Redox-Linked Genetic and Biological Ageing Signals in Rapid Pain Progression of Knee Osteoarthritis: A Hypothesis-Generating Analysis in the Osteoarthritis Initiative

1. Introduction

Osteoarthritis (OA) is a chronic and heterogeneous disease and one of the leading causes of pain and disability worldwide. In 2020, an estimated 595 million people were affected (≈7.6% of the global population), representing a marked increase since 1990 [1]. In Spain, the EPISER2016 study estimated a high prevalence of symptomatic OA, including knee OA [2]. With population ageing and the persistence of modifiable risk factors, OA prevalence is projected to increase substantially over coming decades, with major health-economic consequences due to clinical care, pharmacological treatments, and joint replacement surgery [1,3].

Although structural changes are central to OA pathophysiology, pain is the symptom that most strongly drives disability, impaired quality of life, and health-care use. The course of pain in OA is heterogeneous; while some patients maintain stable trajectories or even improve, others experience rapid and sustained progression, suggesting the existence of distinct clinical subphenotypes or patterns [4,5,6,7]. Recent work has standardised operational definitions of rapid progression using 24-month windows, improving comparability across cohorts and supporting the development of predictive models [8,9].

Genetic studies focusing on pain rather than structural damage in knee OA remain limited, but they already suggest that pain has a partly distinct genetic architecture [10,11]. Candidate-gene associations have implicated variants in TRPV1, PCSK6 and COMT in painful versus painless knee OA, supporting nociceptive and monoaminergic contributions that are not fully explained by radiographic severity [12,13,14]. This pain-structure discordance has also been observed in the Spanish PROCOAC cohort, including patients with KL ≤2 who still reported significant pain [15]. More recently, variation in TACR1 and COL11A2 has been linked to knee OA pain and experimental pain sensitivity, consistent with a role for pain sensitisation pathways [16,17]. In addition, functional variation in SCN9A provides a clear proof of concept that germline changes in nociceptive ion channels can shape pain susceptibility [18,19]. At the polygenic level, a GWAS of knee pain in UK Biobank identified loci near GDF5 and COL27A1, indicating that common variants can influence both joint damage and the propensity to report pain [20].

Mitochondrial biology offers a plausible mechanistic bridge between joint pathology and pain amplification. Mitochondria regulate cellular bioenergetics and are major sources of reactive oxygen species (ROS), which act as redox signalling mediators in inflammation and stress responses. Mitochondrial DNA (mtDNA) haplogroups have emerged as modifiers of OA phenotypes; several cohort studies and meta-analyses have shown that specific mtDNA haplogroups influence the prevalence, incidence and progression of OA, including data from the Osteoarthritis Initiative (OAI) and other populations [21,22,23,24]. These effects are consistent with experimental evidence showing that mitochondrial dysfunction, mtDNA damage and oxidative stress contribute to chondrocyte apoptosis, matrix degradation and joint inflammation [25,26,27]. Importantly, mitochondrial energetics and redox imbalance can also modulate neuroimmune pathways and nociceptor sensitisation, raising the possibility that mitochondrial DNA (mtDNA) background influences pain trajectories in knee OA [28,29].

Markers of biological ageing provide an additional layer of vulnerability relevant to redox biology. Leukocyte telomere length reflects cumulative exposure to oxidative stress and systemic low-grade inflammation, processes implicated in OA and in peripheral/central sensitisation of pain pathways [30,31,32].

Combining inherited genetic variation with ageing-related biomarkers may help define a biologically meaningful subphenotype of rapid pain worsening that is plausibly linked to redox imbalance and oxidative stress, processes that can modulate both joint tissue homeostasis and nociceptor sensitisation. To address this hypothesis, we performed a hypothesis-generating exploratory analysis in knee OA participants from the OAI, combining three complementary layers of information: (i) an exploratory single-cohort GWAS to screen for nuclear variants associated with rapid pain progression, (ii) an association analysis of mitochondrial DNA haplogroups as proxies of mitochondrial bioenergetics and redox balance, and (iii) in a smaller subsample, an evaluation of leukocyte telomere length as a marker of biological ageing and cumulative oxidative/inflammatory burden. Together, these analyses were designed to identify candidate genetic and ageing-related signals consistent with an oxidative stress–related contribution to rapid pain progression, providing a framework for future replication and mechanistic studies.

