Table 2 Characteristics of microarray studies evaluating gene expression profile in acute allograft rejection biopsies in humans.

Tannapfel et al. [14]

2001

Liver (biopsies)

Atlas human cDNA

~ 600 genes

Aim. To investigate the expression of multiple inflammatory and apoptosis related genes in acute allograft rejection.

Design. (Adults) 62 patients, 97 biopsies: acute allograft rejection (n = 32), HCV reinfection (n = 18), CMV infection (n = 5), acute rejection and HCV infection (n = 3), stable graft function (n = 30) and after treatment of acute rejection (n = 9).

Statistics. Not available.

Sreekumar et al. [15]

2002

Liver (biopsies)

Affymetrix HU 6800

~ 6, 400 genes

Aim. To study intragraft gene expression patterns in acute cellular rejection and during recurrence of HCV in HCV infected recipients.

Methods. (Adults) 8 patients and biopsies: HCV infection and acute cellular rejection (n = 4), HCV infection without acute cellular rejection (n = 4).

Statistics. T-tests and fold change threshold.

Inkinen et al. [16]

2005

Liver (biopsies)

Turku Centre of Biotechnology human immunochip

~ 4, 600 genes

Aim. To determine and compare gene signature of CMV infection and acute rejection.

Methods. (Adults) 7 patients and biopsies: CMV infection (n = 4), patients with acute rejection (n = 3).

Gene expression of CMV and acute rejection samples were compared to that of liver graft after reperfusion.

Statistics. Not available.

Asaoka et al. [28]

2009

Liver (biopsies)

AceGene Human chip

~ 30, 000 genes

Aim. To identify genes characteristic of acute cellular rejection in patients with recurrent HCV infections.

Methods. (Adults) 21 HCV positive patients, 22 biopsies: acute cellular rejection (n = 9), without acute cellular rejection (n = 13). The expression of some transcripts (CASP8 and BMP2) was validated through qRT-PCR in this data set and also in a validation set: 32 biopsies from 25 HCV positive patients.

Statistics. Class discovery: unsupervised clustering analysis. Class comparison: Mann Whitney U test, supervised cluster analysis. Biological explanation: networks were built by Ingenuity Pathway Analysis (IPA).

Gimino et al. [17]

(Minneapolis Dataset)

2003

Lung (BAL)

Affymetrix Human Genome U133A

~ 18, 000 genes

Aim. To determine markers of acute rejection in lung recipients.

Methods. (Adults) 26 patients, 34 samples: acute rejection (n = 27), without diagnosis of rejection (n = 7).

Statistics. Class comparison: significance analysis of microarray. Class description: supervised clustering analysis.

Patil et al. [18]

(Minneapolis2 Dataset)

2008

Lung (BAL)

Affymetrix Human Genome U133A

~ 18, 000 genes

Aim. to improve acute rejection diagnostics by identifying genes whose expression best classifies acute rejection versus no rejection

Methods. (Adults) 32 patients, 32 samples: acute rejection (n = 14), without diagnosis of rejection (n = 18). Expression of some transcript was also assessed through qRT-PCR.

Statistics. Class comparison: Significance analysis of microarrays.

Class prediction: prediction analysis of microarrays, method of nearest shrunken centroids with 10 fold cross validation. Biological explanation: Gene Ontology and GoHyperG.

Karason et al. [19]

2006

Heart (biopsies)

Affymetrix Human Genome U133A

~ 18, 000 genes

Aim. To utilize microarray analysis to search for potential biomarkers of cardiac allograft rejection.

Design. (Adults). 20 patients, 14 patients experienced acute rejection episodes. 3 patients with acute rejection and biopsy available at three different time-points (before: normal histology, during: biopsy with acute rejection episode, after: biopsy with normal histology after the rejection episode) were profiled. qRT-PCR was performed for selected genes (CXCL9, CXCL10, NNPA). Serum levels of CXCL9 and -10 in 10 patients at three time points were also determined.

Statistics. Gene clustering according to time-point analysis: self organizing map (SOM) algorithm. Biological explanation: Gene Ontology (GO) and Netaffx.

Akalin et al. [21]

2001

Kidney (biopsies)

Affymetrix

HU 6800

~ 6, 400 genes

Aim. To analyze gene expression profile using microarrays in acute allograft rejection.

Design. (Adults) 10 biopsies: histological evidence of acute cellular rejection (n = 7), without evidence of rejection (n = 3).

Statistics. Each acute rejection sample was compared with each control sample. Genes with a > fourfold increase in the majority of the samples were selected.

Sarwal et al. [22]

(Stanford Dataset)

2003

Kidney (biopsies)

Lymphochip

> 12, 000 genes

Aim. To investigate the possibility that variations in gene-expression patterns in allograft-biopsy samples from patients with acute rejection and related disorders could identify molecularly distinct subtypes of acute rejection to possibly explain differences in clinical behavior.

Design. (Pediatric patients) 50 patients. 67 biopsies: biopsies during acute or chronic allograft dysfunctions (n = 52) and at the time of the engraftment or when graft function was stable (n = 15). The possibility of different sampling of the medullary and the cortical regions was also addressed.

Statistics. Class discovery: unsupervised clustering analysis. Class comparison: significance analysis of microarray. Survival analysis: Kaplan-Meyer/Cox log-rank method. Biological explanation: enrichment of specific functional groups through evaluation of hypergeometric distribution. The exclusion of data from genes whose expression was correlated with the depth of biopsy did not change the cluster analysis.

