• aging;
  • daf-2;
  • gene expression;
  • individuals;
  • microarray nematodes


  1. Top of page
  2. Summary
  3. Introduction
  4. Results
  5. Discussion
  6. Experimental procedures
  7. Acknowledgments
  8. References

We compare the aging of wild-type and long-lived C. elegans by gene expression profiling of individual nematodes. Using a custom cDNA array, we have characterized the gene expression of 4–5 individuals at 4 distinct ages throughout the adult lifespan of wild-type N2 nematodes, and at the same ages for individuals of the long-lived strain daf-2(e1370). Using statistical tools developed for microarray data analysis, we identify genes that differentiate aging N2 from aging daf-2, as well as classes of genes that change with age in a similar way in both genotypes. Our novel approach of studying individual nematodes provides practical advantages, since it obviates the use of mutants or drugs to block reproduction, as well as the use of stressful mass-culturing procedures, that have been required for previous microarray studies of C. elegans. In addition, this approach has the potential to uncover the molecular variability between individuals of a population, variation that is missed when studying pools of thousands of individuals.


  1. Top of page
  2. Summary
  3. Introduction
  4. Results
  5. Discussion
  6. Experimental procedures
  7. Acknowledgments
  8. References

It is of significant interest to identify gene expression changes important to aging, both to further our understanding of the aging process, and to serve as biomarkers of physiological age. The nematode C. elegans is an excellent model system for such studies. Its aging has been well characterized, and numerous mutants are available that significantly increase the nematode mean and maximum lifespan. We hypothesize that comparing age-related gene expression in wild-type animals to that in a long-lived mutant will provide insight into molecular mechanisms associated with wild-type aging. Arguably the best-studied long-lived C. elegans mutant is daf-2, an insulin-receptor homolog involved in dauer formation that, when mutated, can double the lifespan of C. elegans (Kenyon et al., 1993). Despite numerous elegant studies revealing that daf-2 likely acts through an insulin-signalling type pathway, ultimately activating the transcription factor daf-16 (Dorman et al., 1995; Kimura et al., 1997; Ogg et al., 1997; Gems et al., 1998; Paradis & Ruvkun, 1998; Tissenbaum & Ruvkun, 1998; Wolkow et al., 2000), the mechanism by which lifespan is extended in daf-2 mutants is still not understood.

Microarray analysis allows the parallel study of the expression of thousands of genes. This technology has been used to generate whole-genome characterizations of ‘normal’ aging in the model systems C. elegans (Lund et al., 2002), and Drosophila melanogaster (Pletcher et al., 2002). Several studies have characterized gene expression in dauer larvae (Jones et al., 2001; Holt & Riddle, 2003; Wang & Kim, 2003) or young daf-2 adults (McElwee et al., 2003; Murphy et al., 2003). We analyse gene expression in N2 and daf-2 nematodes through 19 days of age, approximately 40% of the daf-2 lifespan, and compare age-related gene expression changes between the two genotypes to facilitate the identification of genes important to normal aging.

To date, microarray studies using C. elegans have required pooling thousands of individuals to obtain sufficient RNA. When co-ordinated changes in gene expression are expected, and when synchronized populations are relatively easy to obtain (e.g. during development), studying large pools is desirable. However, gene expression changes related to aging are not expected to be synchronous since aging in the nematode exhibits a significant stochastic component (Herndon et al., 2002). Additionally, obtaining large populations of synchronized nematodes of advanced age is practically difficult, requiring mass culture of a terrestrial organism in liquid or solid media, in the presence of fertility-preventing drugs, or use of temperature-dependent sterile mutant strains. Therefore, we have developed a method to obtain gene expression profiles of individual nematodes.

Our experimental approach using individual C. elegans has the additional advantage of avoiding the potential masking of intra-individual variability that can occur by pooling; when studying a pool, expression of a gene (an antimicrobial gene, for example) might be abnormally high in a subset of a population (individuals infected by a pathogen, for example), such that the detected gene expression value for the population is skewed. For this reason, it has been argued that pooling of samples is inadvisable as a general practice (Yang & Speed, 2003). By studying multiple individuals, one can identify expression patterns that are common to the aging process between individuals, as well as differences. Analysing gene expression profiles of sufficient individuals may allow the detection of different molecular subclasses within an aging population, as has been demonstrated by the identification of subclasses of cancer within morphologically indistinguishable tumour samples (Alizadeh et al., 2000).

Here, for the first time, we report the analysis of gene expression profiles from individual nematodes, both N2 wild-type and the long-lived strain daf-2, over the wild-type C. elegans lifespan. Using statistical and analytical tools developed specifically for microarrays, we identify classes of genes that change with age in both strains of worms, as well as genes that differentiate normal- from long-lived aging nematodes. The latter are candidates to play a role in modulating lifespan in the nematode.


  1. Top of page
  2. Summary
  3. Introduction
  4. Results
  5. Discussion
  6. Experimental procedures
  7. Acknowledgments
  8. References


We collected gene expression data for 4–5 individuals at 4 distinct ages throughout the lifespan of wild-type N2 nematodes, and at the same ages for 4–5 individual long-lived daf-2 nematodes using a custom-made ‘stress’ cDNA array containing 921 genes, each spotted in quadruplicate (for details, see methods). Agreement among the technical replicates (replicate spots) was extremely tight (median per gene coefficient of variance = 6.0%).

Our experimental design involved assaying each individual nematode against a common reference RNA made from a mixed-stage culture of nematodes. This common reference design allowed us to compare all of the array results to each other. To minimize any effect of day-to-day experimental variation, labelling reactions and array hybridizations were timed, such that representative worms from each of the time points were processed simultaneously. This approach distributes any potential method-related temporal effects equally across all groups. After normalization, the 37 arrays exhibit a similar distribution and range of differential expression values (Fig. 1). As a control, we hybridized one sample six different times on six replicate arrays throughout the course of the experiment, allowing us to have high confidence that there were no changes in array quality or expression quantification over time (data not shown).


