Molecular Correlates of Scarring in Kidney Transplants: The Emergence of Mast Cell Transcripts

Abstract

In the Banff consensus, infiltrates in areas of scarring are ignored. This study aimed to characterize the molecular correlates and clinical significance of scarring and inflammation in scarred areas.

We assessed the extent of interstitial infiltrates, tubulitis and scarring in 129 clinically indicated renal allograft biopsies, and correlated the results with microarray expression data and allograft survival. Findings were validated in 50 additional biopsies.

Transplants with scarring had a worse prognosis if the scarred area showed infiltrates. Infiltration in unscarred and scarred areas was associated with reduced death censored graft survival. In microarray analysis, infiltration in unscarred areas strongly (>r ± 0.4) correlated with 484 transcripts associated with cytotoxic T cells, interferon-gamma, macrophages and injury. Scarring correlated with a distinct set of 172 transcripts associated with B cells, plasma cells, and others of unknown significance. The strongest correlation was with four mast cell transcripts. In biopsies with scarring, high expression of mast cell transcripts was associated with reduced graft survival and poor functional recovery.

In renal allograft biopsies, infiltrates in scarred areas have implications for poor outcomes. Scarring is associated with a distinct pattern of inflammatory molecules, including B cell/immunoglobulin but particularly mast cell-associated transcripts, which correlated with poor outcomes.

Abbreviations: 

ABMR

antibody-mediated rejection

BATs

B-cell-associated transcripts

CATs

cytotoxic T–cell-associated transcripts

GRITs

interferon-γ-dependent rejection-induced transcripts

IFNG

interferon-γ

IFTA

interstitial fibrosis and tubular atrophy

i-IFTA

inflammation in areas with interstitial fibrosis and tubular atrophy

i-Banff

inflammation in areas without interstitial fibrosis and tubular atrophy

IGTs

immunoglobulin transcripts

IRITs

injury and repair induced transcripts

MACATs

mast cell- associated transcripts

MATs

macrophage-associated transcripts

KTs

kidney transcripts, decreased in rejection

PBT

pathogenesis-based transcript set

TCMR

T-cell-mediated rejection

Introduction

Assessment of biopsies from troubled kidney transplants is crucial for mechanistic insight, patient management and clinical trials. Infiltration by mononuclear cells has been recognized as a central histopathological feature of rejection from the first kidney allografts performed (1). According to the current Banff consensus for classification of transplant histopathology (2–4), assessment of interstitial mononuclear infiltration in renal allograft biopsies should include only infiltration in cortical areas without scarring. Infiltration in more than 25% of unscarred cortex accompanied by inflammatory cells in tubules (tubulitis) is the basis of the diagnosis of T-cell-mediated rejection (TCMR). With current immunosuppression, histopathology of renal allograft biopsies has changed: TCMR is less common whereas scarring—interstitial fibrosis and tubular atrophy (IFTA)—is more frequent (5). Scarring is the histological correlate of chronic functional deterioration of renal allografts, as it is in native kidneys (6–9). Areas of scarring are often infiltrated by mononuclear cells, which might have a role in the pathogenesis (reviewed in (10)). But in renal allografts, inflammation in areas of scarring is considered nonspecific and ignored in terms of rejection diagnosis and therapy, as are perivascular and nodular infiltrates (2). In protocol biopsies, the simultaneous presence of scarring and inflammation in unscarred areas is associated with an inferior prognosis compared to those renal allografts with isolated scarring or isolated infiltrates in unscarred cortex, respectively (11–13). However, protocol biopsy studies have not specifically addressed the role of infiltrates in scarring areas and other types of ‘nonspecific’ infiltrates. Furthermore, the significance of infiltrates in areas of scarring in clinically indicated biopsies is unknown.

This study aimed to characterize the prognostic significance and molecular correlates of scarring and of inflammation in scarred areas in kidney transplant biopsies for clinical indications. We assessed the extent of infiltrates in unscarred (i-Banff) and scarred (i-IFTA) cortex, as well as the extent of tubulitis, scarring and of nodular and perivascular infiltrates. Histology findings were correlated with microarray gene expression data and allograft survival.

