• Open Access

Use of microarray biomarkers to identify longevity therapeutics


Stephen R. Spindler, PhD, Department of Biochemistry, University of California, Riverside, Riverside, CA 92521, USA. Tel.: 951-827-3597; fax: 951-827-4434; e-mail: spindler@ucr.edu


A number of lines of evidence, including nonhuman primate and human studies, suggest that regulatory pathways similar to those invoked by caloric restriction (CR) may be involved in determining human longevity. Thus, pharmaceuticals capable of mimicking the molecular mechanisms of life- and health-span extension by CR (CR mimetics) may have application to human health. CR acts rapidly, even in late adulthood, to begin to extend life- and health-span in mice. We have linked these effects with rapid changes in the levels of specific gene transcripts in the liver and the heart. Our results are consistent with the rapid effects of caloric intake on the lifespan and/or biochemistry and physiology of Drosophila, rodents, rhesus macaques and humans. To test the hypothesis that existing pharmaceuticals can mimic the physiologic effects of CR, we evaluated the effectiveness of glucoregulatory drugs and putative cancer chemopreventatives in reproducing the hepatic gene-expression profiles produced by long-term CR (LTCR). We found that 8 weeks of metformin treatment was superior to 8 weeks of CR at reproducing the specific changes in transcript levels produced by LTCR. Consistent with these results, metformin reduces cancer incidence in diabetic humans and ameliorates the onset and severity of metabolic syndrome. Metformin extends the mean and maximum lifespans of female transgenic HER-2/neu mice by 8% and 13.1% in comparison with control mice. Phenformin, a close chemical relative of metformin, extends lifespan and reduces tumor incidence in C3H mice. These results indicate that gene-expression biomarkers can be used to identify promising candidate CR mimetics.


Although reducing morbidity and mortality in the elderly is a major goal of pharmaceutical research, there are presently no authentic longevity pharmaceuticals. This situation exists because there has been no rapid assay for identifying such drugs, not because they do not exist or cannot be developed. Drug discovery and development usually involves rapid, surrogate assays for screening candidate compounds. Most therapeutics for human diseases were discovered using such surrogate assays, usually without any knowledge of the molecular mechanism of the disease for which they were intended. Thus, we do not need to fully understand aging to search for such medicaments. We need to develop and validate rapid, surrogate assays. The development of such assays can benefit from the known methods of extending the maximum lifespan of mammals. The lifespan of experimental animals can be extended robustly with at least two interventions, caloric restriction (CR) and dwarfism, and these interventions are additive in their effects (Bartke et al., 2001).

CR is a nutritional intervention capable of consistently extending average and maximum lifespan and reducing the incidence and severity of most age-related diseases, including cancer. CR is effective whether it is initiated in young or older mice (Merry, 2002; Dhahbi et al., 2004). Maximum and average lifespan can also be extended by genetic manipulations, which reduce signaling through insulin and/or insulin-like growth factor I (IGF-I; Bartke, 2005). Reduced insulin–IGF-I signaling also postpones age-related development of neoplastic diseases, immune system decline, and collagen cross-linking (Bartke & Brown-Borg, 2004). CR and the Ames dwarf mutation act together in mice to additively increase lifespan and alter the expression of specific hepatic genes (Bartke et al., 2001; Tsuchiya et al., 2004). Other long-lived mouse mutants also have been identified, suggesting that additional, potentially manipulable pathways exist for extending lifespan and ameliorating age-related disease (Liang et al., 2003).

We found that 8 weeks of CR (CR8) begins to extend life- and health-span in older mice (Dhahbi et al., 2004; Spindler, 2005). Lifespan extension resulted primarily from reduced tumor-related mortality, especially fewer liver tumors. We also found over this same time interval that CR induces nearly three quarters of the gene expression effects of long-term CR (LTCR) in liver (Cao et al., 2001; Dhahbi et al., 2004; Spindler, 2005). Almost all of these changes in transcript levels dissipate 8 weeks after switching CR mice to a control diet (Dhahbi et al., 2004, 2005a). These results indicate that the gene-expression changes induced by CR are closely linked to its health and longevity benefits.

The close linkage between the gene expression and physiologic effects of CR suggests that microarrays can be used to screen potential therapeutics for their ability to mimic the effects of CR on mammalian physiology. Such assays would avoid the expense, time and theoretical problems associated with using lifespan as an end point. To test this hypothesis, we evaluated four potential therapeutics for their ability to reproduce the LTCR-related gene-expression profiles found in mouse liver. Using Affymetrix microarrays, we identified one such candidate mimetic, metformin. These results suggest that microarray profiling can be used to identify candidate CR mimetics.