Figure 1. Boxplots showing collective differential expression values for all individual nematodes used in the study (N = 37 animals). For each individual array, M, a standard representation of differential expression (log2(Experimental/Reference) (Bowtell & Sambrook, 2003)), is displayed after global loess normalization. Each box represents the interquartile range (difference between upper and lower quartiles (Bowtell & Sambrook, 2003)). The first four groups are N2 (groups are different shades), while the second four are daf-2. Each boxplot is derived from a single animal.

Download figure to PowerPoint

Analysis of gene expression with age and genotype

We began analysis with an unsupervised approach, i.e. using unbiased methods to reveal patterns in the data rather than searching out the behaviour of specific candidate genes. A nested F-test run in the package ‘Limma’ (Linear Models for microarray data) in the R computing environment was conducted to rank genes that best differentiated the two genotypes with respect to time (Table 1). In addition, linear modelling and an empirical Bayes analysis (see methods) was applied to pairwise comparisons to identify genes likely to be significantly differentially expressed at older ages relative to four-day-old animals of each genotype (Table 2), and differentially expressed between genotypes at each age (Table 3).

Table 1.  Genes that differentiate N2 and daf-2 with respect to age
Primer pairAnnotationF-statistic
F28D1.5Member of the thaumatin-like [sweet-tasting] protein family27.78
T27E9.2Small protein with strong similarity to human and bovine ubiquinol-cytochrome C reductase complex 11 Kd protein precursors13.19
Y38A8.2Member of the proteasome subunit protein family11.60
T04C12.6Member of the actin and actin-related protein family 9.12
F28H6.1Serine/threonine protein kinase that transduces signal from AGE-1 to antagonize DAF-16 transcription factor 8.43
F36D3.9Member of the cysteine protease protein family 8.27
F09C6.5Member of the serine protease inhibitor protein family 8.18
F55F3.1Protein with strong similarity to beta subunits of human AMP-activated protein kinase (AMPK) 7.98
F40E10.1hch-1; Putative zinc-dependent metalloprotease 7.80
F30F8.2Protein with strong similarity to Rat mitochondrial glutaminase 7.64
M60.1Putative serine proteinase 7.63
F49A5.6Member of the thaumatin-like [sweet-tasting] protein family 7.46
T15B7.4Putative collagen, has similarity to human COL5A2, alpha-2 collagen, type V 7.07
F42A10.7Protein of unknown function, has strong similarity to C. elegans F42A10.6 6.98
W09D10.3Putative mitochondrial L12 ribosomal protein 6.41
K07G5.2Putative DNA repair protein with similarity to human and D. melanogaster XPA excision repair proteins, involved in xeroderma pigmentosum 6.28
R07E5.2Member of the thiol antioxidant (thioredoxin peroxidase) protein family 6.19
F49E12.2Member of the calpain protease protein family 6.11
F49E8.5dif-1; Member of the mitochondrial carrier protein family involved in embryonic differentiation 5.97
Table 2.  Genes differentially expressed as a function of time
A. N2
Primer pairAnnotation9 days14 days19 days
M*P-value B-statisticM*B-statisticP-valueM*P-valueB-statistic
F28D1.5Member of the thaumatin-like [sweet-tasting] protein family1.810.0006575.41   −1.722.46E-04 4.97
F49A5.6Member of the thaumatin-like [sweet-tasting] protein family1.480.0104942.56      
M60.1Putative serine proteinase   −1.291.89E-0611.57−0.954.97E-04 4.00
Y46H3 AeMember of the C. elegans hsp-16 family    1.713.58E-05 8.15   
C11E4.7Protein of unknown function    1.573.58E-05 7.63   
T27E4.2hsp-16.11    1.623.58E-05 7.51   
Y46H3 Adhsp16–2    1.504.66E-05 7.06   
F36D3.9Member of the cysteine proteaseprotein family    1.896.15E-05 6.63   
F42A10.7Protein of unknown function, has strong similarity to C. elegans F42A10.6   −1.531.80E-04 5.48−1.542.25E-04 5.59
W07B8.5cpr-5; Member of the cysteine protease protein family      −2.045.23E-0712.79
K07G5.2Putative DNA repair protein with similarity to human and D. melanogaster XPA excision repair proteins, involved in xeroderma pigmentosum       0.855.43E-05 7.78
F44C4.3cr-4 cysteine protease      −2.187.11E-05 7.15
Y39B6B.gasp-1; Aspartyl protease of the protease protein family      −1.791.16E-04 6.41
F20D1.5GTP-binding protein of the arf family ras superfamily       1.512.46E-04 5.00
Y40H7 A10Member of the thiol protease protein family      −1.062.46E-04 4.96
C55B7.4Member of the acyl-CoA dehydrogenase protein family      −1.562.46E-04 4.96
F21F8.4Member of the protease protein family – possible aspartyl protease      −1.562.83E-04 4.73
Y39B6B.gAspartyl protease of the protease protein family      −1.864.34E-04 4.24
W07B8.4Member of the cysteine protease protein family      −0.924.97E-04 3.96
Y47H9C.6Protein of unknown function, has a region with similarity to the ICE-like protease [caspase] p10 domain      −1.085.27E-04 3.84
H22K11.1Possible aspartyl rotease and an ortholog of human cathepsin D      −1.741.75E-03 2.66
C52E4.1gcp-1; cpr-1; Cysteine protease expressed in the intestine      −1.882.10E-03 2.43
Y38A8.2Member of the proteasome subunit protein family       0.632.98E-03 2.05
B. daf-2
Primer pairAnnotation9 days14 days19 days
  • *M = log2 differential expression relative to 4-day-old N2.

