1 Genetics Division and Genomics Program, Boston, Massachusetts 02115
2 Neurology Department, Childrens Hospital, Boston, Massachusetts 02115
3 Harvard Medical School, Boston, Massachusetts 02115
![]() |
ABSTRACT |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
autopsy; surgical; microarrays; gene expression
![]() |
INTRODUCTION |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
In animal models it has been shown that the postmortem interval, the storage temperature, and the administration of protease inhibitors shortly before death have a significant impact on postmortem modifications in muscle (38). Overall, electron microscopy of rat skeletal muscle at a storage temperature of around 20°C and without protease inhibitor treatment revealed the following: swollen mitochondria by 2 h postmortem; loss of ribosomes and glycogen, rupture of the I bands, and appearance of small membranous bodies by 12 h; and collapse of the I bands, loss of Z lines and appearance of large membranous bodies by 2448 h (11). At the same temperature, rat skeletal muscle myofibrillar protease activity increased linearly with postmortem time of storage, whereas creatine phosphokinase activity decreased linearly with time (27). Glycogen has been shown to degrade rapidly after death, whereas glucose and lactate concentrations increase (4). At the transcriptional level a small number of individual genes have been studied in autopsy of heart and/or skeletal muscle, predominantly in the context of heart disease (17, 35, 47). However, the sequence of transcriptional events that occur postmortem remains unclear.
Several microarray-based studies of gene expression in normal and diseased states have been performed on human skeletal muscles obtained at biopsy or autopsy (1, 8, 20, 21, 33) and on murine skeletal muscle obtained immediately postmortem (30, 32, 37, 40). These studies identified a number of genes in various functional categories that are differentially expressed in the disease states, suggesting that major alterations in tissue have far-reaching consequences for gene expression. Skeletal muscle tissue obtained at autopsy would be expected to have similarly extensive changes compared with tissue obtained at biopsy. However, no gene expression studies addressing this particular issue have been published to date, either in humans or mice.
We studied seven autopsy and seven surgical human skeletal muscle specimens using oligonucleotide Affymetrix GeneChip arrays (U95Av2) that contain representative sequences from over 12,000 known genes and expressed sequence tags (ESTs). The data sets from 12 of these 14 samples have been previously published as part of the normal control baseline in a study on nemaline myopathy (33). In that study, significant disease-related gene expression changes were identified, particularly in the glycolytic energy metabolism pathway (decreased), in connective tissue-related genes (increased), and in muscle satellite cells (increased). The aim of the current study was to understand gene expression changes that occur in normal human muscle postmortem and to relate these to histological reports and previous studies at the protein and metabolic levels. Remarkably, our findings provide evidence for a very active transcriptional phase in skeletal muscle during the first 46 h postmortem. Although these tissues are clearly subject to ischemia and hypoxia, genes for only some components of the ischemia and hypoxia-response pathways are transcriptionally upregulated under these conditions. These data provide an important baseline for future studies of diseased muscle obtained surgically and at various portmortem intervals.
![]() |
MATERIALS AND METHODS |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
|
Data analysis.
The data sets originating from the 14 specimens were first processed by the Affymetrix MicroArray Suite 5.0 software (MAS5.0), and signal values (reflecting expression levels) and "present/absent" calls (an Affymetrix computed measure representing confidence in gene expression presence) were computed for each probe set. The signal values were not normalized at this stage, although signal values were adjusted to an overall target intensity of 1,500. The correlation coefficients (r) were calculated for all 14 data set pair wise comparisons.
For hierarchical clustering, we used average linkage as previously described, with centered linear correlation as a measure of similarity (19). This approach clusters both specimens and probe sets on two different axes. The selected filters removed probe sets with an "absent" call across all 14 samples or a standard deviation (SD) 1,000. The SD filter is applied to facilitate the clustering of samples, by removing probe sets with very similar expression across all samples that are therefore likely to act as noise in the specific type of analysis. The resulting 2,116 probe sets were processed using the Cluster and TreeView Software (18).
A permutation analysis was performed to evaluate the robustness of the sample clustering. Hierarchical clustering was run seven additional times, each time removing one randomly chosen data set from each of the two main clusters that were originally identified (autopsy and surgical). Thus two different data sets were removed for each of the seven permutations.