2. Materials and Methods

2.1. Cohort Description

For this study, we used data from the OAI, a prospective, multicentre cohort designed to investigate the onset and progression of knee OA, with publicly accessible longitudinal clinical data, imaging and biospecimens (https://nda.nih.gov/oai) [33]. Participants were followed annually for up to ten years using standardised protocols to quantify changes in knee health over time, capturing the onset and progression of symptoms and functional limitations, the development and worsening of structural abnormalities, and longitudinal variation in imaging and biomarker-based indicators of knee OA, with ancillary assessments in other joints when available.

For the descriptive characterisation of the study cohort, we included key clinical variables available in the OAI dataset. Specifically, sex, age and body mass index (BMI) were recorded at baseline for OAI participants.

All participating clinical centres complied with the Declaration of Helsinki and obtained local ethics approval. All participants provided written informed consent.

2.2. Rapid Pain Progression Criteria

We operationalised rapid pain progression using scaled WOMAC pain scores (0–100) and standardised 24-month windows, following previously published approaches [8,9]. Participants were classified as rapid pain progressors if, in any valid 24-month window, they met at least one of the following criteria:

• Increase of at least 10 points over 24 months and substantial pain at the end of the window (WOMAC pain ≥40); OR
• Increase of at least 20 points over 24 months with end pain ≥35; OR
• Sustained substantial pain (WOMAC pain ≥40 at both the beginning and end of the 24-month window).

A 24-month window was considered valid if WOMAC pain scores were available at all annual visits within the window. Participants were classified as non-rapid pain progressors only if they had at least one valid observation period of 48 months or longer in which they did not meet rapid pain progression criteria in any embedded 24-month interval. This conservative control definition was used to reduce misclassification of participants with intermittent rapid worsening or missing follow-up assessments.

2.3. Genetic Analyses

Genetic information for this study was obtained from two different sources: nuclear genome-wide data and mitochondrial data.

2.3.1. Nuclear Genome-Wide Data

We used nuclear genome-wide genotype data generated as part of the Genetic Components of Knee Osteoarthritis (GeCKO) study [34] in OAI participants. These data had been previously deposited in dbGaP (accession phs000955.v1.p1) and were used to extract single nucleotide polymorphisms (SNPs) across the genome for association analyses.

Genotyping was performed in 4,219 OAI participants using the Illumina HumanOmni2.5 BeadChip. Quality control (QC) was conducted with PLINK 2.0 [35] and excluded single nucleotide polymorphisms (SNPs) with low call rate, duplicates, monomorphic variants, minor allele frequency (MAF) < 0.05 or significant deviation from Hardy–Weinberg equilibrium (p < 1×10⁻⁶). At the sample level, individuals with >3% missing genotypes were removed; at the variant level, SNPs with >2% missingness were filtered out. We further excluded duplicate samples, heterozygosity outliers (>±4 SD from the mean) and ancestry outliers (>±8 SD from the European cluster in the 1000 Genomes reference principal component analysis (PCA) space). Post-QC VCF files were imputed on the Michigan Imputation Server (Minimac4), retaining variants with imputation quality (Rsq) >0.8, yielding approximately 8 million SNPs.

2.3.2. Mitochondrial Data

Mitochondrial genetic variation was captured through European mitochondrial DNA (mtDNA) haplogroups, which had been previously assigned by our group to OAI participants following a single base extension (SBE) assay-based protocol described elsewhere [21].

2.4. Telomere Length Assay

A subset of (N=301) OAI participants was used to examine the influence of baseline telomere length on the risk of rapid pain progression. The telomere length of peripheral blood leukocytes (PBLs) from this subset of patients had been previously characterised [31] using a validated quantitative PCR-based assay, as described elsewhere [36]. Briefly, this method quantifies the average ratio of telomere repeat copy number to a single-copy reference gene (36B4) in each sample (T/S ratio) and provides an indirect estimate of mean telomere length.

2.5. Statistical Analyses

Data were analyzed using IBM SPSS Statistics, version 29 (IBM Corp., Armonk, NY, USA) and R software, version 4.4.1 (R Foundation for Statistical Computing, Vienna, Austria).

The genome-wide association study (GWAS) was performed using PLINK 2.0 for the post-imputation processing, and rvtests [37] for the association analyses under an additive genetic model, adjusting for age, sex, body mass index (BMI) and the first three principal components of ancestry derived from PCA on a pruned set of hard-called SNPs using stricter missingness and relatedness thresholds. The final dataset comprised 2,946 QC-passed OAI participants of European ancestry and 7,762,204 SNPs, with rapid pain progression (yes/no) as the outcome. Genome-wide significance was defined as p<5×10⁻⁸. To summarise the results and prioritise independent association signals, we applied linkage disequilibrium (LD)-based clumping in PLINK 2.0 (r² > 0.2 within the study sample), retaining the lead SNP with the lowest p-value in each linkage disequilibrium cluster. Lead variants with p<5×10⁻⁶ were considered suggestive and were reported as hypothesis-generating candidates.

