The Nature of Biopsies with “Borderline Rejection” and Prospects for Eliminating This Category


Philip F. Halloran,


In kidney transplantation, many inflamed biopsies with changes insufficient to be called T-cell-mediated rejection (TCMR) are labeled “borderline”, leaving management uncertain. This study examined the nature of borderline biopsies as a step toward eventual elimination of this category. We compared 40 borderline, 35 TCMR and 116 nonrejection biopsies. TCMR biopsies had more inflammation than borderline but similar degrees of tubulitis and scarring. Surprisingly, recovery of function after biopsy was similar in all categories, indicating that response to treatment is unreliable for defining TCMR. We studied the molecular changes in TCMR, borderline and nonrejection using microarrays, measuring four published features: T-cell burden; a rejection classifier; a canonical TCMR classifier; and risk score. These reassigned borderline biopsies as TCMR-like 13/40 (33%) or nonrejection-like 27/40 (67%). A major reason that histology diagnosed molecularly defined TCMR as borderline was atrophy-scarring, which interfered with assessment of inflammation and tubulitis. Decision tree analysis showed that i-total >27% and tubulitis extent >3% match the molecular diagnosis of TCMR in 85% of cases. In summary, most cases designated borderline by histopathology are found to be nonrejection by molecular phenotyping. Both molecular measurements and histopathology offer opportunities for more precise assignment of these cases after clinical validation.


antibody-mediated rejection


alternative macrophage activation transcripts


acute tubular necrosis


calcineurin inhibitor


donor-specific antibody


glomerular disease


interferon-gamma-inducible transcripts


panel-reactive antibodies


T-cell-mediated rejection


Histological assessment of biopsies is crucial to the management of kidney transplants because diagnoses are used to guide treatment strategies. Central to this is the Banff classification, which grades lesions by using consensus definitions and arbitrary thresholds (1). The defining lesions of T-cell-mediated rejection (TCMR) are interstitial inflammation (i-Banff score ≥2), tubulitis (t-score ≥2) and intimal arteritis (v-lesions). Because v-lesions are uncommon, the i- and t-score are the critical features. However, these lesions are inherently nonspecific and associated with various conditions in native and transplanted kidneys such as acute kidney injury (2). This may explain some of the lesions observed in protocol biopsies from stable patients (3). As a result of reliance on nonspecific features, many inflamed biopsies manifest subthreshold changes, creating the ambiguous diagnosis of “borderline changes”: tubulitis (t1–3) with subthreshold interstitial inflammation (i0–1) or interstitial inflammation (i2–3) with subthreshold tubulitis (t1). Some borderline biopsies probably reflect early TCMR and progress to more severe lesions (4,5), whereas others simply reflect acute kidney injury. It is also possible that TCMR may exist in mild “forme fruste” versions that do not require treatment, as recently suggested in a protocol biopsy study (6). Tubulitis and inflammation are particularly difficult to score when there is extensive atrophy-scarring: according to current Banff consensus, inflammation is only interpretable in nonscarred areas and tubulitis can only be scored in nonatrophic or mildly atrophic tubules. Whether a tubule is too atrophic to score requires subjective judgment by the pathologist and is difficult to reproduce (7–9).

The Banff process always intended that new technologies would eliminate the borderline category (1,10–12). Molecular phenotyping using microarrays has the potential to reclassify borderline cases into TCMR and nonrejection, thereby eliminating the need for an ambiguous category. The objectivity of microarray measurements can also improve reproducibility (7–9). This study analyzed the molecular phenotype of the borderline biopsies, in comparison to biopsies with TCMR and nonrejection. This required inclusion of biopsies with no major histological abnormality. This approach offers the possibility of an integrated histomolecular diagnosis. We hypothesized that adding histo-molecular phenotype to histologic assessment would reclassify the borderline cases by their histo-molecular phenotype and (b) allow us to reexamine the histologic criteria, particularly in challenging biopsies with atrophy-scarring. This project used three published molecular features of TCMR: T-cell burden, canonical TCMR and the rejection classifier (13–15). In addition, we discovered that a previously described risk score classifier for assigning the probability of failure in kidneys with progressive diseases (16) was abnormal in TCMR cases because it detects the nephron distress, which is induced by TCMR as well as by progressive diseases. We measured these four molecular features in biopsies with borderline histopathology changes to determine which were like TCMR-like and nonrejection-like.

