks) (β = 0.15, p = 0.03) remained in the final model as covariates for logistic regression. After adjusting for batch and preoperative pain along with age, sex and race, DNAm at 646 CpG sites were significantly associated with CPSP based on β methylation values (FDR BH < 0.1). The overall distribution of differential methylation effects is shown in the volcano plot (Supplementary Fig. 3). Following internal validation using nonparametric bootstrapping with 1000 replications at the 646 CpG sites, 289 CpG sites met the predefined robustness criteria (Davison and Hinkley, 1997; Wang et al., 2025; Steyerberg et al., 2001), including consistent directionality across bootstrap samples (>99%), moderate to high p-value stability (>70%), and bootstrap confidence intervals excluding the null value (β = 0/OR = 1), and were retained for further downstream analysis (Supplementary file 3; Supplementary Fig. 4. Across these 289 loci, hypomethylation remained dominant (n = 203, 70.2%; mean beta difference = −0.04, SD = 0.02), while hypermethylation accounted for 86 sites (26.2%; mean beta difference = 0.04, SD = 0.02, indicating slightly stronger methylation shifts than observed in previous step. Annotation of the these CpG's again revealed that most signals were enriched in intergenic and intronic regions, with limited representation in promoters, exons, and long noncoding RNA elements (Fig. 4A). When examined relative to CpG-island context, majority (≈93%) of sites were localized to inter-CpG regions, with few within islands, shores, or shelves (Fig. 4B).

Fig. 4.

Genomic annotation and positional distribution of differentially methylated CpG sites (n = 289) across gene features. (A) Distribution of CpG sites differentially methylated for CPSP (hypomethylated (red) and hypermethylated (blue)) across annotated gene features, showing predominant enrichment in intergenic and intronic regions. (B) Proportion of CpG sites across CpG-island features, showing the majority within inter-CpG regions, with minimal representation in islands, shores, or shelves. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

2.4. Gene enrichment analysis of significant CpG sites using CPSP-specific background set

The 289 significant CpG sites associated with CPSP were annotated to their nearest genes using the ChIPseeker package (version: 1.44.0; function annotatePeak). The complete list of CpG–gene annotations is provided in Supplementary File 4. This annotation linked the CpG sites to 57 unique genes (Table 2), Among these 57 genes, 32% involved noncoding RNAs (n = 18), 33% protein-coding genes (n = 19), and 32% pseudogenes (n = 18), with the remainder comprising small RNA loci (rRNA = 1 and snoRNA = 1). These mapped genes were used for both GO and KEGG pathway enrichment analysis. Enrichment was performed using a CPSP-specific background gene set that included 1305 candidate genes and 31 training genes identified and characterized in (Antoun et al., 2022; Davison and Hinkley, 1997), as explained in methods section Significant GO enrichment was observed across all three GO categories (Biological Process (BP), Molecular Function (MF), and Cellular Component (CC)), highlighting neuronal and ion channel–related processes (Fig. 5A–C). GO enrichment was performed using a topology based approach - topGO “weight01” algorithm - which accounts for relationships among neighboring GO terms. Fisher's exact test p values for GO enrichment do not need separate FDR correction (Alexa and Rahnenfuhrer, 2007). Within the BP category, some of the Cellular response to camp (p = 0.001), neuronal action potential (p = 0.01), sodium ion transmembrane transport (p = 0.03), potassium ion transmembrane transport (P = 0.035) and, regulation of postsynaptic membrane potential (P = 0.041). MF terms were dominated by Intracellular camp-activated cation channel activity involved in regulation of presynaptic membrane potential (p = 0.001), cAMP binding (p = 0.003), voltage-gated sodium channel activity (p = 0.01), and voltage-gated potassium channel activity (p = 0.02). CC enrichment identified HCN channel complex (p = 0.002), presynaptic active zone membrane (p = 0.006), and postsynaptic membrane (p = 0.049). Similarly, GO enrichment using a whole genome background also identified similar top gene domains related to HCN and ion processes (Supplemental file 7).KEGG and Reactome Pathway enrichment using EnrichR (version: 3.4) showed no statistically significant pathways after FDR correction. Supplementary Table 3 and Supplementary Table 4 show the full list of Kegg and Reactome terms identified in the analysis.

Table 2.

List of 57 unique genes annotated to the 289 significant CpG sites associated with CPSP.

