This single-center, cross-sectional study conducted at Hospital Clínic Barcelona analyzed serum and histological samples from KTRs who underwent graft biopsies between 2018 and 2023. Inclusion criteria required age >18 years, a pathological diagnosis other than borderline rejection, and HLA-DRB1 alleles with available commercial probes. Exclusion criteria included pregnancy, multi-organ or dual kidney transplants, donor–recipient pairs identical for all four HLA-DRB1 alleles that could not be distinguished using this panel, and borderline rejection due to heterogeneity and limited diagnostic reproducibility. In patients with previous grafts, prior HLA-DR was not shared with the current donor.

Biopsies were performed for routine surveillance at 3 and 12 months post-transplantation or for clinical indications suchs as declining renal function, increasing proteinuria, or newly detected donor-specific antibodies (DSAs). Both surveillance and for-cause biopsies were included to evaluate whether dd-exoDNA correlates with histologic rejection, regardless of clinical presentation. Analyses were stratified by biopsy indication to account for potential differences in biomarker performance.

Serum samples were collected immediately before biopsy in serum separator tubes with gel separator (BD Vacutainer SST), aliquoted, and stored at −80 °C until processing.

Comprehensive demographic, clinical, laboratory, and immunological data were collected from electronic health records. The Pathology Department at Hospital Clínic de Barcelona carried out histological assessments of all biopsy samples, strictly adhering to the 2019 Banff classification criteria [31].

The study adhered to the Declaration of Helsinki. Pre-analytical procedures for sample collection, handling, and storage adhered to the guidelines established by the International Society for Extracellular Vesicles [32]. The local Ethical Committee approved the study (registry code HCB/2022/0044). All transplant procedures were performed in accordance with the Declaration of Istanbul.

Exosomes were isolated using the precipitation method with the ExoGAG kit (Nasasbiotech, Spain), which interacts with surface glycosaminoglycans [33]. Serum samples were centrifuged at 2000g for 10 min. 2 mL of supernatant were mixed with ExoGAG reagent at a ratio of 1:2, incubated at 4 °C for 5 min, and centrifuged at 23,500 g for 30 min. EV pellet was resuspended in 200 μL of 0.1 μm-filtered phosphate-buffered saline (PBS).

Particle concentration and size distribution were determined by nanoparticle tracking analysis (NTA) using a NanoSight LM10 instrument (Malvern, UK) equipped with a 488 nm Blue laser and a SCMOS camera. For each sample, five 50-second videos were captured. Data were analyzed with NTA software v3.4. Exosomes’ morphology was assessed by transmission electron microscopy (TEM), as previously described by our group [34, 35].

Phenotypic characterization was performed using the MACSPlex Human Exosome Kit (Miltenyi Biotec), following the manufacturer’s recommendations. A total of 1 × 10⁹ EVs were incubated with 15 µL of antibody-coated beads overnight, followed by APC-labeled antibodies against CD9, CD63, and CD81 for 1 h. Data were acquired with a BD LSRFortessa SORP cytometer and analyzed using FACS DIVA software (BD Biosciences). Median fluorescence intensity (MFI) for each marker was normalized to average tetraspanin levels per sample.

ExoDNA was extracted using QIAamp DNA Mini Kits (Qiagen), following the manufacturer’s instructions, including protease treatment to reduce non-vesicular protein–DNA complexes potentially co-isolated during ExoGAG precipitation. No DNase I treatment of intact EVs was performed prior to lysis; therefore, the assay captures total EV-associated DNA, including both intravesicular DNA and DNA associated with the vesicle surface. Concentration and DNA integrity were assessed using Qubit single-strand molecular probes (Invitrogen) and an Agilent 4200 TapeStation System (Agilent), respectively, at the Functional Genomics Platform of the August Pi Sunyer Biomedical Research Institute (IDIBAPS). Quality thresholds were verified, with a main peak at ∼160–180 bp. DNA quantity was normalized to EV yield and stored at −80 °C.

