Ultrastructural Characterization of Proteinuric Patients Predicts Clinical Outcomes

Significance Statement

Glomerular features ascertained by electron microscopy are underreported in clinical practice, and their value in predicting outcome is unclear. This study is the first comprehensive investigation of the association of clinical outcomes with 12 glomerular electron microscopy descriptors reflecting the status of podocytes, endothelial cells, and glomerular basement membranes, individually and as electron microscopy profiles after descriptor-based consensus clustering. The authors demonstrate that severe effacement and microvillous transformation, individually and as a component of clusters, were associated with proteinuria remission, whereas prominent endothelial cell and glomerular basement membrane abnormalities were associated with loss of renal function. These findings highlight the importance of a standardized and comprehensive ultrastructural analysis, and that use of quantifiable structural changes in assessing patients with proteinuria might have important clinical implications.

Abstract

Background

The analysis and reporting of glomerular features ascertained by electron microscopy are limited to few parameters with minimal predictive value, despite some contributions to disease diagnoses.

Methods

We investigated the prognostic value of 12 electron microscopy histologic and ultrastructural changes (descriptors) from the Nephrotic Syndrome Study Network (NEPTUNE) Digital Pathology Scoring System. Study pathologists scored 12 descriptors in NEPTUNE renal biopsies from 242 patients with minimal change disease or FSGS, with duplicate readings to evaluate reproducibility. We performed consensus clustering of patients to identify unique electron microscopy profiles. For both individual descriptors and clusters, we used Cox regression models to assess associations with time from biopsy to proteinuria remission and time to a composite progression outcome (≥40% decline in eGFR, with eGFR<60 ml/min per 1.73 m², or ESKD), and linear mixed models for longitudinal eGFR measures.

Results

Intrarater and interrater reproducibility was >0.60 for 12 out of 12 and seven out of 12 descriptors, respectively. Individual podocyte descriptors such as effacement and microvillous transformation were associated with complete remission, whereas endothelial cell and glomerular basement membrane abnormalities were associated with progression. We identified six descriptor-based clusters with distinct electron microscopy profiles and clinical outcomes. Patients in a cluster with more prominent foot process effacement and microvillous transformation had the highest rates of complete proteinuria remission, whereas patients in clusters with extensive loss of primary processes and endothelial cell damage had the highest rates of the composite progression outcome.

Conclusions

Systematic analysis of electron microscopic findings reveals clusters of findings associated with either proteinuria remission or disease progression.

Electron microscopy (EM) analysis is routinely applied in clinical practice to enhance, clarify, or confirm a diagnosis of glomerular disease; however, ultrastructural features are rarely used to predict clinical outcomes. Additionally, reporting of EM features is conventionally restricted to a few parameters, which, in the setting of podocytopathies such as minimal change disease (MCD) and FSGS, are often limited to the amount of foot process effacement (FPE).^1–3 Although the proportion of podocytes exhibiting FPE has been proposed to carry diagnostic weight in differentiating FSGS etiology,^4,5 podocytes exhibit a variety of other underrecognized structural changes that may contribute to better understanding of the disease status, mechanisms, and disease progression.⁶ Furthermore, little attention is given to other components of the glomerular barrier, such as endothelial cells (ECs) and glomerular basement membranes (GBM), which may be equally relevant in understanding mechanisms and contribute to determining outcome in nephrotic syndrome, including treatment response. Although pathologists have historically focused on one element at a time as hallmarks of disease or as a reflection of mechanism of proteinuria (i.e., FPE), the combination of a variety of structural changes affecting podocytes, ECs, and GBMs may better characterize the heterogeneity in progression and response to therapy among patients conventionally grouped under the names MCD and FSGS. Thus, a rigorous and comprehensive morphologic analysis is critical to explore the predictive value of structural changes of the renal parenchyma at the cellular and extracellular level.

The Nephrotic Syndrome Study Network (NEPTUNE) pathology working group developed the NEPTUNE Digital Pathology Scoring System (NDPSS) to capture the complexity of histologic and ultrastructural changes (“descriptors”) in the renal parenchyma of patients with proteinuria and nephrotic syndrome.⁷ The NDPSS applies quantitative and semiquantitative metrics to measure the presence and severity of these changes.⁸ One of the main barriers for the use of morphologic data in clinical research and practice remains reproducibility of observations. Only a limited number of rigorous studies in renal pathology has tested reproducibility of morphologic diagnoses or individual features before their use in establishing classifications or testing their predictive value.^9–11,12

In this study, we performed a rigorous morphologic analysis inclusive of all the discrete elements reflecting structural glomerular damage at the cellular and extracellular level, using the NDPSS, and tested their reproducibility. The goal of this study was to test how these ultrastructural changes associate with outcomes and could contribute to the categorization of patients with proteinuria.

Methods

Case Selection

Patients were selected from the NEPTUNE cohort, a North American, multisite, observational study of children and adults with >0.5 g/d of proteinuria (>1.5 g/d in the second study phase), enrolled at the time of a clinically indicated renal biopsy.⁸ The renal biopsies, including histology and EM images, are collected in the NEPTUNE Digital Pathology Repository (DPR) as previously described.⁷ EM images were taken by an EM technician or a renal pathologist without an standardized protocol across laboratories. The NEPTUNE DPR contained a total of 242 cases of MCD and MCD-like (n=119) (MCD-like defined by the absence of segmental sclerosis and presence of partial effacement and/or global sclerosis exceeding that expected for age),^4,13 and FSGS (n=123) that were eligible for the current study (more than five EM images and follow-up data). Cases included in this study had between five and 83 EM images, with a mean of 18.2 per case.

