We obtained the expression profiles of 7 cardiac tissue samples from the cardiac transplant operation (explant/failed heart) and 13 from a rapid-autopsy program (autopsy/non-failed heart) harvested under the "cold-" and "warm-24 hour autolysis" conditions (Figure ). The left and right ventricular tissues showed highly correlated gene expression in both the heart samples (Pearson's coefficient, r > 0.94), and we therefore analyzed the heart regions together. To assess the expression array performance for each sample, we used the ROC AUC summarization score, which is a measure of overall sensitivity and specificity of the expression profile. The AUC scores for the 20 gene expression profiles harvested at increasing autolysis delay all ranged from 0.80 to 0.90, which is within the parameters suggested by Affymetrix [16
]. While RIN values are typically used to predict sample performance, we only observed a weak correlation between RIN values and ROC AUC (r2
= 0.27, Figure ), and note that even samples with RIN values as low as 2.7 displayed robust hybridization. The means of the RIN values in the "warm" and "cold" 24 hour autolysis samples were 5.1 (SD = 1.5) and 6.9 (SD = 1.5), respectively, though this difference was not statistically significant (P = 0.11) (Additional File 1
, Figure S1). We further observed that the RIN values were independent of the autolysis time (r2
= 0.002, Figure ). While this observation is based on a small dataset and is at odds with many, but not all, reports of RIN quality correlating with post mortem interval [21
], we note that in our analyses, RIN was not a strong predictor of array performance.
Figure 2 RNA quality and hybridization performance. The RIN number, a measure of RNA quality, ranged from 2.7-8.6 in the heart samples. The ROC AUC, a hybridization quality metric, ranged from 0.80 to 0.90, which is within the parameters suggested by Affymetrix. (more ...)
Autolysis interval and RNA quality. The autolysis interval, which ranged from 0 to 24 hours, did not correlate with RNA integrity, measured as RIN numbers (r2 = 0.002).
Autolysis Stability of Gene Expression
As an additional general test of expression array performance with autolysed tissue, we used principal component analysis (PCA) of gene intensities to cluster the 20 autopsy and explant heart samples, along with Hela controls run in triplicate, as well as reference liver and heart data obtained from Affymetrix. Irrespective of the time and temperature of tissue harvest, the expression profiles of the two hearts could clearly be differentiated from non-heart tissues, as observed by the PCA analysis of ~15,000 gene intensities (Figure ). In Figure , we demonstrate that there is a strong correlation between the baseline explant and autopsy hearts compared to their later timepoints (Pearson's correlation ranges from 0.92-0.98). As a comparison, the Pearson's correlation(r), between any two HeLa controls (technical replicates) was 0.98. Additionally, the Pearson's coefficient for HeLa controls and reference liver with respect to a high RIN heart (RIN = 8.6) was 0.65 and 0.66, respectively.
Figure 4 Principal component analysis (PCA) of heart and non-heart gene expression intensities. A scatterplot of the PC1 and PC2 from the PCA analysis of gene intensities of ~15,000 transcripts from the 20 autopsy and explant heart samples, along with reference (more ...)
Figure 5 Correlation between baseline and autolyzed gene expression. a and b. Scatterplots of expression intensity later time-points (x-axis) and the 0 hour time-point (y-axis) for (a) autopsy and (b) explant heart tissues. Red lines represent the linear regression (more ...)
The concordance in expression profiles in autopsy tissue was also investigated by pair-wise comparisons of all cardiac expressed genes, expressed as the fold change between time-points, (Figure ). Specifically, 8,400 genes expressed in the cardiac tissue were retrieved from GNF Expression Atlas 2, which summarizes the expression patterns of human, mouse and rat genes in several selected tissues using whole-genome microarray experiments. To establish a baseline level of expected concordance for technical replicates, we compared expression profile of two HeLa controls, and observed 94.9% of genes with ≤1.25 fold change in expression (red line, Figure ). The baseline explant and autopsy and their later timepoints, showed similar concordance in expression and we saw no difference from technical replicates (95.9% and 93.4% genes with ≤1.25 fold change, respectively) (purple and brown lines, Figure ). We observed a difference between the autopsy and explant expression profiles, with only 86.8% genes with ≤1.25 fold change (black line, Figure ). Comparing, either heart sample to the HeLa controls demonstrated a striking difference in expression, with only 48% with ≤1.25 fold change (green and blue lines, Figure ). Thus, global gene expression analysis suggests that RNA expression is reproducible over 24 hours of autolysis and we could distinguish between heart and non-heart samples.
Figure 6 Concordance in expression profiles in heart tissues. Pair-wise comparisons of all cardiac expressed genes (~8,400 based on the GNF anatomical system data: heart), expressed as the fold change between time-points. To establish a baseline level of expected (more ...)
Autolysis Fluctuations of Gene Expression
After, establishing that global expression is largely unaffected by autolysis interval, we looked at the small fraction of genes that were sensitive to autolysis conditions. We selected transcripts with > 2 fold-change (2-FC) between the baseline and later timepoints from each heart to evaluate the change in gene expression during the 24 hours of autolysis in a time dependent manner. For this analysis, we focused on the "cold-24 h autolysis", as this represents a typical autopsy scenario. Only 2.25% (345/15320) of transcripts from the autopsy hearts had a > 2-FC when we compared the transcript levels from the 4 later time-points (6, 12, 18, 24) to the baseline levels. We queried these genes in the DAVID annotation database to find significantly overrepresented GO terms. Table , provides the functional categories overrepresented with a > 2-fold enrichment, Benjamini-Hochberg p-value < 0.05 and the broader parent GO Slim term. There was significant overrepresentation of GO terms which associated with two GO Slim categories 1) "cellular metabolism" (76/354) and 2) "response to stimulus" (42/354).