Haplogroup and allele frequencies between rapid pain progressors and non-rapid pain progressors were compared using chi-square tests from contingency tables. Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated to assess the odds of carrying each allele in rapid pain progressors compared with non-progressors. For the haplogroup analysis, we followed a previously described approach, comparing each haplogroup with all remaining haplogroups pooled into a single reference category. In addition, a regression model adjusting for the confounder variables of sex, age and BMI was constructed to assess independent associations between mtDNA haplogroups and rapid pain progression.

PBL telomere length (T/S ratio) was first compared between rapid and non-rapid pain progressors using the non-parametric Mann-Whitney U test. We then performed a logistic regression, adjusting for sex, age and BMI, to evaluate PBL telomere length as a potential risk factor associated with rapid pain progression.

2.6. Declaration of Generative AI and AI-Assisted Technologies in the Writing Process

During the preparation of this work the author(s) used ChatGPT in order to adjust the different R scripts. After using this tool, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the publication.

3. Results

3.1. Study Population

The complete dataset included a total of 3,395 OAI participants of European ancestry, of which 1,906 (56.1%) were women and 1,489 (43.9%) were men. Overall, 1,260 (37.1%) participants met the criteria for rapid pain progression and 2,135 (62.9%) were classified as non-rapid pain progressors. A descriptive analysis of this population showed that rapid pain progressors were more likely to be female (p<0.001) and to have higher BMI (p<0.001) than non-rapid pain progressors (Table 1).

3.2. Nuclear Genome-Wide Analysis

We conducted a GWAS to identify nuclear genetic variants associated with rapid pain progression. As described in the Methods section, the final GWAS dataset included 2,946 QC-passed OAI participants of European ancestry and 7,762,204 imputed SNPs, with rapid pain progression (yes/no) as the outcome. Overall, the GWAS showed no evidence of substantial test statistic inflation (genomic inflation factor λ = 0.995), and the quantile-quantile (QQ) plot suggested adequate control of population stratification and relatedness (Supplementary Figure S1). The Manhattan plot is shown in Figure 1.

No SNP reached the conventional threshold for genome-wide significance (p < 5×10⁻⁸). However, we identified 6 independent loci that met the predefined suggestive significance threshold (p < 5×10⁻⁶) and were therefore considered candidate signals for further evaluation (Table 2). The strongest association signal was observed for rs73631790 on chromosome 19q13.43 (p = 7.195×10⁻⁷), located in the proximal 5′ flanking region of DUXA, a double homeobox transcription factor gene. Another prominent signal, rs2862774 on chromosome 2q14.1 (p = 1.03×10⁻⁶), maps intronically within IL36B, which encodes the interleukin-36β cytokine. Among the remaining loci, rs9912678 on 17p11.2 (p = 3.37×10⁻⁶) lies in the upstream flanking region of RAI1; rs57987665 on 14q21.3 (p = 3.62×10⁻⁶) falls in an intergenic region with the closest annotated transcript being the long non-coding RNA LOC105378178; rs34699810 on 3p26.1 (p = 4.11×10⁻⁶) is intergenic and nearest to the antisense lncRNA LMCD1-AS1; and rs62147861 on chromosome 2 (p = 4.06×10⁻⁶) is located in proximity to the uncharacterised non-coding RNA LOC124907827. Effect sizes for these variants were modest to moderate (Table 2), as expected for complex traits.

Given the limited sample size and the absence of genome-wide significance findings, these association signals should be interpreted as exploratory and hypothesis-generating. Replication in independent cohorts and functional follow-up will be required to confirm their role in rapid pain progression.

3.3. Mitochondrial DNA Analysis

Haplogroup data were available for 3,357 OAI participants, of which 1,239 (36.91%) were rapid pain progressors and 2,118 (63.09%) were non-rapid pain progressors. The frequency distribution of major European mtDNA haplogroups ranged from 9% for haplogroup J to 40.9% for the most common haplogroup H (Table 3). In unadjusted analyses, only mtDNA haplogroup H showed a nominally significant over-representation in the rapid pain progressors group (43.4%) compared with non-rapid progressors (39.4%) (OR = 1.179; 95%CI = 1.023-1.359; p = 0.023). In addition, the pooled “other” haplogroup category (individual haplogroup frequency <5%) was nominally associated with a lower risk of rapid pain progression (OR = 0.792; 95%CI = 0.650-0.966; p = 0.021); however, given its heterogeneous composition, this finding was considered exploratory and was not interpreted further. No other haplogroup showed a statistically significant difference in frequency (Table 3).