Ultrastructural Analysis

A NEPTUNE EM electronic scoring form was created to capture the ultrastructural features (Figure 1). EM descriptors are illustrated in Figure 2. Each scoring sheet was labeled with the NEPTUNE study identifier. Each scoring form indicated the total number of EM images to score per case and a quality control indicator where pathologists could mark if images were of poor quality. For each descriptor, pathologists could indicate whether they had any uncertainty in their rating/score.

Figure 1.

Electronic scoring sheet detailing the electron microscopy descriptors examined in this study. The database contains a scoring sheet (A) and corresponding form (B) for each case. Each scoring sheet was labeled with the NEPTUNE study identifiers, and contained the total number of images to score per case and the parameters listed in Table 1 for scoring. Pathologists open the EM Scoring Form (B), which contains the biopsy identifier and number of available images at the top of the form for a single case (C). The form contains an option to indicate that all EM images available have poor quality and the case therefore cannot be evaluated (D). Every EM descriptor is listed in the form with each scoring option labeled. For every score, the form also contains an option to indicate uncertainty in scoring (E). For subepithelial electron dense deposit descriptors (e.g., stage I, stage II, etc.), scoring can only be implemented if pathologists first mark the subepithelial electron dense deposits as “present” (F). Error messages will appear otherwise. Notes can be made in the free text box at the bottom of the form (G). To move to a different record, pathologists can use the navigation buttons at the bottom of the screen (H). To search for a specific case, the biopsy identification number can be entered into the search box at the bottom of the screen and the form will automatically jump to that case (I). To open data for all records at once, pathologists can open the EM Scoring Sheet (A).

Figure 2.

Illustrations of multiple electron microscopy descriptors evaluated and scored by the study pathologists. (A) Black arrow illustrates FPE with condensation of the actin-based cytoskeleton; black asterisk indicates microvillous transformation of the podocytes; short thick black lines indicate thin GBM; white arrow indicates preserved fenestration of EC (magnification ×10,000). (B) black arrows indicate FPE; white chevrons indicate EC honeycombing; white arrow indicates preserved fenestrations of EC; short thick black line indicates thin GBM (magnification ×10,000). (C) Black arrow illustrates foot processes effacement; black arrowheads indicate preserved primary processes; white arrows indicate loss of EC fenestrations; white chevrons indicate loss of EC honeycombing; short thick black lines indicate thick GBM (magnification ×6000). (D) Black arrow indicates loss of podocyte primary processes and foot processes effacement, white asterisks indicate podocyte detachment with newly formed, extracellular matrix; white arrowheads indicate electron densities/hyalinosis (magnification ×3500).

A protocol with instructions for accessing the images in the NEPTUNE DPR, and the EM descriptor manual (Table 1), were distributed to four scoring NEPTUNE pathologists, all with ≥10 years of experience, along with an example scoring form template. Before initiation of the morphologic data collection, training sessions were conducted iteratively and included conference webinars for discussion, pilot scoring tests, and revisions to the scoring form and protocol. Two pilot scoring tests were conducted, each comprising ten images scored by every pathologist, and followed by webinar discussion to resolve any problematic scoring issues and amend the protocol as needed.

Table 1.

List of EM descriptors and metrics

Descriptor	Metric
FPE	Loss of foot processes. Exclude areas where two GBMs touch. Percentage of glomerular capillary surface area affected by effacement will be recorded as a semiquantitative value (0=0%–10%; 1=11%–25%; 2=26%–50%; 3=51%–75%; 4=76%–100% of the outer GBM surface).
Condensation of the actin-based cytoskeleton	Electron-dense cytoskeleton is reorganized and condensed at the GBM aspect of epithelial cell (podocyte) foot processes. Exclude areas where two GBMs touch. Percentage of glomerular capillary surface area affected by effacement will be recorded as a semi-quantitative value (0=0%–5%; 1=≤50%; 2=>50% of the outer GBM surface).
Microvillous transformation	Cytoplasmic projections into the urinary space that emanate from the luminal side of epithelial cell (podocyte) membrane are present. Percentage of glomerular capillary surface area affected by effacement will be recorded as a semiquantitative value (0=0%–5%; 1=≤50%; 2=>50% of the outer GBM surface).
Loss of primary processes	Epithelial cell (podocyte) body sits directly on underlying GBM. This is generally accompanied by complete effacement (loss of foot processes). It will be recorded as present or absent (0=present–normal; 1=absent–loss).
Epithelial cell (podocyte) detachment	Detachment of epithelial cells from underlying GBM is present with intervening new loose basement membrane material (halo). Recorded as present or absent (0=absent, 1=present).
Thickening and thinning of the GBM	GBM thickness will be assessed on ten cross-sections of capillary loops at foci where there are no capillary wall deposits or in segmentally collapsed and sclerotic (solidified) areas. If only sclerotic or collapsed areas are present, select option 999 (cannot determine).
	No change is selected when none of the following changes occur.
	Decreased thickness is selected when at least 25% of the GBM appear thinner than normal.
	Increased thickness is selected when at least 25% of the GBM appear thicker than normal.
	Mix pattern (thick and thin) is selected when thin and thick areas are present within the same biopsy.
GBM abnormal texture	Presence of basket-weave appearance, electron-lucent areas alternating with granular or curvilinear electron-dense areas, the presence of microspherule, microparticles different from organized deposits or rests of invaginating cells within the lamina densa of the GBM. Scored as normal or abnormal (0=normal, 1=abnormal).
Tubuloreticular inclusions	Presence of at least one subcellular organized inclusion in EC cytoplasm is recorded (0=absent, 1=present).
Glomerular EC fenestration	Absence of typical fenestration resulting in a solid rim of EC cytoplasm away from the perinuclear region (0=0%–5%; 1=≤50%; 2=>50% of the inner GBM surface).
Endothelium honeycombing-like appearance	Presence of cribriform or reticular organization of the EC cytoplasm, most often, but not exclusively, present at the mesangial side of the capillary lumen. The threshold for selecting “present” for honeycombing is: at least two layers of “holes” and at least six holes or more. Scored as present or absent, assuming that the absence of it is the pathologic event (0=present–normal; 1=absent–loss).
Electron densities/hyaline material	Any (at least one) ill-defined, electron-dense material; lack of sharp edges, or large bubbly collection of electron-dense material in the subendothelium/intracapillary (hyaline material) will be recorded as absent or present (0=absent, 1=present).