Drug discovery

The vast majority of the drugs now in use were developed and tested in the absence of any clear molecular understanding of the pathology they were intended to treat or the mechanism of action of the drug. In fact, mechanism-based drug development only recently has begun to produce effective therapeutics, and many such drugs are encountering unexpected problems. Virtually all the safe and effective therapeutics in use today were developed using empirical assays, usually with little or no knowledge of the mechanism by which the assay or the compound acts. Cancer therapeutics are examples of this. There are many effective cancer treatments that were originally identified and developed using empirical, surrogate-screening strategies. In fact, the major use of most chemotherapeutics is not against the tumors for which they were tested and approved during clinical trials.

The development of metformin as an antidiabetic provides another example of the discovery and development of a highly effective therapeutic in the absence of any knowledge of the molecular basis for the disease or the mechanism of action of the pharmaceutical (Witters, 2001). Beginning with the early Egyptians and continuing today, the use of medicaments for diabetes mellitus preceded, often by centuries, any knowledge of their physiologic mechanisms of action. The French lilac, Galega officinalis, was used in medieval times to relieve the excessive urination accompanying diabetes mellitus. Centuries later, the active ingredient of the weed was identified as isoamylene guanidine. Independently, through a fortuitous study involving a mistaken assumption, guanidine infusion was found to lower blood glucose concentrations. This led in due course to the development in the 1950s of biguanides, including metformin and phenformin, which were less toxic than guanidine itself. Following its medicinal use in Europe for 20 years, metformin was finally approved for use in the USA in 1995. It remains a key treatment and preventative for type 2 diabetes to this day. Nevertheless, the molecular mechanism of metformin action remains mysterious. Only recently has it been shown to involve, in part, activation of AMP kinase signaling (Zhou et al., 2001).

Will CR work in humans and does it matter?

Some have argued that it is unlikely that CR will prolong human health- and lifespan (Hayflick, 2004). Some suggest that the life history of humans presents little selective pressure for a robust CR response (De Grey, 2005; Demetrius, 2005; Phelan & Rose, 2005). These hypotheses and their underlying assumptions can be debated. However, an unequivocal resolution of the debate is unlikely in the near future. Regardless, CR mimetics developed in mice are likely to have applications to human health.

A number of lines of evidence, including nonhuman primate and human studies, suggest that CR-like processes may be involved in human longevity (Weyer et al., 2000; Roth et al., 2002). An analysis of data from the Baltimore Longitudinal Study of Aging suggests that physiological biomarkers associated with CR in monkeys also are associated with enhanced lifespan in humans (Roth et al., 2002). These biomarkers are low plasma insulin (probably due to high tissue insulin sensitivity), low body temperature and high plasma dehydroepiandrosterone sulfate. The first two have also been associated with CR in rodents, although dehydroepiandrosterone sulfate levels are too low to be measured accurately in rodents. These results suggest that the molecular-genetic processes leading to lifespan extension in CR animals also operate to extend human lifespan. Human genetic variability may lead to activation of some of these processes in particular individuals. The limited physiological evidence presently available for humans suggests that CR produces effects very similar to those found in rodents, dogs, and nonhuman primates (Verdery & Walford, 1998; Walford et al., 1999; Kealy et al., 2000; Weyer et al., 2000).

CR produces changes in human physiology closely associated with enhanced longevity. The well-known study of Kagawa (1978) found that the 20% reduced energy consumption by Okinawan adults in the 1970s was accompanied by a 40% reduction in the mortality rate of 60- to 64-year-olds relative to the rest of Japan. They also found that death rates from cerebral vascular disease were reduced to 59%, malignancy to 69%, and heart disease to 59% of the frequencies found in the rest of Japan. The incidence of centenarians on the island was also 2 to 40 times greater than that of other Japanese communities. Younger Okinawans who consume a Western-type diet had mortality and morbidity rates more similar to those of the West.

Sixty healthy volunteers over 65 years of age (average age 72 years) participated in a 3-year study in which they received 2300 calories every other day and on alternate days 1 L of milk with 500 g of fresh fruit (Vallejo, 1957). A control group of elderly people at the same institution received 2300 calories everyday. The CR group spent less time in the infirmary than the controls (123 vs. 219 days; P < 0.001) and had fewer deaths than the controls (6 vs. 13; not statistically different; Stunkard, 1976).

A longitudinal human CR study serendipitously conducted on eight healthy nonobese humans eating a low-calorie (1750–2100 kcal/day) diet for 2 years in Biosphere II found physiologic, hematologic, hormonal, and biochemical changes in the subjects which resembled those of CR rodents and monkeys (Walford et al., 2002). More recently, significantly reduced risk factors for atherosclerosis were found in individuals who had restricted food intake for an average of 6 years (Fontana et al., 2004).

CR mimetics should be useful whether or not CR extends lifespan in humans

Even if CR does extend human lifespan, it will probably not be the ideal way for humans to enjoy its benefits. Natural episodes of human malnutrition are associated with short stature and late reproductive maturation, lower baseline gonadal steroid production, suppressed ovarian function, impaired lactation, reduced fecundity, weakened immune function, lower basal metabolic rate, reduced body temperature, and hording behavior (Lee, 1979; Prentice et al., 1983; Weindruch & Walford, 1988; Lager & Ellison, 1990, 1996; Pasanisi et al., 2001; Somes et al., 2002; Smith et al., 2004). These maladies are probably not the result of poor-quality diets, since most also have been reported in CR rodents (Weindruch & Walford, 1988; Smith et al., 2004).