Patients and Methods

Biopsies and histopathological scoring

From 104 patients, 129 clinically indicated renal allograft biopsies (the training set), obtained between January 2004 and October 2006 from consenting patients were included in this study (approval by the University of Alberta Health Research Ethics Board, Issue 5299). Biopsies were taken between 1 week and 20 years posttransplant (median 19 months). Histopathological reevaluation of the 129 biopsies was done by one observer (MM). All samples fulfilled the minimal criteria for adequacy and were stained (including C4d on frozen sections) and scored according to the current Banff classification (2,4,14). In addition, the extent of four different types of interstitial infiltrates was semiquantitatively assessed in renal cortex: (1) infiltrates in unscarred areas (i-Banff); (2) infiltrates in areas of scarring (i-IFTA); (3) nodular infiltrates and (4) perivascular infiltrates (15). Analogously, the extent of scarring was assessed independent of whether it was inflamed or not. In contrast to the current Banff rules (2,4), these five histological features were assessed as absolute percentages related to the whole cortex as 100% (Figure 1), to make the numbers for their extent comparable.

Figure 1. Scoring inflammation in renal allograft biopsies. This cartoon illustrates the histological assessment of inflammation in renal allografts. First the extent of the two different morphological compartments, unscarred and scarred was assessed as absolute percentage in terms of 100% cortex present. Analogously, the extent of infiltrates was semiquantitatively assessed as absolute percentages, whereas infiltrates in unscarred and scarred areas were counted separately.

For validation purposes, another 50 biopsies for cause from 50 different patients obtained after the above-described 129 biopsies (between November 2006 and August 2007), were analyzed. This validation set included 17 biopsies from a different centre (University of Illinois at Chicago; approval by the University of Illinois Institutional Review Board, Protocol 2006–0544).

Immunohistochemistry

The following antihuman antibodies were applied to paraffin sections: anti-CD3 (polyclonal, DAKO), anti-CD68 (clone PGM1, DAKO), anti-CD20 (clone L26, DAKO), anti-CD138 (clone MI15, DAKO) and anti-mast cell tryptase (clone AA1, DAKO). Stains were done on a Ventana Benchmark™ automated stainer. For each marker, the percentage of stained cells relative to all inflammatory cells was semiquantitatively assessed for the i-Banff and the i-IFTA compartment.

Microarray experiments

One additional 18-gauge biopsy core was collected for gene expression analysis. The tissue was placed immediately in RNALater and stored at −20°C. RNA extraction, labeling and hybridization to the HG_U133_Plus_2.0 GeneChip (Affymetrix, Santa Clara, CA) were carried out according to protocols published at http://www.affymetrix.com. Microarrays were scanned using GeneArray Scanner (Affymetrix) and processed with GeneChip Operating Software Version 1.4.0 (Affymetrix). Microarray data were preprocessed by robust multi-array analysis (RMA), implemented in Bioconductor version 2.2., and fold changes were calculated relative to native kidney samples taken from unaffected areas of the cortex of eight tumor nephrectomies.

Pathogenesis-based transcript sets

We recently developed a system of summarizing large-scale genome wide expression data as pathogenesis-based transcript sets (PBTs). Microarray gene expression results obtained from tissue samples are collapsed into PBT scores as the geometric mean of fold changes across all probe sets within each PBT. Recently, we were able to show the utility of these gene sets for diagnosing rejection in renal transplant biopsies (16). PBTs reflect the major biological processes in allografts: cytotoxic T-cell-associated transcripts (CATs (17,18)), interferon-γ-dependent rejection-induced transcripts (GRITs (18,19)), kidney transcripts with decreased expression during rejection (KTs (20)), injury and repair-induced transcripts (IRITs (21)), immunoglobulin transcripts (IGTs (22)) and B-cell-associated transcripts (BATs (22)). Probesets in each previously published PBT are available at http://transplants.med.ualberta.ca/. The PBT annotation of a probeset thus acts as a rapid way of understanding the biological process represented by changes in that probeset.

Statistical analysis and clinical follow-up

Spearman correlations were used to assess the relationship between the different histological features and gene expression values. Means were compared using a 2-sample t-test. For allograft outcome analysis, patients in the training set were followed after biopsy for a mean of 24.6 ± 8.7 months (range = 3–41) and either death-censored or censored for end of follow-up. Patients in the validation set were followed for 11.7 ± 3.2 months (range = 1.3–16.3) after biopsy. An event was defined as either graft loss or persistent low eGFR (estimated Glomerular Filtration Rate by the Cockcroft–Gault formula (23)) defined as at least 3 months of eGFR <30 mL/min (time to event = time to end of 3 months of low eGFR). Survival analysis was performed on the last biopsy from each patient using Kaplan Meier analysis with log-rank testing (SPSS 15.0 software; SPSS Inc., Chicago, IL).

