Authors: 1 Thomas McLarnon, 1 Donya Ghazinader, 2 Frank McCarroll, 2 Jack Duncan, 2 Laura Lennox, 1 Steven Watterson, 1 Taranjit Singh Rai

Affiliations: 1 School of Medicine, Personalised Medicine Centre, Ulster University, Londonderry, BT48 7JL , UK, 2 Western Health and Social Care Trust, Altnagelvin Area Hospital, Londonderry, BT47 6SB, UK

Background/ Introduction: Chronic kidney disease (CKD) is a progressive, irreversible condition affecting over 14% of the global population and is directly associated with 1.48 million deaths annually. The current CKD staging does not provide patient-level estimates of disease-related mortality risk, which are already applied across multiple chronic disease conditions such as cardiovascular disease (SCORE2), chronic obstructive pulmonary disease (BODE index) and liver disease (MELD Score). Outside the current clinical staging, cellular senescence has been identified as a key physiological driver of CKD progression, comorbidity burden and mortality risk. Senescence also has an inflammatory phenotype: the senescence associated secretory phenotype (SASP), resulting in the secretion of pro-inflammatory signatures that worsen patient health via disease progression, organ dysfunction and activation of inflammatory cascades. Despite the large amount of mechanistic evidence that implicates SASP components as main drivers of worsening disease severity and mortality, there has been no study to date that investigates these markers in CKD related mortality. To our knowledge, this is the first study to investigate senescence-associated plasma proteins as independent predictors of disease-related mortality in a CKD diagnosed cohort.

Material & Methods: The UK Biobank recruited 500,000 participants within the United Kingdom between 2006–2010 for baseline assessment, with approximately 54,000 participants having proteomic profiling carried out using the OLINK Explore 3072 platform. From this cohort, we identified participants with prevalent CKD (ICD-10 code N18) diagnosed prior to baseline blood sample collection who subsequently died from CKD-related causes during follow-up, yielding 468 participants for analysis. The 9-year time point was selected as it provided perfect class balancing (n=234 vs n=234) while maintaining mean statistical power >80% across most proteins. A validation cohort comprised 155 CKD patients (mean age 59 years; 63% male; CKD stages 2–5) recruited from Altnagelvin and Letterkenny University Hospitals between 2018–2020 was also employed. Proteomic differential expression analysis was carried out using Welch’s two-tailed t-test, with volcano plots, heatmaps, PCA and violin-boxplots employed to visualise proteomic differences between cohorts. Pathway analysis was carried out using the pathfindR package. Supervised machine learning for classification utilised linear SVM, radial SVM, logistic regression and random forest with greedy-forward feature selection, 70:30 train/test split and 10-fold cross-validation. Cox proportional hazards models were constructed in both univariate and adjusted forms (age, sex, CVD, diabetes, eGFR). Kaplan-Meier curves were generated using median expression groupings. Pearson correlation was used to test biomarker relationships with eGFR across both cohorts. The optimal model was benchmarked against a clinical model (age, sex, eGFR, CVD, diabetes) and the Charlson Comorbidity Index (CCI).

Results: Senescence associated proteins ADM, PGF, EDA2R and NT-pro-BNP were all identified as the most statistically significant biomarkers indicative of mortality. Pathway analysis revealed key inflammatory and SASP pathways such as cytokine–cytokine receptor interaction, NF-kappa B signalling and chemokine signalling pathways. Multiple machine learning frameworks were tested across all 372 proteins using forward feature selection, resulting in a linear SVM utilising expressional values of EDA2R, NT-pro-BNP, CTSZ, CLEC4C, HMOX2, NTRK3, REN, CCL3 and CLEC6A. The optimal model yielded a training AUC of 0.832 (sensitivity 0.752, specificity 0.785) and an external AUC of 0.711, making 46 true negative and 43 true positive predictions. Senescence associated proteins EDA2R (HR 2.05), NT-pro-BNP (HR 1.32), REN (HR 1.34) and CTSZ (HR 2.27) were identified as the most statistically significant proteins associated with 9-year disease-related mortality. Remarkably, when adjusted for age, sex, CVD presence, diabetes mellitus presence and eGFR, these proteins all remained statistically significant `EDA2R (adjusted HR 2.02), NT-pro-BNP (adjusted HR 1.2), REN (adjusted HR 1.19) and CTSZ (adjusted HR 1.7)`, demonstrating their predictive capabilities independently from clinical factors. Kaplan-Meier survival curves demonstrated that an increase in protein expression of EDA2R, NT-pro-BNP, REN and CTSZ all correlates with poorer survivability within the 9-year timepoint (all p<0.0001). Pearson correlation confirmed EDA2R, NT-pro-BNP, REN and CTSZ all having inversely proportional relationships with eGFR, preserved within the UKBiobank and across the validation cohort, with EDA2R being the most significant and conserved (r=−0.745, p=9×10⁻²⁹). The proteomics SVM outperformed both the CCI and clinical benchmark models, achieving an AUC of 0.711, compared to 0.613 (clinical) and 0.637 (CCI), and outperformed both comparators across most patient subgroups.

Conclusion: Overall, this study demonstrates that circulating senescence-associated plasma proteins are independent predictors of disease-related mortality in CKD, most notably with EDA2R and CTSZ identified as novel and previously uncharacterised contributors to long-term mortality risk and renal function correlates. These proteins together were able to outcompete clinically centred frameworks such as a clinical model utilising eGFR, sex, age, diabetes and CVD presence, as well as the full ICD10 Charlson comorbidity index. Circulating SASP components are fully detectable in peripheral blood and provide significant utility as prognostic biomarkers, offering a non-invasive alternative to invasive renal biopsies. Additional studies in larger, more heterogeneous CKD cohorts are needed to validate these associations and to determine whether senescence biomarker profiling can be translated into clinical risk stratification frameworks to improve patient outcomes.