Cases were randomly assigned across the four study pathologists such that each case was scored at least three times so intra- and inter-reader concordance for any pair of pathologists could be assessed; 25% of cases were scored once by each of three or four pathologists, and the remaining 75% of cases were scored twice by one pathologist and once by a second pathologist. Scoring was implemented in four rounds, each 4–6 weeks apart. The time between two readings of the same case by the same pathologist was at least 4 weeks.

Statistical Analyses

Reproducibility

Inter- and intra-rater reproducibility was assessed for each EM descriptor, overall and by pathologist or pairs of pathologists. The Gwet agreement coefficient (AC) was used to estimate agreement.¹⁴ For descriptors measured by semiquantitative (ordinal) metrics, ordinal weights were used such that scores matching perfectly received a weight of 1; weights decreased until scores were furthest apart, in which case they received a weight of 0. Several sets of sensitivity analyses were conducted: (1) in addition to the Gwet AC, we also used the Fleiss κ statistic to estimate agreement; (2) we applied the Kendall coefficient of concordance to estimate agreement for four podocyte descriptors to compare data from this study to those previously published by this group¹²; (3) we re-estimated agreement after excluding cases marked as having all “poor-quality images;” (4) we stratified agreement estimates by the number of EM images available for each case; and (5) we dichotomized ordinal descriptors to assess whether poor reproducibility was driven by difficulties distinguishing between any two particular scoring options (for example, dichotomizing condensation of cytoskeleton as absent/focal versus diffuse or as absent versus focal/diffuse).

To create a single set of descriptor scores for each patient and to capture a more robust score in case of low reproducibility, scores from multiple pathologists were averaged before subsequent analyses. These averaged values therefore represent a best estimate of the descriptor score for each individual even with potentially low inter-rater agreement. Sensitivity analyses using scores from each individual rater were also conducted to evaluate this approach.

Individual EM Descriptors and Outcome Prediction

Associations between individual EM descriptors and immunosuppressive medication, urine protein-to-creatinine ratio (UPCR), and eGFR at the time of the biopsy were estimated using Pearson correlation coefficients. Associations between individual EM descriptors and clinical outcomes were assessed after adjustment for patient demographics and clinical characteristics, including immunosuppressant use. Three clinical outcomes of interest were included: (1) time from biopsy to a composite disease progression outcome of ≥40% decline in eGFR with eGFR<60 ml/min per 1.32 m² or ESKD, (2) time from biopsy to first complete remission of proteinuria (defined as UPCR<0.3 mg/mg), and (3) eGFR over time from biopsy. For the time-to-event outcomes, Cox proportional hazards models were used. Because of limited numbers of outcome events and to avoid overfitting, variable selection was performed across all EM descriptors, patient demographics, and clinical characteristics using elastic net penalized regression with ten-fold cross-validation and regularization mixing parameter α=0.75. For eGFR over time, we used linear mixed models with random intercepts and random slopes for each patient to account for repeated eGFR measures within each individual. Interactions between each EM descriptor and time were tested to assess whether eGFR slope differed by ultrastructural changes. We conducted sensitivity analyses that included immunosuppressant use after biopsy as a binary time-varying indicator in models as a potential mediating variable.

Clustering and Outcome Prediction

The EM descriptors were used in resampling-based patient consensus clustering to identify subgroups of patients sharing similar combinations of EM features.¹⁵ To choose the number of clusters, we used the proportion of ambiguous clustering (PAC) with an assessment interval of (0.2, 0.8). PAC ranges from 0 to 1, with lower scores indicating more distinct clusters.¹⁶ We then used radar plots to display the cluster profiles where each spoke represents an EM descriptor and the length of the spoke is the average value of that descriptor divided by the possible maximum or range of that descriptor, i.e., scaling all descriptors to the same 0–1 scale. Heat maps were used to facilitate a more direct comparison of each descriptor across clusters, with shading representing the scaled average values on the same 0–1 scale and with means and SDs shown on the original scale. P values from Kruskal–Wallis tests that assess whether an EM descriptor differs significantly across clusters were also shown on the heat map.

Finally, we linked clusters with clinical outcomes in analyses both unadjusted and adjusted for demographics and clinical characteristics, as described above. For the composite progression outcome, only unadjusted analyses were performed, because of a limited number of outcome events and already having five degrees of freedom accounted for by the six cluster groups. Interactions between cluster and time from biopsy in the longitudinal eGFR model were tested.

Consensus clustering was conducted using Matlab version 2018b (MathWorks, Natick, MA) statistical software and penalized regression was conducted using R version 3.4.2 (R Development Core Team, Vienna, Austria). All other statistical analyses were conducted using SAS version 9.4 (SAS Institute Inc., Cary, NC).