Significance analysis of microarrays (SAM) was used to identify significant fold changes between autopsy and surgical specimens as described (41). Prior to this analysis all data sets were normalized to a slope of 1 with a reference data set. The reference data set was created by taking the average expression of each probe set between the two data sets (one autopsy and one surgical) with the highest average correlation coefficient to all other data sets in their group (i.e., within the autopsy or surgical sample group, respectively). A two-class unpaired data analysis was performed using a threshold of 1.72 [the "
" parameter as described by Tusher et al. (41), enables the user to examine the effect of the false-positive rate in determining significance] and a fold threshold of 2 [where "fold" is calculated as (average expression in autopsy specimens)/(average expression in surgical specimens)]. Probe sets were considered significantly changed in autopsy compared with surgical samples if they were selected at false discovery rate (FDR) cutoff of
1.66%.
The SAM analysis was repeated using only quadriceps-derived specimens (n = 9), to control for muscle type-dependent changes. The list of significantly changed genes in autopsy vs. surgical quadriceps samples was highly similar to the list obtained for all 14 specimens, with the same functional categories showing significant change (3% size variation between them) and
50% overlap between individual probe sets. The high correlation seen between specimens of different muscle types by hierarchical clustering (Fig. 1) and correlation coefficient analysis (r = 0.840.98), together with the high similarity in the significantly changed molecular pathways in all 14 specimens compared with the 9 quadriceps samples, suggest that the muscle types studied are directly comparable. Therefore, for the purpose of this study all 14 specimens were considered together.
|
Supplementary Tables S1 and S2, containing complete results of the SAM and the correlation coefficient analyses, are available online at the Physiological Genomics web site1 and at http://www.chb-genomics.org/beggslab/.
![]() |
RESULTS |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Hierarchical clustering identified two major sample clusters: one consisting predominantly of surgical and one of autopsy samples. However, some overlap was evident between the two sample types, as one member of each group clustered with the other groups (Fig. 1). This overlap was reproducible and seen in all seven dendrograms generated during the permutation analysis. The postmortem interval of the autopsy samples did not appear to have an impact on their correlation coefficients. Hierarchical clustering demonstrated that age, muscle type (see MATERIALS AND METHODS), and postmortem interval did not appear to have a major role, and therefore surgical and autopsy samples of all ages and types were considered together for the purpose of further analysis. It is of particular interest that autopsy specimens ranging from 16 to 46 h postmortem had such similar expression patterns, suggesting that further major transcriptional changes do not occur during the postmortem interval from 16 to 46 h. An independent calculation of the correlation coefficients showed a similar range of r values within each group (r = 0.830.98 for surgical and r = 0.860.96 for autopsy) and slightly lower across groups (r = 0.760.96), comparable to previous analyses of patient and control specimens (33).
The SAM program was used on the 14 normalized data sets to identify significant transcriptional differences between surgical and autopsy specimens. We identified 140 probe sets as significantly overexpressed (FDR 1.66%) in autopsy vs. surgical samples (Supplementary Table S1). None of the downregulated probe sets were considered significantly changed at this FDR cutoff. The fold change of the upregulated probe sets ranged between the selected lower cutoff of 2-fold [seen for the genes tubulin-specific chaperone E (TBCE); eukaryotic translation initiation factor 3, subunit 6 (EIF3S6); and S-adenosylmethionine decarboxylase 1 (AMD1)] and 27.33-fold [seen for the gene v-maf musculoaponeurotic fibrosarcoma oncogene homolog F (MAFF)]. The 140 upregulated probe sets were assigned to functional categories according to their Gene Ontology (GO) annotations. The four main categories included cell death (n = 11), metabolism (n = 35), cell maintenance (n = 46), and cell communication (n = 21) (Fig. 2). Because of the large size of certain subgroups within the four categories, some of these are presented as independent categories. In total seven functional categories were created: metabolism (n = 35 probe sets), transcription (n = 26), cell cycle (n = 23), cell communication (n = 13), apoptosis (n = 11), other (n = 22), and unknown (n = 15) (Supplementary Table S1). Of the seven functional categories, some had major distinct subgroups, which are represented in Fig. 2. Genes with more than one molecular function were placed under all relevant functional categories. The metabolism-related genes were predominantly involved in protein metabolism such as biosynthesis (n = 9), protein targeting/modification (n = 7) and proteolysis (n = 5), amine metabolism (n = 4), and nucleic acid metabolism (n = 4). Although GO usually classifies transcription under nucleic acid metabolism, we have created a separate group, due to their large number, to include all transcription-related probe sets. The majority of cell cycle-related genes were associated with cell cycle arrest (n = 6) and regulation (n = 7), and a small number were M- or S-phase related (n = 2 and 5, respectively). The cell cycle category had the highest average fold change (4-fold) compared with all other categories, with cell cycle regulation and arrest showing the highest fold changes in their category. Several cytoskeleton-related (n = 8) and metal ion binding probe sets (n = 5) were also significantly upregulated, but because of the small size of these groups they were classified under "Other".