However, there is no a priori reason to expect that the life- and health-span effects of CR are inseparable from its negative effects. Coselection of most of the responses to CR itemized above are easily reconciled with its evolutionary advantages. Thus, there is no reason to assume that they all are mediated by one signal transduction pathway. Further, animals with longer lifespans than those of humans do not appear to suffer from such maladies. Thus, it should be possible to identify pharmaceuticals that act to induce the positive effects of CR while avoiding the negative ones.

The apparent maximum lifespan of humans may be far from the limits of mammalian longevity. Since 1981, Inupiat hunters in villages along the north coast of Alaska have found six harpoon points made of ivory and stone in the blubber of freshly killed bowhead whales (Rogers, 2000). Such harpoon points reportedly have not been used by native Alaskan whalers since the 1880s, suggesting these whales were at least 100 years old when taken. Analysis of aspartic acid racemization in 94 eye-lens nuclei from 84 individual bowhead whales found that five males from the group far exceeded 100 years of age (113, 136, 160, 174, 213 years; George et al., 1999; Rosa et al., 2005). Since there are no nursing homes for geriatric whales, at least one mammal appears to be capable of considerably longer lifespans than humans, in the wild. The bowhead whales’ adaptation for extreme longevity may involve qualitative changes in genes or quantitative changes in gene expression. In either case, the molecular pathways responsible for extreme mammalian longevity should be susceptible to small molecule therapeutics in humans. Such therapeutics are being developed for genetic diseases caused by inappropriate levels of gene expression and the expression of mutant proteins. Thus, we should be able to target the pathways responsible for extreme mammalian longevity whether or not they are due to quantative or qualatative differences in gene expression, without inducing the negative effects of CR or dwarfism.

The first requirement for development of lifespan enhancing medicaments is a rapid assay for identifying their effects. We have begun to explore strategies for such screening using a rapid surrogate assay.

Widespread views of aging have slowed the development of longevity therapeutics

A longevity medicament should have therapeutic potential for older people. Aging, however, is widely assumed to result from the gradual, age-related accumulation of essentially irreversible oxidative damage to macromolecules. In this context, CR is often viewed as preventing or slowing the accumulation of such damage, thereby slowing the process of aging (Bokov et al., 2004). The interpretation of most published CR studies has been strongly influenced by this view. Virtually all cross-sectional studies of mammalian aging have been interpreted as though they were performed longitudinally (e.g. see the studies and discussion of the studies in Van Remmen et al., 1995). The effects of LTCR are almost always assumed to have resulted from incremental resistance to essentially irreversible damage. Hence, CR is often said to ‘retard’ aging.

A CR mimetic capable only of decreasing the rate of molecular damage would not provide the kind of rapid improvement in health and longevity among the elderly that would characterize a successful medication. Further, the drug would need to be taken beginning early in life to be effective. Instead, an ideal longevity therapeutic would rapidly reduce age-related morbidity and mortality, even when taken in old age.

We observed that data from many CR studies, interpreted as evidence for the prevention of age-related damage, could also be interpreted as evidence for increased turnover or repair of damaged macromolecules. To test this hypothesis, we conducted a genome-wide microarray survey of the effects of rapidly shifting very old, control-fed mice to a CR diet for 4 weeks (CR4; Cao et al., 2001). Because most mice of the strain used die of liver tumors, we focused our initial studies on this organ. We found that CR4 reproduced 55% of the gene-expression effects of LTCR in elderly mice. For age-responsive genes, it reproduced ∼70% of the effects of LTCR. These results suggested to us that CR could rapidly initiate its health- and lifespan effects.

CR can act rapidly, even late in life, to initiate the extension of lifespan

CR had long been known to extend lifespan and reduce cancer incidence when it is initiated in 12-month-old mice (Weindruch & Walford, 1982). However, studies using rats appeared to show that CR is much less effective or even ineffective in older animals (Lipman et al., 1995, 1998). Our microarray results led us to test the possibility that CR could rapidly initiate its lifespan effects during late adulthood. We shifted control-fed mice to CR immediately before the beginning of the accelerated mortality phase of the lifespan curve (Fig. 1). We reasoned that this would reveal the rate of onset of the longevity effects of CR.

Figure 1.

The effects of CR begun immediately before the onset of the accelerated mortality phase of the lifespan curve. (A) The longevity of mice switched from control to CR at 19 months of age. The percentage of mice remaining alive at the end of each month is shown for CR mice (open circles) and control mice (filled circles). Shown is the mortality trajectory of the mice at the beginning of the experiment (solid line), the slope of the accelerated mortality phase of the survival curve for the control mice (dashed line), the approximate mortality trajectory of the CR mice between the first and second breakpoints in the survival curve (dotted line), and the mortality trajectory after the second breakpoint (dotted dashed). (B) A theoretical survival curve assuming that mortality of the CR mice results from a reduced rate of tumor growth and a constant rate of tumor formation. The lower-case letters designate different parts of the curve. (C) A theoretical curve describing the results expected if mortality in the CR group results from reduced rates of tumor formation and a constant rate of tumor growth. The figure is adapted from Spindler (2005), with permission. For further explanation of the theoretical curves shown in B and C, see Spindler (2005).

