Genomic Profiles, Allograft Dysfunction, and Liver Transplant
Genomic Profiles, Allograft Dysfunction, and Liver Transplant
Using published clinical criteria for EAD we identified a total of 40 deceased donor graft EAD recipients, and compared them to 36 recipients without EAD, matched for age, gender, donor and/or recipient HCV status and MELD score. All patients were transplanted between 2005 and 2010 at the Hospital of the University of Pennsylvania and provided written informed consent for this study. All protocols and consents were approved by the Institutional Review Board of the University.
Clinical data were collected from the electronic transplant clinical database including donor and recipient demographics, intra-operative details, and postoperative liver function. Cold ischemic time of the liver graft was defined as time from donor cross-clamp to removal from ice prior to placement into the recipient. Warm ischemic time was defined as time from removal from ice to the first of either arterial or portal reperfusion.
Two core liver biopsy specimens were obtained from each deceased donor graft at the time of transplantation. The first specimen was taken on the backbench, after cold storage (COLD) and prior to implantation. A subsequent biopsy was taken approximately 1 hour after reperfusion, following completion of the biliary anastomosis and prior to closure of the abdomen (POST). Biopsies were collected in RNAlater (Qiagen, CA) and stored at −80°C until further processing.
Total RNA was extracted using Trizol (Invitrogen, CA) and purified using the RNeasy kit (Qiagen, CA), according to the manufacturer's instructions. Microarrays to investigate global gene expression profiles were performed in 30 EAD and 26 non-EAD patients using Affymetrix Human Genome U133 Plus 2.0 GeneChips (Affymetrix, Santa Clara, CA) and standard protocols.
We validated selected genes from the microarray signatures on the Panomics Quantigene Platform (Affymetrix, Santa Clara, CA) using a Luminex-100 bead-based multiplex assay. These assays represent a non-PCR based technology that eliminates amplification bias and allows multiplexing (36 genes/well). Panomics validations were done on an independently collected cohort of 10 EAD and 10 non-EAD patients supplemented with 5 EAD and 5 non-EAD patients taken from the microarray cohort. Signals were normalized against the geometric mean of the average expression values for three housekeeping genes: HPRT1, PPIB, and ACTB.
Donor characteristics and clinical data are shown (wherever applicable) as either median with range or mean ± SD (Table 1). Univariate clinical data analysis was performed using the Mann–Whitney test and Fisher's exact test (p-values <0.05, significant) (Graphpad Prism software).
Microarray signal intensities from the Affymetrix Human Genome U133 Plus 2.0 arrays (54,675 probesets) were normalized using RMAExpress. Partek Genomics Suite 6.6 and BRB ArrayTools were used for statistical analyses. Study gene expression data are posted at Gene Expression Omnibus (GEO) http://www.ncbi.nlm.nih.gov/geo.
Power calculations were based on detecting a desired fold change of ≥1.5 with a <20% false discovery rate, desired power of 80% and standard deviation of 0.7. This gave us an optimal sample size of 15 samples per group with an alpha value of 0.4846.
Class comparisons between EAD and non-EAD samples were performed with an ANOVA model using Method of Moments at Bonferroni-corrected p-values of <0.001 (diagnostic) and <0.005 (biology). Contrasts between the groups were done using Fisher's Least Significant Difference. Functional analyses were performed using Ingenuity Pathway Analysis (Ingenuity Systems, Redwood City, CA).
Correlations between the array data and clinical study parameters were tested using a multivariate logistic regression model with an adjusted (Wald test) p-value and a local false discovery rate calculation (q-value). http://labs.genetics.ucla.edu/horvath/htdocs/CoexpressionNetwork/tmp/LiverTransplant/UPennStudy-Results.zip
For the predictive signature the Nearest Centroids (NC) and Diagonal Linear Discriminant Analysis (DLDA) algorithms were used. Receiver Operating Characteristic (ROC) curves were constructed using pROC in R and the Area Under the Curve (AUC) was calculated using the formula: ([True Positives] TP/TP + [False Negatives] FN + [True Negatives] TN/[False Positives] FP + TN)/2.
Since such a signature can be subject to statistical over-fitting that would inflate the claimed predictive results, it was tested using Harrell's optimism correction bootstrapping. For the Panomics assays, fold changes between normalized gene expression data of COLD and POST biopsies were determined using the 2 method.
