The aim of this retrospective cohort study was to analyze the course of renal allograft function in a group of 46 transplant recipients who underwent both MRI and transplant biopsy simultaneously. Thirty-two of those patients were included in our previous prospective study, which focused on assessing correlations between T₁ mapping, Banff lesion scores, and conventional graft function parameters [8]. The other fourteen patients underwent MRI before the initial study due to clinical indications and as part of a quality assurance protocol to test its basic feasibility.

Patients were screened for study inclusion at our outpatient clinic. Detailed inclusion criteria are provided in the study from Beck-Tölly et al. [8]. All suitable renal transplant patients scheduled for protocol or indication biopsies were actively asked for study participation. The main inclusion criteria were: age over 18 years and an estimated glomerular filtration rate (eGFR) of more than 10 mL/min/1.73 m² (calculated using the Modification of Diet in Renal Disease formula). Exclusion criteria included MRI-incompatible metallic implants or pacemakers, claustrophobia, and pregnancy. Recruitment took place from December 2017 to January 2019. Non-contrast MRI scans were performed shortly before or after the biopsy, using a whole-body 1.5 T MR system (MAGNETOM Avanto Fit; Siemens Healthineers; Erlangen, Germany).

The primary endpoint was the course of graft function after assessment of baseline MRI T₁. Longitudinal graft function was calculated based on serum creatinine levels measured in a three-month interval over the period of 24 months after the MRI. To further quantify changes in kidney function, the eGFR delta (ΔeGFR) and eGFR slopes were calculated for each observation period.

The secondary endpoint was the frequency of death-censored graft loss in relation to baseline T₁. Graft loss was defined as the resumption of dialysis. All participants provided informed consent. Ethical approval for the study was granted by the institutional ethics committee (Approval No. 1893/2017). The study adhered to Good Clinical Practice guidelines, the principles of the Declaration of Helsinki, and the Declaration of Istanbul.

MRI protocols and methods used in this study have been described in detail elsewhere [8]. In short, we extracted T₁ measurements from our multiparametric MRI images, measured across three paraxial (cranial, middle, caudal) and three paracoronal (anterior, middle, posterior) planes, involving six independent regions of interest per plane. The median of those 36 measurements was defined as the overall median T₁ cortical relaxation time. The choice to focus this current analysis on T₁ was based on results from preceding research, which estimated kidney function based on T₁ in patients with glomerulonephritis [21], as well as one study quantitatively evaluating renal function and renal fibrosis in patients with chronic kidney disease [22].

Morphologic lesions were assessed on formalin-fixed paraffin-embedded sections using standard methodology [8]. Banff single lesions and rejection phenotypes were scored based on the Banff 2017 scheme [23]. In addition to Banff criteria, chronic structural damage in kidney grafts was assessed using the chronicity index as described by Haas et al. [24]. This index combines four key histological features: interstitial fibrosis (ci), tubular atrophy (ct), vascular fibrous intimal thickening (cv), and chronic glomerulopathy (cg). Each feature was scored on a scale from 0 (no changes) to 3 (severe changes), with the chronic glomerulopathy score being doubled. The total chronicity index ranged from 0 to 15, with higher scores indicating more significant chronic injury.

Continuous variables were reported as means with standard deviations (SD) or medians with interquartile ranges (IQR). Categorical variables were summarized as counts and percentages. The median split method was employed to divide patients into two groups of equal size based on the overall T₁. Hence, the “T₁ ^high” group referred to patients with T₁ values above and the “T₁ ^low” group for patients with T₁ values below the median. Spearman´s correlation coefficients were used to analyze the associations between T₁ and baseline variables, including transplant age, baseline eGFR, and the histological parameters ci, ct—as well as the chronicity index. To compare the predictive validity of Banff ci scores with T₁, Fisher’s Z transformation was performed.

The linear mixed-effects model was performed to analyze the changes in the estimated glomerular filtration rate (eGFR slope) over time between the groups.

To compare graft survival, the Kaplan-Meier survival curve and log-rank test were calculated. To address the loss of graft function and the subsequent missing data points in our longitudinal follow-up, we implemented the “last observation carried forward” (LOCF) imputation method. Additionally, the Receiver Operating Characteristic (ROC) analysis was performed to evaluate the ability of T₁ to predict the occurrence of allograft loss. The p-value of <0.05 was considered statistically significant.

