Comparative analysis of multiple microarray datasets may provide insight not obtainable through analysis of any one dataset individually. This study presented a side-by-side comparison of microarray datasets generated from more than ten different studies of mouse liver organs, along with an analysis of each dataset using a consistent statistical methodology. The results provide a comprehensive view of how the effects of life-extending dwarf mutations, caloric restriction (CR), CR mimetic compounds, low fat diets, gender and aging are related at the level of gene expression. Comparisons among these various effects, based on expression patterns of more than 8500 genes, are of interest from multiple perspectives and shed light on a range of issues related to mammalian aging [e.g., see discussions in [16
]]. A total of 43 candidate longevity-associated genes were identified based upon common differential expression patterns among four long-lived dwarf models, or between dwarf models and CR treatments. Each gene was identified across multiple independent experiments, and is therefore very unlikely to be a false positive identification. The set of longevity-associated genes identified in this analysis therefore provides well-supported targets for future experimental investigation as well as potential clues for understanding the functional basis of lifespan extension in mammalian systems.
mutation was associated with much smaller transcriptional effects than mutations carried by Snell, Ames and Little mice. This result is surprising, particularly since GHR-KO and Little mice both carry mutations that specifically affect GH signaling. Little mice are GH deficient because the pituitary is unresponsive to GHRH, while GHR-KO mice have elevated GH levels and lack GH receptor. Transcriptional changes associated with both mutants should therefore reflect inhibition of GH signaling and, on this basis, it might be expected that the lit
mutations have similar effects on gene expression patterns. This was not the case, however, given that 500–600 genes were differentially expressed with respect to Little mice, while fewer than 50 genes were differentially expressed with respect to GHR-KO mice. This difference does not reflect sample size and statistical power disparities between experiments (see Table ), and moreover, agrees with an earlier study that found very few genes (10 of 2352) differentially expressed between GHR-KO mice and normal controls [23
]. This observation is also consistent with phenotypic data, which indicate that Little and GHR-KO mice exhibit dissimilar magnitudes of lifespan extension [9
], as well as differing developmental weight-gain patterns [15
]. It is possible that some disparity between differential expression signatures of GHR-KO and Little mice can be attributed to technical differences in sample preparation and array hybridizations. Another possibility is that the difference is due to levels of circulating growth hormone, which is nearly absent in Little mice and elevated in GHR-KO mice [9
]. This difference would influence gene expression patterns in liver if GH has systemic or local effects independent of the GH receptor. These considerations suggest that, in comparison to other dwarf mutants, GHR-KO mice may provide a more useful model for the purpose of elucidating the mechanisms of longevity extension in mammals. GHR-KO mice exhibit lifespan extension that is comparable to that associated with Snell, Ames and Little mice. However, since the downstream transcriptional effects of the GHR-KO mutation are much less extensive than those of other long-lived models, understanding how this mutation contributes to longevity extension may prove less difficult.
Dwarf mutations and CR both extend lifespan in mice, but the degree to which this effect is mediated by common mechanisms remains unclear [23
]. With regard to mouse liver tissue, this study reveals both similarities and differences between the effects of dwarf mutations and CR. The overall similarity between differential expression signatures associated with dwarf mutations and CR is significantly larger than expected on the basis of chance (Fig. ). This is reflected by the 33 genes that exhibit parallel differential expression patterns in most long-lived mutants and under at least one CR treatment (Fig. ). On the other hand, results of this analysis suggest that dwarf mutations have a much larger and more consistent effect on IGF-I expression than CR treatments (as well as potential CR mimetic compounds). This result was surprising, since several studies have shown that IGF-I expression and protein levels decline under CR [36
]. It was further surprising that of the four CR treatments, IGF-I expression was only decreased significantly when CR was applied to dwarf mice (which already exhibit IGF-I deficiency). Taken together, these results suggest that life-extending dwarf mutations have mechanisms in common with CR, but that IGF-I suppression may not necessarily be included among these shared mechanisms. A limitation related to this inference is that the effects of CR treatments on expression patterns vary considerably among different studies, depending on the duration of caloric restriction, age at which necropsies are performed and the laboratory in which CR is carried out (see Figs. and ). A robust generalization regarding how the transcriptional effects of dwarf mutations relate to those of CR would therefore require even more data from CR mice than was analyzed in the present study.
