Telomere Length Dynamics as a Biomarker of Individual Radiation Sensitivity and Pneumonitis in Lung Cancer Patients Receiving Thoracic Radiation Therapy

Abstract

Background:

Telomere shortening is a biomarker for genome instability and aging, and the vulnerability of telomeric DNA to oxidative damage suggests its potential role in mediating radiotherapy (RT) side effects. This study evaluates telomere length (TL) as a biomarker for clinical radiosensitivity and adverse outcomes in thoracic RT-treated patients.

Methods:

Cancer patients receiving thoracic RT (2019–2022) were prospectively enrolled at Brigham and Women’s Hospital, Boston, Massachusetts. Peripheral blood mononuclear cells (PBMCs) were collected pre-RT and up to 12 months post-RT. TL was measured using quantitative PCR, and multi-pathway DNA repair capacity (DRC) was simultaneously assessed by fluorescence multiplex host cell reactivation (FM-HCR) assays. RT outcomes included patient-reported quality of life and radiation pneumonitis. Linear mixed-effects models were used to analyze TL dynamics; risk prediction models for RT outcomes were evaluated using area under the curve.

Results:

Pre-RT TL decreased with age (0.44% lower per year, 95% CI: 0.12%, 0.77%) and advanced cancer stage (6.87% lower per step increase of stage, 95% CI: 3.45%, 10.16%). Radical RT was associated with telomere shortening (3.7% lower, 95% CI: 0.27%, 7.07%) in PBMCs, detectable up to 6 months post-RT. Pre-RT TL strongly predicted post-RT changes, and TL dynamics outperformed static measures in predicting symptom burden and radiation pneumonitis. Positive associations were observed between TL and DRC against oxidative lesions, with A:8oxoG repair capacity mediating 12.8% of RT-induced TL shortening.

Conclusions:

Lymphocyte TL can reflect individual radiosensitivity and interact with oxidative damage repair. Longitudinal assessment of TL dynamics provides additional predictive value for adverse RT outcomes compared to static measures. Further studies are needed to fully determine the clinical utility of TL.

Introduction

Radiotherapy (RT) is a standard treatment for many solid tumors, providing tumor control through the targeted delivery of typically fractionated doses of ionizing radiation (IR). Advanced RT techniques, including stereotactic body RT (SBRT) and intensity-modulated RT (IMRT), are employed to optimize dose distributions based on specific patient conditions, tumor geometry, and treatment intents.^1,2 RT is particularly critical in the management of lung cancer, which remains the leading cause of cancer-related mortality worldwide, primarily because many patients present with advanced-stage disease at diagnosis.^3,4 More than half of lung cancer patients ultimately receive RT at varying points in their treatment journey, whether for curative, adjuvant/neoadjuvant, or palliative reasons.⁵ However, patient responses to RT exhibit marked variability, with some experiencing severe toxicity while others have milder or delayed adverse events.⁶ In thoracic RT, common toxicities include radiation dermatitis, esophagitis, and pneumonitis.^7,8 These side effects can limit significantly delivery of higher RT doses and thus treatment effectiveness as well as impose a temporary or long-term impact on quality of life. A deeper understanding of the biological underpinnings that drive inter-individual differences in RT response is therefore warranted.

RT exerts much of its cytotoxic effect by inducing DNA damage, thus implicating pathways that maintain genome stability as key determinants of both tumor and normal tissue radiosensitivity. Oxidative damage, in particular, plays an essential role in RT-induced tissue injury.⁹ Given that guanine-rich chromosomal regions are especially vulnerable to reactive oxygen species (ROS), telomeres are thus prime targets for radiation-induced oxidative stress with its structure composed of tandem TTAGGG repeats.^10,11 Telomere length (TL) is widely recognized as a hallmark of aging and cancer, and emerging data suggest it can also serve as a biomarker of radiation sensitivity due to its links with DNA damage and repair processes.^12–16 Specifically, active telomere maintenance involves preventing spurious end-fusions through inhibition of non-homologous end joining (NHEJ),^17,18 and the integrity of telomeric DNA may be further safeguarded by base-excision repair (BER) pathways that mitigate oxidative lesions, such as 8oxoG.^19–21 Yet, the extent to which these mechanisms operate during RT and across diverse clinical settings remains unclear, necessitating systematic investigation.

Lymphocyte TL can be measured longitudinally in samples collected via minimally invasive blood draws and has been studied extensively as an aging biomarker in large population cohorts.^22,23 Prior research links human telomere shortening to various environmental radiation exposures, including those experienced by astronauts and atomic-bomb survivors.^24–27 In clinical practice, lymphocyte-based indicators of genomic integrity have long been employed to infer radiosensitivity in solid tissues, offering insight into how systemic cell populations respond to DNA damage.^28,29 Nevertheless, previous RT-specific studies often lack robust longitudinal assessments, sufficiently large cohorts, or detailed patient-level clinical profiles and toxicity information.^30,31 Specifically, no study has evaluated whether telomere attrition correlates with RT toxicity in lung cancer patients, nor has any research integrated TL measurements with functional assays of multiple DNA repair pathways.

Here, we addressed these knowledge gaps by investigating a diverse cohort of lung cancer patients undergoing RT, measuring lymphocyte TL before and after treatment. By integrating longitudinal TL data with patient characteristics, detailed toxicity outcomes, and multiplexed assays of DNA repair capacity (DRC), we characterized aging-associated and RT-induced TL dynamics. We further identified patient-specific factors that modulate these changes and how these changes contributed to subsequent adverse events. Additionally, we examined how distinct DNA repair pathways intersect with TL dynamics.

Material and Methods

Study Design and Patient Cohort

This study enrolled adult patients treated with RT at the Radiation Oncology Department of Brigham and Women’s Hospital between 2019 and 2022 in Boston, Massachusetts, USA. Eligible participants included those with primary or metastatic tumors receiving RT targeting sites within the lung. Patients were excluded if they were deemed temporarily or permanently unfit to consent to research by their treating radiation oncologist or if they were medically unstable at the time of RT mapping. Among the 177 eligible patients, those with insufficient blood cell counts for research sampling were excluded, resulting in a final cohort of 100 patients. Figure 1 presents an overview of our study design. All participants provided written informed consent prior to enrollment. The study was approved by the Institutional Review Board at Brigham and Women’s Hospital (MGB IRB #2016P001582).

Figure 1. Study design and workflow.