HLA alleles in exoDNA were detected and quantified via dPCR using a panel of six TaqMan probes targeting specific HLA-DRB1 alleles (Bio-Rad): HLA-DRB1*01, HLA-DRB1*07, HLA-DRB1*08, HLA-DRB1*11, HLA-DRB1*13, and HLA-DRB1*15, based on Zout et al [36]. Each probe had HEX and FAM versions, enabling simultaneous detection of donor and recipient alleles. Of note, the HLA-DRB1*15 assay was also specific for the HLA-DRB1*16 allele. As previously described, 2 ng of exoDNA were pre-amplified using SSoAdvancedPreAmp Supermix (Bio-Rad) [36, 37].

Digital PCR assays were conducted using the QuantStudio Absolute Q platform (Thermo Fisher Scientific), with microfluidic array plates for sample partitioning, following the protocol reported by Zou et al [36] and the HLA Expert Design datasheet. Each 9 µL reaction included 1X dPCR Master Mix, allele-specific probes, and exoDNA. PCR cycling conditions: 95 °C for 3 min, 40 cycles of 94 °C for 10s and 55 °C for 30s, followed by 72 °C for 5 min. Data were analyzed using the Absolute Q Digital PCR Software.

The percentage of dd-exoDNA was calculated according to the following formula, as previously described for dd-cfDNA [36]: dd ‐ exoDNA fraction = donor   copy   number total   copy   number = donor   copy   number donor   copy   number + recipient   copy   number

The number of copies for alleles present with two copies per genome was adjusted to a single copy per diploid genome. For alleles shared by the donor and recipient, calculations were adjusted assuming equal allele contributions per genome.

Due to the exploratory nature of the study and the absence of prior dd-exoDNA data, no formal sample size calculation was performed. The cohort included 100 patients with well-defined histological diagnoses, comparable to those in early-phase biomarker studies [38]. The Kolmogorov-Smirnov test was used to assess normality. Normally distributed variables are presented as mean ± standard deviation (SD), and non-normally distributed variables as median and interquartile range (IQR). Qualitative variables are expressed as absolute and relative frequencies.

Between-group comparisons used Student’s t-test, Mann-Whitney U, or Kruskal-Wallis tests, and χ²or Fisher’s exact test as appropriate. When the Kruskal–Wallis test was applied, post-hoc pairwise comparisons were conducted using Dunn’s test with Bonferroni correction for multiple comparisons. ROC curves were used to evaluate diagnostic performance (AUC, 95% CI), with optimal thresholds determined by Youden’s Index. Associations were assessed with Spearman correlation.

For the MACSPlex analysis of 37 surface markers, unadjusted p values were first calculated using the Mann–Whitney U test. To control the false discovery rate, p values were adjusted using the Benjamini–Hochberg procedure; both unadjusted and adjusted p values are reported.

The clinical utility of dd-exoDNA was evaluated using Decision Curve Analysis (DCA) with logistic regression models including DSA and eGFR, with or without dd-exoDNA. Covariates for multivariable logistic regression were selected a priori based on clinical relevance. Sensitivity analyses confirmed robustness. No correction for multiple comparisons was applied for Banff lesion analyses due to the exploratory design; therefore, these findings should be interpreted as hypothesis-generating.

Statistical analyses were performed using IBM SPSS Statistics v31.0 (IBM Corp., Armonk, NY), except for DCA, which was conducted using the rmda package in R (v4.4.1). Graphs were generated with GraphPad v9.5.1 (GraphPad Software, La Jolla, CA). Two-tailed p < 0.05 was considered statistically significant.

This study included serum and allograft biopsies from 100 KTR, comprising 34 surveillance and 66 clinically indicated biopsies performed at a median of 4 [1–13.5] months post-transplant. The baseline characteristics of the included patients are summarized in Table 1.

The mean age of this cohort at the time of transplant was 58.71 ± 14.6 years, with predominantly male sex (64.0%) and recipients of a first transplant in 80.0% of cases. Thirty-eight patients had positive panel reactive antibodies (cPRA) ≥50% at the time of biopsy.

Histopathological analysis revealed rejection in 47 biopsies. Among these, 15 were TCMR, and 32 were classified as antibody-mediated rejection (ABMR). No mixed rejection cases were identified. Of the ABMR cases, 11 met Banff criteria and had detectable DSAs. The remaining 21 cases were classified as ABMR based on the presence of microvascular inflammation and/or positive C4d staining. TCMR cases were further classified into 8 acute and 7 chronic-active. Regarding ABMR, 22 cases were active, and 10 were chronic-active.

