No Time for Cancel Culture

Since 1991, the Banff Allograft Pathology Criteria have been utilized to interpret histologic findings in organ allografts. Banff criteria are developed in an international, collaborative, and inclusive environment to create a standardized but constantly evolving interpretation of allograft histology to facilitate appropriate clinical management across transplant practices and further provide continuity in end points for drug development.¹ This biennial meeting is open to clinicians, scientists, and pathologists, and criteria were initially based on standard histologic staining methods. More recently, criteria incorporate additional tests, including C4d staining, donor-specific antibody (DSA), and gene expression of the biopsy. While dedicated working groups constantly generate data and explore novel approaches including machine learning and integration of molecular phenotyping to continue to refine the classifications,² the Banff Criteria do have limitations. There has not been a large-scale validation of the criteria, and there is known interobserver and intraobserver variability with a dependency on adequate tissue.³ Moreover, the foundational criteria are not specific and overlap between relevant diagnostic entities and thus require contextual clinical interpretation. Despite limitations, the Banff criteria are accepted across the world, and incorporated in research studies and trials of new therapies and biomarkers.

In this issue of JASN, Vaulet et al. utilized a combined series of three publicly available datasets, including their own, of whole transcriptomic microarray analysis of bulk RNA extracted from a total of 909 kidney allograft biopsies.⁴ On the basis of Banff criteria, biopsies included were reduced into the following diagnostic categories: biopsies without rejection, with antibody-mediated rejection (ABMR), T cell–mediated rejection (TCMR), and mixed (TCMR and ABMR) rejection. The authors excluded HLA-DSA–negative ABMR and mixed rejection biopsies to provide consistency in the three datasets. Cell fraction estimates and phenotypes using this transcriptomics data were deconvoluted with the CIBERSORTx algorithm and the gene expression signature matrix KTB18, which were previously constructed from single cell data.⁴ In addition, to corroborate this deconvolution process, in a subset of 18 biopsies, they analyzed the phenotypes using actual observed immune cells from spatial single-cell multiplexed immunohistochemistry on the basis of the Multiple Iterative Labeling by Antibody Neodeposition.⁵ This technique uses a panel of immune cell markers to identify different cell populations while providing a direct count of the observed cells. These approaches revealed that the composition of immune cells in allograft biopsies correlated poorly with the rejection categories as denoted by the Banff criteria and further identified that terminally differentiated effector memory CD8 cells (TEMRA CD8) within the allograft primarily associated with graft failure, regardless of Banff phenotype. These findings confirm those of others linking TEMRA CD8 with graft outcomes,⁶ further suggesting a new potential therapeutic target.

These provocative findings, at first blush, appear to invalidate the international Banff process and resulting consensus on histologic interpretation of allograft biopsies. They highlight the limited specificity of Banff diagnostic categories and further suggest that more advanced techniques and unsupervised computational approaches beyond basic histology are necessary for accurate biopsy interpretation to improve graft outcomes.

In contrast to the authors' interpretations, we believe that their findings confirm the Banff approach and point toward the disconnect between the causal trigger of an alloimmune response and the immune phenotype of the Banff class of rejection. For example, development of HLA-DSA requires cognate B-cell activation through T follicular helper cells (T_FH). The resulting HLA-DSA may lead to activation of multiple effector mechanisms that ultimately converge on mononuclear cell inflammation, a common feature in all rejection scenarios. In the setting of TCMR, the initiators are cognate donor-specific T cells, which eventually utilize the same downstream effector mechanisms. Thus, regardless of the instigating event, inflammation (“i”) is present. Interestingly, both liver and heart Banff categories utilize “acute cellular rejection” and not TCMR specifically. This is in recognition of the convergence of global inflammation of the organ during rejection.

There are some notable features of this study to consider. The cohorts themselves differ in indication for biopsy; one dataset (GSE147089) includes a substantial proportion of surveillance biopsies obtained on functionally stable patients (Table 1), while both public dataset 2 (GSE21374) and public dataset 1 (GSE36059; which includes an unknown subset of biopsies from GSE21374) only include indication biopsies for graft dysfunction or proteinuria. Various iterations of Banff criteria were utilized across all three datasets, which may affect the ABMR diagnoses, and the classification of borderline rejection. Moreover, they overlook any interpretation of chronic-active TCMR or total inflammation (ti) or inflammation in areas of interstitial fibrosis and tubular atrophy (i-IFTA) scoring, which may have been grouped into “nonrejection” category. These phenotypes all have important clinical implications. Indeed, total-i and i-IFTA may be seen in all types of rejection⁷ and better correlate with outcomes than most Banff lesions used to render diagnosis. In particular, i-IFTA is best associated with molecular injury scores and thus disease activity.⁸ It would have been useful to determine if a combination of either of these histologic features with TEMRA CD8 perfected prognostic ability.

Table 1.