This study enrolled 100 adult patients undergoing thoracic radiotherapy (RT) between 2019 and 2022, along with a control group of 24 healthy individuals. Peripheral blood mononuclear cells (PBMCs) were collected at three time points: pre-RT (baseline), mid-RT (~6 weeks into treatment), and post-RT (1 month to 1 year after treatment). Follow-up visits concluded in September 2023, capturing acute and late RT-related outcomes through validated patient-reported questionnaires (EORTC QLQ-LC13, PRO-CTCAE) and medical records. Baseline demographics, cancer diagnostics, RT regimens, and medical history were collected from medical records and patient assessments. Genome integrity biomarkers, including telomere length (TL) and DNA repair capacity (DRC), were measured in PBMCs. TL was assessed using monochrome multiplex qPCR (MM-qPCR) and validated via flow-FISH and whole-genome sequencing (WGS). DRC was quantified using fluorescence multiplex host cell reactivation (FM-HCR) assays, covering all major DNA repair pathways. The study employed three analytical approaches: association analysis was used to identify factors influencing baseline TL and its response to RT; risk prediction was conducted to evaluate the predictive value of TL dynamics for RT-related health outcomes using cross-validation and receiver operating characteristic (ROC) analysis; mediation analysis was performed to assess whether DRC modulates the impact of RT on TL.

Clinical Assessments and Radiotherapy Outcomes

Demographic and clinical characteristics including age, sex, race, active smoking status, primary cancer diagnosis, histology, and staging based on the AJCC 8th Edition Staging Manual, as well as treatment history and concurrent chemotherapy, were collected via electronic medical records. Comprehensive RT treatment parameters were extracted from institutional RT records, including dates, technique, prescribed and delivered doses, fractions, and target site. Based on these data and clinical intent, RT type was categorized as: stereotactic body RT (SBRT), radical (curative) fractionated RT, and palliative therapy. Lung V20 values were derived from dose volume histograms (DVHs). Blood cell counts were obtained from routine laboratory tests before and after RT concurrent with the blood draws. Data quality checks were conducted by radiation oncologists and trained staff to ensure consistency.

Patient-reported outcomes (PROs) were collected during or within 3 months post completion of RT via the European Organization for Research and Treatment of Cancer Quality of Life Questionnaire for Lung Cancer 13 (EORTC QLQ-LC13)³² and PRO version of the Common Terminology Criteria for Adverse Events (PRO-CTCAE) v5.0³³ questionnaires, to assess acute RT-related side effects. Three PROs were analyzed: overall symptom burden (EORTC QLQ-LC13 Quality of Life Lung Cancer Score, n = 36), difficulty swallowing (PRO-CTCAE, n = 22); and radiation skin reaction (PRO-CTCAE, n = 24). Following clinical relevance and prior literature,⁸ each was dichotomized into none-mild (≤ grade 2) versus moderate-severe (> grade 2). Late RT side effects, specifically radiation pneumonitis (with radiographic evidence), were extracted from medical records and available for 93% of patients.

Sample Collection and Biomarker Measurements

Blood samples were sequentially collected before RT (pre-RT), during RT (mid-RT), and after RT (post-RT), with most analyses focusing on pre-RT and post-RT visits; in a random subset of six patients, mid-RT samples were included for exploratory analyses on early telomere responses. While prior studies have largely relied on fixed post-treatment time points, limiting insights into the temporal telomere dynamics, we hypothesized a priori that telomere response to RT may evolve both during treatment and throughout the post-RT recovery period. Our study lent itself to capture this variability, since patients were sampled according to their ability to provide a blood sample and the timing of medically necessary follow up appointments. Consequently, post-RT samples were prospectively collected across a wide follow-up interval ranging from 1 month to over 1 year, with the majority obtained within 6 months after RT. The random allocation of sampling time points was supported by the absence of any observed associations between post-RT timing and clinical or demographic covariates (Figure S1A). We obtained blood samples from 24 healthy individuals, 1:1 matched to patients by age (±5 years) and sex for preliminary case-control comparisons.

Peripheral blood mononuclear cells (PBMCs) were isolated from whole blood via standard density gradient centrifugation. TL was assessed by monochrome multiplex qPCR (MM-qPCR)³⁴ on genomic DNA extracted using the QIAGEN DNeasy Blood & Tissue Kit (Qiagen, Cat: 69504). Briefly, each sample’s telomere repeat copy number (T) was measured relative to a single-copy reference gene (S, beta-globin). Relative TL (T:S ratio) were obtained by dividing T by S, using a reference DNA from a randomly chosen patient sample for standard curves. We additionally performed flow cytometry based fluorescence in situ hybridization (flow-FISH)³⁵ on a random subset of samples (n=14) (Figure S2) and whole genome sequencing (WGS)-based TL estimates by the TelSeq algorithm (v0.0.2)^36,37 for a majority of samples (n=97) to verify reproducibility of TL measures (Figure S3A). The detailed experimental conditions and parameters are provided in Supplementary Methods.

DRC was assessed concurrently with telomere measurements in the full cohort at two timepoints (pre-RT, post-RT) using a fluorescence multiplex host cell reactivation (FM-HCR) assay, which has been extensively validated in human lymphocytes and cell lines.^38–40 Briefly, multiplex plasmid cocktails were prepared to measure the repair activity of several DNA lesions, including uracil (U:G), hypoxanthine (Hx), 8oxoG:C, A:8oxoG, and tetrahydrofuran (THF, abasic site analog) lesions repaired by either short- and long-patch base excision repair (LP-BER); double strand breaks repaired by non-homologous end joining (NHEJ); mismatch bases (G:G) repaired by mismatch repair (MMR); ultraviolet light-induced damage repaired by nucleotide excision repair (NER), and ionizing radiation (IR)-induced lesions measured by IR (3000 Gy) plasmid reporter. The detailed data quality control strategy we employed for pre-processing FM-HCR data has been reported previously.⁴¹ The processed data were represented as pathway-specific z-scores, where a value of 0 corresponds to the cohort's average repair capacity for each pathway, and positive or negative values reflect deviations above or below the mean, expressed in units of standard deviation.

