A Step toward Understanding the Story Behind the Pictures: Molecular Diagnostics and the Banff Classification of Renal Allograft Pathology

Kidney transplantation is mired in mediocrity. Acceptance of current outcomes thwarts innovation; over the past two decades there has been no significant improvement in long-term allograft survival or approval of new immunosuppressant drugs. A barrier to progress is our limited understanding of disease processes that lead to allograft failure on the basis of descriptive histology alone. Novel molecular diagnostics provide powerful new tools to potentially understand complex disease processes. At the XV Banff conference for allograft pathology, a major step was taken to accelerate the use of molecular diagnostics in allograft pathology: the introduction of the Banff Human Organ Transplant (B-HOT) gene panel, which includes nearly 800 genes expressed in the most common histologic diagnoses encountered in clinical practice.¹ To promulgate the use of the B-HOT panel, a partnership with NanoString Technologies was established. This technology allows multiplex transcript quantification from formalin-fixed, paraffin-embedded biopsy samples that is paired with analytical software and allows individual laboratories to integrate the technology into clinical workflows. To enable multicenter validation, a data integration platform was created with an initial goal of analyzing B-HOT transcripts in ≥1000 clinical biopsy samples.

In this issue of JASN, Rosales and colleagues provide the first report of work related to the Banff group’s bold vision.² In this retrospective analysis of 326 archived biopsy specimens enriched for chronic active antibody-mediated rejection (CAMR; n=120) and T cell–mediated rejection (TCMR; n=49) obtained an average of 4 years post-transplant, the authors correlated Banff histopathology scores with B-HOT gene transcripts. The B-HOT panel distinguished CAMR or TCMR in aggregate using AMR and TCMR gene sets identified and validated in earlier microarray studies. The investigators hypothesized pathogenic insights would be revealed by correlation of AMR gene expression pathway scores with individual Banff histologic findings. Only one Banff pathology lesion, peritubular capillaritis, was correlated with AMR pathways in CAMR. Glomerular inflammation did not correlate strongly with AMR scores and there was no correlation of C4d or other Banff scores with AMR scores in the CAMR diagnostic group. Graft failure in patients with CAMR or borderline/suspicious TCMR was associated with multiple damage pathways, confirming previous observations that damage, rather than AMR pathway activity, was more relevant to CAMR outcome. Patients with multiple biopsy specimens without CAMR on the initial biopsy sample, but who developed CAMR on a subsequent biopsy sample within 5 years, showed elevation of a ten-gene set associated with donor-specific antibodies, suggesting CAMR might be predicted by molecular transcripts far earlier than histologic CAMR.

The study demonstrates the potential and challenges of the Banff group’s approach. Investigators are now able to perform transcriptomic, proteomic, and metabolomic analysis that can provide information at the cellular and subcellular level, in addition to microscopic review. To date, most transcriptomic studies have used homogenized tissue or pooled samples obtained from microdissection studies. These seminal studies have identified groups of genes within molecular pathways that associate with specific Banff histologic findings.³ The disadvantages of microarrays include loss of anatomic localization, which precludes “histomolecular integration,” the requirement for a dedicated biopsy sample, an inability to examine archived samples, and limited feasibility to integrate the technology into routine clinical practice because of cost and the requirement to ship samples to a commercial reference laboratory.

In situ techniques, such as NanoString, have the potential to overcome many of these issues but also have important limitations. The ability to analyze only a limited number of genes inherently constrains understanding of disease processes to known genes and established mechanistic pathways. The confirmation that B-HOT genes differentiate TCMR and CAMR in the Rosales et al.’s study is anticipated and self-fulfilling because the examined genes are limited to those already associated with these findings in microarray studies. An in silico assessment of the B-HOT gene panel used three bioinformatic strategies to determine if the B-HOT genes could distinguish the major Banff diagnostic groups including (1) a supervised approach, using the diagnostic and pathogenesis-based transcripts sets identified in previous microarray studies; (2) a semisupervised approach using genes in NanoString pathways associated with mechanisms likely to be important in allograft pathology; and (3) an unsupervised data-driven approach using a principal component analysis.⁴ This study revealed the B-HOT panel successfully identified rejection and nonrejection diagnoses, but there was a highly variable pattern of misclassifications per sample related to heterogeneity of gene expression within a given diagnostic category. Rosales et al.’s findings may not be reproducible because the overall sample size and number of cases within diagnostic categories is limited and the disease categories examined included a spectrum of severity and aggregate diagnoses. Gene expression in this study may be heterogenous within Banff disease categories because of the variation of the timing of the biopsies relative to appearance of donor-specific antibodies and change in allograft function or previous treatment. Heterogeneity in gene expression may also be related to variation in histology distribution because the 20-µm tissue sections used for molecular analysis may include transcripts from cells not visualized on thinner diagnostic histology sections. Additionally, histologic diagnoses may be dependent on findings that do not produce a strong transcriptomic signature. Transcripts for subtle but important pathologic processes may be obscured by abundant ones that are not causally related to disease. Areas of special interest may contain a mix of cell types that will not be captured without adequate transcriptomic resolution.

Rapid advances in spatial transcriptomics now allow single-cell sequencing information to be captured along with spatial information so that challenges related to resolution should not limit future studies. However, interpretation of the vast amounts of transcriptomic information produced from such high-resolution studies is a daunting challenge that is compounded by the need to also integrate clinical and laboratory data to either generate new hypotheses or confirm existing disease models. Rosales and colleagues demonstrate that potential of the Banff group’s vision. However, the vision must be even bolder: The path forward will require the formation of new collaborations between clinicians, pathologists, and bioinformaticians along with patient engagement to conduct carefully designed, prospective studies with serial protocol biopsies that integrate clinical and laboratory information to understand the cellular and subcellular basis of the diseases that limit allograft survival.

Disclosures

J.S. Gill reports serving in an advisory or leadership role for the American Society of Transplantation; receiving research funding from Astellas; and having consultancy agreements with, and receiving honoraria from, Takeda and Veloxis. All remaining authors have nothing to disclose.

Funding

None.

Author Contributions

M.L.Z. Bissonnette, A.M. Cunningham, and M. Riazy reviewed and edited the manuscript; and J.S. Gill wrote the original draft.