Kidney transplantation (KTx) represents the best option for patients with end-stage renal disease (ESRD). Acute kidney allograft rejection (AR) is one of the main causes of allograft loss (1). The gold standard for the diagnosis of AR currently relies on renal transplant biopsy (2). The systematic histological analysis of a renal sample allows for the identification of both type and degree of rejection according to the conventional Banff classification, which provides a standardized scoring of kidney injury (3). More specifically, the Banff classification semi-quantifies both acute and chronic inflammation in five histological compartments, either individually or collectively: glomeruli, tubules, interstitium, peritubular capillaries, and arteries (3).

Although an ultrasound-guided core needle biopsy of kidney allograft is regarded as relatively safe, it remains an invasive procedure with a ∼7% rate of complications (4) including bleeding or development of an arteriovenous fistula. Furthermore, inter-observer variability and sampling errors limit its benefits (5). Therefore, the development and validation of noninvasive methods to detect T cell-mediated rejection (TCMR) and antibody-mediated rejection (ABMR) would be highly valuable. Urinary and plasma biomarkers of acute and chronic renal rejections have been identified and are currently under clinical validation (6–8). Imaging techniques have significantly improved over the past decades, thereby proposing another approach to the noninvasive diagnosis of AR. More specifically, ¹⁸F-fluorodeoxyglucose [(¹⁸F)FDG] positron-emission tomography combined with computed tomography (PET/CT) as first-line examination with a high negative predictive value may help noninvasively prevent avoidable transplant biopsies in kidney transplant recipients (KTRs) with clinically suspected AR but no histological lesions (9, 10). Indeed, the AR-associated immunological reaction against donor antigens induces the recruitment of mononuclear leukocytes into the renal transplant, which corresponds to the core of the Banff classification. The boosted metabolism of these inflammatory cells can be assessed by PET quantification of the renal uptake of the radiotracer, [¹⁸F]FDG. The lack of detectable signals, therefore, suggests the absence of active inflammation.

In the present post hoc analysis of our prospective studies, we test the correlation of biopsy-based Banff vs. PET-based scores of acute inflammation in renal transplants. Furthermore, we assess the prognostic performance of the renal uptake of [¹⁸F]FDG regarding allograft outcomes at one year post per cause biopsy.

Various non-invasive tools are under investigation to rule out the diagnosis of AR in order to mitigate the risks of systematic use of renal allograft biopsies. More specifically, we and others have suggested that imaging-based approaches may be helpful in both structural and functional assessments of the entire renal allograft, which contrasts with the biopsy-based information limited to small-size samples. [¹⁸F]FDG-PET/CT is routinely used for the characterization, staging, and follow-up of inflammatory processes of various origins (14). The usefulness of [¹⁸F]FDG-PET/CT in the management of KTRs presenting with suspected AR has been demonstrated in both training and validation cohorts (9, 10). In our first pilot study published in 2015, sensitivity and specificity of [¹⁸F]-FDG PET/CT in diagnosing acute rejection were 100% and 50% respectively, with a meanSUV threshold of 1.6. In this 2015 cohort, characterized by a 25% prevalence of biopsy-proven AR, the corresponding negative and positive predictive values were 100% and 43.75%, respectively. The proposed threshold of 1.6 of mSUV_mean was then validated in a second study of 86 patients. The ROC area under the curve was 0.86. Test sensitivity and specificity corresponding to the threshold value of 1.6 were 100% and 30%, respectively. In the present post-hoc analysis of these two prospective studies including 105 patients who have undergone a per cause biopsy, we show a positive and significant correlation between mSUV_mean and the acute Banff score. The ¹⁸F-FDG uptake by the renal graft also correlates with inflammation at the microvascular level. Moreover, increasing grades of “total i” scores were associated with increasing mSUV_mean. These observations suggest that the renal uptake of the radiolabeled glucose analog [¹⁸F]FDG is correlated with the severity of biopsy-proven inflammation in the renal allograft. By contrast, the extent of fibrosis and tubular atrophy does not affect the renal [¹⁸F]FDG uptake. Of methodological note, similar results were found when focusing on 96 patients after excluding the eventual second biopsy. The radiotracer [¹⁸F]FDG follows a tissular distribution and cellular metabolism comparable to the glucose: it enters into cells via the glucose transporters (GLUT1), where it is phosphorylated by the hexokinase into [¹⁸F]FDG-6-phosphate and trapped in the cells. Hence, the intracellular accumulation of [¹⁸F]FDG-6-phosphate reveals a high cellular metabolism and is detectable by PET/CT (5, 15). Stimulated inflammatory cells are metabolically activated and show an increased expression of glucose transporters, which increases the uptake and accumulation of [¹⁸F]FDG in tissue sites of inflammation (14, 16). Our correlation observed between mSUV_mean and increasing grades of the “total i” score may thus reflect the accumulation of leucocytes metabolically active in the renal cortex, notably in the case of AR. Conversely, fibrotic tissue and atrophic tubules are less metabolically active, with no detectable impact on [¹⁸F]FDG uptake.