Among the 53 biopsies without evidence of rejection, the most frequent findings were non-specific changes (28 cases) and interstitial fibrosis with tubular atrophy (IFTA, 16 cases). Other diagnoses included chronic vascular changes (7 cases) and diabetes-related complications (2 cases).

Serum exosomes were quantified by NTA, and TEM revealed their typical round shaped membrane particles (Figures 1A,B).

Exosomes concentration was 1.26 × 10¹¹ [0.883 × 10¹¹-1.67 × 10¹¹] particles/mL. When stratified by biopsy results, exosome concentrations were numerically higher in cases with rejection compared to those without (1.30 × 10¹¹ particles/mL (9.8 × 10¹⁰–1.84 × 10¹¹ particles/mL) vs. 1.16 × 10¹¹ particles/mL (8.3 × 10¹⁰–1.51 × 10¹¹ particles/mL), although this difference did not reach statistical significance (p = 0.19). Further differentiation between cellular and humoral rejection revealed a similar trend toward higher exosome concentrations in ABMR compared to TCMR (1.4 × 10¹¹ [1.16 × 10¹¹–1.98 ×10¹¹] vs. 1.1 × 10¹¹ [8.3 × 10¹⁰–1.3 × 10¹¹] particles/mL; p = 0.065) (Figure 1C).

Further exosome characterization was performed using the MACSPlex kit, which analyzes 37 common surface and immunological exosome biomarkers. Several markers showed unadjusted differences between rejection and non-rejection groups, with lower median expression levels observed in the rejection group for CD3, CD8, CD2, CD40, MCSP, and CD86 (Figure 1D). When stratified by rejection type, CD40 serum exosome antigen was also significantly lower in patients with ABMR than in those without rejection. However, after adjustment for multiple comparisons, no marker remained statistically significant (Supplementary Table S1).

The average DNA fragment length of exo-DNA ranged between 160 and 200 base pairs (bp) (Figure 1E), while the median exoDNA concentration was 0.917 [0.48–1.50] ng/µL. No differences were observed in terms of exoDNA concentrations between patients with and without rejection (1.01 [0.49–1.56] ng/μL for control versus 0.83 [0.37–1.48] ng/μL for rejection (p = 0.62). Additionally, exoDNA concentrations were higher in TCMR than in ABMR (1.32 [0.58–2.33] vs. 0.75 [0.36–1.40] ng/μL), although this difference was not statistically significant (p = 0.32).

Dd-exoDNA was consistently detected in all patients 0.92 [0.47–1.51] × 10⁻³). No differences were observed between living donors and deceased donors (1.02 [0.58–3.6] ×10⁻³ vs. 0.92 [0.40–4.56] ×10⁻³, p = 0.35), nor among the different types of deceased donors. In patients with any-type rejection, dd-exoDNA was significantly higher (2.66 [0.56–7.10] × 10⁻³) compared to patients without rejection (0.69 [0.28–1.71] × 10⁻³), with a p = 0.004 (Figure 2).

Dd-exoDNA levels showed substantial dispersion within the TCMR group, particularly in chronic-active TCMR, with a right-skewed distribution characterized by a subset of cases exhibiting markedly elevated values. In line with this heterogeneity, the global Kruskal–Wallis test did not reach statistical significance (p = 0.09); however, post-hoc pairwise comparisons identified a significant difference between the no-rejection and TCMR groups (p = 0.022), while no significant differences were observed for the remaining comparisons (Table 2).

To evaluate whether extreme observations drove these findings, a sensitivity analysis excluding outliers identified using an IQR–based rule applied to log-transformed dd-exoDNA values was performed. After exclusion, the difference between the TCMR and non-rejection groups was attenuated, indicating that a limited subset of TCMR cases accounted for the elevated dd-exoDNA levels.

Among patients with ABMR, no significant differences in dd-exoDNA levels were observed between DSA-negative and DSA-positive cases (2.63 [0.40–5.43] × 10⁻³ vs. 2.02 [0.77–4.06] × 10⁻³, respectively; p = 0.755).