Risk Prediction Model Development

We first evaluated the predictive value of pre-RT TL alongside key demographic and clinical factors (age, sex, active smoking status, cancer stage, concurrent chemotherapy, lung V20, lymphocyte %, and neutrophil %) for TL response to RT using Random Forest (RF) and XGBoost, with feature importance quantified using RF variable importance and SHAP values in XGBoost. To predict adverse events, including PROs and radiation pneumonitis, we applied logistic regression, elastic net regression (GLMNET), and kernel support vector machine (KSVM) using a structured two-stage modeling framework to mitigate overfitting in limited-sample settings while preserving clinically important covariates. In stage one, each telomere predictor (pre-RT TL, post-RT TL, or TL change) was evaluated across 100 stratified bootstrap samples using GLMNET; the penalty parameter was selected via 10-fold cross-validation (fewer folds if <8 events). Covariates with non-zero coefficients at the minimum-deviance λ were tallied, and the top predictors were retained under a constraint of k ≤ n/10. Established clinical predictors (e.g., concurrent chemotherapy, lung V20) were forcibly retained when not consistently selected. In stage two, these selected predictors were combined with each TL metric and modeled using logistic regression and KSVM. Predictive performance was assessed using the area under the ROC curve (AUC), with leave-one-out cross-validation (LOOCV) as the primary resampling approach. Hyperparameters were tuned via Bayesian optimization within an inner 3-fold CV, and AUC confidence intervals were estimated using DeLong’s method. Continuous predictors were standardized and categorical variables dummy coded prior to modeling.

Statistical Analysis

Generalized linear models (GLMs) were first employed to evaluate associations between patient demographics (age, sex, active smoking status), clinical characteristics (cancer stage, treatment history), technical variables (lymphocyte %, neutrophil %, and batch) with baseline TL measurements. For baseline TL models, analyses were performed on the full cohort, as these samples were collected prior to the initiation of RT, enabling us to capture inter-individual differences in TL attributable to pre-treatment factors. A trend test was performed for cancer stage by modeling stage as an ordinal continuous variable to evaluate stepwise increases in disease severity. To visualize age-related TL patterns, partial residuals of age were extracted from GLMs, which yielded an adjusted measure of TL that accounted for relevant covariates (technical and demographic factors). Since TL is expressed as a relative T:S ratio, it was natural log-transformed to improve linearity when modeled as the response variable across the regression models. Effect estimates were then exponentiated to return to the relative ratio scale.²⁶

Linear mixed-effects models (LMMs) were used to examine the impact of RT on TL dynamics, with individual-specific random intercept to account for repeated measures within the same patient. Analyses were performed separately for different RT type groups (radical RT, SBRT) given the known different modalities. In addition, exploratory stratified analyses were conducted to examine RT-associated TL decline across prespecified patient strata (age, sex, smoking, chemoradiation, lung V20, and cancer stage). Technical factors (lymphocyte %, neutrophil %, and batch) were included in every model. In the multivariable models, patient-level covariates (age, sex, active smoking status, cancer stage, concurrent chemotherapy, lung V20, RT dosage and fraction) were also adjusted to control for possible confounding. In the post-RT analysis, covariate-adjusted TL values were obtained by extracting residuals from the LMMs, which were then used to visualize adjusted TL changes across post-RT follow-up times for comparison with observed data distribution.

To evaluate whether the baseline-dependent pattern of telomere length (TL) change reflected a biologically meaningful effect of radiotherapy (RT) rather than regression to the mean (RTM), we conducted a simulation-based analysis grounded in the joint distribution of baseline and post-RT TL. Specifically, TL measurements were first log-transformed and the means, standard deviations, and empirical correlation between baseline (B) and post-RT (P) log-TL values were estimated: ${\stackrel{-}{\mu }}_B=0.155,{\stackrel{-}{\mu }}_P=0.126;{\widehat{\sigma }}_B=0.131,{\widehat{\sigma }}_P=0.136;{\widehat{\rho }}_{BP}=0.709$. Because the observed correlation may be attenuated by assay imprecision, we corrected for measurement error using a reliability coefficient ($\tau$=0.798) derived from an independent comparison between MM-qPCR and flowFISH assays, yielding an error-adjusted correlation (${\widehat{\rho }}_{BP}/\tau$) intended to more closely approximate the underlying biological association. Then, under a null scenario representing RTM and random variation, but no additional biological normalizing effect, we specified a bivariate normal model for the joint distribution of log-transformed baseline and post-RT TL using the observed means, variances, and adjusted correlation. We generated 10,000 simulated datasets from this model, exponentiated simulated values back to the original scale, and applied the same baseline TL tertile cut-points used in the observed data (low <1.11, middle 1.11−1.21, high ≥1.21). For each simulated dataset, we calculated the mean percent change in TL within each tertile group. The observed tertile-specific mean TL changes were then compared with their corresponding empirical null distributions. For each tertile, the proportion of simulations with mean changes as or more extreme than the observed value was taken as a one-sided p-value. This framework directly evaluates whether the magnitude and direction of the observed TL changes can be attributed to RTM alone or exceed what is expected under a model that preserves the observed visit-to-visit correlation and mean shift but imposes no explicit normalizing mechanism. A low p-value indicates that the observed baseline-dependent TL response is unlikely to arise solely from RTM, consistent with an additional effect of RT.

The relationship between DRC and TL was evaluated by multiple regression models adjusting for demographic and technical covariates to address potential confounding. We additionally examined correlations based on covariate-adjusted TL residuals to visualize pairwise associations at the individual data point level. Mediation analysis was then performed to evaluate whether pathway-specific DRC mediated the effect of RT on TL for pathways strongly associated with TL, using a previously described permutation-based method that accommodates flexible model specifications.⁴² Specifically, the direction of causality, from DNA repair to telomere maintenance, was based on prior mechanistic and biological evidence.^19–21 We decomposed the total effect of RT on TL into a direct effect (not mediated by DRC) and an indirect effect (mediated through DRC). The indirect effect was calculated as the difference between the total and direct effects. To test the significance of the mediated (indirect) effect, we employed a permutation-based method that breaks the mediator–outcome pathway by randomly permuting the mediator, repeating this procedure 1,000 times to generate a null distribution. A two-sided permutation P-value was calculated based on the proportion of permuted differences that were as extreme as the observed. Similar frameworks were used to evaluate whether DRC mediated the relationship between age and TL. All models were adjusted for the same covariates used in the primary analysis.