  • **

    M = log2 differential expression relative to 4-day-old daf-2.

F28D1.5Member of the thaumatin-like [sweet-tasting] protein family 2.613.23E-1019.68 3.511.00E-1531.82 1.761.30E-04 5.45
ZK721.2troponin-I−1.101.08E-04 7.12−1.211.63E-05 8.05−1.414.34E-0712.34
T04C12.6Member of the actin and actin-related proteinfamily 0.692.23E-04 5.88−0.831.07E-05 8.89−0.802.27E-05 7.74
H05B21.4G-protein coupled receptor of unknown function 1.362.23E-04 5.80−1.881.78E-0713.18−1.931.35E-0714.10
C34D4.15Putative collagen−1.364.10E-04 5.03      
F49E8.5dif-1; Member of the mitochondrial carrier protein family involved in embryonic differentiation 0.765.29E-04 4.63   −0.794.56E-04 4.18
ZC373.7strong similarity to collagen 1.421.13E-03 3.79−1.575.07E-04 4.51−2.067.10E-07 1.49
F14H3.3Protein with moderate similarity to C. elegans F14H3.5−0.781.62E-03 3.32      
C12D12.6Putative reverse transcriptase 1.172.07E-03 2.99   −1.411.30E-04 5.49
F30F8.2Protein with strong similarity to Rat mitochondrial glutaminase 0.493.05E-03 2.53−0.532.56E-03 2.66−0.558.69E-04 3.50
F23H12.4Collagen−1.504.55E-03 2.08−1.652.56E-03 2.67−1.961.03E-04 6.00
Y46H3 Adhsp16–2; Member of the heat shock HSP16-1 protein family    1.335.24E-04 4.35   
T27E9.2Small protein with strong similarity to human and bovine ubiquinol-cytochrome C reductase complex 11 Kd protein precursors   −0.707.86E-05 6.40−0.794.56E-06 9.46
T15B7.4Putative collagen   −2.081.07E-05 8.65−1.935.07E-05 6.81
C07B5.5nuc-1; Endonuclease with strong similarity to human DNase II, involved in DNA degradation during apoptosis      −1.383.47E-03 2.07
M01E11.5Protein containing a cold-shock (RNA-binding) domain      −0.861.05E-03 3.25
T04C12.4actin      −0.661.30E-04 5.66
Table 3.  Genes differentially expressed between genotypes at each age
Primer pairAnnotation4 days9 days14 days19 days
H05B21.4G-protein coupled receptor of unknown function 1.641.63E-059.11         
F40E10.1hch-1; Putative zinc-dependent metalloprotease−1.031.63E-058.67−0.850.0004965.25      
T15B7.4Putative collagen 2.041.63E-058.26         
K02F6.1Member of the protease protein family 1.661.63E-058.12         
Y49A3 A4unk, similar to f48g7.5 1.401.63E-058.05       1.116.92E-04 3.13
M60.1Putative serine proteinase−1.122.06E-057.66         
Y59C2 A1Member of the metallocarboxy peptidase protein family−0.806.94E-056.37         
ZK721.2troponin-I 1.101.29E-045.66         
ZC373.7has strong similarity to collagen 1.631.67E-045.31         
F23H12.4Collagen 1.862.38E-044.88         
C12D12.6Putative reverse transcriptase 1.362.65E-044.69         
F30F8.2Protein with strong similarity to Rat mitochondrial glutaminase 0.574.16E-044.19         
F55A11.4Member of the mitochondrial carrier (MCF) protein family 0.827.27E-043.60         
F46H5.3Member of the arginine kinase, phosphotransferase protein family 1.108.87E-043.34         
C34D4.15Putative collagen 1.301.03E-033.13         
B0228.5Member of the thioredoxin protein family−0.631.40E-032.74         
Y55D5 A_391.adaf-2−1.031.40E-032.72         
F39G3.5Protein with similarity to human cytochrome b-561 (CYB561) 0.581.40E-032.72         
C04F12.9Putative ribonuclease H    1.300.0003366.12    1.293.52E-04 4.87
F49E12.2Member of the calpain protease protein family−1.770.0003366.05−1.999.10E-057.23      
F09C6.5Member of the serine protease inhibitor protein family    1.570.0004965.06      
F49A5.6Member of the thaumatin-like [sweet-tasting] protein family   −1.500.0010744.14      
C15F1.bsod-1    0.670.0027043.13      
K05C4.1Member of the proteasome subunit protein family   −0.960.0029152.92      
F28D1.5Member of the thaumatin-like [sweet-tasting] protein family       2.278.39E-0712.19 2.323.84E-0713.07
E04A4.7Cytochrome c      −0.861.52E-03 4.29−0.816.07E-04 3.34
T27A1.4Putative ligand-gated ion channel       1.266.87E-03 2.65 1.316.40E-04 3.24
VW06B3R.1Protein with similarity tomitochondrial processing peptidases       0.967.63E-03 2.20   
F36D3.9Member of the cysteine protease protein family      −1.507.63E-03 2.19   
F28H6.1Serine/threonine protein kinase that transduces signal from AGE-1 to antagonize DAF-16 transcription factor         −0.691.60E-04 6.76
C36B1.4Member of the proteasome subunit protein family         −1.002.05E-04 6.15
T24D1.3Protein with strong similarity to C. elegans T24D1.2 gene product         −0.942.80E-04 5.59
R05H5.3Putative nuclear thioredoxin         −0.752.94E-04 5.34
B0284.1Protein of unknown function, contains putative coiled-coil domains         −1.533.52E-04 4.78
Y38A8.2Member of the proteasome subunit protein family         −0.743.52E-04 4.73
Y40H7 A10Member of the thiol protease protein family          1.033.83E-04 4.55
C03H5.1Member of the C-type lectin family          1.074.37E-04 4.27
F56F10.1Member of the carboxypeptidase protein family          1.004.37E-04 4.22
F55F3.1Protein with strong similarity to beta subunits of human AMP-activated protein kinase (AMPK)         −0.654.37E-04 4.15
F01G4.2ERAB homolog?         −0.644.37E-04 4.05
F39H11.5Protein with moderate similarity to S. cerevisiae PRE4 (Proteasome subunit beta7_sc)         −0.584.37E-04 3.99
T02G5.6Member of the MutS homolog (MSH) protein family         −0.924.37E-04 3.90
F23F1.6Member of the amino acid permease family          0.884.37E-04 3.89
K07E8.3Protein of unknown function         −0.585.44E-04 3.63
C55B7.4Member of the acyl-CoA dehydrogenase protein family          1.455.99E-04 3.49
T05H10.6Member of the dehydrogenase E1 component protein family         −0.766.07E-04 3.39
Y57G11C.12Protein with strong similarity to human NADH- ubiquinone oxidoreductase subunit CI-B14 (NDUFA6)         −0.766.07E-04 3.34
F35C8.8Protein of unknown function          1.371.23E-03 2.56
W02D3.6Member of the mitochondrial carrier (MCF) protein family         −0.821.25E-03 2.51
T27E9.2Small protein with strong similarity to human and bovine ubiquinol-cytochromeC reductase complex 11 Kd protein precursors         −0.571.71E-03 2.18
*M = log2 differential expression of daf-2 relative to the same age N2.

