1 Departments of Medicine
2 Pharmacology and Physiology
3 Neurology
4 Center for Aging and Developmental Biology, University of Rochester, Rochester, New York 14642
![]() |
ABSTRACT |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
SAGE; serial analysis of gene expression; microarray; aging
![]() |
INTRODUCTION |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
![]() |
METHODS |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
SAGE.
The effect of age on gene expression in human muscle was determined by SAGE as previously reported (26, 28, 29). The RNA for these analyses was extracted from muscle of eight healthy young men (2124 yr old, mean 23 yr) and eight healthy older men (6677 yr old, mean 71 yr). Muscle was obtained by needle biopsy of vastus lateralis under standardized conditions, which included avoidance of any strenuous activity for 3 days before the biopsy and an overnight stay at the Clinical Research Center before the biopsy.
The SAGE method and some of the SAGE results have been published (29). With this method, inventories of short (14-base) ESTs were generated from cDNA produced from two RNA pools: one from the eight younger subjects and one from the eight older subjects. The abundance of any particular gene transcript in the sample is reflected by the number of times its corresponding SAGE tag is detected. The total number of SAGE tags cataloged was similar for the pools of RNA obtained from younger (53,875 tags) and older subjects (53,853 tags). Analysis of several transcripts in individual samples by RT-PCR indicated that the standard deviations for mRNA levels were generally similar among younger and older subjects and that most of the differences suggested by SAGE could be confirmed (29). The SAGE database also has been validated by its similarity to other databases in which the abundance of longer ESTs from muscle libraries has been cataloged (3, 28). The SAGE data are available at http://www.urmc.rochester.edu/smd/crc/swindex.html.
The UniGene summary page provides a link to a program that identifies the SAGE tags associated with the sequences in the UniGene cluster. Most of the SAGE tags used in the present analysis are based on the consensus mRNA sequence, but when this information was not available, the tags predicted by a high proportion of the ESTs were used. Often, several potential SAGE tags were identified. There are several reasons for having several different SAGE tags in a UniGene cluster, including incomplete or erroneous sequences, alternative splicing, alternative polyadenylation sites, or polymorphisms. In the Supplemental APPENDIX, we have presented the SAGE tag most likely to represent each mRNA, in some cases followed by other potential tags (to view the APPENDIX please refer to the Supplemental Material1 for this article, published online at the Physiological Genomics web site).
Analysis of human RNA with the Affymetrix HG-U95A microarray.
The same RNA pools that were used to generate the SAGE inventories were subjected to analysis by the Affymetrix (Santa Clara, CA) HG-U95A high-density oligonucleotide array in the University of Rochester Microarray Core Facility. This array has 12,000 probe sets corresponding to full-length human cDNAs. Each gene is represented by 1620 pairs of 25-mer oligonucleotides that span the coding region. Sequence information for probe synthesis is derived from the UniGene database. Each probe pair consists of a perfect match sequence that is complementary to the cRNA target and a sequence that has a mismatch in a region critical for target hybridization. The mismatched oligonucleotide serves as a control for nonspecific hybridization. Biotinylated target was generated from the RNA samples as specified by Affymetrix. Quantitation of target hybridization was performed with a GeneArray Scanner (Hewlett-Packard/Affymetrix). The arrays were scanned before and after antibody amplification to address potential issues related to the dynamic range of the scanner. Average abundance of the transcripts represented on the array was scaled to the same nominal value for both samples, and the scaling factors were very close (within 3%) for the arrays used in this study. By comparison with the actual transcript concentrations as previously determined by SAGE, there was a ceiling effect for transcripts with a concentration >300 copies per myonucleus (100 SAGE tags detected) for the scan done without antibody amplification and >60 copies per myonucleus (20 SAGE tags detected) for the scan done with antibody amplification. Thus we used the low-sensitivity scan only for transcripts that produced more than 20 SAGE tags in either sample. Only two transcripts pertinent to the present analysis were too abundant for reliable analysis even with the low-sensitivity scan (i.e., >100 SAGE tags), and neither of these is represented on the HG-U95A array.