Materials and Methods

Patient population and specimens

The study was approved by the institutional review boards of the University of Alberta, Edmonton, Canada (issue # 5299), the University of Illinois, Chicago, USA (Protocol # 2006–0544), the University of Minnesota (Protocol #HSC#0606 M87646) and the Hennepin County Medical Center, Minnesota, USA (Protocol #HSR#06–2670). After written informed consent, renal transplant recipients undergoing a transplant biopsy for clinical indications (BFC; deterioration in function, proteinuria and stable but impaired function) as part of standard of care between September 2004 and November 2008 were enrolled in this study. For controls, normal kidney tissue was obtained from histologically unaffected cortical areas in native nephrectomies performed for renal cell carcinoma. Biopsies were obtained and processed for microarray as described in detail before (17).


Biopsies were assessed by a pathologist blinded to the results of the molecular studies and were prepared and graded according to recent Banff criteria (1) including C4d negative antibody-mediated rejection (ABMR), on the basis of our previous description (18). Biopsies diagnosed as nonrejection were defined as having a ci-score (interstitial fibrosis) of <2 and no features of a disease process. Biopsies diagnosed as interstitial fibrosis and tubular atrophy not otherwise specified (IFTA-NOS) were defined as having a ci-score of ≥2 and no features of a specific disease process. From 403 consecutive biopsies for clinical indications in 315 patients, we eliminated those with specific diagnoses (e.g. ABMR, polyoma nephropathy, recurrent or de novo glomerular diseases(GN) or IFTA-NOS) and focused on the 191 biopsies with the histological diagnoses of TCMR, borderline rejection and nonrejection. Not all biopsies were stained for polyoma virus, as this decision relied on local standard of care.

Histological lesions were scored as continuous variables to allow the generation of diagnostic cutoffs. Interstitial infiltrates were assessed using three different methods: i-Banff (interstitial inflammation only in nonscarred cortex), i-IFTA (interstitial inflammation only in scarred cortex) and i-total (all cortical inflammation), as previously described (19). Tubulitis extent (t-extent) was assessed as the percentage of cortical tubules in a biopsy with tubulitis of any degree. All biopsies were deemed sufficient by consensus criteria, with the exception of 10 biopsies without arteries.

Treatment on the basis of the biopsy was at the discretion of the patient's clinician and no molecular features were available at that time. “Treatment” as used here includes corticosteroids but does not include adjustment of the calcineurin inhibitor (CNI) drug doses. Response to treatment was defined as a ≥15% reduction in the serum creatinine at 1 month after antirejection therapy. This time was selected in advance to get complete data and avoid data mining. The use of the reciprocal of the creatinine did not materially change the conclusions.

Analysis of microarray data

Microarray data files were processed as described previously (20). Expression values for transcripts were calculated as fold change versus eight nephrectomies that served as controls (17). Expression and phenotype data, as well the CEL files, are available as supplemental data on the ATAGC website ( A predictive analysis of microarrays rejection classifier score was calculated as previously described (14), with a rejection classifier score of >0.50 indicating a high probability of rejection. The risk score used was similar to the published risk score (16) and it was calculated for all biopsies, irrespective of time and number of biopsies per patient. The mean risk score of the 403 biopsies for clinical indications (mean = 0.002) was applied to the three categories, stratifying them into high- and low-risk score biopsies. High T-cell burdens (13) were defined as a fold change of ≥2.32 (mean T-cell burden score across 403 biopsies) compared to nephrectomy. Canonical TCMR cases were identified as previously described (15).

Statistical analysis

Data analysis was performed using GraphPad Prism 5 (GraphPad Software, La Jolla, CA, USA), Bioconductor version 2.4 and R version 2.9.1. Pathogenesis-based transcript (PBT) scores were calculated as the average fold change across transcripts for a particular PBT set from nephrectomies. Group comparisons for the mean PBT score used a one-way ANOVA with a Dunnett's multiple comparison test on the logarithmic transformed values. A Mann-Whitney t-test was performed for nonparametric data comparing two groups. Time was normalized through logarithmic transformation before performing Pearson correlations. Allograft function was analyzed as estimated glomerular filtration rate (GFR), using the Cockcroft-Gault formula or serum creatinine. Results are given as means with the level of significance set at p < 0.05. Decision tree analysis was performed using the lesions currently used by the Banff classification, in addition to i-total, i-IFTA and t-extent.