Results

Reproducibility

Using the Gwet AC statistic, 12 out of 12 descriptors had good to excellent (≥0.60) intrarater agreement (Figure 3). Good inter-rater agreement was achieved in seven out of 12 descriptors, using the original scoring metrics. Descriptors with inter-rater agreement <0.60 were GBM abnormal texture, condensation of cytoskeleton, thickening of the GBM, microvillous transformation, and loss of EC fenestration. Three of these were ordinal descriptors, so we dichotomized them to assess sources of poor reproducibility. When condensation of cytoskeleton was dichotomized as absent/focal versus diffuse, inter-rater agreement increased from 0.53 to 0.79. Similarly, dichotomizing glomerular EC fenestration increased inter-rater agreement from 0.28 to 0.56. Dichotomizing microvillous transformation did not change agreement estimates (both at 0.48).

Figure 3.

Overall agreement using the Gwet AC statistic showing good to excellent intra-rater agreement and moderate to good inter-rater agreement for most of the descriptors. EM descriptors from left to right were sorted by decreasing inter-rater agreement statistics. The numbers below the graph on the left side of each EM descriptor label are the prevalence of each descriptor, calculated as a percentage for binary descriptors, a percentage of nonzeros for categorical descriptors, and a mean for ordinal descriptors after scaling values to a range of 0–1.

Notably, because three out of 12 descriptors had prevalence <0.10 (tubuloreticular inclusions, podocyte detachment, and loss of primary processes), we cannot adequately assess agreement when lesions were present, although pathologists had excellent agreement when these lesions were absent.

When using the Fleiss κ statistic to estimate agreement in sensitivity analyses, only two out of 12 EM descriptors had good inter- and intra-rater reproducibility. However, as previously demonstrated in earlier studies, the κ statistic assumes that all observations are subject to chance ratings, so when uncertainty in ratings is low such as in this study (0%–3.9% of observations across descriptors), agreement estimates from the Fleiss κ are likely overly conservative.¹⁴ When using the Kendall coefficient of concordance to assess inter-rater reproducibility among four podocyte descriptors, results were similar to a previous study.¹² Results were similar after excluding cases marked as having poor-quality images, with a Pearson correlation coefficient of 0.998 between original agreement estimates and these sensitivity results. We did not observe systematic differences when stratifying analyses by number of EM images available, individual pathologist, or pairs of pathologists.

Individual EM Descriptors Predicting Clinical Outcomes

Patients with microvillous transformation and those without increased thickening or thinning of GBM were more likely to be on immunosuppressive medications at the time of the biopsy (Supplemental Table 1). Patients with more FPE, condensation of cytoskeleton, or microvillous transformation had higher UPCR at the time of the biopsy; patients with loss of primary processes, podocyte detachment, increased thickening of GBM, or electron densities/hyaline material had lower eGFR at biopsy.

Model selection among individual EM descriptors and demographic and clinical characteristics found different predictors for each of the three outcomes (Table 2). Adjusted for Hispanic ethnicity, MCD cohort patients (versus FSGS) and those with more microvillous transformation were more likely to have complete remission of proteinuria (P<0.002 and P=0.02, respectively). Although 95% confidence intervals were wide, there were also large hazard ratios that indicate a trend toward higher likelihood of complete remission for immunosuppressant use at biopsy, patients with more FPE or with EC honeycombing present, or those without podocyte detachment, increased thickening of GBM, or electron densities/hyaline material. Adjusted for black race, disease cohort, and eGFR, patients with absence of EC honeycombing had 4.7 times higher hazards of the composite progression outcome (P<0.001).

Table 2.

Associations between EM descriptors and outcomes, adjusting for demographics and clinical characteristics

Effect	Complete Proteinuria Remission		Composite (≥40% decline in eGFR with eGFR<60 ml/min per 1.73 m2 or ESKD)		Longitudinal eGFRa
	HR (95% CI)	P Value	HR (95% CI)	P Value	Estimate (95% CI)	P Value
Age (per 10 yr increase)					−2.9 (−4.5 to −1.3)	<0.001
Male					−1.8 (−7.0 to 3.3)	0.49
Ethnicity: Hispanic	0.58 (0.36 to 0.93)	0.02			−9.4 (−16.0 to −2.8)	0.005
Race: black			1.65 (0.72 to 3.75)	0.24	−2.4 (−8.3 to 3.6)	0.44
Disease cohort (FSGS versus MCD/MCD-like)	0.52 (0.34 to 0.78)	0.002	4.22 (0.93 to 19.11)	0.07	−4.4 (−10.2 to 1.3)	0.13
Immunosuppressive medication use at biopsy	1.30 (0.86 to 1.95)	0.21			−1.5 (−8.0 to 4.9)	0.64
UPCR at biopsy, mg/mg (categorical form)						0.58
UPCR<0.3 (reference)
UPCR=0.3 to <3.0					−3.8 (−12.5 to 4.8)	0.39
UPCR≥3.0					−4.7 (−13.5 to 4.2)	0.30
eGFR at biopsy (per 10 ml/min per 1.73 m2 increase)			0.84 (0.73 to 0.98)	0.03	5.0 (4.2 to 5.7)	<0.001
Time since biopsy (yr)					−2.5 (−3.6 to −1.4)	<0.001
FPE	1.16 (0.95 to 1.41)	0.14			1.3 (−1.8 to 4.5)	0.40
Loss of primary processes					−1.0 (−16.9 to 14.9)	0.9
Condensation of cytoskeleton						0.43
Absent (reference)
Focal					2.9 (−3.6 to 9.4)	0.38
Diffuse					−2.4 (−13.6 to 8.7)	0.67
Podocyte detachment	0.48 (0.12 to 1.99)	0.31			−3.6 (−19.5 to 12.3)	0.66
Microvillous transformation	1.69 (1.11 to 2.58)	0.02			1.7 (−4.1 to 7.4)	0.57
Increased thickening of GBM	0.75 (0.47 to 1.21)	0.24	0.96 (0.39 to 2.32)	0.92	−6.9 (−13.2 to −0.7)	0.03
Increased thinning of GBM					3.0 (−6.0 to 11.9)	0.52
GBM abnormal texture					−1.1 (−8.8 to 6.5)	0.77
Loss of EC fenestration					−0.5 (−6.3 to 5.3)	0.86
Absence of EC honeycombing	0.74 (0.38 to 1.43)	0.37	4.70 (2.03 to 10.87)	<0.001	−9.0 (−16.7 to −1.3)	0.02
Electron densities/hyaline material	0.73 (0.43 to 1.23)	0.23			−4.6 (−12.2 to 2.9)	0.23

For longitudinal eGFR, no variable selection was done, except for not including tubuloreticular inclusions that did not have enough variability (fewer than five patients endorsed this descriptor). HR, hazard ratio; 95% CI, 95% confidence interval.