Results

Histological features and time after transplantation

The 104 patients providing 129 biopsies included in this study are part of a previously published cohort (for demographics see (16), cases with BK nephropathy (n = 6), inadequate histology (n = 6) and missing slides (n = 2) were excluded)). By Banff criteria (4) there were 19 TCMR; 14 C4d positive antibody-mediated rejection (ABMR) (2 acute, 12 chronic-active ABMR); 20 borderline changes; 76 without histologic criteria for rejection (signs of calcineurin inhibitor toxicity n = 12, C4d-negative transplant glomerulopathy n = 10, acute tubular injury n = 14, recurrent/de novo glomerulonephritis n = 18, IFTA n = 13, no rejection or other diagnosis n = 8, suspicious for posttransplant lymphoproliferative disease n = 1). Within the validation set (n = 50), histopathological diagnosis according to Banff were: 8 TCMR, 3 ABMR (1 acute, 2 chronic-active), 10 borderline and 29 without histologic criteria for rejection (C4d-negative transplant glomerulopathy n = 3, acute tubular injury n = 7, recurrent/de novo glomerulonephritis n = 6, IFTA n = 7, no rejection or other diagnosis n = 6). Thirty-six of the biopsies were followed by a therapeutic intervention (e.g. steroid boli, ATG, IvIg, plasmapheresis, switch in basal immunosuppression) depending on the original histopathological diagnosis and its clinical interpretation by the physician responsible.

In the training set (n = 129), the extent of infiltrates in scarred areas (i-IFTA) correlated with the extent of scarring (r = 0.911, p < 0.0001). In biopsies with scarring, 0–100% of the scarred area showed infiltrates (mean 49.5%) (Figure 2A). The extent of scarring (r = 0.582, p < 0.0001) as well as of i-IFTA (r = 0.554, p < 0.0001) correlated with the time after transplantation and was greater in biopsies taken more than 6 months after transplantation (Figure 2B). In biopsies after 6 months, the extent of i-IFTA still correlated with time posttransplant (r = 0.25, p = 0.021).

Figure 2. (A and B) Time-dependent change of infiltrates in biopsies for cause. Ordering all 129 biopsies for cause according to time posttransplantation (A) illustrates how the early domination of i-Banff is giving way to i-IFTA as the predominant histological finding with advanced time posttransplantation. Comparing early (<6 months post-TX) versus late (>6 months post TX) allograft biopsies (B) shows a significant increase of scarring, i-IFTA and nodular infiltrates with time, while the unscarred compartment decreases but with a stable inflammatory burden.

Infiltrates in unscarred areas (i-Banff) did not correlate with the extent of scarring or i-IFTA, but correlated strongly with the degree of tubulitis (r = 0.85, p < 0.0001). The extent of infiltrates in the unscarred compartment did not significantly change over time, but the unscarred area was smaller in biopsies taken >6 months posttransplantation as scarring increased.

Nodular infiltrates and perivascular infiltrates were present in both unscarred and scarred areas, and were quantitatively minor contributors to inflammation. Nodular infiltrates increased with time while perivascular infiltrates did not.

Inflammation and scarring in biopsies for cause and allograft survival

Since infiltrates in scarred areas were necessarily dependent on the extent of scarring, we studied whether inflammatory infiltrates in scarred areas give additional information compared to scarring alone. We selected the subset of allograft biopsies with little infiltrate in unscarred areas (i-Banff <25%) (n = 77) that also showed at least grade I IFTA according to Banff (i.e. ≥ci1/ct1) (2). We then split these into two groups: biopsies with ≥50% (n = 46) and biopsies with <50% (n = 31) of the scarred area showing infiltrates. Kidneys with increased infiltrates in scarred areas had a worse prognosis than those with minimal infiltrates (Figure 3A: 93.5% vs. 69.6% survival, p = 0.02).