To address missingness in cell-type–specific blood counts (7.3–12.2%), we assumed data were missing at random and used a Bayesian joint modeling framework to impute missing values while propagating uncertainty. Because overall and differential leukocyte counts were right-skewed, all count variables were modeled on the log scale within a multivariate hierarchical structure that jointly modeled overall white blood cell counts and individual cell-type counts. Missing overall counts were imputed as functions of demographic and clinical covariates (age, sex, smoking status, body mass index, cancer type, cancer stage, and blood draw visit number), whereas missing cell-type counts were imputed conditional on both the overall count and the same covariates. Random intercepts were included to capture inter-individual heterogeneity and within-person correlation across repeated visits. Weakly informative priors were used throughout: fixed effects followed Normal(0, 5) priors, intercepts were assigned Student-t(0, 10) priors, random effect standard deviations followed half-Cauchy priors, and residual correlations among cell types were modeled via an LKJ prior. A total of m = 20 imputed datasets were created by sampling from posterior distributions, and Rubin’s rule was applied to pool parameter estimates while reflecting imputation uncertainty.⁴³ Fully adjusted models were re-run on these imputed datasets and compared to the primary analyses to confirm consistency of the findings. Convergence was confirmed via trace plot inspection, R-hat <1.01, and adequate effective sample sizes, and posterior predictive checks demonstrated close agreement between observed and model-generated data (Figure S3B,C), supporting the adequacy of the imputation model.

All statistical analyses were conducted in R (v4.3.0). The following R packages were used: tidyverse (v2.0.0), pdrc (v1.0.1), lme4 (v1.1–35.5), lmerTest (v3.1–3), mitools (v2.4), visreg (v2.7.0), ComplexHeatmap (v2.20.0), mlr3 (v1.2.0), mlr3learners (v0.9.0), mlr3tuningspaces (v0.5.2), glmnet (v4.1.8), xgboost (v1.7.8.1), shapviz (v0.10.1), randomForest (v4.7–1.1), pROC (v1.18.5), brms (v2.21.0), and future (v1.34.0). Statistical significance was set as two-sided p < 0.05 unless otherwise specified. All analytical code used to generate the results and figures is publicly available at https://github.com/NagelLabHub/LC-Telomere.

Results

Patient Characteristics and Baseline Telomere Length

A total of 100 patients were enrolled (49 women, 51 men), with a mean age of 66 years (range, 37–96). The majority self-identified as White (n = 90), and 69 were former smokers. Nearly half (n = 45) had locally advanced cancer, and 29 had oligometastatic disease. Most patients (n = 71) underwent radical RT with a median total dose of 60 Gy delivered in a median of 27 fractions. The median follow-up after RT completion was 2.8 months (IQR 1.3–5.4). Table 1 summarizes the key demographic and clinical characteristics.

Table1.

Patient characteristics

Variable	Full cohort (n=100)	SBRT (n=24)	Radical (n=71)
Age (years)	66 (61–73)	65 (57–74)	67 (63–73)
Female	49	15	33
White race	90	20	66
Smoking
Never	21	8	11
Former	69	15	52
Current	9	1	7
NSCLC	68	13	53
Stage
Early stage	18	5	13
Locally advanced	45	2	43
Oligometastatic	29	17	11
Widely metastatic	8	0	4
Therapy type
SBRT	24	24	-
Radical	71	-	71
Palliative	5	-	-
Dosage (Gy)	60 (54–60)	56 (50–60)	60 (60–60)
Fractions	27 (12–30)	5 (5–8)	30 (27–30)
Lung V20 (%)	19 (8–26)	5 (3–8)	23 (18–28)
Chemoradiation	46	-	46
Prior radiotherapy	15	6	7
Prior chemotherapy	31	8	19
Prior immunotherapy	13	4	6
Prior targeted therapy	17	10	7
Baseline telomere length (ratio)	1.13 (1.03–1.27)	1.09 (0.97– 1.21)	1.14 (1.05–1.29)

For continuous variables, median (interquartile range) is provided, for categorical variables, count is provided. Abbreviations: NSCLC = non-small cell lung cancer, SBRT = stereotactic body radiotherapy.

We first examined the association between age and TL at baseline. After adjusting for sex, smoking status, cancer stage, neutrophil and lymphocyte percentages, and batch effects, TL declined by an estimated 0.44% per year of age (95% CI: 0.12%, 0.77%; Figure 2A). In a matched case-control subset (n = 24 : 24), no substantial differences in TL were found between case-control pairs (Figure 2B). However, within the patient group specifically, older age was associated with a faster TL decline rate (0.7% lower per year, 95% CI: 0.03%, 1.38%), whereas no significant association was observed in controls (Figure 2C).

Figure 2. Telomere attrition vulnerability increases with cancer severity.

A. Telomere length (TL) decreases with participants’ age. TL is reported as an adjusted telomere : single-copy gene (T:S) ratio, derived from partial residuals of age in a generalized linear model (GLM) adjusted for sex, smoking status, cancer stage, lymphocyte ratio, neutrophil ratio, and batch. N = 100. In panels A-E, each dot represents one individual patient. B. No significant difference in measured TL between cancer cases (mauve) and matched healthy controls (purple). The paired comparison was performed using a linear mixed-effects model (LMM), with each linked line representing matched pairs. N = 24 for patients, 24 for controls. C. Age-associated telomere shortening was found only in cancer patients (mauve). Solid lines depict smoothed trends fitted by generalized additive models (GAM) using penalized regression splines, representing the optimal non-linear relationship between age and TL. Shaded areas around each line denote the 95% confidence intervals (CIs). D. TL decreases with increasing cancer stage severity. Each central bold line shows the mean TL of each stage group, with error bars representing 95% CIs. Statistical significance from pairwise comparisons between stage groups is denoted as *p < 0.05, **p < 0.01. A trend test assessing TL across cancer stages was performed by treating cancer stage as a continuous variable in a GLM. N = 18 for early stage, 45 for locally advanced stage, 29 for oligometastatic stage, 8 for widely metastatic stage. E. Age-associated telomere shortening was found in patients with metastatic cancer. Smoothed trend lines represent generalized additive model (GAM) fits with penalized regression splines, depicting the relationship between TL and age in patients with local (blue; includes early-stage and locally advanced, n = 63) and metastatic (green; includes oligometastatic and widely metastatic, n = 37) disease. Shaded areas around each line denote the 95% CIs. F. Association between patient characteristics and TL at baseline. Each panel displays the estimated relative TL changes (ratios and 95% CIs represented by dots and error bars) for each variable based on crude (univariate), adjusted (multivariate), and pooled (multiple imputation) analyses. The vertical dashed line at 1 indicates no change; values <1 indicate relatively shorter TL, and values >1 indicate relatively longer TL corresponding to changes in the variable indicated at left. Statistical significance is indicated by a 95% CI that does not cross 1. For categorical variables, effect estimates represent the ratio of TL relative to the indicated reference category. For continuous variables, estimates represent the relative TL change per one standard deviation (SD) increase. N = 100.