Population demographics and diagnostic classification

Demographics of the 191 biopsies with the histological diagnoses of TCMR (n = 35), borderline rejection (n = 40) and nonrejection (n = 116) are shown in Table 1.

Table 1.  Demographics
 All patientsTCMR patientsBorderline patientsNonrejection patients
Patient demographics(n = 160)(n = 25)(n = 32)(n = 103)
  1. MMF = mycophenolate mofetil.

Mean recipient age at Tx46 (9–81)48 (22–69)49 (9–76)45 (14–81)
Recipient Gender (% male)105 (66%)16 (64%)21 (66%)68 (66%)
 Caucasian101 (63%)17 (68%)21 (66%)63 (61%)
 Black19 (12%)1 (4%)2 (6%)16 (16%)
 Other24 (15%)4 (16%)5 (16%)15 (15%)
 Not available16 (10%)3 (12%)4 (13%)9 (9%)
Primary disease    
 Diabetic nephropathy37 (23%)2 (8%)7 (22%)28 (27%)
 Hypertension/large vessel disease17 (11%)2 (8%)6 (19%)9 (9%)
 Glomerulonephritis/vasculitis47 (29%)10 (40%)9 (28%)28 (27%)
 Interstitial nephritis/pyelonephritis13 (8%)3 (12%)3 (9%)7 (7%)
 Polycystic kidney disease27 (17%)6 (24%)4 (13%)17 (17%)
 Others8 (5%)1 (4%)1 (3%)5 (5%)
 Unknown etiology11 (7%)1 (4%)2 (6%)9 (9%)
Mean donor age43 (16–69)42 (19–68)41 (17–67)44 (16–69)
Donor gender (% male)61 (44%)11 (44%)13 (41%)37 (36%)
Donor type (% deceased donor transplants)69 (43%)12 (48%)14 (44%)43 (42%)
 All biopsiesTCMR biopsiesBorderline biopsiesNonrejection biopsies
Clinical characteristics at time of biopsy(n = 191)(n = 35)(n = 40)(n = 116)
Indication for biopsy    
 Primary nonfunction10 (5%)2 (6%)3 (8%)5 (4%)
 Rapid deterioration of graft function40 (21%)8 (23%)7 (18%)25 (22%)
 Slow deterioration of graft function66 (35%)11 (31%)13 (33%)42 (36%)
 Stable impaired graft function49 (26%)7 (20%)9 (23%)33 (28%)
 Investigate proteinuria11 (6%)4 (11%)4 (10%)3 (3%)
 Follow-up from previous biopsy5 (3%)1 (3%)2 (5%)2 (2%)
 Others5 (3%)0 (0%)1 (3%)4 (3%)
 Indication unknown5 (3%)2 (6%)1 (3%)2 (2%)
Maintenance immunosuppressive regimens at biopsy    
 MMF, tacrolimus and steroids69 (36%)15 (43%)16 (40%)38 (33%)
 MMF and tacrolimus24 (13%)0 (0%)3 (8%)21 (18%)
 MMF, cyclosporine and steroids32 (17%)7 (20%)8 (20%)17 (15%)
 MMF and steroids13 (7%)8 (23%)2 (5%)3 (3%)
 Azathioprine, cyclosporine and steroids2 (1%)0 (0%)0 (0%)2 (2%)
 Others51 (27%)5 (14%)11 (28%)35 (30%)

Functional and histological comparison among histologic TCMR, borderline and nonrejection

Interstitial inflammation (i-Banff), tubulitis and scarring were compared across the disease categories (Figures 1A–C). There was a gradient of inflammation (mean Banff i-score) from TCMR (1.74) to borderline (0.63) to nonrejection (0.06; p < 0.001; Figure 1A). A similar gradient was seen in the i-total score (Figure 1A). However, the Banff t-score was similar in TCMR (1.97) and borderline (1.85; p = ns), although both were different from nonrejection (0.0 by definition; Figure 1B). The degree of interstitial fibrosis was similar between TCMR (= 1.0) and borderline (= 1.0) but both were higher than nonrejection biopsies (= 0.58; p < 0.001; Figure 1C). Likewise, tubular atrophy was similar in TCMR (= 1.2) and borderline (= 1.1), but both were higher than nonrejection biopsies (= 0.7; p < 0.01).