|
|
![]() |
DISCUSSION |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Hierarchical clustering and correlation coefficient analysis enabled us to perform a primary global analysis of the 14 data sets. These analyses classified the specimens into autopsy and biopsy dominant clusters (Fig. 1), consistent with the slightly higher intragroup (r = 0.830.98) than intergroup (r = 0.790.96) coefficient values. However, two samples, T172 and T289, clustered in the "wrong" category. There were no features unique to these two samples that could account for their misclassification. T172 represented the oldest individual in the biopsy group, but several subjects in the autopsy group were older. T289 had the longest postmortem interval, but another sample was nearly as old. Both sample sites were quadriceps, the most common site from which muscle was obtained for this study.
Overall, the similarities within each class of specimens outweigh the similarities between classes. Although previous studies on bovine muscle have suggested that postmortem degradation of certain proteins was slower in older animals (22), age did not play a role in the clustering of our 14 specimens. A potential limitation of this study is the heterogeneity of muscle sampling sites. The scarcity of control muscle samples precluded an analysis of this size involving only one muscle type. However, the predominance of quadriceps samples and the clustering of all the nonquadriceps samples in the "correct" category (autopsy vs. biopsy) indicate that the heterogeneity of our samples did not have a significant impact on our results. Moreover, an independent analysis of quadriceps specimens alone yielded highly similar results (data not shown).
The most prominent and unexpected change in postmortem muscle was the overall upregulation in protein synthesis. Overexpression was seen in 26 transcription factors or transcription-related genes and an equal number of protein metabolism-related genes. The latter category included RNA binding proteins, ribosomal proteins, translation factors, amino acid/tRNA synthetases, and protein modification and targeting genes. This evidence indicates that despite expectations of an exclusively and rapid degenerative process, skeletal muscle goes through a very active transcriptional/translational phase during the first 46 h postmortem, most likely reflecting a cellular adaptive response to extracellular stimuli.
Specific transcriptional differences between the autopsy and surgical specimens were identified. The 140 probe sets with significant differences at an FDR of 1.66% were grouped into 7 functional categories based on GO annotations. The highest fold changes were seen in the cell cycle-related category, including genes involved in cell cycle regulation and cell cycle arrest (Supplementary Table S1). Overexpression of multiple apoptosis-related probe sets was also observed. Some examples include GADD45A (7.23-fold change), which participates in cell cycle arrest but also responds to environmental stress and leads to apoptosis via the p38/JNK pathway, and NFKBIA (2.24-fold), which can lead to apoptosis by forming a complex with p53 or inducing the activation of caspase-3 (5, 7, 36). Caspase-3 cleaves and activates the upregulated STK3 (seen upregulated by 2.18-fold), which in turn contributes to the apoptotic response (16). Another overexpressed gene, MYC (3.58-, 2.3-, and 2.22-fold, for probe sets 37724_at, 1973_s_at, and 1936_s_at, respectively), has been shown to regulate the switch from cytostatic to apoptotic p53-dependent cellular response, by inhibiting p53 from activating p21 (CIP1) (34). Overall, autopsy skeletal muscle displays alterations in several different cell cycle and apoptosis-related pathways in response to the multiple and complex cell environment changes.