Cancer Stage is Associated with Telomere Length in a Dose-Dependent Manner

Given the heterogeneous nature of our cohort, which included primarily NSCLC (n = 68) but also patients with metastatic disease involving the lung (non-lung primaries), we next explored how cancer stage might affect baseline TL. Cancer stage emerged as a strong driver of inter-patient variability (Figure 2D). A significant trend toward shorter telomeres with more advanced stage was observed (5.01% lower per step increase in stage, 95% CI: 2.03%, 7.90%), becoming even stronger after adjusting for other patient characteristics (6.87%, 95% CI: 3.45%, 10.16%). Furthermore, age-related telomere shortening was pronounced among patients with oligometastatic or widely metastatic disease (0.41% lower per year, 95% CI: 0.06%, 0.75%, n = 37; Figure 2E), but not in patients at early or locally advanced stages (n = 63).

Figure 2F provides a comprehensive overview of these associations in three models: unadjusted, adjusted for covariates, and pooled estimates from multiple imputations for missing values. In all models, cancer stage remained strongly associated with shorter telomeres (relative TL ratio compared to early stage: in crude model, locally advanced = 0.95 [95% CI: 0.88, 1.03], oligometastatic = 0.89 [95% CI: 0.82, 0.96], widely metastatic = 0.88 [95% CI: 0.79, 0.98]; in adjusted model, locally advanced = 0.91 [95% CI: 0.83, 0.99], oligometastatic = 0.85 [95% CI: 0.78, 0.94], widely metastatic = 0.81 [95% CI: 0.71, 0.92]; in imputed model, locally advanced = 0.94 [95% CI: 0.87, 1.01], oligometastatic = 0.87 [95% CI: 0.80, 0.95], widely metastatic = 0.85 [95% CI: 0.75, 0.96]). Prior therapy history was also explored and had limited influence on baseline TL, thus removed from main analysis to preserve statistical power (Table S1).

Radiotherapy-Induced Telomere Dynamics and Modifying Factors

We next investigated changes in TL following RT separately among radical RT (n = 71) and SBRT (n = 24) groups given their distinct modalities. Whereas baseline TL did not differ meaningfully between the two groups (Table 1), only patients undergoing radical RT exhibited a substantial post-treatment decline in TL (overall relative decline by 3.15%, 95% CI: 0.84, 5.40%; Figure 3A). Individualized TL profile also revealed distinct trajectories that varied in a manner that depended upon baseline TL. Patients with longer baseline telomeres showed a decrease in TL ratio post-RT (−6.68% on average in the highest tertile), whereas those with shorter baseline telomeres exhibited a modest increase (+2.01% in the lowest tertile), a pattern that cannot be explained by RTM alone as confirmed by simulation analyses. In a subset of patients (n=6) with an additional mid-RT blood draw, an early decline in TL was observed, suggesting that telomere shortening can manifest early during treatment (Figure 3B).

Figure 3. Radiotherapy (RT) effects on telomere length (TL) and blood cell composition in patients receiving radical therapy. A. TL changes pre- and post-RT, and comparison with simulated null distributions.

In panels A-E, each dot represents a single measurement, and each line (if applicable) connects longitudinal measurements from the same patient. The left panel of A displays TL measurements before and after RT, presented as the telomere-to-single-copy gene (T:S) ratio normalized to a reference sample. Color indicates RT-related ratio changes (blue: greater TL decline, red: greater TL increase, as shown in the “RT effect” color scale). N = 71. In the middle panel, patients are stratified into tertiles based on pre-RT TL (high [n = 24], middle [n = 23], low [n = 24]). Pre-RT values are normalized to 1, and post-RT values are shown as relative ratios to illustrate trends in TL change within each tertile. Titles summarize the overall trend based on linear mixed-effects models (LMMs). *p<0.05 for RT effect (post- vs pre-RT) in crude LMMs. The right panel displays simulated null distributions of percent TL change for each tertile, assuming changes are explained solely by regression to the mean (RTM) and random variation. The observed average TL change (dashed line) is overlaid for comparison. The y-axis represents the percent TL change (0% = no change). A significant deviation of the observed change from the simulated distribution suggests an additional biological effect beyond RTM. p values correspond to one-sided tests comparing observed changes to the simulated null distributions. B. TL trajectories among a subset of six patients with additional mid-RT measurements. .p=0.05, *p<0.05 for RT effect (post- vs pre-RT) in crude LMMs. Lines link individual TL patterns across three time points for the same patient. N = 6. C. Relative TL changes following RT, grouped by month since RT completion. The bold line in each group denotes the mean relative TL change, with error bars indicating 95% confidence intervals (CIs). The horizontal dashed line at 1 represents no change from baseline. Groups on the x-axis reflect time since RT completion:1 = within 1 month; 1–3 = 1 to 3 months; 3–6 = 3 to 6 months; 6–12 = 6 to 12 months; 12 = more than 12 months post-RT. N = 71. D. Relative TL change calculated as the ratio of adjusted TL post-RT to pre-RT, grouped by month since RT completion. Adjustment includes RT, age, sex, smoking status, cancer stage, concurrent chemotherapy, lung V20, RT dosage, fraction, neutrophil ratio, lymphocyte ratio, and batch. N = 71. E. Blood cell counts in patients pre- and post-RT. The left panel shows total white blood cell (WBC) count, and the right panel shows lymphocyte count. ***p<0.001 for RT effect in a crude LMM. N = 71. F. Summary of changes in blood cell composition from pre- to post-RT, and their relationship with TL. The numeric values are effect estimates of the RT variable in crude LMMs, indicating changes in absolute count (1000 cells/μl) and percentage (%) of each cell type associated with RT that reached statistical significance. For the correlations between measured TL and WBC and each cell type specific count, Spearman’s correlations were calculated at both pre-RT and post-RT visits, but no statistical significance was reached. N = 71. G. RT-associated TL changes overall and within patient strata. Dots and error bars present RT-associated changes in TL (ratios and 95%CIs) based on crude (univariate), adjusted (multivariate), and pooled (multiple imputation) analyses. The “Overall” row represents the average RT effect on TL across the entire study cohort. All remaining rows reflect strata-specific RT effects by different patient characteristics split by category or median. Sample size of each stratum is indicated on figure. The vertical dashed line at 1 indicates no change; values below or above 1 suggest relatively shorter or longer TL post-RT, respectively. Statistical significance is indicated by a 95% CI that does not cross 1.