In multivariable logistic regression, carriers of haplogroup H remained more likely to be rapid pain progressors (OR = 1.172; 95%CI = 1.012-1.357; p = 0.034), and rapid pain progression was also independently associated with female sex (OR = 1.494; 95%CI = 1.289-1.732; p < 0.001), age (OR = 1.009; 95%CI = 1.001-1.017; p = 0.031) and higher BMI (OR = 1.113; 95%CI = 1.095-1.131; p < 0.001) (Table 4).

3.4. Telomere Length Analysis

In the subset of 301 OAI participants with available PBL telomere measurements, 103 (34.22%) were classified as rapid pain progressors and 198 (65.78%) as non-rapid pain progressors. Baseline PBL telomere length (T/S ratio) was 0.825 ± 0.268 in rapid pain progressors and 0.985 ± 0.375 in non-rapid pain progressors (p < 0.001, Mann–Whitney U test). In multivariable logistic regression adjusting for sex, age and BMI, PBL telomere length was inversely associated with rapid pain progression (OR per 1-unit increase in T/S = 0.260; 95%CI = 0.098-0.693; p = 0.007) (Table 5).

4. Discussion

This hypothesis-generating study aimed to investigate redox-linked genetic and biological ageing determinants of rapid pain progression in knee OA using participants from the OAI. Using a standardised 24-month window definition, we combined a discovery GWAS with mtDNA haplogroup profiling, and additionally evaluated baseline PBL telomere length in a subset of participants. As an exploratory GWAS without external replication or meta-analysis, these findings should be interpreted cautiously: no variants reached the conventional genome-wide significance, but six suggestive loci were identified as hypothesis-generating candidates. In parallel, mtDNA haplogroup H was nominally over-represented among rapid pain progressors, and shorter baseline telomere length was associated with increased odds of rapid pain progression after adjustment for sex, age and BMI.

To our knowledge, none of the six lead SNPs identified in this study has been previously reported as associated with OA or pain phenotypes, nor do they overlap with the established set of genome-wide significant loci for structural OA described in recent large-scale meta-analyses [11,38,39,40]. Notably, our GWAS focused on a strictly defined longitudinal phenotype of rapid pain worsening, whereas most prior OA pain genetics has examined cross-sectional symptomatic status or broader pain traits; together with limited power in a single-cohort design, this may partly explain the lack of overlap with previously reported loci.

At the gene level, IL36B encodes interleukin-36β, a member of the IL-1 cytokine family that stimulates the production of IL-6, IL-8 and matrix metalloproteinases by synovial fibroblasts and articular chondrocytes, and IL-36 family cytokines are increasingly recognised as mediators of joint inflammation [41,42,43]. Moreover, spinal IL-36γ/IL-36R signalling has been shown to participate in the maintenance of chronic inflammatory pain in preclinical models, supporting a plausible link between this locus and pathways of inflammatory nociception [43,44]. Notably, IL-36 signalling has also been shown to modulate glutathione homeostasis and reactive oxygen species (ROS) resolution in experimental systems, providing a potential mechanistic bridge between inflammatory activation and redox imbalance [45].

RAI1, the nearest gene to the 17p11.2 signal, is the dosage-sensitive transcription factor responsible for Smith–Magenis syndrome, a neurodevelopmental disorder characterised by altered temperature and pain sensitivity, and RAI1 dosage modulates pain sensitivity in mouse models [46,47]; in addition, RAI1 has recently been prioritised as a putative effector gene at an all-OA locus in integrative analyses that combine OA GWAS with osteoclast regulatory data [48]. Beyond nociception-related phenotypes, patient-derived and cellular models of RAI1 haploinsufficiency have reported both mitochondrial and autophagy alterations accompanied by increased oxidative stress/ROS signatures, further supporting a redox-linked biological plausibility for this locus [49].

By contrast, DUXA is a paralogue of DUX4, an embryonic transcription factor causally involved in facioscapulohumeral muscular dystrophy, and although members of the DUX family are important regulators of early myogenic programmes, no direct role of DUXA in OA or pain has yet been reported [50,51]. However, DUX4 has been tightly connected to oxidative stress biology in muscle, including evidence that oxidative stress can upregulate DUX4 expression and that DUX4-related phenotypes involve ROS/DNA damage pathways that can be mitigated by antioxidant strategies, providing an indirect redox-relevant context for this gene family [52].