Statistical computations and analyses were conducted using SPSS for Mac Version 20 (SPSS Inc., Chicago, IL), GraphPad Prism (GraphPad Prism 10.0.3 (217) Macintosh Version by Software MacKiev ^© 1994–2023 GraphPad Software, LLC), R (R Core Team, 2023) and RStudio (2022 by Posit Software, PBC).

Forty-six patients were included, 30 (65%) were male; the mean age at transplantation was 54.3 ± 14.8 years (mean ± SD). Baseline parameters of the total group and the subgroups (T₁ ^high and T₁ ^low) are displayed in Table 1. The majority of patients (80.4%) received deceased donor kidneys. The median time from transplantation to study inclusion was 3 years (IQR 0.7–11.2). Six (13%) participants underwent magnetic resonance imaging before [4 ± 2.5 days, (mean ± SD)] and 38 (82.6%) after (7.9 ± 9 days) the biopsy. Two (4.4%) patients had the MRI on the day of the biopsy. The median cortical T1 was 1,369 ms (IQR 1,279–1,511). The median eGFR at the time of biopsy was 30.8 mL/min/1.73 m² (IQR 20.1–49.6). Fourteen (30.4%) patients reached the endpoint graft loss. Four patients (8.6%) were lost to follow-up before the end of our observation period of 24 months.

Thirty-seven biopsies (80.4%) were performed based on clinical indications, primarily due to the deterioration of graft function, while the other nine biopsies (19.6%) were protocol biopsies. In 14 (30.4%) biopsies, graft rejection was diagnosed (see Table 1). The overall rate of rejections was equally distributed between the T₁ ^high and T₁ ^low groups (30.4% each, p > 0.99). Antibody-mediated rejection (AMR) was numerically but not significantly higher in the T₁ ^high group (26.1% vs. 13%, p = 0.45). The T cell-mediated rejection (TCMR) frequency also did not differ significantly between both groups (4.3% vs. 17.4%, p = 0.34). Twenty-six allografts (56.6%) exhibited high-grade interstitial fibrosis (ci 2 or 3), and in 18 kidneys (39.1%), high-grade tubular atrophy (ct 2 or 3) was found (Supplementary Table S1). Allografts in the T₁ ^high group had more severe interstitial fibrosis: 47.8% with ci 3 compared to 21.7% in the T₁ ^low group (p = 0.044). Tubular atrophy was also more advanced in the T₁ ^high group (ct 3: 30.4% versus 8.7% in the T₁ ^low group, p = 0.031). Although not statistically significant, arterial intimal thickening showed higher severity in the T₁ ^high group (52.2% at cv 2 compared to 34.8% in the T₁ ^low group, p = 0.059). The severity of glomerular basement membrane double contours (cg), did not differ between the groups; cg grades 2 or 3: 22.7% in the T₁ ^high group vs. 14.2% in the T₁ ^low group (p = 0.14). Chronicity index differed significantly between the groups: T₁ ^high 8.5 (5–11) vs. 3 (IQR 2.5–6.5) in the T₁ ^low group, p < 0.01.

There was a significant positive correlation between median T₁ and interstitial fibrosis (ρ = 0.36, p = 0.01) as well as tubular atrophy (ρ = 0.45, p < 0.01). Further on, the chronicity index correlated positively with T₁ (ρ = 0.46, p < 0.01). No significant correlation was found between median T₁ and the time since transplantation (ρ = 0.20, p = 0.16). T₁ did not correlate with median eGFR at baseline (ρ = −0.25, p = 0.09, see Figure 1).