Figure 1.

(A) Continuous i-Banff and i-total scores across the Banff diagnostic categories. (B) Tubulitis. (C) Scarring, atrophy and inflammation in scarred areas. The TCMR category has higher i-Banff and i-total but similar levels of scarring and tubulitis to the Borderline category. One-way ANOVA with Dunnett's multiple comparison testing: ** p-value < 0.01, *** p-value < 0.001.

The mean estimated glomerular filtration rate (GFR) was similar in all the three groups at the time of biopsy and 1 and 6 months later (Table 2). Function improved in some grafts after biopsy regardless of the histological diagnosis and recovery of function was not associated with anti-rejection treatment (p = ns, Fisher's exact test; Table 3). Improvement independent of anti-rejection treatment probably reflects patient management changes by the attending nephrologist (e.g. volume replacement and adjustment of the CNI dose). Thus, recovery of function cannot be used to distinguish true TCMR from other findings in biopsies for cause.

Table 2.  Functional comparisons of current categories assigned by Banff criteria
  1. All values are means unless stated. ***p < 0.001.
    One-way ANOVA with Dunnett's multiple comparison compared to borderline. eGFR = estimated glomerular filtration rate.

Number of biopsies3540116
Days to biopsy median (range)148 (8–3878)176 (7–3584)76 (6–9056)
eGFR at biopsy43.944.648.3
eGFR at 1 month57.3***53.0***57.4***
eGFR at 6 months58.4***63.5***59.6***
Table 3.  Relationship of diagnosis and antirejection treatment to subsequent improvement in kidney function
 Treated [n = 71]Not treated [n = 107]
Histological diagnosis [n =]Functional improvement1No improvementFunctional improvement2No improvement
  1. Fisher's exact test. No effect of treatment on improvement in kidney function in each histological category.
    1Functional improvement is defined as ≥15% decrease in serum creatinine at 1 month.
    2Antirejection treatment is defined as any treatment administered in addition to the patients maintenance immunosuppression, e.g. steroids, antithymocyte globulin
    3Two cases are missing creatinine values at 1 month.
    4Thirteen cases are missing creatinine values at 1 month.

TCMR [35]1811 3 3
Borderline [40]3 9 9 812
Nonrejection [116]4 8143348

Molecular features recognize the gradient in TCMR, borderline and nonrejection.

We explored molecular differences between the three histological groups using the rejection classifier (Figure 2A) and the T-cell burden (Figure 2B). In each graph, the biopsies meeting the molecular definition of canonical TCMR are indicated as solid black symbols. The results demonstrate a gradient in the molecular features from Banff TCMR to Banff borderline to nonrejection, similar in some respects to the i-Banff score. The biopsies designated as canonical TCMR by molecules had consistently high rejection classifier scores and T-cell burdens. Some Banff TCMRs with high rejection classifier scores and T-cell burdens were not classified molecularly as canonical TCMR, indicating that the canonical TCMR category was less sensitive but more specific. The histological borderline category showed great variability in its molecular features, overlapping histological TCMR and nonrejection.

Figure 2.

(A–C) The molecular features of TCMR, borderline and nonrejection demonstrate a gradient from highly inflamed TCMR cases to uninflamed nonrejecting cases, with borderline bridging the spectrum: (A) rejection classifier; (B) T-cell burden score and (C) the molecular risk score. Canonical TCMR cases are shown in each figure in solid points.

The risk score captures the molecular gradient from rejection to nonrejection

The risk score classifier was derived to predict which kidneys undergoing late (>1 year) biopsies were at risk of failing. However, we found that the risk score was also high in TCMR, which does not fail. This is because the risk score detects epithelial distress and matrix remodeling (Figure 2C) that are strongly induced in TCMR. This made the risk score classifier another feature that separated TCMR from nonrejection, with the risk score highest in histologic TCMR, especially those with the molecular features of canonical TCMR. Again, the risk score shows that borderline biopsies were heterogeneous, some displaying TCMR-like and others nonrejection-like features.