The remaining loci map to poorly characterised long non-coding RNAs: LMCD1-AS1, which has been implicated as an oncogenic lncRNA in osteosarcoma, cervical and thyroid cancers, and the uncharacterised ncRNAs LOC105378178 and LOC124907827, for which no specific function in musculoskeletal disease or nociceptive pathways is currently known [53,54].

Taken together, our findings did not identify definitive genome-wide significant variants for rapid pain progression, but they highlighted a small set of biologically plausible candidate loci, particularly at IL36B and RAI1, with potential links to inflammatory–redox mechanisms that merit further investigation. In line with the single-cohort design and the use of a suggestive significance threshold, these association signals should be regarded as exploratory and primarily hypothesis-generating, and replication in independent cohorts and functional studies will be required to confirm whether these loci contribute to the genetic architecture of rapid pain worsening in knee OA.

Our finding that mtDNA haplogroup H was nominally over-represented among rapid pain progressors is consistent with a growing body of evidence indicating that mitochondrial genetic background can modulate OA phenotypes. In European-descent populations, mtDNA haplogroups have been linked to knee OA prevalence and severity, as well as to disease incidence and progression across cohorts including the OAI, supporting the concept that mtDNA lineages may influence joint vulnerability beyond traditional risk factors [21,23,55,56]. Mechanistically, mtDNA haplogroups differ in oxidative phosphorylation performance and redox balance; haplogroup H has been associated with higher oxygen consumption and oxidative damage compared with haplogroups such as J, which has been discussed as a more uncoupled/low-ROS background in OA-related contexts [22,57]. Given the established links between mitochondrial dysfunction/oxidative stress and neuroimmune signalling in pain sensitisation, mtDNA haplogroups could plausibly contribute not only to structural trajectories but also to pain trajectories in knee OA [28,58]. Nevertheless, because the observed association is nominal and the “Other” haplogroup category aggregates heterogeneous lineages, these results should be interpreted cautiously and require replication in independent datasets with comparable longitudinal pain phenotyping.

In the telomere substudy, rapid pain progressors exhibited shorter baseline leukocyte telomere length, supporting the notion that biological ageing and cumulative systemic stress may modulate symptomatic worsening in knee OA. Prior work in the OAI has linked shorter leukocyte telomere length to radiographic knee OA and greater structural burden, and longitudinal analyses have further suggested that accelerated telomere attrition is associated with incident knee OA during follow-up [31,59]. Importantly, telomere length has also been related to the biological burden of chronic knee OA pain: individuals with higher chronic pain severity, and/or high perceived stress in the context of chronic knee OA pain, show shorter telomeres, consistent with an accelerated ageing signal in more severe symptomatic presentations [30,60]. Conceptually, telomere length is an integrative marker of biological aging, influenced by oxidative and inflammatory stressors that are relevant to OA pathophysiology and pain sensitisation [32]. While our results align with this framework, they derive from a subset of patients and, therefore, warrant cautious interpretation and replication, ideally in cohorts with repeated telomere measures and harmonized longitudinal pain phenotypes.

This study has several limitations. First, the genome-wide component was conducted in a single discovery cohort without external replication or meta-analysis, and no associations reached conventional genome-wide significance; therefore, the identified loci should be regarded as exploratory and hypothesis-generating. Second, although we used a strict and clinically meaningful definition of rapid pain progression, misclassification remains possible given the inherent variability of self-reported pain and the influence of time-varying factors such as analgesic use, intra-articular treatments and comorbidities. Third, the telomere analysis was restricted to a relatively small subset of participants, which limits power and may introduce selection bias, and the “other” mtDNA haplogroup category aggregates heterogeneous lineages, reducing biological interpretability. As contingency measures, we report effect sizes with confidence intervals, emphasise internal consistency across complementary genetic layers (nuclear GWAS, mtDNA haplogroups and telomere length), and interpret all signals cautiously within an explicitly exploratory framework. Future work will prioritise replication in independent cohorts with harmonised longitudinal pain phenotyping and, where possible, repeated telomere measurements and functional follow-up to strengthen causal inference and clarify underlying mechanisms.

5. Conclusions

This hypothesis-generating analysis (Figure 2) suggests that rapid pain progression in knee OA may be associated with biological signals consistent with a redox-linked biology framework, spanning nuclear immune–neuroimmune genetic variation (including a suggestive IL36B signal), mitochondrial genetic background, and systemic biological ageing indexed by leukocyte telomere length. Together, these findings provide hypothesis-generating leads for replication and mechanistic studies; however, as oxidative stress was not directly measured and the genome-wide component lacked external replication, independent validation, functional follow-up and integrative multi-omics studies will be required to confirm these associations and to identify actionable redox and mitochondrial targets for preventing rapid pain worsening.