In the T₁ ^high group, eGFR levels consistently declined over time. At baseline, the T₁ ^high group had a median eGFR of 25.6 [19.6–43.3 (median, IQR)], compared to 37.9 (22.1–53.2) mL/min/1.73 m² in the T₁ ^low group (p = 0.21) in the T₁ ^low group (p = 0.20). Across all other time points, the T₁ ^high group experienced a significant and steady decrease in eGFR (Figure 2). The ΔeGFR between various time points (0–3, 0–6, 0–12, and 0–24 months) indicated a significant decline in graft function in the T₁ ^high group over all time points. At 3 months, the ΔeGFR was −6.3 (−11.4 to 0.0) mL/min/1.73 m² in the T₁ ^high group and 1.6 (−2.84 to 5.62) mL/min/1.73 m² in the T₁ ^low group (p < 0.01). At 24 months, the T₁ ^high group had a ΔeGFR of −13.0 (−25.3 to −7.48) mL/min/1.73 m² compared to the T₁ ^low group with 0.6 (−11.80 to 6.68) mL/min/1.73 m² (p < 0.01, see Table 1).

We analyzed the correlation between median T₁ and eGFR values over time. A significant inverse relationship was found between T₁ and eGFR at different time points. At 3 months, the correlation between T₁ and eGFR was moderate (ρ = −0.42, p < 0.01). This negative correlation continued at 6 months (ρ = −0.38, p < 0.01), 12 months (ρ = −0.43, p < 0.01), and remained stable at 24 months (ρ = −0.41, p < 0.01). Fisher’s Z transformation analysis between T₁ and ci association with graft function revealed no significant differences, showing that T₁ is similarly correlated with kidney function as the established ci score (details see Supplementary Table S2). In the subgroup, including only patients who underwent protocol biopsies, we also found significant correlations between T₁ and eGFR at months 3 (ρ = −0.71, p = 0.047), 9 (ρ = −0.81, p = 0.015), 15 (ρ = −0.81, p = 0.015), 18 (ρ = −0.74, p = 0.037), 21 (ρ = −0.81, p = 0.015), and 24 (ρ = −0.83, p = 0.010) (see Supplementary Tables S3, S4 for details).

The baseline (month 0) eGFR intercept for the T₁ ^low cohort was 39.9 mL/min/1.73 m², while the T₁ ^high group had a baseline eGFR intercept that was 9.20 units lower (p = 0.096). Over time, the T₁ ^low group showed a slight, non-significant decline in eGFR at a rate of 0.06 mL/min/1.73 m² per month (p = 0.63). In contrast, the T₁ ^high group experienced a significantly steeper decline, with an additional 0.41 units per month (p = 0.016) compared to the T₁ ^low group. This resulted in a total eGFR decline of 11.31 mL/min/1.73 m² for the T₁ ^high group and 1.40 mL/min/1.73 m² for the T₁ ^low group over 24 months (Figure 3).

We used ROC analysis to assess if T₁ can be used as a predictive marker for renal allograft loss (Figure 4). T₁ above the median resulted in a PPV for predicting graft loss of 52.2% with an AUC of 0.75, p = 0.007. Conversely, the NPV was 91.3%. T₁ demonstrated a sensitivity of 100% across the lower cutoff values, specifically from “>1,126 ms” to “>1,317 ms”. At the cutoff of “>1,317 ms”, the sensitivity slightly decreased to 92.9%, while the specificity saw a substantial increase, indicative of fewer false-positive results. At “>1,337 ms”, sensitivity is still 92.86%, but specificity has increased to 53.1%. At “>1,352 ms”, the sensitivity remained at 92.86%, and the specificity increased further to 62.5%. The analysis identifies T₁ “>1,352” ms as an optimal cutoff point in our patient cohort for balancing sensitivity and specificity in a clinical setting.

The Kaplan-Meier survival analysis revealed significant differences in graft survival between the groups (Figure 5). After 12 months, all kidney transplants in the T₁ ^low group were still functioning, compared to 91.3% in the T₁ ^high group. This difference became more pronounced over time, with survival rates of 91.3% versus 60.9% at 21 months and 87.0% versus 52.2% after 24 months (Log-rank test, p = 0.0015, Figure 5). A T₁ above the median was a significant risk factor for graft loss (HR 7.3, 95% CI: 2.6–21.0). The cortico-medullary difference of the T₁ (ΔT₁) was available in 32 patients. Patients without graft loss had a mean ΔT₁ of −337.13 ms, while those with graft loss had a mean of −251.81 ms, with no significant differences (p = 0.417).