Predicting Future Outcomes from Kidney Biopsies with MCD/FSGS Lesions: Opportunities and Limitations

• nephrotic syndrome
• proteinuria
• focal segmental glomerulosclerosis
• human genetics
• pathophysiology of kidney disease and progression

In this issue of JASN, Zee et al. identified a subset of 48 histologic and 13 ultrastructural morphologic features (descriptors) that predict a series of outcomes of minimal change disease (MCD) or FSGS.¹ They extracted 224 patients from the deeply phenotyped Nephrotic Syndrome Study Network (NEPTUNE) cohort’s central Digital Pathology Repository and analyzed n=196 for the outcome “disease progression,” n=185 for “disease remission,” and n=88 for “response to immunotherapy.” The authors used several machine-learning models to optimize the outcome prediction of which random forest models consistently yielded the highest prediction discrimination and were used for subsequent analyses.

Among the descriptors predicting time to disease progression were expected features such as segmental glomerulosclerosis, absence of minimal changes, periglomerular fibrosis and interstitial fibrosis, and interstitial foam cells. The ultrastructural parameter foot process effacement added little prediction power. Descriptors predicting the time to complete proteinuria remission also included segmental glomerulosclerosis, absence of minimal changes, periglomerular fibrosis and interstitial fibrosis, and, unexpectedly, ultrastructural diffuse microvillous transformation of podocytes. In general, model performance increased when the individual histologic descriptors were ranked and the top 15 predictors included in the model. The analysis of time to treatment response included only n=88 patients from the NEPTUNE study and the prediction model leveled off when considering the top ten predictors, which unexpectedly included acute tubular injury. For all three outcome scenarios, considering grouped descriptors was equipotent in outcome prediction. The authors concluded that known and novel morphologic descriptors most predictive of key clinical should be included in standardized clinical pathology reports describing MCD and FSGS histopathology.¹

This study confirmed the relevance of previously well-known general and disease-unspecific features of chronic kidney damage, but, more importantly, showcases how machine-learning approaches can help to identify the most predictive among numerous morphologic features. For example, foot process effacement itself, which is required for pathologic diagnosis of podocytopathies, was among the lowest-ranked predictors, whereas, for example, the villous transformation of podocytes, which is often observed but previously not defined as a necessary diagnostic criterion, was among the highest. The identification of such morphologic findings might also support insights into disease pathomechanisms, such as suggesting that glomerular endothelial cells play a more important role in the progression of podocytopathies, as previously thought. Endothelial cell honeycombing and fenestrations, as suggested by Zee et al., may have been missed so far because detailed electron microscopy analysis is required to observe these. It will be interesting to see if others can validate these findings. Concerning the response to immunotherapy, it is interesting that acute tubular injury, but not its extent, was ranked among the most predictive features, which would require a mechanistic explanation.

The authors suggest reporting ≤15 descriptors in a qualitative or semiquantitative manner to further improve pathology reports, which would imply practical feasibility and interobserver variability as concerns. For example, Zee et al. name the number of normally appearing glomeruli as predictive of a good outcome. However, interobserver variability in recognizing even normal glomeruli is known to be problematic.² Grouping descriptors might reduce manual overheads and somewhat improve reproducibility. A promising solution could be the use of digital pathology and artificial intelligence–augmented computer vision for the automated extraction of numerous morphologic features.³ This would also enable a more holistic approach by deriving quantitative morphologic data from all compartments of the biopsy, not limited to a prespecified set of features.

The clinical relevance of a more detailed pathology report on MCD and FSGS is questionable because the clinical management of patients with podocytopathies relies on the underlying causative diagnosis, something kidney biopsy is unable to provide. The adherence to the terms MCD and FSGS beyond the description of lesion tissue patterns is a persistent source of confusion because neither MCD nor FSGS name a disease.⁴⇓^–6 Nowadays, the management and prognosis of podocytopathies are determined by the specific causes of podocyte injury, on the basis of additional diagnostic tools (Figure 1). For example, whole-exome sequencing can define a molecular diagnosis in ≤58% of children with steroid-resistant nephrotic syndrome, indistinguishable by clinical phenotype or kidney biopsy results.⁷ Thus, it is likely the diagnosis underlying proteinuria in approximately 60 of the 104 children studied by Zee et al. was a genetic podocytopathy or one of its monogenic phenocopies not requiring any immunotherapy.^5,6 Of note, genetic causes of MCD and FSGS are not restricted to children. In contrast, such patients benefit from inhibitors of the renin-angiotensin system and the sodium-glucose transporter-2.^7,8 The same applies to patients with podocyte injury due to excessive adaptation to the various causes of glomerular hyperfiltration (adaptive FSGS).^4,5,9 Thus in clinical practice, predicting “disease progression” in podocytopathies integrates many sources of information and does not rely on unspecific histomorphologic consequences of podocyte injury. Prognosis depends on the response to an initial course of corticosteroids (in patients who are pediatric), the presence or absence of nephrotic syndrome, risk constellations for adaptive podocyte injury, and results from genetic testing, which also help select the correct treatment, for example, inhibitors of the Renin-Angiotensin-System for all patients with genetic and adaptive podocyte injuries versus immunotherapy only for (auto)immune podocytopathies, with antinephrin antibodies as one potential novel biomarker.^4,10

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

The role of kidney biopsy in the diagnosis and prognosis prediction of podocytopathies. Kidney disease categories keep developing; this offers novel opportunities to overcome the hurdles related to MCD and FSGS, which are not diseases. Although defining a precise molecular diagnosis for these podocytopathies has remained challenging, whole-exome sequencing, together with reverse phenotyping for subtle manifestations of monogenic syndromal disorders and the identification of antinephrin antibodies as a cause for an autoimmune podocytopathy, offer promising possibilities for a definite diagnosis in the majority of patients. Prognosis prediction from histologic lesion patterns benefits from analyzing the biopsy at a high granularity and employing machine-learning tools. The future will be adding other diagnostic tools to predict prognosis (and define treatment) on the basis of identifying the underlying disease-causing podocyte injury.

The study by Zee et al. is a good example of how machine-learning tools can be used to derive those morphologic descriptors from a kidney biopsy that associate well with certain outcomes. In the future, digital pathology and computer vision may use automated algorithms to overcome remaining technical hurdles such as quantitations, interobserver variability, and manual overhead for pathologists. From a clinical perspective, however, the limitations of kidney biopsy in defining the underlying cause of the injury patterns of MCD or FSGS mandate the use of other diagnostics to define the causative diagnosis of podocyte injury, to guide treatment decisions, and predict the response to this treatment, and the long-term prognosis.⁴

• Copyright © 2022 by the American Society of Nephrology