As might be expected in light of the overall process, genes involved in proteolysis appear to be upregulated in autopsy specimens (Supplementary Table S1), a finding that contrasts with observations on injured or intermittently hypoxic muscle samples (9, 46). One of the multiple roles of proteolysis is the regulation of apoptosis (25), achieved by the proteolytic activation of a range of key apoptotic pathway members (Fig. 4)( 42). At least two of the apoptosis-related genes overexpressed in our analysis (STK3, 2.18-fold; and GADD45A, 7.23-fold) are known to undergo proteolytic activation (16, 26). Two additional proteolysis-associated genes (PA2G4, 2.08-fold; and PSMD8, 2.47-fold for 32584_at and 2.16-fold for 1312_at) are thought to be involved in cell cycle regulation. Furthermore, proteolysis is important for cellular homeostasis by maintaining protein concentrations at optimal levels. The upregulation of certain proteolytic pathways might therefore be expected given the apparent increase in protein biosynthesis that is suggested by the overexpression of nine related probe sets (fold change range 2.007.23) (Supplementary Table S1). Consistent with elevated protein synthesis, multiple protein modification-related and protein targeting-related genes were also overexpressed (fold change range 2.022.58). What appears to be transcriptional evidence for a potential increase in protein biosynthesis was accompanied by a widespread overexpression of transcription factors and other transcription regulatory genes. These 26 probe sets had fold changes in the range of 2.04 and 4.28. These data suggest that during the initial 46-h postmortem interval, human skeletal muscle undergoes a very active phase of increased transcription in what may be an aborted attempt to modulate translation and protein synthesis/modification/targeting in response to extrinsic (e.g., hypoxia) and intrinsic (e.g., initiation of apoptotic process) stimuli.
|
In association with this upregulation of cell cycle- and apoptotic-, proteolytic-, and antioxidant-related transcripts, two DNA repair genes (GADD45A, 7.23-fold; and RAD51, 3.90-fold) are also induced, suggesting a coordinated cellular response involving multiple pathways (Fig. 4). RAD51 has multiple interactions at the protein level with metabolic (especially nucleic acid metabolism) and cell cycle genes, as revealed by the interaction network analysis performed based on the Saccharomyces Genome Database (http://www.yeastgenome.org/) (Fig. 3). We hypothesize that the DNA damage induced postmortem triggers the expression of DNA repair genes such as RAD51, which in turn stimulates a range of intracellular cascades including cell cycle regulators (14). At least parts of these cascades are likely to be regulated at the protein level as suggested by the interaction networks and the absence of high correlation between the transcription patterns of these genes. Accordingly, 14 cell cycle regulation-, G1/S-, S-, and G2/M phase-related genes were overexpressed in autopsy skeletal muscle (fold range 2.04 to 27.33) (Supplementary Table S1), and multiple members of the CBF2 and SEC6 interaction networks have roles in the cell cycle (Fig. 3). Increased expression of cell cycle genes has also been described in relation to cardiotoxin-induced muscle injury (46). Additional gene expression changes in injured muscle have been reported for DNA replication factors, cyclins, Rb/E2f/DP, and myogenic transcription factors, as well as structural and receptor proteins (46). Five of these genes had a >1.5-fold change (CDKN1A, CHRNA1, IFG2, PEG3, VEGF, and CDKN1C), but none of them appeared significantly changed in our postmortem specimens.
Unlike previous studies on autopsy muscle specimens, no highly significant (for our thresholds) expression changes were seen for genes related to energy metabolism, muscle structure, Ca2+ homeostasis, or Ca2+-dependent proteases such as previously described at the histochemical level (2, 4, 27, 38, 45). There were also no significant changes in stress proteins (heat shock proteins), which are usually synthesized in response to muscle injury (10). This apparent disparity may be due in part to different postmortem time points used in these studies, differences between transcript and protein level/activation, as well as the highly stringent criteria applied to selection of "significantly changed" probe sets in the present study.