NCBI GeneID	Symbol	Description	Gene Type	Accession Number	Genomic Range Start	Genomic Range Stop
7514	XPO1	exportin 1	PC	NC_000002.12	61,477,849	61,538,612
114899	C1QTNF3	C1q and TNF related 3	PC	NC_000005.10	34,017,858	34,244,724
348980	HCN1	hyperpolarization activated cyclic nucleotide gated potassium channel 1	PC	NC_000005.10	45,254,948	45,696,380
55112	DYNC2I1	dynein 2 intermediate chain 1	PC	NC_000007.14	158,839,245	158,958,698
100302167	MIR1299	microRNA 1299	ncRNA	NC_000009.12	40,929,010	40,929,092
7179	TPTE	transmembrane phosphatase with tensin homology	PC	NC_000021.9	10,521,583	10,605,716
55672	NBPF1	NBPF member 1	PC	NC_000001.11	16,562,423	16,613,564
103352539	LINC01410	long independently transcribed non-coding RNA 1410	ncRNA	NC_000009.12	62,801,465	62,813,486
133418	EMB	embigin	PC	NC_000005.10	50,396,192	50,443,345
79937	CNTNAP3	contactin associated protein family member 3	PC	NC_000009.12	39,064,710	39,288,167
109910381	RNA5-8SN3	RNA, 5.8S ribosomal N3	rRNA	NC_000021.9	8,395,606	8,395,762
9721	GPRIN2	G protein regulated inducer of neurite outgrowth 2	PC	NC_000010.11	46,541,736	46,557,596
80157	CWH43	cell wall biogenesis 43C-terminal homolog	PC	NC_000004.12	48,986,275	49,062,079
7539	ZFP37	ZFP37 zinc finger protein	PC	NC_000009.12	113,038,377	113,056,724
85319	BAGE2	BAGE family member 2 (pseudogene)	PSEUDO	NC_000021.9	10,413,497	10,518,274
441581	FRG2B	FSHD region gene 2 family member B	PC	NC_000010.11	133,623,895	133,626,795
441234	ZNF716	zinc finger protein 716	PC	NC_000007.14	57,450,177	57,473,559
646096	CHEK2P2	CHEK2 pseudogene 2	PSEUDO	NC_000015.10	20,282,744	20,291,558
654342	LSP1P4	LSP1 pseudogene 4	PSEUDO	NC_000002.12	91,636,683	91,659,949
647121	EMBP1	embigin pseudogene 1	PSEUDO	NC_000001.11	121,519,112	121,571,888
653786	OTOAP1	OTOA pseudogene 1	PSEUDO	NC_000016.10	22,545,698	22,576,865
102724951	LINC03105	long independently transcribed non-coding RNA 3105	ncRNA	NC_000021.9	7,430,668	7,469,008
403323	LERFS	lncRNA negative regulator of fibroblast-like synoviocyte migration, SYNCRIP interacting	ncRNA	NC_000009.12	62,857,856	62,898,087
124149	ANKRD26P1	ankyrin repeat domain 26 pseudogene 1	PSEUDO	NC_000016.10	46,469,337	46,569,097
149992	ANKRD30BP2	ANKRD30B pseudogene 2	PSEUDO	NC_000021.9	13,038,166	13,118,251
102723475	LOC102723475	potassium voltage-gated channel subfamily E regulatory subunit 1B	PC	NC_000021.9	7,816,677	7,829,632
26080	FAM230C	family with sequence similarity 230 member C	ncRNA	NC_000013.11	18,195,297	18,232,016
100289097	FRG1CP	FSHD region gene 1 family member C, pseudogene	PSEUDO	NC_000020.11	28,580,633	28,602,665
441058	LINC01667	long independently transcribed non-coding RNA 1667	ncRNA	NC_000021.9	9,781,848	9,821,061
646066	TAF11L5	TATA-box binding protein associated factor 11 like 5	PC	NC_000005.10	17,525,235	17,525,831
100132352	FRG1HP	FSHD region gene 1 family member H, pseudogene	PSEUDO	NC_000009.12	40,992,261	41,039,103
400238	LINC00642	long independently transcribed non-coding RNA 642	ncRNA	NC_000014.9	90,455,230	90,458,905
728034	BMS1P23	BMS1 pseudogene 23	PSEUDO	NC_000002.12	94,192,531	94,208,419
101927050	KMT2CP4	lysine methyltransferase 2C pseudogene 4	PSEUDO	NC_000002.12	89,625,445	89,637,907
103908605	FAM88E	family with sequence similarity 88 member E	ncRNA	NC_000009.12	42,567,859	42,569,349
442421	PTGER4P2-CDK2AP2P2	PTGER4P2-CDK2AP2P2 readthrough, transcribed pseudogene	PSEUDO	NC_000009.12	62,838,445	62,847,206
100101266	HAVCR1P1	hepatitis A virus cellular receptor 1 pseudogene 1	PSEUDO	NC_000019.10	24,162,193	24,163,447
102723316	FRG1DP	FSHD region gene 1 family member D, pseudogene	PSEUDO	NC_000020.11	29,080,390	29,096,832
102723390	FRG1EP	FSHD region gene 1 family member E, pseudogene	PSEUDO	NC_000020.11	29,475,879	29,497,334
102723780	FRG1FP	FSHD region gene 1 family member F, pseudogene	PSEUDO	NC_000022.11	10,940,597	10,961,529
729732	LOC729732	uncharacterized LOC729732	ncRNA	NC_000008.11	12,537,079	12,665,611
650226	LOC650226	ankyrin repeat domain containing 26 pseudogene	PSEUDO	NC_000007.14	56,423,704	56,448,375
646652	LOC646652	integral membrane glycoprotein-like pseudogene	PSEUDO	NC_000005.10	34,189,581	34,194,204
124903442	LOC124903442	uncharacterized LOC124903442	PC	NC_000015.10	20,295,653	20,332,127
101929373	LOC101929373	uncharacterized LOC101929373	ncRNA	NC_000010.11	42,242,512	42,276,842
102724200	LOC102724200	trafficking protein particle complex subunit 10-like	PC	NC_000021.9	5,155,774	5,165,464
102724207	LOC102724207	ankyrin repeat domain-containing protein 36C-like	PC	NC_000016.10	33,741,790	33,758,001
105372345	LOC105372345	uncharacterized LOC105372345	ncRNA	NC_000019.10	27,681,072	27,687,970
105372733	LOC105372733	uncharacterized LOC105372733	ncRNA	NC_000021.9	9,996,066	10,119,490
105374848	LOC105374848	uncharacterized LOC105374848	ncRNA	NC_000002.12	87,319,419	87,326,756
105375937	LOC105375937	uncharacterized LOC105375937	ncRNA	NC_000008.11	85,495,693	85,534,760
107984037	LOC107984037	uncharacterized LOC107984037	ncRNA	NC_000022.11	11,825,277	11,835,968
107985200	LOC107985200	uncharacterized LOC107985200	ncRNA	NC_000001.11	143,382,482	143,424,762
112268291	LOC112268291	uncharacterized LOC112268291	ncRNA	NC_000022.11	15,600,587	15,605,196
124901826	LOC124901826	small nucleolar RNA U2–19	snoRNA	NC_000007.14	63,175,940	63,176,019
124903767	LOC124903767	putative uncharacterized protein MGC34800	PC	NC_000016.10	34,388,895	34,389,821
124903955	LOC124903955	uncharacterized LOC124903955	ncRNA	NC_000017.11	26,998,080	26,999,949

Each CpG site that was significant (289) was annotated to its nearest gene using the ChIPseeker package (annotatePeak function). The table lists NCBI Gene IDs, gene symbols, functional descriptions, gene biotypes, and genomic coordinates (based on the hg38 reference genome). Abbreviations: PC, protein-coding; ncRNA, non-coding RNA; rRNA, ribosomal RNA; snoRNA, small nucleolar RNA.