^aThere is a significant interaction between loss of EC fenestration and immunosuppressive medication use at biopsy in the longitudinal eGFR model.

Patients with increased thickening of GBM and absence of EC honeycombing had an average of 6.9 and 9.0 lower eGFR during follow-up, respectively, compared with those without increased thickening of GBM and with EC honeycombing present (Table 2). No significant interaction between EM descriptor and time was found with the longitudinal eGFR outcome. Sensitivity analyses that included immunosuppressant use after biopsy showed little difference in results, implying little evidence that the effects of EM descriptors on outcomes are owing to the use of subsequent immunosuppressant treatment. Sensitivity analyses using individual pathologists’ descriptor scores rather than averaging also yielded similar results.

EM Profile Clusters

The optimal number of EM profile clusters was six (PAC=0.13). Each cluster corresponded to a specific EM profile. Cluster 1 included patients with the lowest amount of observable morphologic changes overall (Figure 4), and therefore was named the “fewer changes” cluster. Cluster 2 contained patients with more frequent thinning of the GBM, and was named the “thin GBM” cluster. Cluster 3 was composed of patients with more prominent FPE and microvillous transformation, and was named the “classic MCD” cluster. Like cluster 3, cluster 4 had more prominent FPE, but also had loss of EC fenestration and abnormal GBM texture, and was named the “abnormal GBM texture” cluster. Cluster 5 included patients with the most observed thickening of the GBM and extensive microvillous transformation, and was named the “thick GBM” cluster. Cluster 6 had extensive podocyte and EC damage, reflected by FPE, loss of primary processes, loss of EC fenestration and absent EC honeycombing, and was named the “EC changes” cluster. Electron densities, most likely representing hyaline material, were also more prominent in cluster 6. All EM descriptors were significantly different (P<0.001) across at least two cluster groups, with the exception of tubuloreticular inclusions (Figure 5).

Figure 4.

Radar plots showing how patients clustered into six distinct EM cluster profiles, each having a specific set of EM features. Each spoke represents an EM descriptor and the length of the spoke is the average value for that cluster divided by the possible maximum or range of that descriptor, i.e., scaling all descriptors to the same 0–1 scale.

Figure 5.

Heat map of EM descriptors by cluster profiles showing that all EM descriptors (except tubuloreticular inclusions) are significantly different across at least two cluster groups. Shadings represent the scaled average values on a 0–1 scale. Means and SDs in each cell are on the original scale. For each descriptor, a test for differences in mean values among the clusters is shown, with P values on the basis of the Kruskal–Wallis test. Cluster 1: fewer changes, cluster 2: thin GBM, cluster 3: classic MCD, cluster 4: abnormal GBM texture, cluster 5: thick GBM, and cluster 6: EC changes.

Patients in fewer changes cluster 1 and classic MCD cluster 3 were younger on average, whereas patients in thick GBM cluster 5 and EC changes cluster 6 were oldest. Classic MCD cluster 3 had the highest fraction of pediatric patients (71%), whereas thick GBM cluster 5 and EC changes cluster 6 had the fewest pediatric patients (29% and 23%, respectively). Thin GBM cluster 2 had the fewest black patients, whereas all other clusters were composed of at least 25% black patients. Clusters 1–3 (fewer changes, thin GBM, and classic MCD clusters) all had around 30% Hispanic patients, whereas clusters 4–6 (abnormal GBM texture, thick GBM, and EC changes) contained <20% Hispanic patients. Average eGFR and UPCR were highest in classic MCD cluster 3. Thick GBM cluster 5 had the lowest eGFR, followed by EC changes cluster 6. About 25% of patients in each cluster were on immunosuppressive medications at biopsy, except for classic MCD cluster 3, in which over half were on immunosuppressants (Table 3).

Table 3.