For the genotype–temporal interaction, the individual expression values from each array were plotted for the five genes with the highest F-statistics (Fig. 2). All of these genes also exhibited a high B-statistic in at least one of the pairwise comparisons (Fig. 2). High values of the B-statistic coupled with an appropriate experimental design indicate a strong likelihood that such genes are significantly differentially expressed (Smyth, 2004). Therefore we also plotted the expression profiles of several genes with high B-statistics to visually confirm differential expression across time or between genotypes (Fig. 3).


Figure 2. Top five ranked genes that differentiate aging N2 from daf-2, determined by the F-test. Individual M-values are plotted (log2) for each age-group, 4–5 individuals per age, per genotype. Squares are N2, circles are daf-2; dashed lines show the mean expression levels over time for daf-2, while solid lines show mean expression for N2 with age. On the right are tables showing the B-statistics (log-odds, derived from empirical Bayes analysis, see methods for details) resulting from the pairwise comparisons at each age (d = days of age). A blank field indicates the B-statistic was less than 2. A grey field indicates the comparison was not conducted. Tables 2 and 3 have additional information relevant to this Figure.

Download figure to PowerPoint


Figure 3. Expression values for genes identified by pairwise comparisons through empirical Bayes analysis as having a significant likelihood of differential expression as a function of age and/or genotype. Shown are three families of genes. Top row:collagens, second row: hsp-16 s; third row: proteases. A more complete list can be seen in Table 2. Individual M-values are plotted (log2) for each age-group, 4–5 individuals per age, per genotype. Squares are N2, circles are daf-2; dashed lines show the mean expression levels over time for daf-2, while solid lines show mean expression for N2 with age.

Download figure to PowerPoint

To identify groups of genes with similar temporal expression patterns, we calculated the average expression values for each group of individual nematodes, and clustered the genes for N2 (Fig. 4a) and daf-2 (Fig. 4e) with respect to time. Two clusters were readily apparent in the N2 data (Fig. 4b), which corresponded to one group of genes enriched in proteases (28/80), and a small group enriched in hsp-16 family members (4/6). The expression values were averaged for all members of each of these groups, and the means plotted for each genotype (Fig. 4c). The group containing hsp-16 genes also formed a cluster in the daf-2 data (Fig. 4d), but the group enriched in proteases did not have a similar expression profile in daf-2 (Fig. 4c).


Figure 4. Hierarchical clustering of genes expressed in N2 and daf-2 with respect to age identifies one common cluster and one cluster unique to N2. The mean M for all individuals in a given group (age and genotype) was calculated for each gene, and the genes clustered. Genes were clustered for the N2 data set (a) or daf-2 data set (e). Two readily observable clusters of genes with similar decreasing or increasing temporal expression patterns are displayed on an expanded scale (b). The expression values for the members of these clusters were averaged for N2 and daf-2, and plotted (c) (Solid squares–N2, open circles-daf-2). The hsp-16 enriched cluster also appears independently in the daf-2 cluster results (d).

Download figure to PowerPoint

For a global look at gene expression over all time points and genotypes, we conducted hierarchical clustering of all 37 arrays. The arrays clustered primarily according to genotype, rather than by age (Fig. 5).


Figure 5. Hierarchical clustering reveals that genotype is a greater influence on gene expression profiles than age. Clustering of individual animals reveals segregation by genotype, rather than age. A clear separation can be seen between N2 and daf-2, with no apparent clustering of age effects.

Download figure to PowerPoint

Analysis of variance among individuals

One advantage of studying gene expression in individuals is the potential to identify dysregulation or loss of regulation of genes with age, detected as a change in the variance of gene expression among individuals in a group. A generalized linear model (GLM) detected no global trend in the data towards increased variance in gene expression with age (data not shown). To identify any genes for which the variance in expression values differed across time points, we applied a Bartlett test to the data. Only one gene was identified in each genotype: for N2 this gene was T12A2.3 (a translation initiation factor), whose variance in gene expression peaks at 14 days, and for daf-2 the outlier gene was F25H2.13, a helicase, whose variance is highest at 4 days.