Materials and Methods
Patient Population and Clinical Data
Using published clinical criteria for EAD we identified a total of 40 deceased donor graft EAD recipients, and compared them to 36 recipients without EAD, matched for age, gender, donor and/or recipient HCV status and MELD score. All patients were transplanted between 2005 and 2010 at the Hospital of the University of Pennsylvania and provided written informed consent for this study. All protocols and consents were approved by the Institutional Review Board of the University.
Clinical data were collected from the electronic transplant clinical database including donor and recipient demographics, intra-operative details, and postoperative liver function. Cold ischemic time of the liver graft was defined as time from donor cross-clamp to removal from ice prior to placement into the recipient. Warm ischemic time was defined as time from removal from ice to the first of either arterial or portal reperfusion.
Liver Biopsies
Two core liver biopsy specimens were obtained from each deceased donor graft at the time of transplantation. The first specimen was taken on the backbench, after cold storage (COLD) and prior to implantation. A subsequent biopsy was taken approximately 1 hour after reperfusion, following completion of the biliary anastomosis and prior to closure of the abdomen (POST). Biopsies were collected in RNAlater (Qiagen, CA) and stored at −80°C until further processing.
RNA Preparation and Gene Expression Detection
Total RNA was extracted using Trizol (Invitrogen, CA) and purified using the RNeasy kit (Qiagen, CA), according to the manufacturer's instructions. Microarrays to investigate global gene expression profiles were performed in 30 EAD and 26 non-EAD patients using Affymetrix Human Genome U133 Plus 2.0 GeneChips (Affymetrix, Santa Clara, CA) and standard protocols.
We validated selected genes from the microarray signatures on the Panomics Quantigene Platform (Affymetrix, Santa Clara, CA) using a Luminex-100 bead-based multiplex assay. These assays represent a non-PCR based technology that eliminates amplification bias and allows multiplexing (36 genes/well). Panomics validations were done on an independently collected cohort of 10 EAD and 10 non-EAD patients supplemented with 5 EAD and 5 non-EAD patients taken from the microarray cohort. Signals were normalized against the geometric mean of the average expression values for three housekeeping genes: HPRT1, PPIB, and ACTB.
Statistical Analysis
Donor characteristics and clinical data are shown (wherever applicable) as either median with range or mean ± SD (Table 1). Univariate clinical data analysis was performed using the Mann–Whitney test and Fisher's exact test (p-values <0.05, significant) (Graphpad Prism software).
Microarray signal intensities from the Affymetrix Human Genome U133 Plus 2.0 arrays (54,675 probesets) were normalized using RMAExpress. Partek Genomics Suite 6.6 and BRB ArrayTools were used for statistical analyses. Study gene expression data are posted at Gene Expression Omnibus (GEO) http://www.ncbi.nlm.nih.gov/geo.
Power calculations were based on detecting a desired fold change of ≥1.5 with a <20% false discovery rate, desired power of 80% and standard deviation of 0.7. This gave us an optimal sample size of 15 samples per group with an alpha value of 0.4846.
Class comparisons between EAD and non-EAD samples were performed with an ANOVA model using Method of Moments at Bonferroni-corrected p-values of <0.001 (diagnostic) and <0.005 (biology). Contrasts between the groups were done using Fisher's Least Significant Difference. Functional analyses were performed using Ingenuity Pathway Analysis (Ingenuity Systems, Redwood City, CA).
Correlations between the array data and clinical study parameters were tested using a multivariate logistic regression model with an adjusted (Wald test) p-value and a local false discovery rate calculation (q-value). http://labs.genetics.ucla.edu/horvath/htdocs/CoexpressionNetwork/tmp/LiverTransplant/UPennStudy-Results.zip
For the predictive signature the Nearest Centroids (NC) and Diagonal Linear Discriminant Analysis (DLDA) algorithms were used. Receiver Operating Characteristic (ROC) curves were constructed using pROC in R and the Area Under the Curve (AUC) was calculated using the formula: ([True Positives] TP/TP + [False Negatives] FN + [True Negatives] TN/[False Positives] FP + TN)/2.
Since such a signature can be subject to statistical over-fitting that would inflate the claimed predictive results, it was tested using Harrell's optimism correction bootstrapping. For the Panomics assays, fold changes between normalized gene expression data of COLD and POST biopsies were determined using the 2 method.
Source...