Historically, the lesions selected for scoring AR were the infiltration of the interstitium (i) and the tubules (t) by mononuclear cells, which are regarded as representative of the severity of TCMR (13). The “ti character” was added in Banff 2017 and is defined as inflammation in the total parenchyma, including scarred and non-scarred cortex, whereas the “i score” is limited to the non-scarred cortex (3). Compared with “i and t scores”, the “ti score” better reflects the molecular phenotypes of the tissue and is a better predictor of graft survival outcomes than the “i score” in cases where at least mild IFTA is present (13, 17). The correlation between [¹⁸F]FDG uptake and the severity of the histological lesions has been previously shown in a rat model of kidney AR. Reuter et al. (18) showed the most intense [¹⁸F]FDG signaling in animals with biopsy-proven AR characterized by large infiltrates of inflammatory mononuclear cells in the renal cortex (which mimics the “ti” score).

Although [¹⁸F]FDG uptake significantly correlates with the degree of histological inflammation, we could not demonstrate that mSUV_mean predicts the 1-year outcomes of kidney allografts. This dichotomy could be explained by the small number of cases in the different groups and the inherent low statistical power limiting the interpretation of these results. Another limitation of this study is the wide range of time between the transplantation and the date of the per cause biopsy [i.e., 272 (32; 1,721) days]. In the present cohort, the histological score “total i” does not predict the outcome of the graft, in contrast with the Mengel study including 104 patients with per cause biopsy (13). Note that “an event” in Mengel et al. was defined as (i) either graft loss or (ii) persistent (at least 3 months) eGFR <30 ml/min estimated by the Cockcroft–Gault formula. Prognostic studies currently performed on non-invasive urinary markers are also limited. Rabant et al. (19, 20) have shown that the CXCL10: Cr ratio, measured at the time of a per cause biopsy showing ABMR, can identify patients at high risk for kidney allograft loss. They also showed that CXCL9 and particularly CXCL10 urinary levels quantified at early post-transplantation might predict immunologic quiescence in clinically and histologically stable kidney recipients. Another study demonstrated that high urinary CXCL10 was associated with inferior endpoint-free graft survival and it was an independent predictor of long-term renal allograft outcomes, irrespective of histology results (21). Still, Hirt-Minkowski et al. could not demonstrate that urine CXCL10 monitoring had a beneficial effect on 1-year outcomes in a cohort of patients with protocol biopsy (22). However, the experimental designs of these studies, with repeated biomarker assessments and/or highly selected patients, are very different from the present study.

While [¹⁸F]FDG-PET/CT seems to be a promising non-invasive tool to entirely image the renal allograft, there are some limitations to its use: (i) the poor specificity of [¹⁸F]FDG tracer which accumulates in other inflammatory conditions, like tumors or infections; (ii) the restricted availability of PET/CT machine and its cost; (iii) the minor exposure to radiations originating from both PET and CT procedures; (iv) the 3-hour delay between [¹⁸F]FDG i.v. injection and image acquisition (10, 14). Of note, the current development of digital PET/CT would probably improve the image quality, while dropping the financial burden and reducing the radiation dose for the PET and CT portions (23, 24).

The interpretation of our post hoc analysis is limited by the small number of patients with biopsy-proven AR and the single-center nature of the study. Moreover, by purpose, the design of the present study does not take into account the final histological diagnosis but it focuses on the degree of inflammation. Due to the evolution of the Banff classification, the final diagnosis may change over time. Recent works suggest that there is a growing interest to automate the Banff classification combining histological features and omics to improve diagnosis and enable better risk stratification and therefore patient management (25).

The present work suggests an interesting approach to reducing the number of “unnecessary” biopsies (26). Additional prospective studies are needed to evaluate the potential interest of PET/CT in clinical decision-making and practical use, e.g., for monitoring treatment and adjusting the immunosuppressive regimen post-AR, as well as the combination of noninvasive imaging with the new urinary and blood biomarkers (like CXCL10 and CXL9). Moreover, it would be interesting to implement the mSUV_mean in web-based tools designed to classify acute kidney injury in KTRs (27).

[¹⁸F]FDG-PET/CT may help to noninvasively assess the degree of histological inflammation in KTR with suspected AR and clinical indication of kidney allograft biopsy. The prognostic performance of mSUV_mean at 1 year could not be formally demonstrated in the present cohort.