The Microarray Analysis Suite (Affymetrix) generated the comparative analysis. Distinct algorithms made an absolute call (presence/absence for each transcript), a decision about differential expression between samples [no change (NC in the Supplemental APPENDIX), marginal increase (MI), marginal decrease (MD), increase (I), or decrease (D)], and the magnitude of the change (represented as fold change, with a value of 1 indicating no difference). The difference decision and fold change estimates are given in the APPENDIX. The mathematical definitions for each of the algorithms are described in the Affymetrix Microarray Suite Users Guide, Version 4.0. The algorithm for the absolute call excluded outliers (3 standard deviations or more from mean) when averaging differences between perfect match and mismatch probe pairs, but the fold change estimate is based only on probe pairs used for both samples. Occasionally, an increase or decrease in expression is indicated even when the transcript is listed as being absent in both samples, and in these cases we have simply indicated "absent both" in the APPENDIX. In some cases a UniGene cluster is represented by more than one probe set, and these are presented separately in the Supplemental APPENDIX.
Defining concordance or discordance between the murine and human data.
The goal of this analysis was not to make final decisions about which genes are differentially expressed in young and old muscle of both mice and men, but rather to evaluate the overall degree of consistency between the results reported for mice and the results observed in humans. Thus we did not use a stringent P value to define concordance. SAGE results were considered to be consistent with the mouse data if there was a difference in tag counts in the direction predicted by the mouse study and if the probability that the difference in tag counts was due to sampling error was less than 20%. In the case of tags occurring at least five times in the inventories of both young and old muscle, the probability that a difference in SAGE tag counts of the observed magnitude and direction was caused by sampling error was computed from a standard test of the difference between two proportions (9). The Yates continuity correction was used with this test. When tags are detected less frequently, this test can be inaccurate, so the probability that the difference was the result of sampling error was computed as follows. According to the Poisson distribution, P(x) = e-yyx/x!, where x is the number of times a particular tag was detected among the 53,850 tags cataloged for each age group in the SAGE study, and y is the actual frequency per 53,850 tags in the pool from which tags were sampled (1). The value of y is unknown, but under the null hypothesis yyoung = yold. The best initial estimate for y is (xyoung + xold)/2. Several values around this initial estimate (in steps of 0.2) also were used to compute P(xyoung) and P(xold). Decisions were made based on the value for y that resulted in the highest probability that the difference in SAGE tag counts between young and old was caused by sampling error. The chance that x will be n or smaller, P(x n), is P(x = 0) + P(x = 1) + ... + P(x = n). The chance that x will be m or larger, P(x
m), is 1 - P(x = 0) - P(x = 1) - ... - P(x = m - 1). Finally, P(xyoung
n and xold
m) = P(x
n) x P(x
m). For example, suppose that a particular gene is overexpressed in old muscle according to the data obtained in mice. We do not detect its SAGE tag in young human muscle, but we detect its SAGE tag in old muscle two times. Is this tag really more abundant in the sample from older muscle, or is there a good chance that this result would occur even if the true abundance were 1 per 53,850 tags in both samples? If yyoung = yold = 1, P(xyoung = 0) = 0.37, P(xold
2) = 0.26, and P(xyoung = 0 and xold
2) = 0.10. No other value for y results in a larger value for P(xyoung = 0 and xold
2). Thus, in this case, we would conclude that the SAGE data are consistent with the mouse microarray data.
SAGE results were considered to be inconsistent with the mouse data when two conditions were met. The first condition was that there was no difference (less than 1.25-fold change) in tag counts in the inventories from young and old muscle or a difference opposite in direction from what was expected from the mouse data. The other condition was that this apparent discordance was unlikely (<20% chance) to happen, when in reality there was a 1.5-fold or more difference, in the predicted direction, in the abundance of the tag in pools from young and old muscle. We used a 1.5-fold difference for this computation, even though the age-related differences in mice were 1.7-fold, because we were only looking for similar trends in mice and men and because most mean differences in mRNA concentrations
1.5-fold between young and old human muscle are statistically significant when individual RNA samples are analyzed (29). The probability of obtaining a result inconsistent with the mouse data because of sampling error was calculated as described in the preceding paragraph, except for modifications required to estimate the probability of a 1.5-fold or more difference in y (in the predicted direction). For example, suppose that a SAGE tag corresponding to a gene overexpressed in old murine muscle was detected four times in young muscle and only three times in old muscle, an apparent inconsistency with the mouse data. Are these data inconsistent with 1.5-fold or greater overexpression in older human muscle, or is there a good chance that this result would occur even if there actually is overexpression in old human muscle? We tested several sets of possible values for y, in which yold = (1.5 x yyoung), in increments of 0.2 for yyoung, to find the one value yielding the highest probability that the inconsistency between the mouse and human data were caused by sampling error. The highest probability, 0.12, was associated with yold = 4.2 and yyoung = 2.8. Thus we conclude that, given the observed data, it is unlikely that the gene is overexpressed by 1.5-fold or more in old human muscle.