Fig. 5.

Gene Ontology (GO) enrichment analysis of genes annotated to CpG sites with differential DNAm associated with chronic postsurgical pain (CPSP). Bar plots showing enriched GO terms across (A) Biological Process (BP), (B) Molecular Function (MF), and (C) Cellular Component (CC) categories based on genes linked to 289 significant CpG sites. Enrichment was performed using the CPSP-specific background gene set (n = 1336). The red dashed line represents the nominal significance threshold (p = 0.05, −log₁₀ scale) for each term. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

2.5. Differentially methylated region (DMR) analysis

To identify clusters of CpG sites showing coordinated methylation changes, the 289 significant CpG sites were analyzed for differentially methylated regions (DMRs) using DMRcate. This analysis identified 21 DMRs, of which 20 remained significant after HMFDR correction (HMFDR <0.05). The 20 significant DMR were mapped to 7 unique genes (Table 3) Genomic coordinates and annotation of all identified DMRs are provided in Supplementary file 5.

Table 3.

List of 7 unique genes mapped to the 20 significant DMR (HFDR<0.05) associated with CPSP.

NCBI GeneID	Symbol	Description	Gene Type	Accession Number	Genomic Range Start	Genomic Range Stop
55628	ZNF407	zinc finger protein 407	PC	NC_000018.10	74,597,870	75,065,671
80157	CWH43	cell wall biogenesis 43C-terminal homolog	PC	NC_000004.12	48,986,275	49,062,079
647,121	EMBP1	embigin pseudogene 1	PSEUDO	NC_000001.11	1.22E+08	1.22E+08
441,058	LINC01667	long independently transcribed non-coding RNA 1667	ncRNA	NC_000021.9	9,781,848	9,821,061
101,929,373	LOC101929373	uncharacterized LOC101929373	ncRNA	NC_000010.11	42,242,512	42,276,842
107,984,037	LOC107984037	uncharacterized LOC107984037	ncRNA	NC_000022.11	11,825,277	11,835,968
107,985,200	LOC107985200	uncharacterized LOC107985200	ncRNA	NC_000001.11	1.43E+08	1.43E+08

Significant DMR (HFDR<0.05) was annotated to its nearest gene using the ChIPseeker package (annotatePeak function). The table lists NCBI Gene IDs, gene symbols, functional descriptions, gene biotypes, and genomic coordinates (based on the hg38 reference genome). Abbreviations: PC, protein-coding; ncRNA, non-coding RNA; rRNA, ribosomal RNA; snoRNA, small nucleolar RNA.

2.5.1. Gene and Motif enrichment analysis of significant DMR

To explore the biological relevance of the significant DMRs, the 7 genes overlapping the 20 FDR-significant DMRs were annotated and subjected to gene enrichment analysis using the CPSP-specific background gene set (Chidambaran et al., 2020; Chidambaran et al., 2021a) and full genome background. Gene enrichment analysis for full genome background are shown in Supplementary file 7 sheet DMR.CPSP background enrichment did not generate any result, as there was no overlap between the 7 unique genes and the CPSP-specific background results. But using motif enrichment analysis we found significant overrepresentation of transcription factors binding motifs involved in transcriptional regulation, stress signaling, and neuronal differentiation (Table 4 shows the significant results, Supplementary file 6 has the full list of TF (sheet 1) and motifs (sheet 2) identified in the analysis). Some of the top enriched motifs included CENPB (q = 1.59 × 10⁻11), NR4A2::RXRA (q = 1.7 × 10⁻⁷), TEAD2 (q = 2.1 × 10⁻⁶), and TEAD4 (q = 4.01× 10⁻⁵); ETV2, SMAD3, SMAD5, and SMAD4 in TGF-β/SMAD-mediated signaling pathways; several STAT family members (STAT6, STAT5A/B, STAT4, STAT1) (q values ranging from 6.0 × 10⁻4 to 0.12), and transcription factors MYCN, RORA, RORB, and NPAS2 (q < 0.05) involved in circadian and neuronal plasticity.

Table 4.

List of significantly enriched transcription factors (TF) and its best motifs, mapped from 20 DMR associated with chronic postsurgical pain (CPSP).

TF	P value	q value (FDR adjusted P value)	Best Motif for TF by q Value
CENPB	1.06635E-14	1.591E-11	MA0637.1
NR4A2::RXRA	2.29418E-10	1.71146E-07	MA1147.1
SMAD5	3.75508E-10	1.86753E-07	MA1557.1
SMAD3	1.35185E-09	5.0424E-07	MA0795.1
TEAD4	1.34544E-07	4.01478E-05	MA0809.2
ETV2	2.56282E-07	6.37289E-05	MA0762.1
NR1H4	7.52475E-07	0.000160385	MA1110.1
TEAD3	1.06956E-06	0.000199474	MA0808.1
TEAD2	2.10057E-06	0.000348227	MA1121.1
FOXH1	3.20741E-06	0.000478546	MA0479.1
Stat6	4.43584E-06	0.000601661	MA0520.1
CEBPG(var.2)	5.98331E-06	0.000743925	MA1636.1
Smad4	9.02184E-06	0.001035429	MA1153.1
RORB	3.79042E-05	0.003770201	MA1150.1
TCF7	6.01299E-05	0.005607115	MA0769.2
RORA	8.89839E-05	0.007375777	MA0071.1
STAT1	0.000152726	0.010357574	MA0137.3
ZNF317	0.000171925	0.011152683	MA1593.1
NR2C2(var.2)	0.000208135	0.012939047	MA1536.1
HNF4G	0.000281399	0.01554989	MA0484.2
HNF4A	0.000368385	0.019629654	MA0114.4
GRHL1	0.000622205	0.029010321	MA0647.1
Stat5a	0.00118853	0.042304188	MA1624.1

Nominal significance was indicated by p < 0.05. Multiple testing adjusted significance was indicated by q < 0.05. Best Motif for TF by q Value is the motif variant that attained the smallest adjusted p value across all motif models mapped to that TF.