We found that CR begins to extend lifespan and reduce tumor incidence within 2 months in 19-month-old, male B6C3F1 mice (Fig. 1; Dhahbi et al., 2004). Linear regression and breakpoint analysis of the survival curve indicated that the CR mice attained a 3.1-fold lower mortality rate within 2 months of initiating a CR diet (Dhahbi et al., 2004; Spindler, 2005). During this phase of their lifespan, tumor-related mortality was lower in the CR mice (67% vs. 80%) and they enjoyed a 42% increase in time to death, as well as an increase in their mean and maximum lifespans (4.7 and 6.0 months, respectively). Sixty-five percent of the control mice and 44% of the CR mice died of large liver tumors during this time. The shapes of the survival curves suggest that the onset of CR primarily reduced the rate of tumor growth (Fig. 1; Spindler, 2005). We think that the fact that later-life CR produced a 42% increase in remaining lifespan is important, since LTCR begun after weaning extends the lifespan of this strain of mice by about 40%, in our hands. Thus, the maximum time required for CR to decelerate mortality and begin to increase lifespan was 2 months.

The reason for the differences between the effectiveness of later-life CR in our studies and those reported by Lipman et al. (1995, 1998) is not known. However, they may be related to species-specific differences in the susceptibility of mice and rats to late-life CR. Alternatively, the differences may be related to subtle methodological differences related to CR in older animals. Examples of such methodological issues are the phased introduction of CR (which was also used in the rat studies cited above), which allowed Weindruch and Walford to obtain lifespan extension when CR was initiated at 12 months of age (Weindruch & Walford, 1982). Previous studies using the abrupt introduction of CR had found shortened rather than extended lifespan. Similarly, differences in the husbandry of dwarf mice determine whether shortened or lengthened lifespan will be observed (Brown-Borg et al., 1996).

Demographic studies in Drosophila have shown that their short-term risk of death, and therefore their lifespan, responds rapidly to food consumption (Mair et al., 2003; Partridge et al., 2005). Shifting the flies from a control to a food-restricted diet decelerated their short-term risk of death within 2 days. Similarly, shifting from a restricted to a control diet accelerated their short-term risk of death in 2 days. The similarities between these results and those we found in mice suggest that the rapid effects of diet on survival have been phylogenetically conserved.

Gene expression biomarkers are closely linked to the physiological actions of LTCR

We determined the kinetic relationship between the gene-expression effects of CR and its health and longevity effects. For the reasons already stated, we focused our initial investigations on the liver. Two, 4 and 8 weeks of CR, initiated in older mice, progressively induced most of the transcription profile associated with LTCR (Spindler, 2005). Two similar patterns of response were found, and the major one is shown in Fig. 2. In 2 weeks, CR induced half of the changes in gene expression produced by LTCR. In 4 and 8 weeks, 58% and 72% of the LTCR-related changes in gene expression were induced. Thus, an LTCR-like pattern of gene expression was substantially induced within the time frame required to initiate the longevity and anticancer effects of CR.

Figure 2.

Dynamics of the gene expression response in old control mice shifted to CR for 2 weeks (CR2), 4 weeks (CR4), and 8 weeks (CR8), and in old long-term CR (LTCR) mice shifted to control feeding for 8 weeks (CON8). The most common pattern of gene expression found is shown. The figure is adapted from Dhahbi et al. (2004), with permission.

Importantly, shifting 32-month-old LTCR mice to control feeding for 8 weeks returned 90% of the LTCR-responsive genes to control expression levels. All of the genes which required longer than 8 weeks to respond to CR shifted back to control expression in just 8 weeks of control feeding. Thus, in liver all the LTCR-responsive genes can react rapidly to caloric intake.

To the best of our knowledge, no other changes in gene expression have been as closely linked to enhanced health and longevity in a mammal. The few genes that have been shown to control longevity in mammals and lower eukaryotes are mainly regulatory genes which alter the expression of many effector genes (Hekimi & Guarente, 2003; Liang et al., 2003). To the best of our knowledge, few of the effector genes responsible for actually mediating the lifespan-extending effects of these regulatory genes are known. Thus, these rapidly CR-responsive genes appear to be more directly linked to lifespan enhancement in mammals than any others.