Annotations overrepresented in 2.25% genes fluctuating in the autopsy (non-failed) heart
We performed a similar experiment with the explanted heart, to see which categories of genes fluctuate in response to autolysis under the same conditions. One percent (154/15320) of transcripts from the explanted hearts had a > 2-FC, when we compared the fold changes in the transcripts from the 3 later time-points (6, 12, 18) to the baseline. Significantly overrepresented functional categories from the DAVID annotation database and the associated broader GO Slim parent terms are shown in Table . In the explant heart the significant enrichment of GO terms associated with the "response to stimulus" category was observed annotating 44/154 genes. Three other significant functional categories were mapped to "circulatory system process" (8/154), "development" (15/154) and "cell differentiation" (27/154). Thus, the broad parent GO Slim term "response to stimulus" was significantly overrepresented in both the autopsy and explant hearts, including 21 genes in common (Additional File 2
, Table S1), in this analysis of variation in gene expression during 24 hours of autolysis.
Annotations overrepresented in 1% genes fluctuating in the explant (failed) heart
Gene Expression Profile for a Failing Heart
Having established that the global expression from the individual cardiac tissues is reproducible during 24 hours of autolysis, we next compared the expression profile of the ventricular cardiac tissue from the explant (failing heart) and the autopsy (non-failing heart) by using SAM's two class unpaired test statistic. We focused on the "cold-24 hour autolysis" heart samples, as representative of real-world autopsy conditions. Gene expression profiles from 8 autopsy and 5 explant heart samples were analyzed with the SAM technique, which identified ~480 differentially expressed (DE) transcripts at > 2-FC and q-values < 0.5. The heatmap in Figure demonstrates ~480 transcripts with > 2-FC expression between the two groups with 374 up-regulated and 108 down-regulated transcripts in the explant/failed heart. The list of ~480 genes differentially up- and down-regulated between the explant and autopsy hearts are given in the Additional File 3
, Table S2. The results were unchanged after exclusion of the samples harvested at the baseline 0 hour time-point.
Differential gene expression. Heatmap of transcripts with > 2 fold change in expression between the failing and non-failing heart; 374 genes are up-regulated and 108 down-regulated. (red = up-regulated, and green = down-regulated)
Clinical evidence clearly demonstrates the dynamic nature of the components of the extracellular matrix in response to mechanical unloading of the failing heart [22
] and a quantitative increase in collagen subtypes [24
]. We identified ECM proteins collagen I alpha 1 (COL1A1
), collagen III alpha 1 (COL3A1
) and fibronectin (FN1
) to be upregulated > 2-FC in the failing heart. Importantly, we identified natriuretic peptide A (NPPA
) which was 11-fold higher in the failing heart. NPPA
is part of a conserved adaptive change in molecular phenotype in response to heart failure and serves as both a diagnostic and potential therapeutic marker [25
]. Recently, higher than normal levels of osteoglycin (OGN
, 12-fold increase in the failing heart) were associated with the heart becoming enlarged in rats and mice and in humans [27
]. Periostin, (POSTN
, 8-fold increase in the failing heart) has also been implicated in cardiac remodeling and therefore heart failure [28
]. Additionally, the expression of connective tissue growth factor (CTGF
, 4 fold change), transforming growth factor-beta (TGFB2
, 3.18 fold change), brain natriuretic peptide precursor (NPPB
, 2.29 fold change), Mu-crystallin (CRYM, 2.8
and clusterin (CLU
, 2.5 fold change), were all elevated in the failing heart, consistent with previous reports [29
]. The expression of Alpha-1-antichymotrypsin (SERPINA3
), a known protease inhibitor, responsible for degradation or disassembly of myocardial proteins, was down regulated in the failing heart when compared to the non-failing heart (8 fold decrease), which is consistent with the report by Yang et al [35
We also queried with the DAVID database to look for enrichment of functions from GO biological processes represented by the DE genes. The significantly over-represented functional annotations (> 2-fold enrichment and Benjamini-Hochberg p-value < 0.05) and the associated GO Slim terms are listed in Tables and . Seventy-nine of the 374 up-regulated genes in the failing heart were annotated into overlapping overrepresented GO functional categories, with the associated GO slim terms "cellular metabolism" and "response to stimulus" annotated for 29/374 and 23/374 genes, respectively (Table ). Fifty-nine of the 108 down-regulated genes belonged to overlapping functionally overrepresented GO categories which related to parent GO Slim terms "response to stimulus" (49/108), "regulation of biological quality" (23/108), "cellular process" (9/108), and "transport" (5/108) (Table ). Given that we observed an enrichment of genes involved in "response to stimulus" in the analysis of genes that fluctuate during autolysis, we investigated whether the ~480 DE genes between the failing and non-failing heart included genes identified as fluctuating during autolysis. We observed that 13% of the DE genes showed temporal fluctuation in the autopsy hearts, while 9% showed fluctuation in the explants, including important marker genes like NPPA, NPPB and POSTN. We explored how fluctuation of these genes during autolysis influenced the comparison of gene expression between failed and non-failed heart. In all three cases, these genes were clearly able to distinguish failed from non-failed heart at all time points, with the exception of NPPB at the 12 hr timepoint (Figure ). Thus, despite > 2FC fluctuations in temporal expression in some important genes due to autolysis, it is still possible to clearly identify specific and relevant differentially expressed genes in autolyzed cardiac tissue.
Annotations overrepresented in the up-regulated genes in the explant (failed) heart
Annotations overrepresented in the down-regulated genes in the explant (failed) heart
Differentially expressed and fluctuating genes. Expression profile of (a) NPPA, (b) NPPB and (c) POSTN genes in the failing and non-failing hearts. The genes are differentially expressed and show fluctuation in expression in the failed heart.