Intermittent and chronic hypoxia in skeletal muscle have been shown to affect cellular processes such as the increase of type 1 myosin heavy chain, the increased activity of tricarboxylic acid cycle and the decrease of lactate dehydrogenase (9, 31). Exercise under hypoxic conditions activated the hypoxia-inducible factor 1 (HIF1) and seemed to influence the expression of the vascular endothelial growth factor (VEGF), myoglobin (MB), and phosphofructokinase (PFK) (43). No significant gene expression changes were evident for these pathways in our analysis, but HIF1A and VEGF had a >1.6-fold increase, and phosphofructokinase in yeast has been shown to interact with the RAD51 protein, which was upregulated 3.9-fold in our analysis (Fig. 3). It is still unclear whether hypoxia-inducible factor 1 activity is regulated transcriptionally or posttranscriptionally, with one recent study describing its regulation by the ubiquitin-proteasome pathway and hence supporting the latter model (23, 44). Overall, postmortem skeletal muscle shares only some of the gene expression changes seen in living tissues responding to hypoxia.
In support of the observations above, the 15 probe sets of unknown molecular function identified as significantly overexpressed in our analysis had many high correlations with the 125 known genes (Supplementary Table S2), indicating their highly similar expression patterns across all 14 muscle specimens studied. Moreover, high correlations with additional members of the major affected pathways described are seen, such as transcription, translation, nucleic acid binding, and signal transduction.
Nucleic acids extracted from autopsy specimens are often used in clinical research either because of the nature of the study or lack of suitable biopsies. The present study demonstrates that mRNA remains sufficiently intact for microarray analysis in autopsy skeletal muscle up to 46 h postmortem and that no major transcriptional differences exist between specimens of <20-h and 34- to 46-h postmortem intervals. The list of genes significantly changed between autopsy and surgical specimens is a useful resource for interpreting future studies combining the two types of skeletal muscle specimens. As an example, in our previous study on nemaline myopathy, combining autopsy and surgical specimens, only one of the 206 (0.48%) genes identified there, namely glutamine synthase (GLUL), overlapped with the current list, supporting that all pathways identified were disease related and not due to the surgical/autopsy nature of the specimens (33). Extra care should be taken in those studies where postmortem changes may overlap with disease processes, thus making the distinction of the two a challenging task.
During the initial 46-h postmortem interval, human skeletal muscle undergoes an active transcriptional phase. Genes for proteins involved with transcriptional and translational pathways are transcriptionally induced, potentially leading to synthesis of multiple new proteins. These data suggest activation of several pathways involving protein modification, targeting, and transport. Postmortem muscle also upregulates expression of cell protective (e.g., metallothioneins), cytostatic (e.g., cell cycle regulation- and arrest-related genes) and proteolysis-related genes. The observed overexpression of several apoptosis-related genes suggests the initiation of this process; however, the absence of other key apoptotic markers among the overexpressed genes suggests this process deviates from the physiological apoptotic process seen during life. Finally, although there is some overlap with injured and hypoxic tissue, the specific combination of gene expression changes identified in our analysis of postmortem muscle tissue appears largely to have a distinct profile.
![]() |
ACKNOWLEDGMENTS |
---|
GRANTS
Microarray experiments were conducted in the Childrens Hospital Gene Expression Core Laboratory supported by National Institutes of Health (NIH) Grant NS-40828. This work was also funded in part by NIH Grant AR-44345, Projects 1 and 3 of NS-40828, and generous support from the Lee and Penny Anderson Family Foundation, the Joshua Frase Foundation, and the Muscular Dystrophy Association of the USA. Louis M. Kunkel is an Investigator at the Howard Hughes Medical Institute.
![]() |
FOOTNOTES |
---|
Address for reprint requests and other correspondence: A. H. Beggs, Genetics Division, Childrens Hospital Boston, 300 Longwood Ave., Boston, MA 02115 (E-mail: beggs{at}enders.tch.harvard.edu).
10.1152/physiolgenomics.00137.2003.
1 The Supplementary Material for this article (Tables S1 and S2) is available online at http://physiolgenomics.physiology.org/cgi/content/full/00137.2003/DC1.
![]() |
REFERENCES |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
|
HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Visit Other APS Journals Online |