Elevated sCD25 predicts early infection in patients with major burns Authors: Laura Hurtado-Navarro, Eva Flores-Cabeza, Pablo Mata-Martínez, Manuel Sánchez-Sánchez, Verónica Terrón-Arcos, Carolina Martín-Flórez, Lucía Cachafeirio, Javier Vejo-Gutiérrez, Jesús Fernández-Felipe, Roberto Lozano-Rodríguez, Laura Córdoba, Gonzalo Sáenz de Santa María-Diez, Julia del Prado-Montero, José G. Rigor-Pérez, Rebeca Pérez de Diego, Carlos del Fresno, Francisco J. Cueto, Eduardo López-Collazo Published: 2026 Severe burn injuries are associated with profound immune dysregulation and a high risk of infection, driven by skin barrier disruption, immunosuppression, and frequent invasive procedures [1]. The accompanying systemic inflammatory response reduces the specificity of conventional markers, complicating the early diagnosis of infection [2]. We analysed 308 plasma samples from 34 adult patients with major burns admitted to the Burn Intensive Care Unit at La Paz University Hospital. Figure 1A and Supplementary Table 1 summarise the clinical trajectories of these patients, including total body surface area (TBSA) affected (mean 41.26%; range 2–83), survival (mortality rate of 32.3%), length of hospital stay (mean 42.41 days; range 1–195), and documented infections (in 25/34 patients; 73.53%). As expected, we found a strong association between prolonged hospitalization and detection of infections (Supplementary Fig. 1A-B).Fig. 1Stratification of Infection Risk in Burn Victims admitted to ICU. (A) Swimmer plot showing the characteristics of individual severely burned patients over time since hospital admission, including stay length, infection occurrence (positive microbiological culture result, grey circles), patient death (exitus, black diamond) and percentage of total body surface area affected (red colour gradient). (B-C) Evolution of plasma levels of sCD25 (B) and sCD27 (C) around the infection onset. (D) UMAP representation of individual burned patients based on their first sample immunomodulator levels, where colour depicts whether they had an infection during their hospital stay, diagonal lines indicate the mortality outcome and bubble size reflects the time from admission to first infection or death. (E) Volcano plot illustrating the association between each analyte’s best-performing threshold [4] and patient time-to-infection. Significantly associated analytes are highlighted in mustard. (F–H) Probability of infection curves comparing severely burned patients with high or low concentration of plasmatic sCD25 (F), sCD27 (G) and PCT (H) from hospital admission to infection onset. (I–K) Receiver operating characteristic (ROC) curves showing the predictive performance for infection status at early time points after sample collection (24 h and 48 h) for sCD25 (I), sCD27 (J), and PCT (K). Area under the curve (AUC) values with 95% confidence intervals (CI) are indicated for each time point. (L) Scatter plot showing the bootstrap resampling discrimination performance (c-index) of individual immune mediators for infection risk against their corresponding q-value. (M) Distribution of optimal sCD25 cut-off values derived from 1,000 bootstrap resampling iterations. The vertical line indicates the median estimated cut-off, with shaded areas representing the 95% confidence interval. (N) Probability of infection curves comparing patients with sCD25 levels above or below the bootstrap-derived cut-off. Statistical significance was assessed using the Mantel-Cox log-rank test. *p < 0.05, and **p < 0.01 Stratification of Infection Risk in Burn Victims admitted to ICU. (A) Swimmer plot showing the characteristics of individual severely burned patients over time since hospital admission, including stay length, infection occurrence (positive microbiological culture result, grey circles), patient death (exitus, black diamond) and percentage of total body surface area affected (red colour gradient). (B-C) Evolution of plasma levels of sCD25 (B) and sCD27 (C) around the infection onset. (D) UMAP representation of individual burned patients based on their first sample immunomodulator levels, where colour depicts whether they had an infection during their hospital stay, diagonal lines indicate the mortality outcome and bubble size reflects the time from admission to first infection or death. (E) Volcano plot illustrating the association between each analyte’s best-performing threshold [4] and patient time-to-infection. Significantly associated analytes are highlighted in mustard. (F–H) Probability of infection curves comparing severely burned patients with high or low concentration of plasmatic sCD25 (F), sCD27 (G) and PCT (H) from hospital admission to infection onset. (I–K) Receiver operating characteristic (ROC) curves showing the predictive performance for infection status at early time points after sample collection (24 h and 48 h) for sCD25 (I), sCD27 (J), and PCT (K). Area under the curve (AUC) values with 95% confidence intervals (CI) are indicated for each time point. (L) Scatter plot showing the bootstrap resampling discrimination performance (c-index) of