  1. Top of page
  2. Summary
  3. Introduction
  4. Results
  5. Discussion
  6. Experimental procedures
  7. Acknowledgments
  8. References

Studying individual nematodes provides practical and statistical advantages for the study of aging in C. elegans with microarrays. Using individual worms allows one to avoid the complex culture and isolation protocols that are necessary to obtain sufficient. C. elegans material for traditional studies. In the present study, nematodes were cultured under standard conditions on agar plates with standard densities of animals per plate, and individuals were picked directly into sterile water and flash-frozen until the RNA was prepared. Obtaining the hundreds of thousands of nematodes that are required to obtain sufficient RNA for traditional studies necessitates growing worms at high density, either in liquid culture or periodically washing them from plates and subjecting them to potentially stressful sedimentation and or filtration to remove progeny and dead worms from the population. Under these conditions, nematodes are less healthy, as evidenced by a reduced mean lifespan (Fabian & Johnson, 1994), and age-related gene expression has been demonstrated to be affected by these types of culture conditions (Fabian & Johnson, 1995). In addition, drugs and/or additional mutations are frequently used to make the worms sterile to aid in reducing contamination by young progeny. Using our approach, these time-consuming steps are avoided, and only healthy nonstressed worms uncontaminated by young or dead individuals, are obtained. This is a significant advantage to the study of aging, in which it becomes increasingly difficult to obtain a large, clean, synchronized population at later ages.

Studying individual nematodes is also an advantage to the study of aging because a population of nematodes does not age synchronously, but exhibits a significant stochastic component (Herndon et al., 2002). Using samples of pooled nematodes is an excellent way to detect robust changes that occur concurrently or synchronously in a population, but such an approach will miss interindividual variability that may exist within the population. The single-worm approach will allow the examination of individual samples for the purpose of analysing variance or identifying subclasses of individual aging. When combined with observations of individual nematodes’ behaviour and appearance, gene expression profiles of those individuals may identify correlations between genes and phenotypes.

Using statistical tools developed for microarray analysis (‘Limma’, Smyth (2004)), we have identified those genes on our stress array that best differentiate aging N2 from aging daf-2 (Table 1), and also genes whose expression changes either in response to age (Table 2) or as a result of the daf-2 mutation (Table 3). The genes with the 19 largest F-statistics are listed in Table 1, but we find that genes with F-statistics greater than two (about 170 genes) reveal reliable differences between genotypes. The gene that most strongly differentiates daf-2 from N2 with respect to time is F28D1.5. This gene is a thaumatin family member, a group of allergenic proteins that can have antifungal properties (Krebitz et al., 2003). F28D1.5 displays a similar but delayed expression pattern in daf-2 relative to N2, perhaps marking the delayed aging of daf-2 (Fig. 2a), or revealing a delayed need for antimicrobial defenses in daf-2. Interestingly, expression of F28D1.5 peaks in mid-life then falls, rather than increasing throughout the lifespan. Antibacterial peptides were found to increase with age in drosophila (Pletcher et al., 2002), perhaps indicating that this form of environmental stress is a common aspect of aging between drosophila and C. elegans.

The next top four ranked genes that most strongly differentiate N2 aging from daf-2 aging (those with the highest F-statistic in Table 1) each represent different interesting phenomena. All are more highly expressed in old N2 relative to old daf-2 (Fig. 2). The second ranked gene on the list, T27E9.2, which encodes an electron transport chain (ETC) component of the mitochondria, decreases in expression with age in daf-2, while increasing in N2. This suggests a difference between mitochondrial function or metabolism in aging daf-2 animals. This idea is further supported by the identification, in our pair-wise comparisons, of several other mitochondrial proteins, including another ETC component and a carrier protein, whose expression decreases with age in daf-2, but not in N2 (Tables 2 and 3). The increase in expression in N2 of the T27E9.2 transcript may be indicative of metabolic demand, as an increase in mitochondrial mass is a frequent byproduct of metabolic dysfunction. Previous screens using RNAi have also indicated that mitochondrial function plays an important role in nematode aging (Dillin et al., 2002; Lee et al., 2003).

The third ranked gene revealed by the F-statistic analysis, Y38A8.12, encodes a proteasome component. This increases in expression in N2 but not daf-2 with age, suggesting that N2 may require greater protein degradation capability at older ages than daf-2. This is also supported by the pairwise comparisons, in which additional proteasome components are identified as relatively more highly expressed in 19-day-old N2 than daf-2 (Table 3). Perhaps this transcriptional increase reflects an increase in the abundance of malformed proteins that is accelerated in N2 relative to daf-2. Though our data set covers the lifespan of N2 nematodes, it only covers approximately 40% of the daf-2 lifespan. It will be interesting to determine whether the change in expression of the proteasome is attenuated throughout the lifespan of daf-2, or if the change in expression of these genes reaches the same magnitudes on a delayed time scale in daf-2.

The fourth ranked gene revealed by F-statistic analysis, T04C12.8, encodes an actin. This decreases in expression with age in daf-2 while increasing with age in N2. Two actin-related genes also are identified in the pairwise comparison as changing with age in daf-2 (Table 2). This increase in actin expression in aged N2 might represent a compensatory mechanism in response to the muscle degeneration observed in aged wildtype nematodes (Herndon et al., 2002).