Molecular TCMR features consistently reclassify the borderline biopsies

High risk scores were observed in most histologically defined TCMR (28/35), in several nonrejection biopsies (12/116) and in some borderline biopsies (17/40; Table 4). The rejection classifier, the T-cell burden and the canonical TCMR designation also identified most histologically defined TCMR, a minority of borderline and few nonrejection biopsies (Table 4).

Table 4.  Comparison of feature assignments with current histologic diagnosis
Histological diagnosisHigh risk score1High rejection classifier score2High T-cell burden3Canonical TCMR4
  1. 1High risk was defined as being greater than the mean risk score (0.002) of all 403 biopsies.
    2High rejection classifier rejection score was defined as >0.5 probability of rejection.
    3High T-cell burden was defined as being greater than the mean T-cell burden (2.32) of all 403 biopsies.
    4Canonical TCMR was defined as having both a GRIT1 and AMAT1 score in the top 20th percentile of all 403 biopsies.

TCMR [n = 35]28 (80%)25 (71%)26 (74%)20 (57%)
Borderline [n = 40]17 (43%)13 (33%)14 (35%)3 (8%)
Nonrejection [n = 116]12 (10%)3 (3%)11 (9%) 1 (1%)

Eliminating borderline by using molecular features to distinguish rejection from nonrejection

It was apparent from the above analyses that molecular features might represent an opportunity to add new understanding to the current histologic definitions, particularly in the large borderline group. We selected two TCMR molecular features representing epithelial distress (the risk score) and T-cell burden, to reclassify the biopsies. Histologically defined TCMR, borderline and nonrejection were stratified as having a high- or low-risk score, with their T-cell burden plotted on the y axis (Figure 3).

Figure 3.

Each histological category was stratified into high risk score and low risk score groups before plotting them by their T-cell burden.

Most of the 35 histologic TCMR biopsies had both a high-risk score (28/35) and a high T-cell burden (26/35). All seven histologic TCMR biopsies with a low-risk score had low T-cell burdens. Five of these cases had been designated as TCMR on the basis of isolated v-lesions–a lesion that is potentially not specific for rejection (21; Table S1) and therefore currently being reevaluated by the Banff process (22).

Of 17 borderline biopsies with high-risk scores, 13 had a high T-cell burden. The four exceptions (high-risk score, low T-cell burden) may represent acute kidney injury as three had delayed graft function that subsequently recovered without treatment. Almost all (22/23) borderline biopsies with low-risk scores had low T-cell burden.

Most histologic nonrejection biopsies (110/116) had no molecular features of TCMR.

Proposing and testing a new diagnostic classification

These results suggest a new classification to eliminate the borderline category by assigning the borderline biopsies to either TCMR or nonrejection on the basis of their molecular features (Figures 4A and B). Of the 40 Banff borderline biopsies, 13 were TCMR-like and 27 were nonrejection-like. In addition, of 35 Banff TCMRs, 9 were nonrejection-like.

Figure 4.

(A and B) Representative figures demonstrating the elimination of borderline and low-risk score TCMRs using the risk score and T-cell burden strategy to create the TCMR-like and nonrejection-like categories.

We tested these new categories using the molecular rejection classifier and the molecular definition of canonical TCMR. As assessed by the rejection classifier, TCMR-like biopsies were similar to Banff TCMR and the nonrejection-like biopsies were similar to nonrejection biopsies (Figure 5). Two of the three borderline biopsies with molecular features of canonical TCMR were reclassified as TCMR-like.

Figure 5.

The application of the rejection classifier and canonical TCMR (shown in solid) classifier confirms the validity of the new groups. The TCMR-like category has a similar rejection classifier score to TCMR but higher than either nonrejection-like or nonrejection. *** p-value < 0.001.