Patient characteristics overall and by clusters

Characteristic	Total (n=242)	Cluster 1 (n=51)	Cluster 2 (n=19)	Cluster 3 (n=86)	Cluster 4 (n=22)	Cluster 5 (n=42)	Cluster 6 (n=22)
Age, yr	26.40 (21.43)	21.73 (16.24)	32.00 (22.28)	19.16 (19.57)	30.41 (27.26)	36.69 (20.62)	37.09 (21.26)
Pediatric patients	52% (125)	53% (27)	47% (9)	71% (61)	50% (11)	29% (12)	23% (5)
Adult patients	48% (117)	47% (24)	53% (10)	29% (25)	50% (11)	71% (30)	77% (17)
Male	54% (131)	61% (31)	37% (7)	53% (46)	41% (9)	64% (27)	50% (11)
Blacka	33% (78)	33% (16)	11% (2)	33% (28)	27% (6)	45% (19)	32% (7)
Hispanic	24% (57)	31% (16)	32% (6)	27% (23)	9% (2)	17% (7)	14% (3)
Disease cohort
MCD/MCD-like	49% (119)	51% (26)	58% (11)	66% (57)	27% (6)	33% (14)	23% (5)
FSGS	51% (123)	49% (25)	42% (8)	34% (29)	73% (16)	67% (28)	77% (17)
eGFR at biopsyb	90.39 (44.42)	93.20 (40.31)	93.01 (34.91)	98.35 (38.41)	90.19 (47.65)	71.58 (36.63)	86.57 (77.83)
UPCR at biopsya	5.32 (7.72)	2.18 (3.01)	5.73 (5.95)	7.98 (9.83)	3.61 (4.62)	5.29 (8.90)	3.97 (3.15)
On immunosuppressive medications at biopsy	34% (83)	25% (13)	16% (3)	53% (46)	27% (6)	24% (10)	23% (5)
Rate of complete remission during study follow-up (no. of events per 100 person-yr of follow-up)c	37.57	22.08	29.49	78.32	30.50	28.80	13.75
Rate of ≥40% decline in eGFR with eGFR<60 ml/min per 1.73 m2 or ESKD during study follow-up (no. of events per 100 person-yr of follow-up)d	3.99	5.38	7.09	0.85	2.77	6.70	9.58

Cluster 1: fewer changes, cluster 2: thin GBM, cluster 3: classic MCD, cluster 4: abnormal GBM texture, cluster 5: thick GBM, cluster 6: EC changes.

^aMissing <5%.

^bMissing <10%.

^cn=197 (n=25 had complete remission at biopsy; n=11 missing UPCR at biopsy; n=19 had no follow-up UPCR).

^dn=209 (n=24 missing eGFR at biopsy; n=9 had no follow-up eGFR).

Clusters Predicting Clinical Outcomes

The EM clusters corresponded to different clinical profiles. Patients in EC changes cluster 6 had the highest rates of the composite progression outcome of ≥40% decline in eGFR with eGFR <60 ml/min per 1.73 m² or ESKD and the lowest rates of complete proteinuria remission (Figure 6, Table 3). Patients in classic MCD cluster 3 had the highest rates of complete proteinuria remission. Other clusters appeared to have similar outcomes in unadjusted analyses. Although adjusted analyses were underpowered for time-to-event outcomes, results were largely consistent with unadjusted results (Supplemental Table 2). Similarly, patients in EC changes cluster 6 had the lowest eGFR during follow-up, followed by thick GBM cluster 5, and classic MCD cluster 3 had the highest eGFR during follow-up from adjusted longitudinal models (Table 4). Interactions between cluster and time from biopsy in both unadjusted and adjusted models were not statistically significant, implying little evidence that mean eGFR slope differed across clusters.

Figure 6.

Kaplan–Meier curves showing the probability of not reaching (surviving) a ≥40% decline in eGFR with the event-time eGFR<60 ml/min per 1.73 m² or ESKD (left) and the cumulative probability of complete proteinuria remission (right) by clusters. Cluster 1: fewer changes, cluster 2: thin GBM, cluster 3: classic MCD, cluster 4: abnormal GBM texture, cluster 5: thick GBM, and cluster 6: EC changes.

Table 4.

Associations between clusters and longitudinal eGFR from linear mixed effects models

Clusters	Longitudinal eGFR
	Model 1 (unadjusted)		Model 2 (adjusted)
	Estimate (95% CI)	P Value	Estimate (95% CI)	P Value
Clusters		<0.001		0.05
1: Fewer changes (reference)
2: Thin GBM	−1.0 (−20.0 to 18.1)	0.92	2.4 (−9.7 to 14.6)	0.7
3: Classic MCD	7.7 (−4.6 to 19.9)	0.22	4.6 (−3.4 to 12.5)	0.26
4: Abnormal GBM texture	−5.1 (−22.5 to 12.3)	0.56	−2.6 (−13.9 to 8.7)	0.65
5: Thick GBM	−21.7 (−36.4 to −7.0)	0.004	−5.0 (−14.8 to 4.7)	0.31
6: EC changes	−16.1 (−34.7 to 2.4)	0.09	−13.1 (−25.3 to −0.9)	0.04

Both models included time since biopsy. Model 2 adjusted for disease cohort (FSGS versus MCD/MCD-like), patient age, sex, black race, Hispanic ethnicity, eGFR and UPCR at biopsy, and immunosuppressant use at biopsy.

Discussion

This study addresses some critical issues in nephropathology, including standardization and reproducibility of morphologic observations, the association of individual morphologic features with outcome, and the association of combinations of morphologic features with outcome. These results indicate that comprehensive morphologic profiles should include EM features, along with light microscopy findings, to stratify patients with proteinuria into homogeneous categories that can help elucidate underlying mechanisms of disease.

The robustness and utility of any scoring system relies on comprehensiveness of the parameters used, standardization of the processes, reproducibility of data, and their clinical relevance.⁶ In this study, we performed a meticulous ultrastructural morphologic analysis of MCD, MCD-like, and FSGS renal biopsies from the NEPTUNE cohort,¹³ and tested agreement of observations before correlation with other clinical and demographic parameters. A critical step for reproducible morphologic assessment is a consensus-based establishment of detailed and unambiguous criteria in a reference manual and crosstraining of study pathologists.¹² To have better control of this step, the early phases of this study were dedicated to reviewing and refining the NDPSS definitions, testing them via pilot studies, and conducting webinar-based training. This approach was previously used by our group and proven to increase reproducibility.^12,14 Standardization of preanalytic, analytic, and postanalytic steps, including documentation of the lesions and reporting of ultrastructural features, has some challenges. For example, there is an inherent limitation in the interpretation of EM related to the small size of the material available, the focal and segmental distribution of some morphologic changes, and the image collection and representation of the ultrastructural features, in most cases performed by trained technicians. Additionally, there is often an over-representation of injured portions of the glomerulus, which could bias data analysis. In 2004, the Renal Pathology Society recommendations for EM included the ultrastructural examination of one or two glomeruli, with the collection of low-, medium-, and high-magnification images of both capillary loops and mesangial areas.¹⁷ In the setting of podocytopathies, this allows for a more objective representation of the extent of damage. The lack of standardization of imaging may be problematic in the setting of multicenter studies, calling for the implementation of more rigorous control of the preanalytics and analytics, which still remain under the control of individual laboratories.