The fifth gene on the list, F28H6.1, encodes akt-2, an Akt/PKB serine/threonine protein kinase. Expression of this gene increases with age in N2 but decreases in daf-2 (Fig. 2). This is of interest since this gene is known to be involved in signal transduction from daf-2 to the transcription factor daf-16 (Paradis & Ruvkun, 1998). Normally acting to suppress daf-16 activity, an increase in akt-2 in older N2 might contribute to N2′s shorter lifespan, compared to daf-2. Little is known about the transcriptional regulation of akt-2 in C. elegans.

The pairwise comparisons with respect to age and genotype also identify other classes of genes whose expression changes in interesting ways. Many of the genes with high B-statistics with age in N2 are proteases (Table 2A), though these are not identified by the same comparisons in daf-2 (Table 2B). Proteases are substantially enriched in one node of the clustered genes in the N2 data (Fig. 4): 28/80 (35%) genes in this node are proteases, while 11% of the genes on the array are proteases. The genes in this node decrease in expression over time (Fig. 4). This clustering pattern is not seen in the clustered daf-2 data, and in fact these genes do not display a similar expression profile in daf-2 (Fig. 4). This suggests that maintaining high expression of these proteases is related to the extended lifespan of daf-2 nematodes. Relevant to this point is the recent report that RNAi of an aspartyl protease, ZK383.4, decreased the lifespan of daf-2 significantly, while decreasing N2 lifespan only slightly (McElwee et al., 2003). This protease was not included on our array, but aspartyl proteases make up a significant portion of the genes whose expression both changed with age in N2, and differed between old N2 and old daf-2 on our data.

The pairwise comparisons and the cluster analysis of our data set uncovered a common temporal pattern of small heat-shock protein expression in both N2 and daf-2 nematodes (Tables 2 and 3, Fig. 4). A similar temporal pattern of heat-shock gene expression, in which expression peaks in mid-life then declines at later ages, was previously described by Lund et al. in wild-type C. elegans aging (Lund et al., 2002). Our results would indicate that this pattern plays no role in the different lifespans of N2 and daf-2 nematodes.

Hierarchical clustering of all 37 arrays (Fig. 5) reveals that physiological age has relatively less impact than does the daf-2 mutation on the expression of the genes on our array: rather than clustering with respect to age, the arrays cluster primarily by genotype. This is consistent with the only other microarray study on aging in nematodes (Lund et al., 2002), which found that relatively few genes changed with age.

One previous microarray study of aging Drosophila melanogaster (Pletcher, 2002) examined the question of global dysregulation of gene expression with age. In comparing three populations of 50 flies each, they found no evidence of systematic increases in variance of gene expression. Our approach, looking at gene expression at the level of the individual, may help further characterize this phenomenon. We also found no evidence of a global, age-related increase in the variance of the expression of the genes on our array.

At the level of individual genes, we identified two genes, one in each genotype, whose variance in expression differed between ages. One of these, T12A2.3, is predicted to be a translation initiation factor, which is interesting since variance in such a protein could cause variance in the level of many proteins. The variance in this gene's expression peaked at 14 days in N2, though it had a constant, very low variance in all ages of daf-2. The significantly variant gene identified in daf-2, F25H2.13, is predicted to be a helicase. This gene has a higher variance at 4 days in daf-2, and a constant, low variance in older daf-2 and all N2 populations. This approach to the analysis of variance will be better able to detect differences in a larger study with more individuals per group.

To better characterize how daf-2 mutations extend lifespan, two recent studies have identified genes that are candidates for regulation by daf-16 (the downstream effector of daf-2). Both studies compared gene expression in young adult daf-2 animals to that of daf-2; daf-16 animals (McElwee et al., 2003; Murphy et al., 2003). Although this approach is not the same as comparing daf-2 to wild-type individuals as was done in our study, there are some similarities in results. For example, Murphy et al. found a thaumatin homolog, F28D1.3, to be up-regulated by daf-2 RNAi, and RNAi of this gene reduced daf-2 lifespan slightly (Murphy et al., 2003). Though this gene does not change dramatically in our data, expression of another thaumatin homologue, F28D1.5, as mentioned above, changes significantly throughout the lifespans of both genotypes and significantly differentiates aging daf-2 from N2.

In conclusion, we have demonstrated the feasibility and utility of microarray analysis of gene expression in individual nematodes. This approach has the practical advantage of allowing the collection of healthy, age-confirmed nematodes known not to be contaminated by young or dead animals. Using individual nematodes, we have identified temporal patterns in heat-shock protein gene expression that were previously observed using another experimental approach with C. elegans microarrays (Lund et al., 2002). In addition, we have identified genes that differentiate aging N2 from aging daf-2. Such genes that show a delayed or attenuated response when comparing normal aging to that of a long-lived mutant are implicated in the modulation of aging in wild-type C. elegans.

Experimental procedures

  1. Top of page
  2. Summary
  3. Introduction
  4. Results
  5. Discussion
  6. Experimental procedures
  7. Acknowledgments
  8. References

Nematode culture: C. elegans strains N2 and daf-2 (e1370) were cultured at 20 °C on NGM plates seeded with OP50 bacteria as described (Wood, 1988). Worm strains and OP50 were provided by the Caenorhabditis Genetics Center, which is funded by the NIH National Center for Research Resources (NCRR). Synchronized populations of nematodes were initiated by conducting timed lays, during which 35–40 gravid adults were allowed to lay eggs on NGM plates for 2.5 h, after which time the adults were removed. The resulting eggs were allowed to hatch and develop at 20 °C. On the third day after hatching (day 3), the young-adult worms were picked to new plates for use in the survival experiments. Worms were cultured at a density of 100 per 10 cm plate, with initial numbers of 200 N2 and 100 daf-2 per experiment. Worms were picked to fresh plates at least every other day, at which time they were scored as dead or alive based on response to gentle prodding with a platinum wire. At the desired time points, 10 individuals were picked individually into 200 µL PCR tubes containing 10 µL H2O and frozen immediately in dry ice. Per time point, all individuals were harvested within 90 s.