Candidate longevity-associated genes were identified based on one of either two premises. The first premise is that lifespan extension in Snell, Ames, Little and GHR-KO mice is due to shared GH/IGF-I alterations, such that genes exhibiting parallel transcriptional changes in all four models are likely to play an important role in longevity. The second premise is that longevity extension in dwarf models and under CR is achieved through common mechanisms, suggesting that genes with similar expression changes in multiple dwarf models and under CR are potentially important for longevity. There is evidence in favor of both of these premises, and both have been endorsed in previous analyses [17
]. It should be noted, however, that since the functional basis of mammalian lifespan is not well understood, the validity of each premise has not been established with certainty. Furthermore, all data examined in this study were generated from liver tissue. Liver tissue is the primary manufacturing site of IGF-I, but it is nonetheless possible that expression patterns in other tissues are of equal or greater importance in determining longevity. For instance, fibroblasts of long-lived mice are more stress-resistant than those of normal controls, suggesting that dwarf mutations affect a wide range of tissue and cell types [10
]. In fact, reduced IGF-I levels in circulation may be the main factor behind extended lifespan in dwarf mice, such that with respect to liver, the only expression change consequential for longevity determination is that of the IGF-I transcript. These considerations should be weighed when evaluating the potential role of candidate genes in determining lifespan and rates of aging in mice and other mammalian species.
Only three genes were differentially expressed with respect to all four long-lived dwarf models with corresponding effects under at least one CR treatment (IGF-I, Igfals and Lifr). Both Igf1 and Igfals have a well-documented role in lifespan determination [39
], but a potential role of leukemia inhibitory factor receptor (Lifr) in longevity has not been widely explored. Lifr expression was downregulated at all ages in four long-lived dwarf models, by short-term CR and by low-fat diet. Determining whether Lifr downregulation in liver contributes to longevity extension represents a difficult task, since Lifr is highly pleiotropic and has been associated with a wide range of biological effects. While there are several molecules that Lifr may interact with [41
], Lifr expression is a primary determinant of cellular responsiveness to leukemia inhibitory factor (Lif) [44
]. The Lif glycoprotein is a member of the IL-6 type cytokine family and exhibits a wide range of effects among different cell types. At a systemic level, excess Lif has been associated with low body weight, hypermotility, overgrowth of bone, calcification of several organs, loss of spermatocytes in males and severe loss of adipose tissue [44
], and in the liver specifically, Lif has been found to stimulate triglyceride secretion [45
]. Interestingly, however, Lif(-/-)
mutants have been found to develop normally and possess generally good health, although pregnancy does not occur in females due to problems with blastocyst implantation [46
]. To some degree, inhibition of Lif/Lifr signaling may be countered in long-lived mice by down-regulation of Socs2 (suppressor of cytokine signaling) [47
], which was also down-regulated in all four long-lived dwarf models. Declines in Lifr expression may increase sensitivity to drug-induced liver disease [49
], which in general is contrary to the stress resistance characteristics of long-lived mice [11
], but is consistent with one study that found decreased resistance to acetaminophen toxicity in multiple dwarf strains [50
Fmo3 and Cyp2f2 were differentially expressed with respect to most (3/4) long-lived dwarf models and also under the majority (3/4) of CR treatments that were examined. In addition, the expression level of both genes varied with mean lifespan among 21 BxD recombinant inbred mouse strains in a direction consistent with differential expression analyses. Since both genes are monooxygenases with established roles in drug metabolism, their expression in liver could mediate the life-extending effects of dwarf-mutations and CR through similar mechanisms. Fmo3 is a flavin containing monoxygenase for which expression levels are elevated in long-lived dwarf mutants, female mice, CR mice and mice treated with each of three different CR mimetic compounds (metaformin, glipizide and rosiglitazone). The Fmo3 gene has received considerable attention in studies of human populations, since it exhibits extensive polymorphism among individuals and may be a factor promoting differential drug response [51