Time-stratified analyses further revealed that telomere shortening appeared most pronounced between 3 and 6 months post-RT (Figure 3C). To characterize further post-RT telomere dynamics, we further fitted models adjusting for individual-level covariates and detected a substantial RT-induced telomere shortening from as early as 1 month through 6 months post-treatment, supporting a sustained RT effect. The residuals were then extracted from the adjusted models for comparison with crude estimates (Figure 3D).

In parallel, we observed substantial shifts in white blood cell subsets: decreased lymphocyte counts/percentages and increased neutrophil counts/percentages after RT, whereas total WBC counts remained unchanged (Figure 3E, F). Although no significant direct correlation was noted between TL and these cell counts (before or after RT), we adjusted for lymphocyte and neutrophil percentages to account for potential confounding due to RT-induced changes in circulating immune composition.^26,44

The TL decline observed in the radical RT group remained substantial after adjusting for multiple covariates, suggesting an independent RT influence on TL (overall relative decline by 3.7%, 95% CI: 0.27, 7.07%; Figure 3G). Exploratory stratified analyses conducted across key patient characteristics indicated that the magnitude of RT-associated TL change appeared greater in male patients, smokers, and those receiving concurrent chemotherapy. By contrast, lung V20 (which has previously been associated with radiation-induced lymphopenia^45,46) or cancer stage exhibited similar decline profiles across strata. This suggests that heterogeneity in RT-induced TL change may be more strongly influenced by systemic and host factors rather than localized RT dose distribution, warranting confirmation in larger, more targeted cohorts.

Due to limited sample size (n = 5), patients receiving palliative RT were excluded from the main analysis. Exploratory models adjusting for technical covariates suggested a modest post-RT TL increase in this group, but the interpretation is limited by sample size (Figure S1B). No consistent TL change was observed in the SBRT group, either in crude (by 1.71%, 95% CI: −6.40, 2.76%) or adjusted (by 1.09%, 95% CI: −4.90, 6.74%) models.

Telomere Length as a Predictor of Radiotherapy Outcomes

Based on these observations, we examined next the predictive value of TL for RT-related outcomes. In the radical RT group, we categorized patients as “decliners” (n = 44) versus “increasers” (n = 27) based on post-RT changes in TL. Decliners had higher pre-RT TL (mean [SD]: 1.20 [0.17]) compared to increasers (1.12 [0.14]; Figure 4A). Logistic regression showed that a 1-SD increase in pre-RT TL nearly doubled the likelihood of being a decliner (OR = 1.84, 95% CI: 0.99, 3.41), and risk prediction models further indicated that pre-RT TL outperformed most clinical covariates except age, based on both variable importance scores and SHAP values (Figure 4B). Thus, pre-RT TL may serve as an informative biomarker for anticipating how telomeres respond to radiation exposure, complementing age as a known determinant of telomere dynamics.

Figure 4. Biomarker potential of telomere Length (TL) dynamics for radiotherapy (RT) side effects. A. Distribution of TL by its response to RT.

RT effect was defined by “Dec” (blue): declined post-RT (n = 44), or “Inc” (red): increased post-RT (n = 27). Both bar counts and density curves are shown to depict the distribution in each group. B. Predictor importance evaluation in classifying patients by their telomere response to RT. Predictors are ordered from most to least informative and the top 3 predictors are bolded. The left panel shows variable importance scores from a Random Forest (RF) model, measured by Mean Decrease in Accuracy (higher = more important); the right panel displays SHAP values from an XGBoost model, where higher absolute SHAP values indicate greater impact on model predictions. C. TL distribution by RT-induced toxicity status. Adverse events include swallowing difficulty (n = 13), radiation skin reaction (n = 14), and overall symptom burden (n = 23), categorized as None-Mild (grade ≤ 2, grey) vs Moderate-Severe (grade > 2, rose); radiation pneumonitis (n = 68) is categorized as No (grey) vs Yes (rose). Both bar counts and density curves are presented for each group. D. Predictive performance of TL measures for RT side effects. Receiver operating characteristic (ROC) curves compare the performance of three TL metrics: pre-RT TL (purple), post-RT TL (blue), and RT-induced TL change (green) and a baseline model using only clinical covariates (No TL, grey), for predicting radiation pneumonitis (RP) in patients treated with radical RT overall (n = 68), radical RT with concurrent chemoradiation (n = 46), radical RT only (n = 22), SBRT (n = 24). Classification models included logistic regression and kernel support vector machine (KSVM) at stage two of the two-stage framework. ROC curves were generated using leave-one-out cross-validation. Mean area under the curve (AUC) values are reported for each model.

To assess whether telomere metrics were also associated with patient outcomes, we evaluated three radiation toxicity PROs within the radical RT group, including skin reaction (n = 14), swallowing difficulty (n = 13), and overall symptom burden (n = 23), graded as none-mild (≤2) or moderate-to-severe (>2). Radiation pneumonitis data were available for a larger subset of patients (n = 68). No meaningful differences in pre-RT TL, post-RT TL, or their changes were observed between severity levels (Figure 4C).

Despite the lack of significant univariate associations, incorporating TL dynamics into predictive models improved outcome prediction, particularly for radiation pneumonitis (Figure 4D). Among patients treated with radical RT (either with concurrent chemoradiation or without), integrating RT-induced TL changes outperformed those relying solely on single-timepoint TL measures (Figure 4D, Table S2–3). Exploratory modeling within the SBRT subgroup also demonstrated improved predictive performance when using TL change rather than static TL metrics. Moreover, TL-inclusive models outperformed those using only clinical covariates in both the radical combined and chemoradiation subgroups. While performance remained adequate in the radical-only and SBRT subgroups, the limited sample sizes likely constrained predictive gains.

For PROs, due to limited event counts, predictive modeling was limited to the overall symptom burden endpoint. Even within this constraint, TL change improved prediction in both the SBRT and radical RT subsets compared to static TL measures or clinical covariates only, albeit inconsistent due to small sample size (Table S4). These findings imply that TL metrics, particularly the numerical changes in TL measurements induced by RT, may enhance the early identification of patients at elevated pneumonitis risk, warranting further investigation in larger cohorts.