Construction of microarray: The custom cDNA ‘stress’ array consisted of 923 PCR products prepared under standard conditions, representing 921 genes, as well as negative controls (printing buffer, without any DNA target), each spotted in quadruplicate. Genes were chosen based on potential mitochondrial localization, involvement in a stress response, and interest to researchers at the Buck Institute. PCR primers were synthesized by Gorilla Genomics, based on sequences available at Spotting of PCR products and PCR reaction setup was facilitated through use of a volumetric robot (Genesis, RSP150, Tecan), and verification of the amplification of unique products for each gene by gel electrophoretic sizing of all amplicons was carried out. Amplified products were transferred to 384-well plates and purified using Millipore purification plates (Millipore) according the manufacturers protocol. The purified PCR products were transferred to 384-well print plates (Genetix), dried, and re-suspended in 6 µL print buffer (50% DMSO in 0.04X SSC). Spotted cDNA microarrays were generated on commercially available glass slides (UltraGAPS, Corning) with a commercial microarrayer (Omnigene, GeneMachines) at the Genomics Facility of the Buck Institute. PCR failures, scored as no product or multiple products, were filtered out from analysis, leaving 820 genes. Annotations were compiled from WormBase Release WS108 (

Preparation of RNA: Individual C. elegans were collected into PCR tubes containing 10 µL of nuclease-free water and frozen at −80 °C until needed. Total RNA was prepared from each nematode using the Absolutely RNA Nanoprep kit according to the manufacturer's instructions (Stratagene). The common reference RNA was prepared from a mixed-stage culture of N2 grown on NGM plates using Trizol, according to the manufacturer's instructions (Gibco-BRL).

Preparation of aRNA: The entire yield of total RNA from each nematode, or 2 µg of the common reference RNA, were subjected to 2 rounds of linear amplification using the MessageAmp kit as described (Ambion).

Microarray: 200 ng of the aRNA was labelled using the 3DNA Array 350RP kit according to the manufacturers instructions (Genisphere), including the optional Qiagen clean-up step after the ligation. Hybridization was carried out according to the Genisphere protocol. Hybridization was at 42 °C overnight, and washes at 50 °C according to the Genisphere protocol. After scanning with a Packard Bioscience scanner at a resolution of 10 µ (Perkin Elmer), spot intensities were measured using GenePix (Axon).

Analysis: Image files were quantified in Genepix using the default settings (Axon). These files were then imported into GeneTraffic (Iobion) for quality control and archival purposes. Spots that were judged as substandard by visual examination of each slide were flagged and excluded from further analysis. For example, spots that had dust artifacts or spatial defects were manually flagged and excluded (Bowtell & Sambrook, 2003). All microarrays had a low background, and hence background subtraction was not used (Bowtell & Sambrook, 2003). All statistical and graphical analysis was carried out in the R computing environment (1.81, Raqua on an Apple Macintosh) using the Linear models for Microarray data package (Limma, 1.3.13 (Smyth, 2004)), which is part of the Bioconductor project ( (Gentleman et al., 2003).

Within Limma, global loess normalization was carried out for each microarray, which showed that between microarray normalization was not necessary due to the approximate equivalence of the interquartile range between all arrays within the study (37 arrays) (Yang et al., 2002). The 4-replicate spots per gene in each array were used to maximize the robustness of each genes differential expression measurement via the ‘lmFit’ function within Limma (Smyth, 2004). This step uses a pooled correlation estimate to generate a more robust estimate of the gene expression across replicate spots, compared to a straight average of replicate spots. Differential expression of genes was determined using an empirical Bayes (EB) approach within Limma (Smyth, 2004). A discussion of the theory behind empirical Bayes analysis is beyond the scope of this article; for a detailed discussion of the method, and its utility in various microarray paradigms, see (Smyth, 2004). After EB analysis within Limma, genes are ranked as being differentially expressed in decreasing order of the B-statistic (essentially the log-odds of differential expression). No cut-off for significance is specified by this type of analysis, and cut-off for genes which are differentially expressed is best determined empirically through volcano plots, which allows one to see when selected genes are well separated from the data cloud (Yang & Speed, 2003). Limma also facilitates nested F-test analysis, which allowed us to identify genotype–time interactions. The F-tests were carried out using a 2 × 4 factorial design, utilizing contrasts of genotype vs. age via the ‘classifyTests’ command within Limma. Briefly the moderated t-tests that resulted from our EB analysis in combination with our stated contrasts (i.e. time by genotype) were used via the classifyTests command to generate F-statistics that were then ranked in order of decreasing magnitude (see the guide within Limma for additional details of this procedure). False discovery (FDR) for both EB analysis and the nested F-tests were carried out within Limma using the method of Benjamini & Hochberg (Benjamini & Hochberg, 1995).

The Strain–Time interaction can be represented as follows.

4 day old N29 day old N214 day old N219 day old N2
4 day old Daf29 day old Daf214 day old Daf219 day old Daf2

From these interactions, the following contrasts were constructed which were then used in the nested F-test analysis:

9doDaf-9doN2–4doDaf2 + 4doN2; 14doDaf2–14doN2–4doDaf2 + 4doN2; 19doDaf2–19doN2–4doDaf2 + 4doN2.

Using these three contrasts in conjunction with the FDR method described above generated a list of genes that differentiated genotype by time. The top ranked genes in this list are shown in Table 1.

As an additional visual confirmation of differential expression, genes that were identified as being likely to differentiate the two genotypes across time by the F-test, were plotted to empirically determine the validity of the ranked test (Fig. 2).