]. With regard to aging processes, a plausible hypothesis is that elevated Fmo3 levels in liver contributes to extended longevity by increasing stores of glutathione (GSH) and glutathione-S-transferase (GST), which then leads to enhanced resistance to oxidative stress. This hypothesis is appealing, since Fmo3 has an important role in sulfoxidation of methionine [52
], and it has been found that long-lived Ames mice exhibit elevated methionine metabolism [54
], which may underlie increased GSH, GST and oxidative stress resistance in long-lived models [54
Cyp2f2 is a cytochrome P450 enzyme, and in contrast to Fmo3, its expression was downregulated in long-lived and CR mice. Several other cytochrome P450 enzymes exhibited differential expression, with slightly less consistency, among multiple long-lived models and under CR treatments (Cyp4a14, Cyp2c38, Cyp2c39 and Cyp4a12b). Although Cyp2f2 displayed the most consistent differential expression pattern, it is unlikely that Cyp2f2 would have a unique role in mammalian aging apart from other cytochrome P450 enzymes. Cytochrome P450 enzymes are involved in xenobiotic detoxification and are catalysts for a large number of metabolic reactions. It is unclear how this system may influence aging in mammals, but multiple P450 enzymes are also differentially expressed in long-lived C. elegans daf-2
mutants, which has led to the suggestion that P450 enzymes slow aging by reducing damage generated from toxic compounds [55
]. From the standpoint of mammalian aging, it is of interest that some P450 enzymes, including Cyp2f2, are regulated by peroxisome proliferator-activated receptors (PPARs) [57
]. Cytochrome P450 enzymes may therefore be part of a broader cellular response that has previously been associated with longevity in long-lived and CR mice [59
]. An interesting pattern is that, in many cases, P450 enzyme expression changes in long-lived mice and under CR were mirrored by expression differences in females relative to males. In a comprehensive analysis of 41 cytochrome P450 enzyme genes, it was found that 14 genes exhibited this pattern to some degree (data not shown). This result may be attributable to the role of GH as a factor regulating cytochrome P450 expression [60
The correlation structure of expression patterns among genes may be as informative as differential expression analyses but is often an underexploited aspect of microarray datasets. Similarity among expression patterns across many conditions can be used, for example, to make inferences and generate testable hypotheses regarding interactions between genes [61
]. Along these lines, genes with expression patterns among contrasts that closely corresponded with those of the IGF-I transcript were identified in this study. This analysis identified Snm1 (DNA cross-link repair 1a; also called Dclre1a) as a potential IGF-I regulated or co-regulated gene (see Fig. and additional file 2
). The induction pattern of Snm1 is opposite that of IGF-I across contrasts, such that its expression is increased in proportion to IGF-I inhibition. Snm1(-/-)
mice have reduced lifespan, which is primarily due to elevated mortality from bacterial infection and cancer, suggesting that Snm1 is a tumor suppressor with an immunological role [63
]. At the cellular level, overexpression of Snm1 in yeast increases resistance to genotoxic stress agents that induce DNA cross-links [64
], and stem cells derived from Snm1(-/-)
mice are sensitive to the cross-linking agent mitomycin C [65
]. These results suggest that elevated Snm1 expression may be a causal factor underlying resistance to the DNA-alkylating agent methyl methanesulfonate that has previously been found in cell lines derived from long-lived mice [10
]. It would thus be worthwhile to evaluate whether Snm1 expression is elevated in such cell lines (as in hepatic tissue), and if so, whether these cells are resistant to DNA cross-link inducing agents (e.g., nitrogen mustard, cisplatin). Interestingly, recent studies report that DNA repair mutations leading to accelerated senescence phenotypes are, like long-lived dwarf mutations, associated with IGF-I inhibition [66
]. It is therefore surprising that in long-lived and CR mice, IGF-I inhibition is associated with elevated Snm1 expression, which may enhance DNA repair and promote genomic stability.