Figure 3. (A and B) IFTA and inflammation in biopsies for cause and allograft survival. Kaplan–Meier curves show that allografts with scarring lacking considerable inflammation in this compartment have better outcome than those with extensively inflamed scarring (A). Figure 3B shows that inflammation in either cortical compartment (unscarred areas and scarring) above the current Banff threshold for rejection (i.e. >25%) is associated with an inferior prognosis compared to allografts with infiltrates below this threshold. Events are defined as either allograft loss with return to dialysis or persistent (>3 months) low (<30 mL/min) eGFR. Given is the time between the last biopsy of a patient and an event.

We compared outcome of allografts with infiltrates in unscarred areas (i.e. above the current Banff threshold for rejection, > 25% of cortex involved) with those showing predominantly infiltrates in scarred areas (i-IFTA) (Figure 3B). Without considering therapeutic intervention as a variable, allografts with >25% of the cortex showing infiltrates either in the unscarred areas (69%, p = 0.05), or in scarred areas (60%, p = 0.002) had worse survival than those with minimal infiltrates (less than 25% in both compartments (89%)).

Thus, infiltrates in scarred areas have prognostic relevance, whether compared to grafts with minimal infiltrates or to grafts with uninflamed scarring.

Transcripts correlated with scarring and with infiltrates in unscarred areas are mutually exclusive

We examined the correlation between gene expression and histological features using 54 676 probesets on the HG_U133_Plus_2.0 GeneChip. We set a threshold correlation coefficient of >r ± 0.4 and a p-value of <0.001 for a probeset to be considered strongly correlated with a histological feature. This approach identified 484 probesets correlated with i-Banff, 249 with Banff t-score, 172 with scarring, 202 with infiltrates in scarring, 34 with nodular infiltrates and none with perivascular infiltrates. Remarkably, the 484 transcripts that correlated with i-Banff and the 249 that correlated with tubulitis showed no overlap with the 172 transcripts correlated with scarring or the 202 transcripts that correlated with infiltrates in scarring (Figure 4A).

Figure 4. (A and B) Correlations between individual genes and histological lesions and their overlap. Considerable overlap was seen between i-Banff and the Banff t-score and between i-IFTA and scarring. None of the 493 transcripts correlating with tubulitis or i-Banff overlapped the 242 transcripts correlating with scarring or i-IFTA (A). At an arbitrary threshold (r > 0.4 and p < 0.001) 484 genes were associated with i-Banff, 249 with Banff t-score, 202 with i-IFTA, 172 with scarring, 34 with nodular infiltrates and none with perivascular infiltrates (B): i-Banff showed the largest enrichment of cytotoxic T cell-associated transcripts (35%) and interferon-γ-dependent rejection-induced transcripts (14%), followed by macrophage-associated transcripts (19%) and injury and repair-induced transcripts (10%). Only 18% of the correlated transcripts were not annotated as PBTs. For i-IFTA and scarring, the majority of associated transcripts was not annotated as PBTs. Most of the annotated transcripts were B cell or immunoglobulin-associated transcripts (23% of probesets for i-IFTA and 16% for scarring). As a kind of positive control nodular infiltrates showed the strongest association with B cell/immunoglobulin-associated transcripts.

The PBT annotations of the probesets correlating with the extent of each histological feature are indicated in Figure 4B. The extent of i-Banff correlated mostly with cytotoxic T-cell-associated (35%) and interferon-γ-dependent rejection-induced transcripts (14%), followed by macrophage-associated (19%) and injury and repair-induced transcripts (10%). Only 18% of the correlated transcripts were not annotated as PBTs. No B cell/plasma cell transcripts correlated with i-Banff and/or tubulitis. Considerable overlap (Figure 4A) of probesets was seen between i-Banff and tubulitis: 240 probesets simultaneously correlated with the extent of both lesions.

There was also considerable overlap between the transcripts correlating with scarring and i-IFTA: 132 probesets were correlated with both features (Figure 4A). Most annotated transcripts were B cell-associated or immunoglobulin transcripts (16% of probesets for scarring and 23% for i-IFTA are annotated as B cell or immunoglobulin-associated transcripts). Besides B cell and plasma cell-associated transcripts a distinct subset of injury and repair-associated transcripts correlated with the extent of scarring and infiltrates in scarred areas: the so called ‘intermediate’ injury and repair-associated transcripts (21), comprising predominately transcripts related to embryogenesis and organ development. Decreased expression of renal parenchymal transcripts, predominantly representing solute carriers and metabolic transcripts in the tubular epithelium, was also correlated with scarring. More than 50% of the scarring and/or i-IFTA-associated probesets were not annotated as PBTs (70% for scarring and 59% for i-IFTA, details see below).