Rapid effects on gene expression are found in the heart, a largely postmitotic tissue

The regulatory priorities of largely postmitotic tissues such as the heart appear to be different than those of mitotically competent tissues like liver. We found that only 20 of 106 LTCR-responsive heart genes changed expression after 8 weeks of CR (Dhahbi et al., 2005a). These results are similar to those found in the largely postmitotic adipose tissue (Higami et al., 2004). Taken together, these results suggest that the physiologic and gene-expression effects of CR are established more rapidly in mitotically competent tissues such as liver than in postmitotic tissues such as the heart and adipose tissue.

This interpretation is consistent with the different regulatory imperatives at work in postmitotic and potentially mitotic tissues. In rodents, and especially in most laboratory strains of mice, neoplasms are the most common cause of death. For this reason, lifespan studies conducted in mice are primarily measurements of the rate of formation and growth of spontaneous neoplasms. Even CR mice die primarily of cancer (Dhahbi et al., 2004). Most neoplasms arise from mitotically competent cells, and not from postmitotic tissues such as the heart and adipose tissue. Therefore, CR may rapidly and strongly affect the molecular biology of mitotically competent tissues, and produce fewer rapid effects in postmitotic tissues.

The gene-expression effects of LTCR dissipate rapidly

The effects of CR are thought to dissipate after a return to control caloric intake (Merry, 2002). However, there are very few studies of this transition. In liver, shifting old, LTCR mice to control feeding for 8 weeks returned 90% of the LTCR responsive genes to control expression levels (Dhahbi et al., 2004; Spindler, 2005). Similarly, in the heart, 8 weeks of control feeding returned 97% of the LTCR-responsive transcripts to control expression levels (Dhahbi et al., 2005a). These results indicate that most LTCR-responsive genes are rapidly responsive to caloric intake in both mitotically competent and largely postmitotic tissues.

CR appears more proactive than protective

The results reviewed above suggest that the gene expression and physiological effects of LTCR in mice do not result from the prevention of molecular damage. Instead, they are probably due to the rapid repair or elimination of such damage. This view is consistent with results from other laboratories. The demographic studies in Drosophila cited above demonstrate a rapid linkage between food restriction and lifespan in this species (Mair et al., 2003; Partridge et al., 2005). The cells of adult Drosophila are postmitotic, and flies are not known to suffer from cancer. Thus, phylogenetically conserved, tissue-specific mechanisms for rapid enhancement of life- and health-span exist for both mitotic and postmitotic tissues across broad species differences. Other published biochemical and physiological studies in rodents and humans support this view (reviewed in Spindler, 2005). Together these results suggest that the most important effects of CR may be in repairing or eliminating damage in mitotically competent tissues, in part through apoptosis. Likewise, the suppression of apoptosis and necrosis in largely postmitotic tissues may allow an enhanced opportunity for repair.

What is the best way to assay for potential lifespan-extending medicaments?

Development of assays for longevity medicaments has proven to be difficult. Screening pharmaceuticals using short-lived organisms such as nematodes and files is of unknown efficacy, and the results of such assays appear to be inconsistent (Melov et al., 2000; Bayne & Sohal, 2002; Keaney et al., 2004). It seems that lifespan studies in a short-lived mammal such as the mouse should be informative. Vigorous mouse strains, however, live for more than 40 months, making lifespan studies time-consuming and expensive. Short-lived or enfeebled rodent strains introduce confounds into such studies. In addition, most rodent strains die of one or a few characteristic diseases. For mice, these are often a few, strain-specific types of cancer. Unless the therapeutic tested delays aging in the organs and cell types responsible for the characteristic causes of death, these screening protocols will fail, even though they might be effective in other organs or tissues.

Gene expression patterns are useful surrogate biomarkers for drug discovery

Reliable and reproducible biomarkers for the longevity effects of CR, dwarfism and other longevity-associated interventions would be of great utility in drug discovery. Until recently, efforts to identify such biomarkers have met with limited success. The close temporal linkage between the effects of CR on longevity and gene expression suggests a reciprocal cause-and-effect relationship between them. Thus, we hypothesized that the gene-expression changes induced by CR could be used in drug discovery as surrogates for its health and longevity effects.

To test this hypothesis, we evaluated the ability of five potential CR mimetics to reproduce the gene-expression profiles associated with CR in mouse liver using Affymetrix microarrays (Dhahbi et al., 2005b). Because of the linkage between CR, insulin, IGFI, and the rate of aging (Bartke, 2005; Kurosu et al., 2005), we studied the effects of the glucoregulatory compounds metformin (MET), glipizide (GLIP), GLIP plus MET (GM), and rosiglitazone (ROS). MET, a biguanide, increases insulin sensitivity in the liver and muscle, and decreases hepatic glucose production and output (Radziuk et al., 2003). Rosiglitazone, a thiazolidinedione, increases peripheral insulin sensitivity (Stumvoll & Haring, 2002). Glipizide, a sulfonylurea, increases insulin secretion by pancreatic β cells (Rendell, 2004). Because of the importance of tumorigenesis in determining the lifespan of mice, we also tested a putative chemopreventative, soy isoflavone extract (SOY), for its LTCR-like gene-expression effects (Ricketts et al., 2005). Soy isoflavones have been proposed to exert their biological effects in part through the peroxisome proliferator activated receptors (PPAR), like fibrates and thiazolidinediones. Recent microarray studies suggest that 19% of the effects of CR on gene expression may be mediated by activation of the PPARα receptor signal transduction pathway (Corton et al., 2004).