Current recommendations for glomerular EM analysis and reporting (postanalytic phase) of native kidney biopsies are restricted to a limited set of features, including the extent of podocyte FPE, the absence/presence and location of electron dense deposits, the GBM thickness and appearance, and the presence of endothelial tubuloreticular inclusions.¹⁷ The assessment of these parameters, however, is often subjective, partly because of unclear consensus definitions, and therefore is poorly reproducible. Although current recommendations are useful for conventional categorization of MCD or FSGS, our results suggest that reporting of additional ultrastructural changes might have important clinical implications. Still, reproducibility of observations needs to be considered before analysis. To overcome the suboptimal reproducibility in some of the descriptors, scores from multiple pathologists were averaged before use for clustering and for outcome prediction analyses. Our study results are therefore on the basis of best estimates of descriptor scores across multiple pathologists. We also discovered some reasons for disagreement by dichotomizing ordinal descriptors, i.e., that there was more disagreement between absence and focal amounts of condensation of cytoskeleton and loss EC fenestration. Although additional training has been shown to improve reproducibility, the human limitations to replicate data can only be overcome by automated assessment.

In the setting of MCD/FSGS, investigators’ and clinicians’ attention has been historically focused on podocytes as major contributors in the pathogenesis of nephrotic syndrome. The ultrastructural alterations of podocytes described in nephrotic syndrome include FPE, microvillous transformation, and podocyte vacuolization and detachment from the GBM.^18–21 Podocytes respond to injury in part by rearrangement of their cytoskeleton, typically altering their morphology. However, most studies evaluating the status of podocytes in FSGS focused on FPE, with discordant results regarding the clinical value of this finding.^5,22,23 Other studies have evaluated podocyte abnormalities using morphometric methods with the intent of providing objective measurements and increase accuracy.^3,22,23 The amount of effacement or foot process width was also used to discriminate between FSGS etiologies. We performed an extensive ultrastructural assessment of podocytes and demonstrated that additional podocyte parameters associate with increased protein excretion (microvillous transformation and potentially higher extent of FPE and condensation of cytoskeleton) and lower eGFR (loss of primary processes and podocyte detachment) at the time of the biopsy, and predicted complete remission (higher extent of FPE and microvillous transformation) or the lack of remission (podocyte detachment) during follow-up. These findings clearly support the value of a more rigorous and extensive ultrastructural evaluation of podocytes in clinical practice. Our data, however, also indicate that the status of podocytes is not the only determining factor in predicting outcomes.

Glomerular ECs appear to contribute significantly to the physiologic characteristics of the glomerular filtration barrier,²⁴ and there is a growing body suggesting that the glomerular endothelium contributes to the pathogenesis and progression of glomerular diseases.^25–32 The role of the EC in the development and maintenance of the filtration barrier is in part attributed to a tightly regulated podocyte–to–EC crosstalk through paracrine signals.³² The development of EC injury and thrombotic microangiopathy in patients receiving anti-vascular endothelial growth factor (VEGF) therapy has been linked to impaired VEGF signaling. In the past decade, animal models have shown that the disruption of the fine balance between the podocyte-derived VEGF and antiangiogenic factors, such as VEGF receptor-1, leads to endothelial injury and to a clinical syndrome similar to preeclampsia.^33–35 Furthermore, the contribution of endothelial dysfunction and injury in the development of diabetic nephropathy has been demonstrated in patients with both type 1 and type 2 diabetes through the measurement of different vascular factors, including vWf.^36–39 Ultrastructural studies of diabetic nephropathy have shown a reduction in glomerular capillary fenestration of EC.^30,31 Loss of fenestration, FPE, and subendothelial widening was also reported in lupus nephritis classes 3 and 4 with nephrotic syndrome, suggesting a causal relationship between podocyte and EC injury.⁴⁰ By contrast, little information exists on the role and morphologic alterations of the EC in FSGS and MCD.^23,41,42 In a recent study, Daehn et al.⁴³ demonstrated reciprocal paracrine signaling between podocytes and ECs leading to the development of podocyte loss and segmental glomerulosclerosis. After adriamycin-induced glomerulosclerosis, transgenic mice with podocyte-specific expression of a constitutively active TGFβ receptor type-1 showed endothelin-1 release by podocytes, which mediated mitochondrial oxidative stress and dysfunction in adjacent ECs. In turn, EC dysfunction promoted podocyte apoptosis, podocyte loss, and development of segmental sclerosis.

Our results extend observations of previous studies showing that glomerular ECs are morphologically altered in some patients with glomerular disease. Moreover, our data indicate that EC injury is associated with poor clinical outcome. Patients with absent EC honeycombing were less likely to achieve complete remission in proteinuria and had a lower eGFR during follow-up. By demonstrating an association between morphologic changes in glomerular ECs and clinical outcomes, our data support the hypothesis that the glomerular endothelium participates in the pathophysiology of FSGS and MCD.

Several associations observed with individual EM descriptors were maintained in the identified clusters: in the EC changes cluster 6, where endothelial ultrastructural changes were most severe, patients had the highest risk of progressive loss of kidney function, and the lowest rate of complete proteinuria remission relative to those included in other clusters. This finding again highlights the important role of the glomerular endothelium in predicting clinical outcomes.