The relationship between dd-exoDNA fraction and individual Banff classification scores was also assessed. The median dd-exoDNA was 0.81 [0.40–2.42] × 10⁻³ in the absence of glomerulitis (g), while it was significantly higher (2.43 [0.50–16.87] × 10³) in those with g ≥ 1 (p = 0.037). Similarly, for peritubular capillaritis (ptc), the median dd-exoDNA was 0.92 [0.29–2.63] × 10⁻³ in samples with no inflammatory activity, increasing to 2.82% [0.58–16.47] × 10⁻³ in patients with ptc ≥1 (p = 0.040). For tubulitis (t), a significant difference was also observed, with values reaching 2.93 [1.19–17.23] × 10⁻³ in cases with moderate to severe (t ≥ 2), compared to 0.89 [0.40–4.36] × 10⁻³ in patients with mild or absent involvement (p = 0.043). In contrast, no significant differences were found regarding complement fraction C4d deposition (p = 0.807), interstitial inflammation (i) (p = 0.45), and vasculitis (v) (p = 0.253) (Figure 3).

To assess the diagnostic performance of dd-exoDNA in distinguishing rejection from non-rejection, the area under the ROC curve (AUC) was calculated. The AUC for discriminating rejection from non-rejection was 0.67 (95% CI: 0.56–0.79). For differentiating TCMR or ABMR specifically from non-rejection, dd-exoDNA demonstrated an AUC of 0.67 (95% CI: 0.51–0.83) and 0.60 (95% CI: 0.48–0.73), respectively (Figure 4).

Compared with DSAs’ diagnostic ability at the time of biopsy, dd-exoDNA showed numerically higher AUC, although the confidence intervals overlapped, as DSAs yielded an AUC of 0.59 (95% CI: 0.47–0.70). Similarly, the estimated glomerular filtration rate (eGFR) showed poor diagnostic accuracy, with an AUC of only 0.37 (95% CI: 0.23–0.51) (Supplementary Figure S1).

Notably, dd-exoDNA performance varied depending on biopsy indication. In clinically indicated biopsies, its discriminative ability improved significantly (AUC 0.70 [95% CI: 0.57–0.83]), whereas for surveillance biopsies, the AUC was lower (0.52 [95% CI: 0.30–0.74]) (Supplementary Figure S2). It should be noted that among the 34 surveillance biopsies, only 5 were rejection cases, compared with 42 among the 66 clinically indicated biopsies.

We aimed to determine a clinically significant threshold for identifying samples potentially indicative of kidney graft rejection. Using a threshold of 1.9 × 10⁻³, dd-exoDNA showed a specificity of 77.4% and a sensitivity of 57.4% for distinguishing rejection from non-rejection. Among the 100 dd-exoDNA determinations, 39 were above the 1.9 × 10⁻³ threshold. Of these, 27 corresponded to patients with biopsy-proven rejection, while the remaining 12 were from patients without rejection. The positive predictive value (PPV) for identifying rejection was 69.2%, while the negative predictive value (NPV) was 67.2%. Among patients without histological rejection, those with exoDNA values above the 90th percentile showed heterogeneous biopsy findings. Two patients presented chronic graft lesions, including moderate chronic vascular changes and moderate interstitial fibrosis with tubular atrophy. In contrast, the remaining three showed no histological evidence of rejection or other relevant abnormalities.

To further assess whether dd-exoDNA fraction was independently associated with rejection, we conducted a multivariable logistic regression analysis adjusting for relevant clinical and immunological variables. In the adjusted model, dd-exoDNA remained independently associated with rejection, with an odds ratio (OR) of 3.68 (95%CI 1.32–10.26, p = 0.013). Biopsies performed for clinical indication were also significantly associated with rejection (OR 7.43, 95% CI 2.21–25.02, p = 0.001). Maintenance immunosuppression with mTOR inhibitors, although associated with a lower risk of rejection in the univariable analysis (OR 0.32, p = 0.007), did not retain statistical significance after adjustment (OR 0.4, p = 0.101). Likewise, eGFR at biopsy and DSAs at biopsy did not remain statistically significant after adjustment (Table 3). Decision curve analysis showed that the model including dd-exoDNA provided a higher net benefit than the baseline model (DSA + eGFR) across a range of probability thresholds (Figure 5).