We administered MET, GLIP, GM, ROS or SOY to mice, cold packed in their diets for a period of 8 weeks (Dhahbi et al., 2005b). A control group received the diet alone. All these groups were isocaloric, and their weights did not vary significantly before or after the studies. The studies also included control groups that were subjected to LTCR or caloric restriction for 8 weeks (CR8). The data were filtered and analyzed statistically using multiclass SAM. The results indicated that MET produced more LTCR-like changes in gene expression (75% of the total LTCR-responsive gene set) than were produced by CR8 (71% of the LTCR-responsive gene set; Table 1). Eight weeks of MET reproduced 92% of the gene expression effects of CR8. The other treatments were much less effective. These results indicate that MET substantially reproduces the gene-expression effects of LTCR and CR8.

Table 1.  Numerical overlap between the transcriptional effects of LTCR, CR8 and each of the drug treatments
Treatment group*Number of LTCR-like responsesNumber of CR8-like responses
  • *

    Number of LTCR- or CR8-responsive gene-expression changes produced by each treatment. MET is metformin, GLIP is glipizide, GM is the combined administration of GLIP and MET, ROS is rosiglitazone, and SOY is soy isoflavone extract. The percentage of changed genes identical to those changed by LTCR or CR8 are shown. This table is adapted from Dhahbi et al. (2005b), with permission.


To further quantify the significance of the overlap between LTCR and each treatment, the overlapping genes were analyzed using Venn Mapper, providing another level of statistical stringency. Venn Mapper reports the number and identity of the up- and down-regulated genes in the overlap between treatment groups, and produces a table of z-values. At a significance of z ≥ 2 (a P-value < 0.05), MET and CR8 yielded the highest number of genes overlapping with those of LTCR (349 and 218, respectively; Fig. 3). The other treatments produced fewer overlapping genes. We also determined the number of LTCR-like changes in gene expression produced by the treatments at increasing fold-change thresholds. MET, followed by CR8, again maintained the most overlap with LTCR (Dhahbi et al., 2005b). These results indicate that MET reproduces LTCR-like effects on gene expression even better than CR8, with the other treatments producing fewer such effects.

Figure 3.

The number of LTCR-like genes induced by each treatment at increasing levels of statistical stringency. Affymetrix data were filtered and normalized using MAS 5.0 and RMA, and subjected to multiclass SAM analysis followed by t-tests to determine the effects of each dietary or drug treatment. The differentially expressed genes were merged and analyzed using Venn Mapper to identify genes significantly affected by LTCR and each of the treatments. The comparisons were performed at a fold-change cutoff of 1.2 and the indicated z-values. The figure is from Dhahbi et al. (2005b), with permission.

As an additional metric for judging the effectiveness of the treatments, we created an unbiased estimate of their overlapping physiological effects using GenMAPP and MAPPFinder (Dhahbi et al., 2005b). The number of gene ontology (GO) terms common to LTCR and each treatment is a quantitative measure of their functional similarities (Dahlquist et al., 2002; Doniger et al., 2003). MET produced the highest number of GO terms overlapping those of LTCR, again outstripping even CR8 (Fig. 4). Together, these results indicate that 8 weeks of MET treatment surpasses even 8 weeks of CR at producing a gene-expression profile which overlaps that of LTCR as regards functional gene groups. We think this result is significant because CR8 initiates lifespan enhancement and reduces tumor incidence in old mice (Dhahbi et al., 2004). Based on these data, MET is a promising candidate CR mimetic.

Figure 4.

Relative abundance of GO terms for genes responsive to LTCR and each drug treatment. The bars represent the total number of GO terms and their distribution across the three highest-level branches in the GO tree: biological processes (P), molecular functions (F), and cellular components (C). Output from GenMAPP and MAPPFinder was filtered to include only GO terms with z-score ≥ 2, P-value < 0.01 and the percentage of genes meeting the criterion = 10. The figure is from Dhahbi et al. (2005b), with permission.