To confirm that the microarray method and SAGE produce the same results when the same samples are analyzed, we identified 20 transcripts for which the Affymetrix algorithm indicated an increase or decrease in expression in older human muscle (limited to transcripts for which the algorithm defined the transcript as present in the sample with higher expression) and for which we could make a decision about concordance or discordance with the SAGE database using the criteria described above. In 18 of 20 cases (90%) there was concordance between methods. This outcome is consistent with our goal of making the correct decision about concordance in more than 80% of the cases. (There were 59 transcripts that were either increased or decreased in older human muscle according to the Affymetrix method, but SAGE data allowed a concordance/discordance decision in only 20 cases. It is beyond the scope of the present report to discuss differences that were not also observed in the study of mice.)
The Affymetrix algorithm for detecting differential expression considers several parameters from the raw data, and the results of these analyses are presented in the Supplemental APPENDIX. However, this decision matrix often indicates that there is no difference between samples even when the best estimate for the magnitude of the difference is 1.5-fold or greater. As noted above, mean differences between young and old of 1.5-fold are often significantly different when individual samples are analyzed (29). Therefore, for the present analysis, we considered the human microarray data to be consistent with the murine microarray data if the difference in expression level between young and old human samples was 1.5-fold or more in the direction predicted by the murine data, even when the algorithm scored the outcome as "no change." If the difference in humans was 1.2- to 1.4-fold in the predicted direction, then no decision was made about concordance or discordance. If the difference was 1.1-fold or less, or was in the opposite direction from that predicted by the murine data, then the results from mice and men were considered discordant.
![]() |
RESULTS |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
The SAGE database had entries for 61 human genes that are homologous to murine genes differentially expressed in young and old muscle of mice. The microarray had probe sets for 70 human genes homologous to the ones differentially expressed in young and old muscle of mice. Based on one or the other method, or both, we categorized 49 results as being either concordant or discordant in mice and humans. We were not sufficiently certain of the other results to make a decision regarding concordance or discordance. The decisions were as follows: 17 results were similar in mice and men and 32 were dissimilar; 6 were classified as "overexpressed in both mice and men," 19 as "overexpressed in mice but not in men," 11 as "underexpressed in both mice and men," and 13 as "underexpressed in mice but not in men." Table 1 lists the genes for which aging appeared to have a similar effect on expression in mice and men. Table 2 lists those for which there was evidence that the effect of age on expression was not the same in mice and men.
|
|
The effects of age that were the largest in mice were no more likely than smaller effects to be observed in humans. The mean effect in mice was 2.4 ± 0.8-fold (mean and standard deviation) for the results concordant with the human data and 2.9 ± 0.9-fold for the results discordant with the human data.
In mice, 63% of the age-related changes in gene expression were either completely prevented or partially inhibited by caloric restriction. However, reversibility by caloric restriction was not a predictor of concordance in the effect of age in mice and men. The effect of age in mice was inhibited by caloric restriction in the case of 10 of the 17 genes (59%) listed in Table 1 (concordant results) and 26 of the 32 genes (81%) listed in Table 2 (discordant results).
![]() |
DISCUSSION |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Genes encoding proteins involved in glucose and energy metabolism tend to be underexpressed in older muscle in both mice and men (15, 29). This finding is consistent with phenotypic, histochemical, and biochemical data indicating a reduced capacity for energy metabolism in older muscle (4, 19, 22). In humans, the age-related reduction in transcripts encoding glycolytic enzymes might be related to a relative reduction in the mass of type 2 fibers (29). The reduced expression in older murine muscle of several genes encoding enzymes involved in protein turnover is consistent with the slower protein synthesis in older human muscle (24, 30). A few genes involved in reinnervation of muscle fibers were found to be overexpressed in older murine muscle, which is consistent with histological evidence for denervation-reinnervation cycles with aging of human muscle (17). Even though some of the gene expression data from mice are consistent with histological or biochemical human data, some of the specific genes involved in protein turnover and reinnervation that were differentially expressed in young and old murine muscle were not differentially expressed in young and old human muscle.