TCMR-like biopsies have higher levels of scarring, interfering with the diagnosis of TCMR

We studied why many TCMR-like biopsies (i.e. formerly borderline by histopathology) were potentially misclassified by the current Banff criteria (Figures 6A–C). Compared to biopsies called TCMR by histology alone, the TCMR-like biopsies displayed lower i-Banff scores (p < 0.001), more scarring (p < 0.01) but similar i-total scores. Tubulitis was similar in TCMR and the borderline classes that were TCMR-like and nonrejection-like (Figure 6B) by definition because this feature was essential to the histologic diagnosis. TCMR-like biopsies differed from nonrejection-like biopsies by the i-total score, ci-score and i-IFTA score (p < 0.001, respectively; Figure 6C).

Figure 6.

(A) Continuous i-Banff and i-total scores across the new diagnostic categories and (B) tubulitis. (C) Scarring, atrophy and inflammation in scarred areas across the new diagnostic categories. The TCMR-like category has higher i-total, i-IFTA, ci- and ct-scores compared to the other categories, indicating that scarring prevents a diagnosis of TCMR. One-way ANOVA with Dunnett's multiple comparison testing :*p-value < 0.05, **p-value < 0.01, ***p-value < 0.001.

Thus, atrophy-scarring and its related inflammation (i-IFTA) were unusually high in the biopsies classified as Banff borderline but with TCMR-like molecular features. These findings indicate that TCMR-like biopsies are true TCMR but fail to be diagnosed by the current histologic criteria for TCMR because the scarring reduces the amount of tissue that can be used for assessment of the i-Banff score. (This does not show up in Figure 1 because of offsetting effects of the two processes making up the current borderline group: true TCMR and nonrejection.)

New inflammation and tubulitis criteria approximate the new histologic molecular phenotype

We studied whether new histologic criteria could be developed that would better approximate the new definition of TCMR versus nonrejection, thus eliminating the borderline category. A decision tree using continuous histological scores identified the i-total score and the extent of tubulitis as the features best approximating the molecular phenotype (Figure 7). Applying the total i-score with a threshold of >27%, 24/26 newly classified TCMRs and 9/13 TCMR-like cases, equating to 85% of cases, were correctly identified (Node 5). Two cases misclassified had been pretreated with steroids before the biopsy and thus molecularly sterilized. Extent of tubulitis (>3%) identified a group with subthreshold i-total but an approximately 50% probability of TCMR (Node 4): with this new histological system, only 11 biopsies were assigned to this group compared to the 40 cases in the previous borderline category.

Figure 7.

Decision tree analysis. The i-total and t-extent lesions are the closest histological approximation of the molecular method of eliminating borderline.


Borderline rejection is a large ambiguous diagnostic category in the current Banff classification, reflecting the difficulty of distinguishing by histology, biopsies with TCMR from those with minor changes, including inflammation induced by the injuries of the transplant process. The magnitude of the problem is indicated by our finding that borderline outnumbers TCMR (40 vs. 35) in this unselected population of biopsies for indications. We developed a combined histological and molecular approach to reclassify and eliminate the borderline category. Using a molecular risk and T-cell burden score obtained from microarray results, we stratified the biopsies already given histologic diagnoses on the basis of high and low molecular scores into TCMR-like versus nonrejection-like. We confirmed this classification by applying two other independently derived molecular tools for diagnosing rejection i.e. a rejection classifier (14) and canonical TCMR (15). Borderline cases behaving molecularly like TCMR had not been called TCMR by current Banff rules because they had advanced IFTA, which interferes with the assessment of interstitial inflammation in the nonscarred compartment (i-Banff score). Our results indicate how the uncertainty in the diagnosis of TCMR can be reduced and the borderline category can be eliminated, not by ignoring histopathology but by adding a molecular dimension to the current histologic features. Because most transplant centers do not have access to molecular analysis of allograft biopsies, we performed decision tree analysis to show how refined histological assessment could better approximate the combined histologic-molecular classes. We found that reliance on the i-total score (i.e. all cortical inflammation including areas of fibrosis and atrophy) plus secondary assessment of the “extent of tubulitis” score could estimate the probability of TCMR with relatively few ambiguous cases.