Interferon-γ-induced transcripts did not correlate with scarring or with inflammation in scarring. Only a few transcripts previously annotated by our group as T-cell-associated (CDCA8, TMEM71, KLRB1, BTBD11, P2RY5, NR3C1) or macrophage-associated (ANTXR1, TK1, UBE2T) correlated with scarring.

B-cell-associated or immunoglobulin transcripts accounted for 53% of the probesets correlated with nodular infiltrates. The gene most highly correlated with nodular infiltrates was CD20 (r = 0.53). The second largest group of correlated probesets was annotated as cytotoxic T-cell-associated transcripts (24%), while 15% had no PBT annotation.

Non-PBT annotated transcripts correlating with the extent of scarring and with i-IFTA

We eliminated probesets not identified as genes by Affymetrix, or as PBTs. In cases with multiple probesets representing the same gene, we retained only the most highly correlated probeset from both overlapping lists. Table 1 shows the 25 genes most strongly correlated with the extent of scarring and with i-IFTA. Four of the top six genes code for transcripts associated with mast cells: carboxypeptidase A3, mast cell tryptase beta 2, tryptase alpha/beta1 and Fc IgE receptor alpha.

Table 1.  Top 25 non-PBT annotated genes correlated with scarring and i-IFTA

Affymetrix ID	Gene symbol	Gene name	Correlation* r-Value
*Spearman correlation with extent of fibrosis/atrophy or i-IFTA, p < 0.001.
205624_at	CPA3	Mast cell carboxypeptidase A3	0.619
207134_x_at	TPSB2	Mast cell tryptase beta 2	0.566
204719_at	ABCA8	ABC transporters—ATP-binding cassette, subfamily A (ABC1), member 8	0.564
205683_x_at	TPSAB1	Mast cell tryptase alpha/beta 1	0.555
205044_at	GABRP	Gamma-aminobutyric acid (GABA) A receptor	0.554
211734_s_at	FCER1A	Receptor for Fc fragment of IgE, high affinity I, expressed predominantly on mast cells	0.547
209173_at	AGR2	Anterior gradient homolog 2	0.523
229461_x_at	NEGR1	Neuronal growth regulator 1	0.509
213974_at	ADAMTSL3	ADAMTS-like 3	0.503
221933_at	NLGN4X	Neuroligin 4	0.499
228310_at	ENAH	Cytoskeleton regulatory protein hMena	0.497
202508_s_at	SNAP25	Synaptosomal-associated protein	0.497
226435_at	PAPLN	Papilin	0.486
228241_at	BCMP11	AGR3 = anterior gradient homolog 3	0.485
219552_at	SVEP1	Sushi, von Willebrand factor type A, EGF and pentraxin domain containing 1	0.484
227088_at	PDE5A	Phosphodiesterase 5A	0.479
219778_at	ZFPM2	Zinc finger protein, multitype 2	0.474
207496_at	MS4A2	Fc fragment of IgE, high affinity I, receptor for; beta polypeptide	0.47
202992_at	C7	Complement component 7	0.468
1558714_at	ROB01	Roundabout, axon guidance receptor, homolog 1	0.467
219867_at	CHODL	chondrolectin	0.466
206336_at	CXCL6	Chemokine (C-X-C motif) ligand 6 (granulocyte chemotactic protein 2)	0.466
210258_at	RGS13	Regulator of G-protein signaling 13	0.462
201489_at	PPIF	Peptidylprolyl isomerase F (cyclophilin F, CYP3; Cyp-D)	−0.47
208321_s_at	CABP1	Calcium binding protein 1	−0.49

Contrary to expectations, typical fibrogenesis-associated transcripts (e.g. TGFβ- or collagen-related genes) did not correlate with the extent of scarring or i-IFTA above the arbitrary threshold, with the exception of the probeset for TGFB2 (r = 0.404).