3. Discussion

In this multisite prospective study, we evaluated DNA methylation as a potential blood-based marker of CPSP risk using EM-sequencing. We identified 289 mostly hypomethylated DMLs located in regulatory regions associated with CPSP, after bootstrapping with stringent criteria. They enriched several biological pathways, molecular functions and cellular components known to be involved in neuronal and ion channel–related processes. We also identified 20 FDR-significant DMRs annotated to 7 unique genes, which significantly enriched transcription factor binding motifs involved in stress signaling, neuronal differentiation, inflammatory processes and circadian biology.

3.1. CPSP is associated mostly with hypomethylation at CpG sites in regulatory regions

In this study we identified 289 DMLs associated with CPSP located in intergenic and intronic regions. This pattern suggests that regulatory elements outside classical promoters, such as enhancers and silencers, may contribute to individual differences in pain vulnerability. Although this might reflect overall distribution of CpG sites, other investigators have also reported epigenetic alterations relevant to CPSP and other pain conditions supporting the genomic distribution observed in our cohort. A genome wide methylation study that used reduced representation bisulfite sequencing to compare DNA methylation differences between neuropathic pain, nociceptive pain, and control groups reported that nearly 80% of significant CpGs were located in intergenic or intragenic regions, with a smaller proportion falling within promoter or exon sites that influence how nociceptive signals are controlled (Stenz et al., 2022). Their DMR analysis also revealed clustered regional changes, with 26 DMRs identified in the nociceptive group and eight DMRs associated with chronic neuropathic pain (Stenz et al., 2022). This distribution mirrors our own results, where we detected 20 DMRs that remained significant after FDR correction, although we were unable to identify similar enriched pathways. Despite examining the same clinical phenotype in a non-overlapping spine surgical cohort, we found no overlap between 637 CpG sites associated with CPSP identified in that study, although we found enrichment of non-promoter regions (Chidambaran et al., 2019). This lack of overlap is potentially because of differences in DNAm platforms used (we used Illumina EPIC array in our prior study and RRBS in the previous cited study above) which cover different genomic regions with different depths (Guanzon et al., 2024). Thus, differences in assay design/genomic coverage, phenotype/cohort and analytical methods likely contribute to different DNAm signals/pathways associated with chronic pain phenotypes, but the overall genomic distribution pattern remains consistent.

In this study, more than 70% of CPSP-associated CpG sites were hypomethylated, a pattern consistent with findings seen in several other chronic pain conditions (Xiong et al., 2024). Work in patients with fibromyalgia reported that about 69% of differentially methylated CpG sites were hypomethylated compared to controls (Ciampi de Andrade et al., 2017). A study of chronic lower back pain also showed a predominance of hypomethylation at pain related CpG sites (Aroke et al., 2020). Additional evidence comes from a large cohort of individuals who developed complex regional pain syndrome (CRPS) after traumatic limb amputation, wherein 48 differentially methylated CpG sites were identified with 85% being hypomethylated compared to those with non-CRPS neuropathic pain (Bruehl et al., 2019). Together, these observations indicate that widespread hypomethylation may play an important role in the transition from acute to chronic pain.

Although global hypomethylation is consistent with our findings, gene specific studies show a more complex pattern (Xiong et al., 2024). This includes our own previous work in which increased methylation at several CpG sites in the OPRM1 promoter was associated with more intense acute pain and a higher likelihood of CPSP in adolescents after spine surgery, suggesting that reduced OPRM1 expression may heighten pain sensitivity (Chidambaran et al., 2017). Furthermore, higher methylation at TRPA1, a gene involved in sensing harmful stimuli, has also been linked with more severe pain in patients with chronic low back pain, postherpetic neuralgia, and Crohn's disease (Sukenaga et al., 2016) and comparable hypermethylation-based effects have been reported for tumor necrosis factor in long lasting postsurgical breast pain and for Secreted Protein, Acidic, Rich in Cysteine (SPARC) in chronic low back pain and in animal models (Stephens et al., 2017). Taken together, these findings suggest that global hypomethylation may relate to overall vulnerability to chronic pain, while hypermethylation at specific genes may influence how nociceptive signals are controlled.

3.2. DMLs functionally enriched neuronal and ion-channel–related GO terms

In our cohort, the differentially methylated sites mapped to 57 unique genes, and a large proportion of these were noncoding RNAs, pseudogenes, or small RNA loci. This distribution is consistent with work showing that pain-related epigenetic variation often involves noncoding regions rather than classical protein coding genes (Chidambaran et al., 2021b), again supporting the idea that noncoding regulatory domains may play an important role in shaping epigenetic sensitivity to pain. Examining function, we found that genes linked to DMLs were enriched for neuronal and ion channel related Gene Ontology terms including regulation of postsynaptic membrane potential, potassium ion transport, and voltage gated channel activity. This pattern is consistent with other chronic pain EWAS (Montesino-Goicolea et al., 2020; Goodin et al., 2022), in which differentially methylated genes are repeatedly enriched for pathways involved in neuronal development, synaptic function, and neuroimmune signaling. This adds to our prior findings that DNAm at specific CpG sites in genes linked to opioid, dopaminergic, and GABA-related pathways was associated with CPSP in children undergoing spine fusion surgery (Chidambaran et al., 2019; Chidambaran et al., 2017). The processes identified fit well with current concepts of chronic pain biology and may support association between DNAm signatures in genes linked to neuronal excitability, synaptic plasticity, and central sensitization, and the transition from acute to persistent pain.