Flechner et al. [23]

(Cleveland Dataset)

2004

Kidney

(biopsies and PBLs)

Affymetrix

HG-U95Av2

~ 10, 000 genes

Aim. To determine gene expression profiling in transplant patients including: normal donor kidneys, well functioning transplants without rejection, kidneys undergoing acute rejection, and transplants with renal dysfunction without rejection.

Design. (Adults) 23 graft recipients and 9 donors. Acute rejection biopsies (n = 7), renal dysfunction without rejection on biopsies (n = 6), biopsies carried out more than one year post transplant in patient with good transplant function and normal histology (n = 10), biopsies from living donor controls (n = 9). PBLs were also collected and profiled. Expression of some transcript was also assessed through qRT-PCR.

Statistics. Class discovery: unsupervised clustering analysis. Class comparison: significance analysis of microarray filtered with limit fold model and MAS 5.0 present/absent calls. Class-prediction: leave-one-out method. Biological explanation: analysis of functional classes of the differentially expressed genes.

Reeve et al. [24]

(Edmonton Dataset)

2009

Kidney (biopsies)

Affymetrix Human Genome U133 Plus 2.0

> 38, 000 genes

Aim. To define a classifier to distinguish rejection vs non rejection using predictive analysis for microarrays.

Design. (Adults) 143 patients, 186 biopsies: acute rejection samples (acute cellular rejection, antibody mediated rejection or mixed) (n = 51), non-rejection samples (n = 135).

Statistics. Class comparisons: Bayesian t-test and false discovery rate. Class prediction: prediction analysis of microarrays. Biological explanation: analysis of functional classes of the differentially expressed genes according to KEGG pathways and to authors' defined pathogenesis-based transcripts.

Morgun et al. [25]

(San Paulo Dataset)

2006

Heart (biopsies)

Qiagen/Operon Array

~ 14, 000 genes

Aim. To analyze gene expression differences between rejection, non rejection and Trypanosoma cruzi infection.

Design. (Adults) 40 patients, 76 biopsies (rejection, no rejection and Trypanosoma cruzi infection recurrence). Expression of some transcripts was also assessed through qRT-PCR.

Statistics. Class comparison: random variance t-test filtered with univariate/multivariate tests for false discovery rates; supervised clustering analysis. Class prediction: 6 different multivariate models models (compound covariate predictor, diagonal linear discriminant analysis, 1- and 3-nearest neighbor predictor, nearest centroid predictor, support vector machine) and leave-one-out cross validation. The authors validated the predictor-set in independent datasets of biopsies (collected on different continents and analyzed with different chip batches). The authors also tested the predictor set by analyzing the data from data from Cleveland (Kidney) Stanford (Kidney) and Minneapolis (Lung) datasets.

Biological explanation: Database Annotation, Visualization and Integrated Discovery (DAVID)/Gene Ontology and KEGG Pathways.

Saint-Mezard et al. [26]

(Paris Dataset)

2008

Kidney (biopsies)

Affymetrix Human Genome U133 Plus 2.0

> 38, 000 genes

Aim. To identify a robust and reliable molecular signature for acute rejection in humans.

Design. (Adults) 45 patients, 47 biopsies: acute rejection (n = 8), acute rejection and chronic allograft nephropathy (n = 8), borderline (n = 3), non rejection (n = 7), and chronic allograft nephropathy (n = 22). Normal kidney tissue was obtained from histopathologically unaffected areas of the cortex of native nephrectomies performed for renal carcinoma was used as control.

Statistics. Genes differentially expressed (Paris Dataset) were intersected with those from with 2 public human datasets: 1) Stanford Dataset and 2) Cleveland Dataset and with one Non Human Primate (NHP) model of acute renal allograft. However, the authors used biopsy microarray data from Edmonton Dataset as in independent confirmation set. Score from the identified classifier was correlated with the histopathological Banff score. Expression of some transcripts was also assessed through qRT-PCR.

Class comparison: ANOVA with or without false discovery rate and additional cutoff based on twofold change. Class discovery: Principal component analysis, supervised clustering analysis (using the genes differentially expressed in all four datasets); Class prediction: leave-one-out cross-validation and 10-fold cross-validation. Biological explanation: Gene regulatory networks were generated using MetaCore.

Rodder et al. [29]

(Tenon/Inelspital Dataset)

2011

Kidney (biopsies)

Affymetrix Human Genome U133 Plus 2.0

> 38, 000 genes

Aim. To identify the expression of metzincins and related genes in allograft rejection biopsies.

Design. (Adults) 41 biopsies: normal histology (n = 20), borderline changes (n = 4), acute rejection (n = 10) and acute rejection and interstitial fibrosis/tubular atrophy (n = 7). Expression of some transcripts was also assessed through qRT-PCR.

Statistics. Class prediction: ANOVA and shrinking centroids methods were used for variable selection and a variety of classification methods were tested. Leave-one-out method was performed as internal cross-validation. Classifier performance was estimated as correct rate after 1-level cross validation. The model was validated in Edmonton, Cleveland and Stanford datasets. Gene set scores from biopsies were also determined and correlated with Banff scores.

Chen et al. [27]

(Stanford2 Dataset)

2011

Kidney (biopsies)

Affymetrix Human Genome U133 Plus 2.0

> 38, 000 genes

Aim. To identify biomarkers across similar conditions through integration of related datasets.

Methods. (Pediatric patients) 36 patients and biopsies: acute rejection biopsies (n = 18), stable function biopsies (n = 18).

Statistics. Class comparison: significance analysis of microarrays and fold change filter. The upregulated genes during acute rejection were intersected with genes upregulated during acute rejection in two other datasets (Cleveland and San Paulo).