Confirmation by immunohistochemistry

Since different tissue cores were analyzed by histopathology and microarray, we stained a subset of 33 biopsies representing the spectrum of histological features and with paraffin embedded material available to confirm the above described correlations between gene expression and histopathology. The percentage of CD20+ B cells was greater in i-IFTA (8.1 ± 7.4% vs. 3.2 ± 3.8%, p = 0.006). The percentage of CD138+ positive interstitial cells (plasma cells) was greater in the i-IFTA, but this difference did not reach statistical significance (8.2 ± 14.1% vs. 4.2 ± 7.1%). The percentage of mast cells was higher in biopsies with scarring, and was greater in infiltrates in the scarred areas than in the unscarred compartment (6.5 ± 4.8% vs. 1.9 ± 1.5%, p = 0.0004).

T cells and macrophages were the dominant cell types in both compartments, but the relative frequency of other cell types (B cells, plasma cells, mast cells) was higher in scarred areas, making the T cell and macrophage percentage relatively lower (i-IFTA vs. i-Banff: T cells 40.8 ± 13.1% vs. 53.2 ±12.4%, p = 0.006; macrophages 17.2 ± 10.1% vs. 27.1 ± 13.3%, p = 0.02).

Mast cell-associated transcript set

To assess expression of mast cell-associated transcripts across all biopsies, we collapsed the expression values of the four mast cell-associated transcripts into a ‘Mast cell-associated transcript’ (MACAT) score for each biopsy. MACAT scores were higher in biopsies with more scarring and i-IFTA (i-IFTA <25%: 0.39 ± 0.71 vs. i-IFTA ≥25%: 1.36 ± 0.67, p < 0.0001), but not in biopsies with more inflammation in the unscarred compartment. The MACAT score correlated with time after transplantation (r = 0.55, p < 0.01), extent of scarring (r = 0.61, p < 0.01) and of i-IFTA (r = 0.63, p < 0.01) but not i-Banff (r =–0.03, p > 0.05). Even in biopsies with at least scarring of Banff grade I, that is >5% of cortex involved, MACAT scores still correlated with the extent of scarring (r = 0.51, p < 0.0001) and i-IFTA (r = 0.55, p < 0.0001).

Biopsies with low mast cell transcript expression (i.e. lowest tertile of MACAT scores in all 104 patients) were associated with better allograft survival (94.3%) compared to high mast cell transcript expression (i.e. the intermediate and highest tertile) (73.9%, p = 0.01). Even when this analysis was restricted to biopsies with scarring (at least Banff grade I, n = 88), high mast cell transcript expression was still associated with a worse allograft survival (71.2%), compared to those biopsies with scarring expression (96.6%, p = 0.01, Figure 5).

Figure 5. Mast cell PBT and allograft survival. Kaplan–Meier curves show that within allografts with scarring (at least Banff grade I) those with high expression of mast cell-associated transcripts have a worse prognosis. Events are defined as either allograft loss with return to dialysis or persistent (>3 months) low (<30 mL/min) eGFR. Given is the time between the last biopsy of a patient and an event.

Analysis of the validation set

The validation set confirmed the association of mast cell transcripts with scarring; three mast cell-associated transcripts (carboxypeptidase A3, mast cell tryptase beta 2, tryptase alpha/beta1) were in the top 10 probesets when microarray expression values were correlated with the extent of scarring and i-IFTA. Mast cell transcript scores were higher in biopsies with more scarring and i-IFTA (<25% vs. ≥25%, p = 0.004) and correlated with the extent of scarring (r = 0.72, p < 0.0001) and i-IFTA (r = 0.65, p < 0.0001). In terms of allograft outcome low mast cell transcript scores (i.e. lowest tertile of MACAT scores in the 50 patients of the validation set) were associated with a better allograft survival (100%) compared to high MACAT scores (76.5%, p < 0.02). Restricting this analysis to biopsies with scarring (at least Banff grade I, n = 40) still showed lower allograft survival for high MACAT scores (70.4%), compared to those biopsies with scarring and low MACAT scores (100%, p = 0.02).