3.3. DMRs did not enrich GO terms but showed transcription factor motif enrichment

Methylome analyses have shown that specific DNA sequence motifs correlate with methylation state and influence transcription factor binding (Rimoldi et al., 2024; Grau et al., 2023), potentially affecting downstream gene regulation relevant to pain sensitivity. However, many DMRs were located in intergenic or non-coding regions, and gene enrichment analyses of DMR-associated genes did not identify robust pathway-level enrichment, limiting biological interpretation. Nevertheless, motif enrichment analysis indicated that CPSP-associated methylation differences were preferentially localized to genomic regions containing binding sites for transcription factors involved in stress-response (e.g., SMAD/TGF-β), immune and inflammatory signaling (e.g., STAT family), circadian regulation (e.g., NPAS2, RORA, RORB), and pathways previously implicated in neuronal differentiation and plasticity (e.g., NR4A2::RXRA, TEAD) (Jakaria et al., 2019). These are consistent with existing literature regarding their roles in pain pathways. For example, neuronal methylation atlas data shows that NR4A2-related transcription factors occupy hypomethylated regulatory regions in neurons and immune/stress-responsive cells, suggesting a role in activity dependent regulation of excitability and adaptation to injury (Bridi and Abel, 2013), and reduced Nr4a expression influenced by translational processes (including DNAm) has been associated with neurodegeneration, autism spectrum disorders, schizophrenia, and epilepsy (Catala-Solsona et al., 2021). TEAD (TEA Domain Transcription Factors) are central effectors of the Hippo signaling pathway, known to regulate neuronal inflammation, differentiation and signaling, with recent discussions of therapeutic potential for diseases such as Alzheimer's⁵⁶. The activity of TEADs depends on co-activators such as YAP/TAZ, which when active, enter the nucleus, bind TEADs, and modulate transcription (Currey et al., 2021; Panciera et al., 2017). Experimental work in neuropathic pain models shows that inhibition of YAP TEAD signaling can reduce mechanical hypersensitivity, indicating that TEAD responsive regulatory programs may contribute to pain-related plasticity (Gessler et al., 2024). Smad proteins are a family of intracellular proteins that act as signal transducers for the transforming growth factor beta (TGF-β) pathway implicated in neural development and neuroplastic responses to injury and pain, with therapeutic potential as shown in neuropathic pain models (Echeverry et al., 2009). In the STAT family, STAT3 (signal transducer and activator of transcription was shown to correlate with clinical parameters of surgical recovery, including functional impairment and pain in adult patients undergoing hip surgery (Gaudilliere et al., 2014). STAT proteins as major regulators of cytokine driven pathways (Busch-Dienstfertig and Gonzalez-Rodriguez, 2013). STAT activation occurs through the JAK STAT cascade, which is used by pro inflammatory cytokines including IL-6, IL-12, and IL-18. Experimental pain models have shown that loss of IL-6 or blockade of JAK STAT3 signaling reduces hyperalgesia and nerve injury induced allodynia (Xiong et al., 2023). Recent transcriptomic analyses of synovial macrophages in osteoarthritis further highlight the prominence of STAT pathways, with most pain related macrophage signatures involving STAT dependent signaling (Blackler et al., 2024). NPAS2 knockout mice display enhanced mechanical and thermal sensitization after inflammatory challenge and NPAS2 deficiency or manipulation changes opioid tolerance, withdrawal, and hyperalgesia in rodent models (Dong et al., 2025). RORA has been linked to circadian modulation of neuroinflammation and migraine-like phenotypes in human and animal studies (Farahani et al., 2020). Given the use of peripheral blood and the absence of strong gene-level enrichment signals, these findings should primarily be interpreted as candidate biomarkers and hypothesis-generating regulatory signatures. Future studies incorporating pain-relevant tissues and multi-omic functional analyses will be required to determine whether these blood-based epigenetic signatures reflect biological mechanisms contributing to pain chronification or serve as preoperative markers potentially associated with CPSP risk.

3.4. Limitations and future studies

This study provides novel insights into the epigenetic architecture of CPSP in adolescents using Next-generation sequencing using a multisite, well-characterized cohort. Although our retention rate was acceptable (80.9%), subjects lost-to follow up differed by sex, race and surgery type, with potential retention bias. Other limitations are that methylation data were generated at relatively low sequencing depth, with an average coverage of about seven reads per CpG site, reducing confidence in individual effect estimates. However, we did verify that this was the optimal coverage for our sample (Supplementary Fig. 2).

Multiple testing corrections were not applied at every level, and the two-stage selection framework may have increased the potential for selection bias and overfitting. The bootstrap analysis showed consistent directionality and associations with CPSP after adjustment for clinical covariates, supporting the internal stability of the retained CpGs. However, because bootstrapping was applied only to the CpGs retained after the second step, it evaluates the stability of those selected loci rather than of the discovery process that produced them. Given the sample size and regression threshold, some degree of selection-related variability and overestimation of effect sizes remains possible, and replication in an independent cohort will be important before establishing the reproducibility of the identified CpGs.

DML enrichment was similar in both CPSP-specific and full human genome background for several GO terms (supplemental file 7). Most retained CpGs were located in intergenic or intronic regions, consistent with the broader regulatory distribution of DNA methylation, though this limits direct gene-level interpretation. In addition, nearest-gene annotation may not accurately capture distal or context-dependent regulatory relationships. Therefore, the GO enrichment results provide biologically relevant pathway-level context but should be regarded as hypothesis-generating until the regulatory targets of these CpGs are functionally characterized. Future studies should consider larger cohorts, deeper sequencing, multimodal integration of genomics and immune data, and longitudinal designs to validate our findings and extend generalizability.

3.5. Conclusions

This DNAm sequencing study provides the first evidence in a multicenter pediatric cohort that genome-wide DNA methylation differences are associated with CPSP. These findings support DNAm as contributors to CPSP biology and open the door for mechanistic studies exploring how epigenetic regulation contributes to pain persistence in children, although replication and formal assessment of predictive performance would be required before any clinical or risk-stratification role can be established. Recently, predictive tools including preoperative pain, type of surgery, education level, and concerns about the procedure, predict CPSP risk in adults with good discrimination (Wildemeersch et al., 2025). However, effective prediction will likely require integrating biological (genetic and epigenetic), psychological, and sensory biomarkers (Berardi et al., 2022). TF motif enrichment pathways identified allow mechanistic hypotheses generation for future studies especially as they may be amenable to potential therapeutic targeting (for example, Hippo and TGF-β signaling). This concept aligns with the broader movement in pain research, emphasized by A2CPS to identify modifiable mechanistic biomarkers that can be targeted to prevent the progression to chronic pain (Berardi et al., 2022).