The present study has demonstrated analytical approaches for the comparative analysis of microarray datasets that may have application in other contexts. Microarray analyses based upon individual datasets often identify an exceptionally large number of genes, which limits the utility of microarray data as a tool for selecting candidates in follow-up studies. Recently, however, public repositories of high-quality microarray data have been established [67
], and statistical methods aimed at comparative analysis of these resources continue to be developed [69
]. Increasingly, therefore, comparative analysis provides a feasible approach for filtering out false-positive identifications and identifying transcripts most consistently supported across multiple experiments. This yields a set of candidate genes that is necessarily smaller and more tractable for subsequent experimental investigation, and moreover, each identified gene is more likely to represent a statistically significant finding. Furthermore, apart from the identification of individual gene candidates, comparison of expression datasets provides unique insight into genome-wide patterns among studies.
The maximum human lifespan is approximately twice as large as the maximum chimpanzee lifespan [73
], which is more than ten times larger than the maximum mouse lifespan [2
]. The evolutionary lineage connecting humans to chimpanzees to mouse may therefore represent a remarkable instance of lifespan extension and delayed rates of aging. Inhibition of IGF-I signaling is (at present) the only known genetic manipulation that extends lifespan in multiple species, so it is tempting to ask whether this pathway has contributed to evolutionary extension of lifespan within the mammalian lineage. It is interesting to note that, contrary to longevity extension via nearly all IGF-I signaling mutations, lifespan extension in the mouse-chimp-human lineage has been accompanied by an increase
in body size. This observation is encouraging with regard to the possibility of developing interventions that delay the onset of age-related disease in humans without undesirable consequences (e.g., dwarfism). Clarke et al. [74
] used sequence data and dN
ratios to identify mouse-chimp-human orthologs that exhibit more rapid evolutionary change than expected based upon neutral substitution models (i.e., positive selection). Supplemental data from their study provides no indication of positive selection with respect to IGF-I, providing little indication that IGF-I sequence changes have been consequential during human evolution. It is interesting to note, however, that Dio1 (iodothyronine deiodinase), which is downregulated in most dwarf mouse strains and under CR (see Fig. ), has been associated with positive selection and accelerated evolution within the mouse-chimp-human lineage (see supplemental data from [74
]). This is also the case for Papp-A (pregnancy associated plasma protein A) [74
], which was not identified in the present study, but is of importance since Papp-A-KO mice exhibit diminished IGF-I bioactivity and extended lifespan [75
]. These results are suggestive, but given the many phenotypic differences between mice, chimpanzees and humans (besides lifespan), this evolutionary criterion does not establish the functional significance of Dio1, Papp-A or other positively selected genes.
These findings provide a useful reference point for future experimental studies of long-lived dwarf mice and mammalian aging. It would be of interest, for example, to determine whether genes identified in the present study are differentially expressed with respect to other long-lived mouse strains for which data is not currently available. Similar gene expression changes may be found in liver tissue of p66shc
], IGF-IR knockouts [77
], Klotho transgenic mice [78
] and Papp-A-KO mice [75
], since each of these long-lived models carry mutations that also inhibit the IGF-I signaling pathway. Because these mutations inhibit IGF-I signals at points further downstream than those considered in this study, expression data generated from these models could be combined with results of the present study to assemble a more comprehensive picture of hepatic IGF-I signaling. Since the currently known life-extending mutations are maintained on differing genetic backgrounds [1
], evaluating potential background effects on lifespan and gene expression patterns will also be critical for elucidating hepatic IGF-1 signaling pathways [79
]. Ultimately, however, it is important to develop a systemic model of GH/IGF-I signaling, with further evaluation of the respective roles of IGF-I and GH in extended longevity. IGF-I is manufactured in a wide variety of tissue types, and in some cases IGF-I production may be independent of GH signaling [80
]. It is therefore important to consider not only local effects of GH reduction on gene expression in liver, but also effects of GH/IGF-I in other tissue types. Studies of long-lived C. elegans mutants, for instance, have suggested that IGF-I signaling in nervous tissue may be consequential in determining lifespan [81
]. It would therefore be useful to evaluate whether genes identified in the present study are also differentially expressed in non-hepatic tissues of Ames, Snell, Little or GHR-KO mice.