This study of functional recovery after biopsies of troubled transplants, over the complete range of times posttransplant observed in the clinic, is limited by the constraints on both “baseline” and “follow-up” analyses, involving multiplicity of intervention, limitations in data availability and potential for subjectivity and data mining. Clinicians often react to troubled kidneys with several overlapping strategies e.g. steroid treatment, adjusting drug doses and correcting volume disturbances. Initial reports from pathologists may not match definitive reports, creating disparities where clinical actions do not seem to match diagnoses. The issue of baseline and follow-up creatinine is highly influenced by time. For follow-up, we prospectively selected 30 days and 6 months in the design of the study as clinically meaningful times and have continued to rely on this, without looking at the data, to avoid data mining. The variation in actual creatinine values observed in individual patients at different times often reflects unrecorded clinical events, making every point that is selected for examination problematic. The use of creatinine values before the biopsy to establish baseline is even more problematic as late patients may not have creatinine measurements for more than three months before the biopsy; early patients are having creatinine measurements daily but the values may be changing rapidly e.g. recovering from acute kidney injury. We attempted to address this in a detailed study of GFRs in the renal transplant population (23) and concluded that the prebiopsy creatinines are highly influenced by time and only the postbiopsy creatinines permit any accurate estimate of the clinical course. We cannot come up with any satisfactory definition of what a prebiopsy “baseline” value is. Thus to change the way the creatinine is assessed in this study is limited by the completeness of observations in each patient, where missing data can profoundly alter the conclusions but also opens the possibility of data mining. Other alternatives could also create a multiple testing as well as incompleteness problem. The use of 1/creatinine is biologically better as an estimate of GFR, but statistically identical as all the variation is the creatinine and did not alter the results.

As indicated in this study, the Banff borderline category is a composite of injury with secondary inflammation and TCMR, the diagnosis of which may be obscured by partial treatment or biopsy sample variability. Severe injury, such as occuring peritransplantation, triggers a wound healing response in which inflammation plays a key role. This remains readily detectable molecularly in morphologically normal mouse isografts for at least 21 days and probably lasts indefinitely as shown in human protocol biopsies (6,24). Core-to-core sample variability is an ongoing problem in all biopsy studies, whether read by histopathology or molecules and may contribute to the discrepancies between the molecular and histopathology readings. Formal studies on the variation in molecular phenotype between centers should eventually be performed. However, we have found in transplant nephrectomies that molecular readings are relatively uniform compared to histopathology and are similar in cortex and medulla (paper in preparation). Furthermore, although histopathology reads two or more cores, it reads only thin sections, whereas the molecular readings read a much greater volume of tissue across the entire 18-gauge needle core. Each technique has its merits, but it seems likely that each will add a dimension that complements the other and reduces the impact of sampling error.

The observed intrinsic heterogeneity of the borderline category also explains the conflicting treatment results in the literature (10–12). This analysis indicates that the histologic assessment of TCMR is more accurate when it relies primarily on interstitial inflammation and less on tubulitis. Tubulitis per se is a nonspecific lesion seen in many nontransplant settings and essentially always seen in atrophic tubules. Therefore, it is consensus that tubulitis must not be scored in atrophic tubules but this requires a difficult decision about whether the affected tubule is too atrophic or not. This ambiguity in the rules for scoring tubulitis causes poor reproducibility with a kappa value of 0.11, not much better than chance, potentially contributing to the mislabeling in the borderline category (7–9). In the transplant setting, a diagnosis of borderline may often indicate acute kidney injury with minor inflammation (25).

Histo-molecular phenotyping has the potential to increase the accuracy of the diagnosis, thus preventing false positive diagnoses with inappropriate antirejection treatment. However, it might be expected that response to antirejection treatment would be a useful readout of true TCMR, we found that improvement is frequent after a BFC regardless of its diagnosis and treatment, probably reflecting clinicians adjusting maintenance therapy or fluid balance in the light of ongoing laboratory tests. Thus, response to antirejection therapy does not confirm the histological diagnosis: many treated TCMR or borderline cases do not recover function and many untreated borderline or TCMR cases do. The borderline diagnosis requires a new approach because the diagnostic system should offer more guidance for therapy (12). Of the 75 biopsies considered as either Banff TCMR or borderline in this study, only 39 (52%) had the combined molecular and histological features to warrant treatment (TCMR or TCMR-like). Once we have a more accurate estimate of whether a true TCMR process is operating, the whole issue of the therapy of TCMR in the current era of immunosuppression should be reexamined: Does it always require therapy and what are the implications of the treatment choices? Is there a self-limited form of TCMR–a “forme fruste”, as suggested by some experience with protocol biopsies? A more accurate assessment system will open the door to such studies.