Our study highlighted which specific ultrastructural changes may associate with different clinical outcomes (remission and progression). It is intuitive that individual structural changes associated with remission may reflect reversible/transient alterations, whereas those associated with worsening of the disease may reflect a more permanent/irreversible structural damage. For example, FPE and microvillous transformation were individually associated with remission. Similarly, patients in the classic MCD cluster 3, characterized mainly by FPE and microvillous transformation, were more likely to achieve remission and had the highest eGFR at follow-up. These findings suggest that both FPE and microvillous transformation are associated with reversible biochemical responses and structural reorganization of the cytoskeleton. By contrast, the cluster with more loss of podocyte primary processes, reflecting a more severe change in cell phenotype,⁴⁴ had the highest rate of progression. Interestingly, our findings suggest GBM thickening as an irreversible ultrastructural alteration with prognostic significance. GBM thickening alone was associated with poorer outcomes (lower rate of proteinuria remission and lower eGFR during follow-up). Similarly, patients with EM profiles characterized by structural abnormalities of the GBM (abnormal GBM texture cluster 4, thick GBM cluster 5, and EC changes cluster 6) also had poorer outcomes (although not statistically significantly different from the fewer changes cluster 1). The concept of reversible and irreversible structural damage is not novel. The relationship between reversible and irreversible podocyte structural changes and progression has been elucidated in animal models of FSGS: FPE and microvillous transformation were described in the early/reversible phases, and when the disease progressed toward the development of segmental sclerosis, there was an incremental percentage of podocyte detachment and death.^23,45–47 GBM thickening is a well recognized early feature of diabetic nephropathy and its presence has been associated with the progression to ESKD.⁴⁸ Studies on diabetic nephropathy have also elucidated how the disruption of the podocyte-GBM interface and the structural abnormalities of podocytes are critical in the development of GBM thickening.^49,50 Further investigations are needed to evaluate the role, mechanisms, and impact of GBM thickening on disease progression of FSGS and MCD.

Our study has a few limitations worth noting. First, the relatively small number of patients included in this study may have reduced our statistical power to identify other EM features significantly associated with renal outcomes. Small sample size also limited our ability to conduct extensive subgroup analyses or adjust for more granular patient characteristics, e.g., specific types of immunosuppression or all individual EM descriptors. Consequently, our findings require validation in larger independent cohorts to determine if our findings are incrementally more predictive than clinical or laboratory data alone. Second, because of the intermediate inter-rater agreement for some descriptors, including microvillous transformation, the associations with outcomes should be interpreted carefully. Although averaging descriptor scores across multiple raters for outcome analyses helped to stabilize the variable values, multiple raters may not always be available. Therefore, further investigations into the reasons for disagreement are needed to improve pathologists’ agreement or to inform automated image analysis algorithms. Finally, all of the EM descriptors used in this study were captured on a qualitative/binary scale or semiquantitative/ordinal scale rather than a quantitative/continuous scale. Although this was done on the basis of pathologists’ training and routine operation, and a previous study showed little difference in reproducibility when using different scales for morphologic assessment, a continuous scale could provide more granularity and power for outcome analyses.

In contrast with current approaches that stratify patients on the basis of the amount of FPE,^5,51 the detailed assessment of EM features using the NDPSS has highlighted the value of podocyte parameters other than effacement, and of other components of the glomerular filtration barrier for clinical outcome prediction. These parameters were relevant to predict outcome and/or proteinuria remission not only individually, but also when EM descriptors were used to stratify patients into clusters. These observations provide new insights into the contributions of a detailed ultrastructural analysis and call for a revision of current approaches and the implementation of better standardized analytics. Imminent studies will focus on correlating these observations with light microscopy findings to establish clinically relevant patient categories, and with molecular mechanisms to better understand the role of structural changes in predicting response to novel, targeted therapies.

Disclosures

Dr. Barisoni reports personal fees from Moderna, personal fees from Protalix, personal fees from Sangamo, personal fees from Vertex, outside the submitted work. Dr. G. Liu reports grants from National Institutes of Health/National Institute of Diabetes, Digestive, and Kidney Diseases (NIH/NIDDK), during the conduct of the study. Dr. Q. Liu reports grants from NIH/NIDDK, during the conduct of the study. Dr. Mariani reports grants from NIH/NIDDK, during the conduct of the study; personal fees from Reata Pharmaceuticals, outside the submitted work. Dr. Zee reports grants from NIH/NIDDK, during the conduct of the study.

Funding

This work was supported by the National Institute of Diabetes, Digestive, and Kidney Diseases through grant R01-DK-118431. The Nephrotic Syndrome Study Network Consortium is a part of the National Institutes of Health Rare Disease Clinical Research Network, supported through a collaboration between the Office of Rare Diseases Research, National Center for Advancing Translational Sciences, and the National Institute of Diabetes, Digestive, and Kidney Diseases (grant U54-DK-083912). Additional funding and/or programmatic support for this project has also been provided by the University of Michigan, the NephCure Kidney International, and the Halpin Foundation.

Supplemental Material

This article contains the following supplemental material online at http://jasn.asnjournals.org/lookup/suppl/doi:10.1681/ASN.2019080825/-/DCSupplemental.

Supplemental Table 1. Correlation coefficients between immunosuppressive medication use, UPCR, eGFR at biopsy, and EM descriptors. Correlation coefficients with P values <0.05 are bolded.

Supplemental Table 2. Associations between clusters and clinical outcomes (≥40% decline in eGFR with eGFR<60 ml/min per 1.73 m² or ESKD and complete proteinuria remission) from Cox models.