Repair of Oxidative DNA Damage Mediates Radiotherapy-Induced Telomere Length Decline

Because oxidative damage plays a central role in RT-induced genotoxicity, we assessed whether TL correlates with DRC, especially in BER pathways. In the full cohort (n = 100), TL residuals from adjusted models showed modest positive correlations with DRC for several oxidative lesions measured in patient lymphocytes (Figure 5A): repair of U:G (Pearson’s r = 0.171, FDR = 0.135) and A:8oxoG (Pearson’s r = 0.245, FDR = 0.014) lesions. Stratified analyses revealed that these associations were primarily driven by patients in the radical RT group (n = 71; Figure 5B). After adjusting for patient characteristics, associations with U:G and A:8oxoG persisted, with 1-SD increase in repair capacity for U:G associated with a 4.2% increase in TL (95% CI: 0.16%, 8.54%) and a 1-SD increase in A:8oxoG repair associated with a 6.1% increase in TL (95% CI: 2.91%, 9.58%), highlighting the role of BER in safeguarding telomere integrity. In SBRT group (n = 24), we observed an inverse relationship between TL and Hx(MPG)-mediated repair, which remained after covariate adjustment (11.1% lower per 1-SD increase, 95% CI: 4.63%, 17.18%). The inverse relationship could reflect instability arising from BER intermediates arising from MPG-dependent initiation of BER, which has been implicated in toxicity, genomic instability and cancer risk in other contexts.^47–50

Figure 5. Associations between telomere length (TL) and concurrent measurement of DNA repair capacity (DRC). A. Correlations between DRC and TL in the full cohort.

DRC was standardized by z-scores and TL was residuals from adjusted models. The smoothed relationship was modeled using a generalized additive model (GAM) with penalized regression splines to allow for potential non-linear trends. Shaded areas around each line denote the 95% confidence intervals (CIs). Pearsons’s correlation coefficient and p value adjusted by false-discovery rate (FDR) are reported in each panel. N = 100. B. Association between DRC and TL in radical RT (n = 71) and SBRT (n = 24) groups. Points represent the estimated TL ratio per one standard deviation (SD) increase in DRC, with corresponding 95% CIs. Crude: adjusted for only technical factors including lymphocyte ratio, neutrophils ratio, and batch; Adjusted: fully adjusted for age, sex, smoking status, radiotherapy status (and concurrent chemoradiation for radical group), lung V20, lymphocyte ratio, neutrophils ratio, and batch; Imputed: pooled estimates from the fully adjusted model across multiple imputations to address missing data. The vertical dashed line at 1 indicates no association; values below or above 1 suggest inverse or positive associations, respectively. C. Directed acyclic graphs (DAGs) illustrate the causal structure of each model examining whether specific DRC mediates the association between RT or age and TL, with arrows representing hypothesized pathways from exposure (RT or age) to the mediator (A:8oxoG or U:G) and to the outcome (TL). The proportion of the total effect mediated by each marker is shown as Prop.Mediated = Indirect effect / Total effect, where Total effect = Direct effect + Indirect effect.

We conducted a permutation-based mediation analysis to explore whether changes in repair capacity partly explain the observed RT-induced telomere shortening in the radical RT group (Figure 5C). Our previous findings indicated that RT reduces repair capacity for U:G and A:8oxoG lesions; furthermore, prior mechanistic studies demonstrated that DNA repair is crucial for telomere maintenance,^19–21 supporting the causal direction of DNA repair against telomere attrition but not vice versa. Reduced A:8oxoG repair capacity accounted for 12.8% of RT’s effect on telomere shortening (p_perm = 0.017), although U:G repair did not show a similar mediating effect (p_perm = 0.184) (Table S5). Neither pathway explained the age-related TL decline, suggesting that other mechanisms contribute to age-associated telomere attrition.

Discussion

Our study demonstrates that TL in patients undergoing RT is influenced by a combination of factors, including age, cancer stage, RT modality, and the functional capacity to repair oxidative DNA damage. Radical RT was associated with a sustained reduction in TL in PBMCs, persisting up to six months post-treatment, with variability by baseline TL, sex, smoking status, and concurrent chemotherapy. Pre-RT TL emerged as a potential biomarker for subsequent post-RT changes, and dynamic measures of TL outperformed static TL measures in predicting radiation pneumonitis, supporting TL’s potential value in assessing individual radiosensitivity. Moreover, changes in TL were partially mediated by diminished repair capacity for oxidized guanine (A:8oxoG), reinforcing the role of BER in protecting telomeres from radiation-induced oxidative stress.

Previous evidence has shown that telomere dynamics in PBMCs vary widely among individuals, influenced by both physiological aging and disease states.⁵¹ Our results refine this understanding by demonstrating that age-related telomere decline is more evident in cancer patients, but not in healthy controls within the studied age range, a pattern more pronounced among those with advanced-stage disease. Patients with advanced malignancies likely experience prolonged oxidative stress and chronic inflammation, potentially explaining their greater vulnerability to telomere erosion in somatic cells.⁵² The observed dose-response relationship between cancer stage and shorter TL aligns with prior reports linking severe disease states to accelerated telomere attrition.⁵³ Although the exact temporal sequence of cancer progression and telomere erosion is not fully established, our data support a model in which cancer status intensifies age-driven telomere shortening.

Although previous studies indicated that white blood cell counts and subtypes can influence measured TL,^44,54 we did not observe direct associations in our cohort. However, adjusting for lymphocyte and neutrophil proportions improved our model fits, highlighting the potential for telomere heterogeneity across immune cell subsets to influence TL estimates. We also observed a substantial decline in lymphocyte counts following RT, consistent with well-documented lymphocyte radiosensitivity.⁵⁵ These fluctuations in blood cell composition may mask or exaggerate net telomere changes in PBMCs, underscoring the need for future studies that leverage cell-type-resolved or single-telomere assays to uncover more granular mechanisms.⁵⁶

After adjusting for potential confounders, radical RT remained a significant driver of telomere shortening, consistent with earlier findings in prostate cancer cohorts.³⁰ Patients with higher baseline TL were more prone to experience post-RT decline, whereas those with shorter TL tended to show modest increases, suggesting a possible depletion of cells with critically short telomeres during RT.³¹ Our simulation analysis further affirmed these trends reflect biological relevance rather than being solely driven by RTM. Similar trends reported by Luxton et al.³⁰ underscore the baseline-dependent response of pre-RT TL has potential utility for stratifying individuals susceptible to accelerated telomere attrition, as revealed by predictor importance from our ML models. Pronounced TL decrease observed specifically in radical RT patients may reflect higher-dose, larger-volume, more prolonged exposure to RT or more aggressive tumor biology, collectively contributing to greater sensitivity toward telomere damage.