Although designed to predict graft failure in kidneys having late biopsies after 1 year, the risk score proved to be a useful additional tool for detecting TCMR, presumably because it detects the epithelial injury that is at the heart of TCMR (16). The risk score classifier uses transcripts that reflect an ongoing response to injury, including epithelial distress, de-differentiation and matrix remodeling, with many of the transcripts already known to be expressed by the epithelium following injury e.g. HAVCR1 (“KIM-1”) and alpha-v-integrin 6 (ITGB6; 26,27), with less reliance on transcripts reflecting inflammation (T-cell or macrophage burden and IFNG effects). The risk score/T-cell burden strategy is more inclusive than our previously published tools to identify TCMR from nonrejection because it actually assesses tissue injury, which is observed in three main situations acute kidney injury, TCMR and progressive diseases (Famulski et al., manuscript in preparation). Thus in the setting of self-limiting injury or a readily treatable disease (e.g. TCMR), the risk score identifies the injury-repair response rather than the risk of graft failure.

In developing a new approach that incorporates time, alternative histological features not currently used to define TCMR can improve the discrimination of TCMR from nonrejection, previously identified using a combined molecular and histological approach. The i-total score that captures all inflammation in the biopsy core and not the current Banff i-score, which only considers inflammation in nonscarred areas, is the best histological feature predicting the molecular phenotype of TCMR, in this population free from ABMR, GN or polyoma virus nephropathy. Because most disease processes in kidney transplants are accompanied by interstitial inflammation (28), the i-total score is not specific for TCMR. But once other major diseases are excluded, the i-total score allows the diagnosis of TCMR in late biopsies with advanced scarring when the current i-Banff score cannot be applied. The i-Banff score was derived at a time when biopsies were performed early and minimal scarring was present. Currently, biopsies are often clinically indicated in damaged kidneys (e.g. extended criteria donors) and after the first year frequently show a degree of scarring. This by definition reduces the size of the unscarred compartment in which the diagnosis of TCMR relies upon (8). In the absence of molecular phenotyping, i-total improves the opportunities for assessing TCMR, detecting 85% of cases that meet the histo-molecular criteria for TCMR. This new approach combining i-total and t-extent (instead of the degree of tubulitis) places most of the 40 biopsies in the Banff borderline category (where the probability of TCMR is only 33%) in either nonrejection or high probability of TCMR, leaving only 11 biopsies with 50% probability of having TCMR.

This study indicates that the consensus system should be revisited to determine whether the current Banff criteria are valid in late scarred kidneys. Readers, editors and referees are used to looking at data from clinical trials where the biopsies and the creatinine values are coordinated in time and all are relatively early. This study captures the real world of the clinic: real-time readout of the entire renal transplant population and such analyses have not been available until recently. This creates some surprising results but it is the true unselected data. We therefore propose that the Banff group should now examine these new rules for diagnosing TCMR with the possibility of incorporating them into the classification, particularly in cases with advanced scarring.


This research has been supported by funding and/or resources from Novartis, Bristol-Myers Squibb, Genome Canada, the University of Alberta, the University of Alberta Hospital Foundation, Hoffmann-La Roche Canada Ltd., the Alberta Ministry of Advanced Education and Technology, the Roche Organ Transplant Research Foundation, the Kidney Foundation of Canada, and Astellas Canada. Dr. de Freitas was supported by the St. Johns Ambulance Air Wing Transplantation Fellowship. Dr. Halloran held a Canada Research Chair in Transplant Immunology until 2008 and currently holds the Muttart Chair in Clinical Immunology.


The authors of this manuscript have the following conflicts of interest to disclose as described by the American Journal of Transplantation. P. F. Halloran is the chief executive officer of Transcriptome Sciences Inc, a company with an interest in molecular diagnostics, and has consulted for Novartis, Astellas, and Bristol-Myers Squibb in the past 3 years. The other authors have no conflicts of interests to disclose as described by the American Journal of Transplantation.