The physiological effects of MET are consistent with its gene-expression effects

Consistent with the results reviewed above, chronic treatment of female transgenic HER-2/neu mice with metformin increased their mean lifespan by 8% and their maximum lifespan by 13%. It also significantly reduced the incidence and size of mammary adenocarcinomas and increased the mean latency of the tumors. Phenformin, a chemical relative of MET, extends the lifespan of C3H mice by ∼23% while reducing tumor incidence by 80% (Dilman & Anisimov, 1980; Anisimov et al., 2003). In humans with type 2 diabetes, taking MET may be associated with reduced cancer risk (Evans et al., 2005). MET also protects hamsters fed a high-fat diet from malignant, hyperplastic and premalignant pancreatic lesions (Schneider et al., 2001). Indirect evidence suggests that signaling through AMP-activated protein kinase may mediate both the antidiabetic and the anticancer effects of MET (Hawley et al., 2003). AMP-activated protein kinase may be involved in regulating the lifespan of Caenorhabditis elegans, Drosophila and yeast (Tschape et al., 2002; Apfeld et al., 2004; Harkness et al., 2004). MET is also effective in the treatment of polycystic ovary syndrome (Homburg, 2004). Finally, MET therapy has been shown to inhibit the development of metabolic syndrome in humans (Orchard et al., 2005). Metabolic syndrome is associated with increased risk of cardiovascular- and diabetes-associated morbidity and mortality. Together, these results suggest that use of microarray biomarkers has identified a very promising candidate CR mimetic.

Hepatic pathways coregulated by MET and LTCR

The coregulation of genes by LTCR and MET implies that they utilize common transcription factors and upstream effectors. Analysis of the coregulated genes with PathwayAssist indicated that the major shared effector pathways include insulin, tumor necrosis factor, fibroblast growth factor 2, epidermal growth factor, and elements of their related signal transduction pathways, including Fos, MAPK1 and 8, and p53. The major cellular processes targeted by the CR-mimicking effects of MET are apoptosis and cell survival, differentiation, cell proliferation, and focal contact. Chaperones were identified as a major effector shared by LTCR and MET.

Some years ago, we found that LTCR, short-term CR, and fasting reduced the levels of most endoplasmic reticulum and some cytoplasmic chaperone mRNAs and proteins in the liver of mice (Spindler et al., 1990; Dhahbi et al., 2002; Dhahbi & Spindler, 2003; Spindler & Dhahbi, 2003). In these studies, we found that MET and LTCR both negatively regulate the chaperones TRA1, HSPA5, GRP58 and DNAJB11, and the related transcription factor XBP1 (Dhahbi et al., 2005b). These chaperones are involved in apoptosis, proliferation, differentiation, inflammation, oxidative stress, and other cellular processes. Chaperone overexpression reduces apoptosis and promotes tumorigenesis, while underexpression enhances apoptosis and prevents tumor formation (Sugawara et al., 1993; Jamora et al., 1996). Chaperone overexpression is also associated with the acquisition of resistance to chemotherapeutics and cell-mediated immunity (Shen et al., 1987; Gomer et al., 1991; Sugawara et al., 1993; Chatterjee et al., 1995; Garrido et al., 2003). Chaperones can inhibit key effectors of the apoptotic machinery, including the apoptosome and apoptosis-inducing factors (Garrido et al., 2003). These overlapping effects are consistent with the anticancer effects of MET and CR. Thus, in a mitotically competent tissue such as liver which is prone to tumor formation, enhanced apoptosis is adaptive. In contrast, in the heart, a predominantly nonmitotic tissue, the gene regulatory priorities are different. In the heart, we find that CR alters gene expression in a manner consistent with lower levels of apoptosis and enhanced cell survival (Dhahbi et al., 2005a).

Other gene-expression biomarkers of longevity

A variety of dietary and genetic manipulations has been reported to extend the lifespan of mammals (see below). In mice, the most robust and best characterized of these is CR and dwarfism. For organisms as diverse as nematodes, flies and mice, lifespan can be regulated by mutational alteration of components of the insulin/insulin-like growth factor receptor signaling pathway (Brown-Borg et al., 1996; Tatar et al., 2001; Kenyon, 2005). In mice, a family of single gene mutations which interfere with growth hormone/IGF-I signaling and reduce insulin signaling, resulting in dwarfism, increase mean and maximal lifespan by 40% to 70% beyond those of their phenotypically normal siblings (Flurkey et al., 2001). The enhanced lifespan of one of these mutant strains of mice, the Ames dwarf, can be further extended by CR (Bartke et al., 2001).

To investigate the molecular basis for these additive effects on lifespan, and to broaden our understanding of longevity-associated biomarkers, we investigated the hepatic gene-expression patterns associated with CR and dwarfism (Tsuchiya et al., 2004). We found additive effects of LTCR and dwarfism on the expression of 100 genes (Fig. 5A). Dwarfism affected the expression of 212 other genes whether or not CR was present. Likewise, CR affected the expression of 77 genes, whether or not dwarfism was present. These results indicate that dwarfism and CR affect overlapping, but distinct, sets of genes. One of a number of possible mechanisms by which such effects could be mediated at the molecular level is shown in Fig. 5B. These results point to sets of genes closely associated with the regulation of lifespan in mammals. These biomarkers should prove useful for identifying potential lifespan-enhancing medicaments.

Figure 5.