Bulk transcriptomic datasets used to estimate immune cell fractions⁴

Dataset	Biopsies, N	Surveillance Biopsy, %	Banff Schema	Transcriptional Assessment	Nonrejection Diagnoses
BIOMARGIN GSE147089 (PMID: 32641395)	224	63	2017	Affymetrix GeneChip 3′ IVT PLUS Reagent Kit	ATN BKPVN GN IFTA PyelonephritisTMA
Public dataset 1 GSE36059 (PMID: 23356949)	403 (includes a subset of GSE21374)	None	2009	Affymetrix HG_U133_Plus_2.0 GeneChip	AKI BKPVN GN IFTA “Other”
Public dataset 2 GSE21374 (PMID: 20501945)	282	None	2007	Affymetrix HG_U133_Plus_2.0 GeneChip	AKI BKPVN GN IFTA “Other”
Combined set 1 GSE36059 + GSE147089	596	Approximately 23%	2009 2017	Affymetrix HG_U133_Plus_2.0 GeneChip+Affymetrix GeneChip 3′ IVT PLUS Reagent Kit	AKI ATN BKPVN GN IFTA “Other” Pyelonephritis TMA
Combined set 2 GSE21374 + GSE147089	506	Approximately 28%	2007 2017	Affymetrix HG_U133_Plus_2.0 GeneChip+Affymetrix GeneChip 3′ IVT PLUS Reagent Kit	AKI ATN BKPVN GN IFTA “Other” Pyelonephritis TMA

ATN, acute tubular necrosis; BKPVN, BK polyomavirus nephropathy; IFTA, interstitial fibrosis and tubular atrophy; TMA, thrombotic microangiopathy.

As we often see, interpretation of any biopsy is dependent in part (as much as we are reluctant to admit) on clinical findings. This is particularly true of these datasets where there is no information about DSA or concomitant nonrejection histologic findings (e.g., drug toxicity, severe hypertensive changes; Table 1). Interestingly, all biopsies studied had some level of inflammation in both deconvoluted and Multiple Iterative Labeling by Antibody Neodeposition datasets; the predicted fraction of immune cells was at least 10.5%, which coincidentally confirms the Banff threshold of “i1.” Our interpretation of this finding is that rejection, unlike the presence of a tumor/cancer, is not a yes/no phenomena, but a continuous process that the Banff schema embraces.

Another issue with clinical implications is the grouping of distinct clinical phenotypes together. Is chronic active TCMR immunologically similar to chronic active ABMR? With bulk transcriptomic data, one cannot tell. Yet, all rejections were combined into the data analysis, regardless of their clinical classification. While the authors support this approach, it is difficult to support regarding clinical treatment; in fact, there is no mention of how recipients were treated, and the impact of those treatments would be an essential element determining graft outcomes. Clinically, acute ABMR and acute TCMR behave differently in treatment success; the former has limited treatment options, and the latter has depended on graft function to indicate complete response although repeat biopsies may show ongoing interstitial inflammation. However, once an inflammatory compartment is established, noncognate T cells and monocytes/macrophages are involved independent of the initial immune pathway (TCMR versus ABMR), causing further injury and propagating inflammation. Establishing one subset of CD8 T cells as a therapeutic target requires broader study in determining efficacy of current treatments to this population.

While this work continues to confirm the role of natural killer (NK) cells in ABMR and hence the possibility that NK markers may serve a potential diagnostic tool (as well as therapeutic target), the presence of NK cells in unsupervised clustering algorithms was surprisingly and, contrary to previous molecular studies, insufficient to distinguish T-cell versus antibody-mediated rejection. This may be because NK cells are quantitatively low compared with all the T cells and monocytes potentially dominating the unsupervised analysis in this study. Alternatively, recent studies suggest that a novel NK cell population infiltrating kidney allografts with ABMR are driven by IL-21–producing T_FH cells, again, demonstrating crosstalk between allo- and innate immunity and collaborative presence of both T and NK cells in ABMR.⁹ The complexity of such immune responses may be further deduced from such molecular studies.

Finally, the authors indicate that they gathered information about cellular elements of the kidney. While these data serve as the denominator of their cell analyses, there is no information provided about the results although they account for more than 50% of the proportion of all cells analyzed. Prior studies have identified the key roles of endothelium and injured tubular epithelium, serving as a nidus for immunological responses directly associated with allograft function and outcome.¹⁰

The mere finding of this study that the immune cell composition is ill-related to the Banff rejection categories does not allow us to conclude which approach is diagnostically more accurate or better suited to manage patients after kidney transplantation. The findings rather propel the well-established Banff process of an iterative learning from new insights informing the Banff rules, criteria, and diagnosis. Already since the very first Banff meeting, it was clear that no iteration of the Banff classification will be absolutely accurate; hence, the international Banff process was established to constantly evolve as new knowledge emerges. The elegant approach Vaulet et al. chose here showed us that bulk molecular studies reveal phenotypes as nonspecific as the Banff lesions they are associated with. Putting all of these into the spatial biopsy (which has been the Banff approach since 1991, so to say “spatial pathology”), serological, and clinical context is rapidly emerging as the promising but more challenging approach toward precision diagnostic in transplantation.

Disclosures

Disclosure forms, as provided by each author, are available with the online version of the article at http://links.lww.com/JSN/E699.

Funding

R.B. Mannon: US Department of Veterans Affairs (I01CX002666) and National Institute of Diabetes and Digestive and Kidney Diseases (U01DK115997).

Author Contributions

Conceptualization: Roslyn B. Mannon, Michael Mengel.

Resources: Roslyn B. Mannon.

Writing – original draft: Roslyn B. Mannon.

Writing – review & editing: Roslyn B. Mannon, Michael Mengel.