Lee et al. (15) listed nine stress response genes that were expressed at a higher level in old muscle than in young muscle of mice. They also noted that increased expression of another gene, sarcomeric mitochondrial creatine kinase, could reflect inactivation of its protein product by oxidative stress. We concluded that 5 of these 10 genes did not have increased expression in older human muscle. No concordance/discordance decision about the other five genes was made, but their expression was not detected in either young or old human muscle by either SAGE or by the microarray. Thus, even if there is some increase in expression of these genes in older human muscle, the absolute level of expression remains very low. Of course, this observation does not mean that accumulation of oxidative damage to macromolecules is not involved in the aging of human muscle, only that the ambient level of oxidative stress under basal conditions is not sufficient to trigger expression of these specific stress response genes.
There were 17 genes for which the effect of aging on expression appeared to be the same in mice and men (Table 1). It must be emphasized that Table 1 is a preliminary list, and that prospective studies are needed to confirm that aging alters the expression of these genes. One of the genes that appears to be more highly expressed in older muscle of both mice and humans, the transcription factor PEA3, has already been found to be expressed at a higher level in senescent muscle of rats (20). Pro-collagen I and III genes appear to be expressed at a lower level in older muscle of both mice and men, which is consistent with an age-related reduction in collagen I and III mRNA in myocardium of rats (25). Another of the genes expressed at a lower level in older muscle of both mice and humans, an ATP synthase subunit, was examined in individual human samples by quantitative RT-PCR (29). The age-related difference was highly significant according to this analysis, as were declines in senescent human muscle of several other transcripts encoding proteins involved in ATP synthesis and mitochondrial electron transport.
Many of the differences in gene expression between young and old murine muscle were not evident in human muscle. There are several possible reasons for the discrepancies other than the obvious one of species difference. These include false positives and false negatives in the databases, differences in the fiber type distributions or pattern of use in the particular muscles examined, the possibility that mice and men were examined at different relative biological ages, the effects of inbreeding on aging of the mice, more stringent selection for healthy aging in the human subjects, and limitations on physical activity imposed by caging the mice. These factors are discussed in the following paragraphs.
Some of the effects of age that were observed in mice might have been false positives that would not be confirmed with another group of mice or with an independent method. Lee et al. (15) reported differences of approximately twofold or more, based on the average of all nine individual comparisons between three mice in each age group. An independent method was not used to validate the microarray data, but in a later study (by the same group) age-related differences in gene expression in the brain were confirmed by quantitative RT-PCR for six of seven genes (16). However, the analysis of large gene array databases is a new field, and it is not clear how many of these differences would be verified in a prospective study.
There could be some "false negatives" in the human SAGE and microarray databases due to analytical imprecision. For the seven discordance decisions based only on SAGE, we were >90% confident that sampling error could not explain the discordance. For the 17 discordance decisions based only on the microarray method, the likelihood of a false negative result is low because strict criteria were used to define discordance. For the other eight discordance decisions, both SAGE and the microarray method were in agreement. Thus, for a few transcripts, we might have falsely concluded that aging did not have the same effect in mice and men. However, the number of such errors would be too small to alter the conclusion that discordant results are more frequent than concordant results when comparing the effect of age on expression of specific genes in mice and men.
The gastrocnemius muscle was studied in mice, and the vastus lateralis was studied in men. Although some of the results might have been different had the same muscle been studied in mice and men, we are most interested in effects that are evident in all muscles that suffer loss of mass and function in both humans and animal models. Nevertheless, it is worth noting some similarities and differences between these muscles both within and across species. Both human vastus lateralis and murine gastrocnemius have a mixture of different fiber types, and both are involved in the locomotion associated with routine daily activities. The aging of human gastrocnemius has been studied less extensively than the aging of vastus lateralis, but it appears to be similar with respect to selective type 2 fiber atrophy and fiber type grouping (4, 14). Although one study reported reduced mitochondrial enzyme activity (including citrate synthase) in gastrocnemius of older subjects (4), similar to what has been observed in vastus lateralis (22), another study indicated that the modest decline (-21%) in citrate synthase activity in gastrocnemius of older men was not evident in vastus lateralis (-11%) (13). There is very little quantitative information on the aging of gastrocnemius muscle in mice. Some effects of aging on this muscle in mice are similar to effects that have been observed in human vastus lateralis, including fiber atrophy, lipofuscin accumulation, and mitochondrial abnormalities (18). In rats, the gastrocnemius and vastus lateralis muscles appear to undergo similar atrophic changes with senescence (8), but such a comparison has not been reported for mice. In human vastus lateralis, loss of bulk with age is related both to loss of fibers and atrophy (17). In various muscles of mice, the relative contributions of fiber loss and fiber atrophy to the age-related sarcopenia are variable (5, 12, 23), but gastrocnemius and vastus lateralis have not been examined.