4. Methods

4.1. Study design

This multisite study is registered with ClinicalTrials.gov: NCT01839461, NCT01731873. Subjects were identified from operating room schedules for recruitment. Participants met following inclusion criteria were recruited between 2016 and 2022 at six participating institutions: eight years of age or older, diagnosed with idiopathic scoliosis undergoing spinal fusion surgery, or pectus excavatum undergoing Nuss surgery, fluent in English, parents or adult patients able to provide informed consent and HIPAA authorization, and children able to give assent. Exclusion criteria included systemic autoimmune disease or preoperative use of immunosuppressive medications, a cancer diagnosis within the previous five years, developmental delay that would interfere with collection or interpretation of pain scores and questionnaires, pregnancy or breastfeeding, opioid use or major surgery within the prior six months, infectious disease within one month, and significant systemic disease (renal, hepatic, or cardiorespiratory).

4.2. Ethical approval

These studies have been approved by Cincinnati Children's Hospital Medical Center (CCHMC) Institutional Review Board. Single IRB (umbrella: CCHMC IRB) was approved by all collaborating site IRBs. Written informed consent from parents or guardians and assent from children was obtained prior to enrollment in each study.

4.3. Data collection

Patient sociodemographic (age, sex, race, ethnicity and residential address) were obtained. Deprivation Index was estimated using census tract level data. It ranges from 0 to 1, with higher values indicating greater deprivation (Brokamp et al., 2019).

Data on preoperative diagnosis, comorbidities, surgical details (type and duration), pain scores, and postoperative analgesic use were extracted from the electronic medical record. Pain intensity was measured with the Numerical Rating Scale (NRS) and were recorded preoperatively, and every 4 h during the hospital postoperative period. NRS/0–10 is a validated subjective measure of pain intensity in children aged 7–17 years and adults (von Baeyer, 2009). Acute postoperative pain was measured by taking the area under the curve (AUC) of NRS-11 pain scores over POD 0 and 1 post surgery. Postoperative opioid exposure was calculated as weight adjusted morphine milligram equivalents (MEq) summed over postoperative days 0 to 1 and included intraoperative intravenous, postoperative intravenous, patient-controlled analgesia (PCA) and oral opioid doses. In addition, Data regarding use of analgesic adjuvants such as acetaminophen, non-steroidal anti-inflammatory drugs (NSAIDs), gabapentin/pregabalin, use of regional anesthesia such as epidural/nerve catheters, methadone, ketamine and dexmedetomidine were collected.

4.3.1. Survey and questionnaires

Questionnaire to assess psychosocial risk (administered preoperatively), pain attributes, functional disability and quality of life (administered preoperatively and 2–6, 10–12 months after surgery) are described below. To support retention, participants received $50 for enrollment and completion of the initial questionnaires and $20 for each follow-up visit at either 2 to 6 months or 10 to 12 months. 4.3.1.1 Pediatric pain screening tool (PPST) is a simple nine item questionnaire with physical and psychosocial sub scores, was administered preoperatively. It was previously validated for rapid identification of risk for poor pain outcomes in youth with pain (Simons et al., 2015) and we found that a PPST score ≥ 2 was associated with high risk (48.94%) of CPSP in a cohort undergoing spine fusion or Nuss procedure (Narayanasamy et al., 2022). 4.3.1.2 PainDETECT (PD) is used to describe neuropathic vs. nociceptive pain (Freynhagen et al., 2006) with high sensitivity, specificity, and positive predictive accuracy. We assessed the response to the questions “How strong was the pain during the past 4 weeks on average (on a scale of 0-10)?” and “How strong was the strongest pain during the past 4 weeks (on a scale of 0-10)?”. 4.3.1.3 The Pediatric Quality of Life Inventory (Peds-QL) measures health-related quality of life using multidimensional child self-report scales consisting of 23 items. Higher total scores indicate higher quality of life (Varni et al., 2001). 4.3.1.4 Functional Disability Inventory (FDI) is a 15-item scale that assesses the extent to which children experience difficulties in completing specific tasks (walking to the bathroom, eating regular meals, being at school) (Walker and Greene, 1991) (Range 5–60).

4.4. Chronic pain outcome

CPSP was deemed positive if any pain score > 3/10 was reported on PainDETECT at 2–6 or 10–12 month after surgery. Occurrence of infection and malignancy in the interim period leading/contributing to chronic pain was ruled out. NRS pain scores >3/10 at three months have previously been described as a predictor for persistence of pain (Gerbershagen et al., 2011). We have previously also shown that CPSP defined thus, is associated with higher functional disability and lower quality of life (Einhorn et al., 2025).

4.5. Patient sample collection and DNA extraction

Preoperative blood samples were collected during clinically indicated intravenous line placement, in EDTA tubes and stored at room temperature for no more than 2 days before DNA isolation. To minimize batch effects, blood collected from all sample sites were shipped on dry ice overnight to CCHMC for DNA isolation and processing methods. CCHMC Molecular Genetics Laboratory (Cincinnati, Ohio) performed genomic DNA isolation on all samples using either automated extraction on the Chemagen MSMI (Perkin Elmer, Baesweiler, Germany) or manual extraction using the Puregene Blood Kit (Qiagen, Louisville, KY, USA). Isolated DNA was stored at −80 °C until analyses.

4.6. DNA methylation library preparation

CCHMC DNA Sequencing and Genotyping Core (Cincinnati, Ohio) performed enzymatic conversion of DNA and prepared sequencing libraries from 200 ng of input DNA using NEB Next Enzymatic Methyl-sequencing Kit (NEB, Ipswich, MA, USA) per manufacturer's protocol for standard input libraries. Compared to bisulfite sequencing, enzymatic conversion of DNA is less harsh and produces libraries with more uniform GC coverage, and greater mapping efficiency with sensitive detection of both 5-methylcytosine (5-mC) and 5-hydroxymethylcytosine (5-hmC) (Vaisvila et al., 2021). Methylation conversion efficiency was demonstrated to be above 99.5% (Supplementary Fig. 5).