Building a classifier

We built a simple threshold classifier from the original set of 129 biopsies based on mast cell transcript scores. The classifier was designed to predict recovery of allograft function (estimated glomerular filtration rate (eGFR)) after biopsy. We studied the change in eGFR between biopsy and 6 months post biopsy, and defined two classes: patients with unchanged or decreasing eGFR (i.e. no recovery of allograft function after biopsy), and patients with increasing eGFR (recovery of function of at least 10% from the value at biopsy). The training set threshold predicted eGFR status in the test set (n = 50) with an accuracy of 60%, sensitivity of 82%, specificity of 32%, positive predictive value of 61% and negative predictive value of 58%. The accuracy was significantly higher than that obtained using randomly shuffled eGFR status labels (56%) in a permutation test (p = 0.001).

Classifiers based on single training set: test set splits are known to make very inefficient use of the available data (24). Therefore, we also built a classifier using the full dataset (n = 179), and estimated error rates using leave-one-out cross-validation (LOOCV). This resulted in the following estimates: accuracy 71%, sensitivity 84%, specificity 47%, positive predictive value 74% and negative predictive value 63%. The accuracy was significantly higher than that obtained in a permutation test (64%, p = 0.001).

Discussion

We undertook these studies to establish the significance and molecular features of scarring (i.e. interstitial fibrosis and tubular atrophy) in transplant biopsies for cause, and of infiltrates in scarred areas, which have previously been regarded as nonspecific and ignored (25). The issue of scarring has become more prominent as immunosuppression has reduced the frequency of TCMR (i.e. i-Banff/tubulitis) and grafts are surviving longer. The results confirm the clinical value of the empirically derived Banff i- and t-scores in unscarred areas, representing the current diagnostic consensus. But we also found that infiltrates in either compartment, unscarred or scarred, are relevant for allograft survival, indicating that currently ignored infiltrates in scarred areas are significant. A genome-wide transcript analysis revealed mutually exclusive associations of sets of genes with the degree of scarring and infiltrates in scarred areas (i-IFTA) compared to infiltrates and tubulitis in unscarred areas. TCMR (i-Banff and tubulitis) was predominately associated with genes representing T cells, IFNG effects and macrophages, as well as a subset of injury-induced genes. Scarring and i-IFTA correlated with transcripts associated with B cells and plasma cells, and with a different subset of injury-induced transcripts and decreased expression of some kidney transcripts, but also with many transcripts not previously annotated for membership in transplantation relevant transcript sets. Within these the surprising finding was the high correlation of mast cell-associated transcripts with scarring. In those allografts showing scarring increased expression of mast cell-associated transcripts was associated with impaired graft survival.

While the typical TCMR inflammatory process is scored in unscarred areas, by histopathology this process also extends to scarred areas. By immunohistochemistry the scarred areas often display the typical TCMR inflammation (T cells/macrophages) when the unscarred areas have these changes. Thus the typical TCMR related transcripts—effector/effector-memory T cells, macrophages and IFNG-associated transcripts—correlated with the infiltrate and tubulitis scored in unscarred areas. But scarring has added complexity, including B cells, plasma cells and mast cells as demonstrated by immunohistochemistry and strong correlation with transcripts related to these cell types. This can explain why T cell, IFNG and macrophage-associated genes correlate highly with i-Banff/tubulitis but not with fibrosis/atrophy or i-IFTA. When a biopsy has extensive fibrosis/atrophy, the more prominent B-/plasma/mast cells-driven infiltrate will become the predominant (i.e. above our arbitrary threshold) correlating component. In contrast, in biopsies without scarring this type of inflammation is virtually absent, thus allowing for high correlations with the TCMR type of inflammation. Thus areas of scarring are not protected from TCMR, but are complex due to other types of inflammation.

As we previously discussed (22), controversies over the prognostic relevance of B-cell clusters and B cell- and plasma cell-associated transcripts in renal allografts reflect the fact that scarring is time-dependent, and such cells accumulate in scarred areas (15,26–30). Both protocol and indication biopsies beyond 6 months posttransplant often manifest an increase in B-cell nodules as well as scarring (15,22,26,28). But B cells also correlate with inflammation (i.e. the i-Banff infiltrate) in unscarred areas (22). Thus, the interpretation of B cells and plasma cells must take into account the intrinsic risk of poor outcomes in kidney transplants that require late biopsies for cause (>6 months posttransplant), due to an increased probability of serious and untreatable disease states compared to kidneys requiring early biopsies. Hence, features associated with late biopsies will automatically be associated with reduced survival (22,28), and whether such features (e.g. B cells) contribute to reduced survival must not be assumed. Some experimental data suggest possible roles for B cells in fibrosis. For example, B-cell knock out mice show decreased liver and skin fibrosis (31,32), and B cells, like mast cells, are capable of producing cytokines linked to fibrosis (e.g. IL4, IL6, IL13), potentially through collagen production and fibroblast proliferation (10,33,34). However, our previous analyses have shown no association of B cells or plasma cells with transplant outcomes once time posttransplant is taken into account (22).