A summary of hepatic gene expression profiling of normal and dwarf mice fed ad libitum or LTCR. (A) Dwarfism changed the expression of 312 genes (212 + 100 genes), LTCR changed the expression of 177 genes (77 + 100), and dwarfism and LTCR together changed the expression of 389 genes (212 + 100 + 77 genes). Of the 100 additively changed genes, 95 showed no statistical evidence of an interaction between dwarfism and CR, while 5 showed evidence of an interaction. (B) A model for the regulation of 212 genes by dwarfism (hypothetical gene 1), 77 genes by CR (hypothetical gene 2), 95 genes additively by CR and dwarfism (hypothetical gene 3), and 5 genes interactively by CR and dwarfism (hypothetical gene 4). The double -headed arrow indicates a physical or other interaction between transcription factors bound to adjacent sites which synergistically alters their activity. The figure is adapted from Tsuchiya et al. (2004), with permission.

CR and dwarfism produced changes in gene expression consistent with increased insulin, glucagon and catecholamine sensitivity, increased gluconeogenesis, protein turnover, lipid β-oxidation, apoptosis, and xenobiotic and oxidant metabolism. They also produced changes consistent with decreased cell proliferation, lipid and cholesterol synthesis, and chaperone expression (Fig. 6). The data also suggest that the combinatorial effects of CR and dwarfism on apoptosis, glycolysis, signal transduction, translation, and RNA splicing are key to their physiological effects. Both treatments also strongly down-regulated genes associated with cholesterol, fatty acid and lipid biosynthesis.

Figure 6.

Cellular processes responsive to dwarfism and CR in mice. Gene ontology classifications of regulated genes were determined manually by examination of the PubMed, GenBank, NCBI, GeneCards, NetAffx, EMBL Bioinformatic Harvester, LocusLink, and MGI online databases. The consensus functional classification of each gene was judged by examination of the relevant literature. Identical gene ontology groups are boxed together.

A number of other genetic mouse models have been reported to enhance longevity, although the evidence is presently less robust than that for dwarfism (Barger et al., 2003; Liang et al., 2003; Kurosu et al., 2005). Nevertheless, these models may provide additional biomarkers for drug discovery. For example, overexpression of the Klotho gene product, which binds to a cell-surface receptor to suppress action of the insulin/insulin-like growth factor-1 signaling pathway, can extend mouse lifespan by 20–30% (Kurosu et al., 2005). Microarray analysis of these mutants may provide additional gene-expression biomarkers for drug discovery.

Dietary manipulations other than CR have been reported to extend the lifespan of mammals, but the evidence for most of these is much less robust than for CR. Methionine restriction extends the lifespan of rats by up to 42% (Orentreich et al., 1993; Richie et al., 1994; Miller et al., 2005). Restriction of dietary tryptophan also extends the lifespan of rats and mice (Segall & Timiras, 1976; De Marte & Enesco, 1986; Ooka et al., 1988). However, interpretation of these results is problematic due to the small sample sizes, high early-life mortality, and modest lifespan effects. Every-other-day feeding (intermittent fasting) was reported to extend maximum lifespan (Goodrick et al., 1983). However, only a single study, published more than 20 years ago, reports a lifespan effect. This study suffers from a number of major confounds including unmeasured food consumption and variable amounts of wheel running among groups.

Distinguishing the beneficial effects of CR from its negative effects

As discussed above, CR has both positive and negative physiological effects, as judged from a human perspective. One aspect of early-stage drug screening that we would like to develop is the capability to distinguish positive from negative physiological outcomes on the basis of gene-expression biomarkers. As a model for interpreting gene-expression profiles, we reasoned that low insulin diabetes might produce a gene-expression signature partially overlapping that of LTCR. Forty percent LTCR in mice is characterized by a 50% reduction in fasting insulin levels (Dhahbi et al., 2001). To explore this hypothesis, we profiled the effect of streptozotocin-induced diabetes (SID) on transcript levels in mouse liver (Dhahbi et al., 2003). We found that SID, like LTCR, enhances the expression of hepatic genes associated with protein degradation and apoptosis (Fig. 6). However, the similarities between SID and LTCR end there. While LTCR enhances transcript levels associated with cell and protein renewal, SID enhances gene expression associated with reduced cell and protein renewal. These results suggest that it is important to consider the entire transcript profile when inferring biological outcomes. The combinatorial effects of genes, and not simply the over- or underexpression of individual genes can lead to very different outcomes.


CR begun relatively late in the lifespan of mice begins rapidly to decelerate mortality, extend remaining lifespan, and delay the onset and/or progression of cancer as a cause of death. These health and longevity effects are coincident with the induction of many LTCR-responsive genes in the liver. These rapid effects appear to be causally linked to the beneficial physiological outcomes of the treatment. CR-related gene-expression profiles were successfully used as biomarkers for identifying a potential CR mimetic. Distinct patterns of gene expression induced by dwarfism and low insulin diabetes indicate that the combinatorial effects of many genes are important in reproducing the health- and lifespan extending effects of CR and dwarfism. The results indicate that longevity enhancing medicaments can be developed or discovered using gene-expression biomarkers.


I would like to thank my colleagues for their excellent work, which I have described above. Many of our studies were supported by unrestricted gifts from the Life Extension Foundation.