4.7. DNA methylation sequencing and methylation calling

CCHMC DNA Sequencing and Genotyping Core (Cincinnati, Ohio) performed sequencing on the Illumina NovaSeq 6000, flow cell type S4, using 150 bp paired-end reads. Approximately 162 ± 2.7 (Mean ± SEM) million raw reads of 2 × 150 bp each were retrieved from each sample with an average Phred quality score = 36. Adaptor removal and trimming of reads were accomplished using Trim Galore 0.6.0. Trimmed reads were aligned to the GRCh38/hg38 (GCF_000001405.26) genome. BAM files were deduplicated, and methylation calling was performed. Alignment, deduplication of reads, and methylation calling of CpG-sites were all accomplished using bismark 0.22.3 (see Code Availability). Average (± SEM) cytosine coverage was 6.7 ± 1.4 per patient.

4.8. Differential methylation of CpG sites

Prior to testing for CpG methylation differences associated with CPSP, we estimated and adjusted for individual cell-type fractions prior to differential methylation analysis. To determine the minimum coverage threshold required for accurate methylation calling, we compared percent methylation values across two technical replicates at multiple minimum coverage cutoffs (2×, 7×, 10×, and 20×). Pearson correlation coefficients were calculated between replicate pairs using SeqMonk, and results were visualized with scatterplots (Supplementary Fig. 5). A minimum coverage threshold of 7× (R² = mean R² for replicates =0.9155). was selected based on the highest inter-replicate correlation and clustering consistency. Processed Bismark cov files were imported into RnBeads (v2.16.0³¹) for preprocessing and differential methylation analysis. CpGs with coverage <7×, located on sex chromosomes, or overlapping known SNPs were excluded. After filtering, 27,260,077 CpG sites remained, and CpGs present in at least 90% of samples were retained for downstream analysis.

4.9. Statistical analyses

The clinical characteristics and demographics of the cohort were described using mean (± standard deviation [SD]), median (interquartile range [IQR]) and frequency (percentage) depending on data distribution. We confirmed CPSP had higher functional disability and lower quality of life at longitudinal timepoints, compared to non-CPSP group. CPSP and non-CPSP cohorts were then compared for preoperative variables including age, weight, sex, race, deprivation index, pain scores for average and strongest pain over the last four weeks, PainDETECT total score, functional disability, quality of life and PPST. In addition, we compared CPSP and non-CPSP groups by batch of DNAm and immune cell types. We also compared the groups for surgery type distribution, perioperative medications and post operative pain AUC. Variables differing between groups at p < 0.1 were considered for stepwise regression to identify covariates for downstream differential methylation analysis, with CPSP as dependent variable (described below).

4.10. Differential DNAm analyses

4.10.1. DML identification

To identify differentially methylated loci (DML), we first used the RnBeads pipeline (v2.16.0) to test associations between DNAm (dependent variable) and CPSP using linear models implemented with LIMMA, fitted with an empirical Bayes approach (Muller et al., 2019; S, 2004). Models were adjusted for demographic variables and technical factors, specifically age, sex, race, and batch. CpG sites with P ≤ 0.05 in this stage were advanced to the next step. They were re analyzed using logistic regression in R (version: 4.5.0) with CPSP status as the dependent variable and the CpG beta as the independent variable. Models included variables (p < 0.1) identified above as covariates. Differential DNAm at a CpG site was considered significant if it showed association with CPSP at Benjamini Hochberg false discovery rate threshold of p < 0.1.

To evaluate the robustness of CpG associations identified in the second step of epigenome-wide association analysis, we performed nonparametric bootstrap resampling. For each CpG site, a multivariable logistic regression model was fit with the binary outcome as the dependent variable and CpG methylation level as the primary predictor, adjusting for monocyte proportion, natural killer cell proportion, CD4+ T-cell proportion, CD8+ T-cell proportion, granulocyte proportion, preoperative pain score, sex, age, race, and batch. Bootstrap validation was conducted using 1000 resamples with replacement at the participant level. For each bootstrap replicate, a sample of N participants (equal to the original analytic sample size, N = 166) was drawn with replacement from the original dataset. The adjusted logistic regression model was then refit to the resampled dataset, and the CpG regression coefficient and corresponding Wald p-value were recorded. This process was repeated 1000 times for each CpG site.

For each CpG, the following bootstrap summary statistics were calculated:

• 1.Median odds ratio (OR): the median of the exponentiated bootstrap regression coefficients, representing the typical effect size across bootstrap samples.
• 2.Bootstrap 95% confidence interval (CI): the 2.5th and 97.5th percentiles of the bootstrap odds ratio distribution.
• 3.Direction consistency: the proportion of bootstrap replicates in which the regression coefficient had the same sign as the coefficient estimated from the original dataset.
• 4.P-value stability (selection frequency): the proportion of bootstrap replicates yielding a nominally significant association (p < 0.05).

Direction consistency was used to assess the stability of the estimated effect direction, whereas p-value stability quantified the reproducibility of statistical significance across resampled datasets. CpG sites were retained for further downstream analysis when they met the predefined robustness criteria, including consistent directionality across bootstrap samples (>99%), moderate to high p-value stability (>70%), and bootstrap confidence intervals excluding the null value (OR = 1) (Davison and Hinkley, 1997; Wang et al., 2025; Steyerberg et al., 2001).

For each CpG, the following statistics were also calculated.

Original regression coefficient (β): the CpG coefficient estimated from the original dataset.

Bootstrap mean coefficient: the mean CpG coefficient across the 1000 bootstrap samples.

Bootstrap standard error (BootSE): the standard deviation of the bootstrap coefficient estimates.

Percentile bootstrap 95% confidence interval (Perc_LCL, Perc_UCL): the 2.5th and 97.5th percentiles of the bootstrap coefficient distribution.

Bias-corrected and accelerated (BCa) 95% confidence interval (BCa_LCL, BCa_UCL): bootstrap confidence intervals adjusted for bias and skewness in the bootstrap distribution.