The unexpected finding was that mast cell transcripts were the strongest correlate of scarring and predicted poorer outcomes when the extent of scarring and herewith indirectly time posttransplant was taken into account. In addition, mast cell-associated transcripts also predicted functional recovery after biopsy, even in biopsies with existing scarring. The association of scarring with mast cell transcripts was confirmed by immunohistochemistry, which showed increased numbers of mast cells in biopsies with scarring, recalling the underappreciated association of mast cells with fibrosis (35–37). How mast cells might be related to the pathogenesis of scarring and atrophy deserves further study (34).

Surprisingly, areas of established scarring did not strongly (i.e. >r ± 0.4) correlate with transcripts expected to be expressed in ongoing fibrogenesis (e.g. TGFβ1 (r = 0.18), collagens (r < 0.3), CTGF (r = 0.19)), indicating that mature scarring is not necessarily a sign of ongoing injury leading to future scarring. Moreover, the injury and repair-associated transcripts (IRITs), annotated in models of injury without residual fibrosis—mouse kidney isografts and ischemic acute tubular necrosis (21), showed distinct associations with inflammation in unscarred areas versus the extent of scarring. ‘Late’ IRITs comprising predominately transcripts related to cell cycle, fibrogenesis (e.g. TGFβ- or collagen-related) and extracellular matrix turnover (21) were associated with i-Banff/tubulits whereas ‘intermediate’ IRITs comprising predominately transcripts related to embryogenesis and organ development were correlating with scarring. Therefore, scarring itself is not a correlate of active fibrogenesis, but a marker of previous injury and repair that failed to restore the tissue to normal.

Inflammation in areas of scarring (i-IFTA) had a worse prognosis than scarring without inflammation. Numerous transcripts correlating with the extent of scarring are expressed by infiltrating inflammatory cells (i.e. B/plasma/mast cells), but these transcripts are completely separate from those correlating with TCMR as it is graded in the unscarred areas. However, besides transcripts associated with infiltrating inflammatory cells, scarring and i-IFTA were associated with decreased transcripts indicating epithelial deterioration (i.e. KTs) as well as some genes (see Table 1) already discussed in the literature as potential therapeutic targets in scarring. For example, inhibition of PDE5 (a phosphodiesterase hydrolyzing cGMP) ameliorates renal injury and fibrosis (38). Thus, potential roles in scarring of other correlating genes within this list are possible.

Scoring of inflammation in areas of scarring should be added to the evaluation of transplant biopsies, particularly those in which the scarring is extensive and where little unscarred cortex is present in the biopsy. This would represent an alignment with biology and provide relevant information in terms of prognosis. Nevertheless, specific diagnosis of the underlying disease process causing the inflammation (e.g. rejection vs. recurrent disease) remains still the primary goal of histopathological evaluation of a renal allograft biopsy. In addition, our retrospective analysis did not allow us to assess whether treatment might affect the different infiltrate types and their associated transcript sets. This has to be studied further in prospective randomized trials including combined clinical, histopathological and molecular readouts of therapy effects.

Acknowledgments

We thank Kara Allanach, Anna Hutton, Vido Ramassar and Robin Stocks for excellent technical assistance. We are grateful to Dr. Deborah James for critical reading and editing of the manuscript. This research has been supported by funding and/or resources from Genome Canada, Genome Alberta, the University of Alberta, Capital Health Edmonton Area, the University of Alberta Hospital Foundation, Roche Molecular Systems, Hoffmann-La Roche Canada Ltd., Alberta Innovation & Science, the Roche Organ Transplant Research Foundation, the Kidney Foundation of Canada, and Astellas Canada. Dr. Halloran also holds a Canada Research Chair in Transplant Immunology and the Muttart Chair in Clinical Immunology.

Competing Interests

None of the authors has to disclose any conflict of interest.