Additionally, we explored whether telomere metrics could predict RT side effects, including skin reactions, swallowing difficulties, and overall symptom burden. Although we did not detect strong associations within our modest sample, ML analyses showed that dynamic TL changes improved risk prediction performance compared to static TL measures alone. This observation aligns directly with known mechanisms of pneumonitis development, which involves persistent inflammation, oxidative stress, and cellular senescence − processes directly linked to accelerated telomere attrition.⁹ Thus, dynamic TL may better reflect real-time physiological changes occurring during pneumonitis progression, offering biological plausibility for further investigation of TL dynamics as potential biomarkers of radiation pneumonitis risk.

Mechanistically, our analyses revealed a direct relationship between TL and BER capacity, particularly for A:8oxoG lesions. In accordance with prior work indicating that oxidative damage drives telomere erosion,^19,20 diminished A:8oxoG repair accounted for approximately 12.8% of the RT effect on TL. Whereas DNA glycosylases that process oxidative DNA lesions contribute to the erosion of telomeres in cancer cell lines,⁵⁷ our data suggest protective effects for BER in non-cancerous cells. By contrast, U:G repair capacity did not mediate the relationship, emphasizing the specific importance of repairing oxidized guanine to maintain telomere stability.^10,58 While conflicting findings exist regarding telomere maintenance mechanisms under different radiation exposures and disease status,^18,24,25,30 our during-RT measurements indicated an immediate decline in response to high-dose thoracic RT, and suggest that telomere erosion due to inefficient BER under radiation-induced oxidative stress may outweigh the protective effects of compensatory telomere maintenance in a high-stress clinical setting.

Our work adds to a growing body of evidence indicating that impaired BER accelerates telomere attrition.⁵⁹ Notably, we did not detect significant age mediation by A:8oxoG repair capacity, suggesting a role for defects in other genome maintenance pathways, especially in the context of advanced cancer patients. While numerous studies indicate that DRC declines with age,⁶⁰ in this cancer cohort, we did not observe an age-A:8oxoG association, possibly because older or advanced-stage patients already exhibit reduced repair capacity. This discrepancy highlights how a cancer-related oxidative stress milieu may overshadow the subtler age-related BER deficiencies. Together with studies noting reduced antioxidant capacity beyond age 60 years,^61,62 our observations support the idea that older adults are inherently more vulnerable to oxidative damage. Taken together with mechanistic studies implicating oxidative DNA damage and repair in telomere attrition,^19,20 we propose that maintaining telomere integrity depends on functional 8oxoG repair throughout the lifecourse, whereas diminished BER in older, more fragile patients further exacerbates telomere attrition under acute oxidative stress conditions, highlighting BER repair as a potential target for preserving TL in these populations.

Despite these promising findings, our study has some limitations. First, heterogeneity in the patient population and modest sample sizes for specific PROs limit statistical power, potentially explaining weaker predictive performance. The pneumonitis outcome in this study was determined using a combination of patient-reported symptoms and radiographic findings, likely resulting in greater inclusion of cases of lower-grade pneumonitis (e.g., grade 1 per CTCAE for radiographic findings alone), potentially diluting stronger underlying associations with higher-grade cases. Second, our observational design inherently carries the risk of residual confounding due to unmeasured or unknown factors. Although extensive adjustment for known clinical and demographic variables, and correlation among these variables may have partially controlled latent confounders (Figure S4), some residual bias inevitably persists, potentially affecting observed estimates in either direction. Third, our BER assays (U:G and A:8oxoG) reflect DNA repair in a sequence context that may not be representative of telomere-localized events, lesion repair efficiency specifically within telomeric DNA requires clarification through future longitudinal and cell-type-resolved studies.

Nevertheless, our longitudinal design across multiple RT timepoints enabled us to identify subgroups most susceptible to telomere shortening. The robustness of our TL measurements was reinforced by complementary validation using advanced flow-FISH and WGS. Our unique ability to integrate TL with simultaneous DRC measures across multiple critical DNA repair pathways provided a comprehensive view of genome integrity, expanding on earlier, smaller studies. As an exploratory study using lymphocyte TL dynamics to predict RT-related adverse events in a clinical patient cohort, these findings should be viewed as hypothesis-generating. Our multifaceted approach lays the groundwork for future studies aimed at rigorously testing telomere dynamics as biomarkers of radiation sensitivity and oxidative stress responses.

Moving forward, larger prospective cohorts with more homogeneous patient populations and controlled conditions should validate the predictive value of both baseline TL and dynamic telomere changes following RT. Mechanistic investigations focused on telomeric DNA damage may further illuminate whether enhancing A:8oxoG repair could potentially preserve TL in high-risk patient populations. Additionally, although we used RT-induced blood lymphocyte depletion, and lung V20, which has been associated with lymphopenia in previously published studies, as proxies for blood-pool irradiation, they might not fully capture the direct cellular response within irradiated lung tissue, introducing potential measurement errors that attenuate true associations. Future research incorporating lung-specific biomarkers or imaging-derived radiobiological metrics may better characterize local tissue responses. A future goal would be to assess the association between exposure of PBMCs to RT and doses to organs-at-risk for side effects.

In summary, our work provides longitudinal evidence that RT-driven telomere attrition is modulated by advanced cancer stage, baseline TL, RT modality, and the capacity to repair oxidative DNA damage. These findings highlight an up to six months post-treatment window during which telomere changes may be particularly pronounced, underscoring the role of BER in maintaining telomere integrity under therapeutic oxidative stress. Future validation and mechanistic exploration are necessary steps before telomere-based metrics can contribute to individualized strategies aimed at preserving genome integrity and potentially improving patient outcomes.

Abbreviations:

AUC

area under the receiver operating characteristic curve

BER

base-excision repair

DRC

DNA repair capacity

FISH

fluorescence in situ hybridization

FM-HCR

fluorescence multiplex host cell reactivation

IR

ionizing radiation

LOOCV

leave-one-out cross-validation

MMR

mismatch repair

NER

nucleotide excision repair

NHEJ

non-homologous end joining

OR

odds ratio

PBMC

peripheral blood mononuclear cells

RF

random forest

ROC

receiver operating characteristic

ROS

reactive oxygen species

RT

radiotherapy

RTM

regression to the mean

SD

standard deviation

TL

telomere length

WGS

whole genome sequencing

Data sharing statement

The raw data generated and analyzed during this study are not publicly available due to patient privacy concerns. However, de-identified datasets can be obtained from the corresponding author upon reasonable request. The full analytical code can be publicly accessed through GitHub: https://github.com/NagelLabHub/LC-Telomere.