4.10.2. DMR identification

To identify differentially methylated regions (DMR), we used DMRcate (version:3.4.1) package in R (Peters et al., 2015) to detect clusters of spatially correlated CpG sites among the significant loci from the refinement step, and statistical significance was determined after FDR correction (FDR < 0.05). Beta-values were used as input, and default smoothing parameters were applied (lambda = 1000, C = 2) to define region boundaries.

4.11. Genomic annotation and enrichment analysis

4.11.1. Genomic annotation

Annotation of the differentially methylated CpG (DM CpG) sites (DML) was performed using the annotatr R package (version: 1.34.0) with the hg38 genome build. Annotated genomic features included promoter regions (0–2 kb upstream of transcription start sites), exons, introns, 3′ and 5′ untranslated regions (UTRs), intergenic regions, and CpG island–associated elements (islands, shores, shelves, and inter-CpG regions). Because CpG loci can map to more than one genomic feature, sites were allowed to carry multiple annotations. Annotation results were summarized to report the frequency of CpG distribution across major genic and CpG island categories and were visualized as bar plots in R (version: 4.5.0). For functional interpretation, CpG sites were assigned to the nearest annotated gene using ChIPseeker annotatePeak (version: 1.44.0). Genomic coordinates and transcript models were supplied by TxDb.Hsapiens.UCSC.hg38.knownGene (version: 3.21.0), and gene annotation mappings such as Entrez identifiers and gene symbols were provided by org.Hs.eg.db (version:3.21.0) Transcription start site proximity was defined as 1500 bases upstream to 500 bases downstream of the TSS. Only genes with valid Entrez identifiers were retained for enrichment testing.

4.11.2. Gene ontology and pathway analysis for DML and DMR

GO enrichment was performed with topGO (version: 2.60.1) using the weight01 algorithm and the Fisher statistic. KEGG pathway enrichment was performed with clusterProfiler (version: 4.16.0) using the enrichKEGG function, and Reactome pathway enrichment was performed with ReactomePA (version:1.52.0). To provide a biologically relevant context, enrichment analyses were performed against a CPSP-specific gene sets and genome-wide background. The ToppFun application of the Toppgene suite, a computational platform for gene list enrichment analyses and machine learning based candidate gene ranking, was used to curate the CPSP-specific gene set as described in prior studies conducted by the corresponding author (Chidambaran et al., 2020; Chidambaran et al., 2021a). The knowledge base for ToppGene Suite is a comprehensive collection (>12 million gene annotations) of diverse gene annotation libraries compiled from a variety of publicly available gene annotation databases and resources. A literature curated “training set” of 31 genes were used as input for enrichment in Toppfun, using fourteen gene ontology (GO) annotation categories downloaded from NCBI Entrez Gene ftp site to do this ranking. The CPSP-specific gene set included the 1305 prioritized candidate genes - and the 31 training genes curated from prior CPSP genomics studies P values were adjusted for multiple testing with the Benjamini and Hochberg method and significance was assessed at an FDR threshold of 0.05.

4.11.3. Motif and transcription factor analysis

Motif and transcription factor enrichment was performed on DMRs, and a biologically relevant background was generated from promoter regions of the CPSP background gene list using TxDb.Hsapiens.UCSC.hg38.knownGene (version: 3.21.0) with promoters defined as 2000 bp upstream to 200 bp downstream of the TSS. Sequences were retrieved from BSgenome.Hsapiens.UCSC.hg38 (version: 1.76.0) and position weight matrices were obtained from JASPAR2020 (version: 0.99.8) and converted with TFBSTools (version:1.42.0). Motif scanning and binary enrichment testing comparing DMR versus background sequences were performed with monaLisa (version:1.10.1). Motif level results were summarized to the transcription factor level by extracting candidate TF names, reporting the minimum p and q per TF, and counting supporting motifs. Background regions were length matched to test regions to reduce composition bias. Enrichment significance was interpreted at q < 0.05 after Benjamini Hochberg correction.

• •20 DMRs enriched TF motifs in stress, inflammation, circadian, plasticity

CRediT authorship contribution statement

Siva Athitya Lakshamana Vijayarajan: Writing – original draft, Visualization, Project administration, Investigation, Data curation. Kristen N. Krolick: Writing – original draft, Visualization, Project administration, Methodology, Data curation. Xue Zhang: Writing – review & editing, Visualization, Formal analysis, Data curation. Lisa J. Martin: Writing – review & editing, Supervision, Project administration, Methodology, Formal analysis. Susan Glynn: Writing – review & editing, Project administration, Investigation, Data curation. Kristi Cole: Writing – review & editing, Methodology, Data curation. Victor Garcia: Writing – review & editing, Resources, Project administration, Investigation. Hong Ji: Writing – review & editing, Investigation, Formal analysis. R.J. Ramamurthi: Writing – review & editing, Project administration, Investigation, Data curation. L.M. Einhorn: Writing – review & editing, Supervision, Project administration, Investigation, Data curation. C.L. Monitto: Writing – review & editing, Supervision, Methodology, Investigation, Data curation. A. Ganesh: Writing – review & editing, Supervision, Project administration, Investigation, Data curation. Vidya Chidambaran: Writing – review & editing, Validation, Supervision, Resources, Project administration, Investigation, Funding acquisition, Conceptualization.

Declaration of competing interest

V. Chidambaran, L. J. Martin, and H. Ji are named inventors on US Patent number 11,618,924 B2 dated April 4, 2023 for “Methods and compositions relating to OPRM1 DNA methylation for personalized pain management” (CCHMC). The remaining authors have no conflicts of interest to declare.

Appendix A. Supplementary data

Supplementary material 1

Supplementary material 2

Supplementary material 3

Supplementary material 4

Supplementary material 5

Supplementary material 6

Supplementary material 7

Supplementary material 8

Data availability

The dataset generated during this study will be available at dbGaP accession number phs000522.v1.p1, study ID 52248 for use under health/medical/biomedical purpose under data use modifier – nonprofit use only with institutional review board approval. Please email the corresponding author. All open-source program options (including the sequencing QC steps, alignment, methylation calling, in-silico cell type deconvolution, and RnBeads) are publicly available and can be accessed at https://github.com/KristenKrolick/Chidambaran_Lab_Manuscript.git.