Lee at al. (15) studied 5-mo-old and 30-mo-old mice. If the average life span of C57BL/6 mice is 30 mo (15) and that of a man living in the United States is 72 yr (11), then a 5-mo-old mouse has lived 17% of its life and a 23-yr-old man (mean age of the young men in this study) has lived 32% of his. A 30-mo-old mouse and a 71-yr-old man (mean age of older men in this study) have lived about as long as they were expected to live at the time of their birth. However, other studies have suggested that the average life span of C57BL/6 mice is closer to 25 mo (10, 27). Thus the mice used in the microarray study might have been unusually old for this strain. Age as a fraction of the expected life span might not be the best indicator of the stage of biological senescence. Perhaps some of the age-related changes occur earlier or later in life (relative to mean life span) in mice than in men. Studies with a more detailed time course for postmaturational changes in gene expression, in both mice and men, would be required to address this issue.
Inbred mice might have some genetic characteristics influencing the aging process that would be rare in outbred mice or in other mammalian species. Such characteristics might influence gene expression in muscle only indirectly, for example, by altering behavior, hormone levels, or nonlethal pathologies. The overall health status of the 30-mo-old mice studied by Lee et al. (15) is unclear. The older men who donated muscle were carefully examined and found to be free of any significant acute or chronic medical problems or neuromuscular abnormalities. It must be emphasized that the selection of healthy older men does not result in a sample of subjects who have escaped the effects of aging on skeletal muscle. In our previous studies using the same selection criteria, we have documented age-related changes in muscle, including loss of mass and strength, slower protein synthesis, and an increased incidence of abnormal mitochondria (manifested as ragged red fibers) (21, 30, 31).
Physical activity must be considered in the interpretation gene expression in muscle. The confinement of mice in cages limits their physical activity. A lifetime of inactivity could influence the impact of aging on gene expression. It would be interesting to see if mice given access to running wheels or other devices to stimulate physical activity have the same changes in gene expression as mice maintained under standard conditions. The men who donated muscle for the SAGE study had no restrictions on their lifetime physical activity, but we excluded men with a recent history of a very high level of activity, and asked them to refrain from any strenuous activities for a few days before the biopsy. Old age often is associated with reduced physical activity (7), which might be more important than age per se in determining the expression of some genes.
Although the number of murine genes is not known, a rough estimate is that 10% of them were sampled by the microarray. Sampling
10% of the genes resulted in the discovery of
100 genes differentially expressed in young and old muscle. The present study suggests that fewer than half of these differences also occur in human muscle. Thus aging probably alters the expression of fewer than
500 genes in both species, assuming that a similar proportion of differentially expressed transcripts and a similar degree of concordance between species would be found among the unsampled genes. Identification of these genes may help to elucidate the etiology of sarcopenia and compensatory adaptations in old muscle.
![]() |
ACKNOWLEDGMENTS |
---|
This research was supported by National Institutes of Health Grants AG-17891, AG-10463, AG-13070, RR-00044, and AG-18254 and by a Paul B. Beeson Physician Faculty Scholar Award to C. A. Thornton from the Alliance for Aging Research.
![]() |
FOOTNOTES |
---|
Address for reprint requests and other correspondence: S. Welle, Univ. of Rochester Medical Center, Endocrinology, Box 693, 601 Elmwood Ave., Rochester, NY 14642 (E-mail: stephen_welle{at}urmc.rochester.edu).
1 Supplemental material (the APPENDIX) to this article is available online at http://physiolgenomics.physiology.org/cgi/content/full/5/2/67/DC1.
![]() |
REFERENCES |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|