individual immune mediators for infection risk against their corresponding q-value. (M) Distribution of optimal sCD25 cut-off values derived from 1,000 bootstrap resampling iterations. The vertical line indicates the median estimated cut-off, with shaded areas representing the 95% confidence interval. (N) Probability of infection curves comparing patients with sCD25 levels above or below the bootstrap-derived cut-off. Statistical significance was assessed using the Mantel-Cox log-rank test. *p < 0.05, and **p < 0.01 Early detection of infection remains challenging in this population, and prophylactic antibiotic therapy is not routinely recommended [3]. We therefore examined the temporal evolution of immunomodulator levels in relation to the onset of the first microbiologically confirmed infection. This analysis identified distinct trajectories surrounding infection onset, including patterns comparable to those observed for white blood cells (WBC), C-reactive protein (CRP) and procalcitonin (PCT) (Fig. 1B-C and Supplementary Fig. 2). These findings highlight the dynamic behaviour of soluble immune checkpoints and cytokines during the post-burn period and suggest that selected biomarkers may provide superior clinical utility for the early prediction of infection. To evaluate their predictive value, we analysed the first plasma sample obtained after admission. A UMAP projection based on all analyte concentrations revealed a clear separation of patients according to time to infection, with those exhibiting shorter latency clustering in the lower right region (Fig. 1D). Time-to-event modelling was then performed, and optimal cut-off values were identified using the KMplot web tool [4]. In this analysis, sCD25 was associated with a markedly increased risk of earlier infection (HR 3.99; 95% CI 1.58–10.09; p = 0.0018), followed by sCD27 (HR 2.60; 95% CI 1.07–6.30; p = 0.0291) and PCT (HR 2.44; 95% CI 1.50–5.92; p = 0.0435), whereas most other soluble mediators showed weaker or non-significant associations (Fig. 1E-H and Supplementary Table 2). These results indicate that, among the soluble mediators analysed, initial sCD25, sCD27 and PCT levels display the strongest associations with early infection onset. In contrast, conventional biomarkers and clinical scores, including APACHE-II (HR 0.99; 95% CI 0.33–2.98; p = 0.9827), CRP (HR 1.72; 95% CI 0.77–3.85; p = 0.1841) and IL-6 (HR 2.09; 95% CI 0.88–4.97; p = 0.0894) did no associate with time to infection, (Supplementary Fig. 3A–C and Supplementary Table 2). To further assess early predictive performance, receiver operating characteristic (ROC) analysis was performed using the infectious status 24 and 48 h after sample collection. sCD25 consistently showed the highest discriminative capacity, with superior area under the curve (AUC) values at both time points compared with all other markers, including PCT and sCD27 (Fig. 1I-K). Although IL-6 and sCD27 displayed similar AUC values, both –together with APACHE-II score and CRP— showed inferior or inconsistent early predictive performance compared with sCD25. Specifically, APACHE-II, CRP and IL-6 did not improve the overall interpretation of the ROC analysis at 24 and 48 hours, in agreement with their lack of significant association in the time-to-event analysis (Supplementary Fig. 3D-F). To evaluate the robustness and reproducibility of these associations, we applied a bootstrap resampling strategy to this dataset. Among all candidate biomarkers, only sCD25 retained a significant association with time to infection with a C-index of 0.7143 and an FDR-adjusted q-value of 0.0310, whereas sCD27 and PCT showed unstable predictive performance and failed to retain statistical significance after resampling (sCD27: C-index = 0.6346, q = 0.2789; PCT: C-index = 0.6656, q = 0.1239; Fig. 1L and Supplementary Table 3). Thus, sCD25 emerged as the only reproducible early predictor in this cohort. To derive a clinically applicable metric avoiding overfitting resulting from the maximisation of the log-rank statistic for cut-off selection, we implemented an internal validation approach based on bootstrap resampling. An optimal sCD25 cut-off of 1514 ng/mL (95% CI, 571.4 to 1720) was identified using midpoint-based candidate thresholds (Fig. 1M). Patients with baseline sCD25 levels above this threshold exhibited a significantly shorter time to infection (Fig. 1N). sCD25, released from activated lymphocytes and regulatory T cells, may also contribute to immune dysfunction by acting as an IL-2 sink. By sequestering free IL-2, sCD25 may reduce cytokine bioavailability, thereby impairing T-cell proliferation and contributing to the functional immunosuppression that predisposes burn patients to infectious complications [5]. In summary, sCD25 emerges as a robust and reproducible early predictor of infection risk that outperforms conventional standard-of-care biomarkers. These findings support further validation of sCD25 as a tool for early risk stratification and clinical monitoring in patients with severe burns. Supplementary Information Additional file 1. Additional file 1.