To identify genes whose variance in expression level differed between time points, we performed a Bartlett test (Bartlett, 1937) for each genotype data set within R, i.e. a test of the null hypothesis that the four variances are the same. Our N for this was either 4 or 5 for each time point. For each gene (N = 820), a Bartlett test statistic was calculated, and a QQ plot generated for each data set for the test statistics against a chi-squared on three degrees of freedom.

Normalized M (log2 experimental/reference) values (Bowtell & Sambrook, 2003), were exported and used in the Cluster 3.0 program, derived from Mike Eisen's program, but written for the Apple Macintosh (Michiel de Hoon, University of Tokyo). Clusters were created by first filtering the data on all arrays via limiting the data to those genes present in at least 80% of the arrays (i.e. 29/37 arrays), and then using Hierarchical clustering across genes and arrays with a similarity metric of uncentred correlation. Visualization of clusters was carried out in Java TreeView on an Apple Macintosh (

The gpr data files used in this report are available from Array express:


  1. Top of page
  2. Summary
  3. Introduction
  4. Results
  5. Discussion
  6. Experimental procedures
  7. Acknowledgments
  8. References

This work was supported by a National Institutes of Health Grant AG18679, and a senior scholar Ellison award from the Ellison Medical Foundation awarded to SM. Special thanks go to Krysta Felkey for microarray work, and Terry Speed and Jing Yi, for valuable advice and helpful suggestions.


  1. Top of page
  2. Summary
  3. Introduction
  4. Results
  5. Discussion
  6. Experimental procedures
  7. Acknowledgments
  8. References
  • Alizadeh AA, Eisen MB, Davis RE, Ma C, Lossos IS, Rosenwald A, Boldrick JC, Sabet H, Tran T, Yu, X et al. (2000) Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling. Nature 403, 503511.
  • Bartlett MS (1937) Properties of sufficiency and statistical tests. Proceedings of the Royal Statistical Society Series A 160, 268282.
  • Benjamini Y (1995) Controlling the false discovery rate: a practical and powerful approach to multiple testing. Journal of the Royal Statistical Society Series B 57, 289300.
  • Bowtell D, Sambrook J (2003) DNA microarrays, a molecular cloning manual. Cold Spring Harbor Lab Press, Cold Spring Harbor, NY, USA.
  • Dillin A, Hsu AL, Arantes-Oliveira N, Lehrer-Graiwer J, Hsin H, Fraser AG, Kamath RS, Ahringer J, Kenyon C (2002) Rates of behavior and aging specified by mitochondrial function during development. Science. 298, 23982401.
  • Dorman JB, Albinder B, Shroyer T, Kenyon C (1995) The age-1 and daf-2 genes function in a common pathway to control the lifespan of Caenorhabditis elegans. Genetics 141, 13991406.
  • Fabian TJ, Johnson TE (1994) Production of age-synchronous mass cultures of Caenorhabditis elegans. J. Gerontol. 49, B145B156.
  • Fabian TJ, Johnson TE (1995) Identification of genes that are differentially expressed during aging in Caenorhabditis elegans. J. Gerontol. 50A, B245B253.
  • Gems D, Sutton AJ, Sundermeyer ML, Albert PS, King KV, Edgley ML, Larsen PL, Riddle DL (1998) Two pleiotropic classes of daf-2 mutation affect larval arrest, adult behavior, reproduction and longevity in Caenorhabditis elegans. Genetics 150, 129155.
  • Gentleman R, Bates D, Bolstad B, Carey V, Dettling M, Dudoit S, Ellis B, Gautier L, Gentry J, Hornik K, Hothorn T, Huber W, Iacus S, Irizarry R., Leisch FC, Maechler M, Rossini AJ, Sawitzki G, Smyth GK, Tierney L, Yang JYH, Zhang J (2003). Bioconductor: a software development project. Technical Report November 2003, Department of Biostatistics, Harvard School of Public Health, Boston, USA.
  • Herndon LA, Schmeissner PJ, Dudaronek JM, Brown PA, Listner KM, Sakano Y, Paupard MC, Hall DH, Driscoll M (2002) Stochastic and genetic factors influence tissue-specific decline in ageing C. Elegans. Nature 419, 808814.
  • Holt SJ, Riddle DL (2003) SAGE surveys C. elegans carbohydrate metabolism: evidence for an anaerobic shift in the long-lived dauer larva. Mech Ageing Dev 124, 779800.
  • Jones SJ, Riddle DL, Pouzyrev AT, Velculescu VE, Hillier L, Eddy SR, Stricklin SL, Baillie DL, Waterston R, Marra MA (2001) Changes in gene expression associated with developmental arrest and longevity in Caenorhabditis elegans. Genome Res. 11, 13461352.
  • Kenyon C, Chang J, Gensch E, Rudner A, Tabtiang R (1993) A C. elegans mutant that lives twice as long as wild type. Nature 366, 461464.
  • Kimura KD, Tissenbaum HA, Liu Y, Ruvkun G (1997) daf-2, an insulin receptor-like gene that regulates longevity and diapause in Caenorhabditis elegans [see comments]. Science 277, 942946.
  • Krebitz M, Wagner B, Ferreira F, Peterbauer C, Campillo N, Witty M, Kolarich D, Steinkellner H, Scheiner O, Breiteneder H (2003) Plant-based heterologous expression of Mal d 2, a thaumatin-like protein and allergen of apple (Malus domestica), its Characterization an Antifungal Protein. J. Mol Biol. 329, 721730.
  • Lee SS, Lee RY, Fraser AG, Kamath RS, Ahringer J, Ruvkun G (2003) A systematic RNAi screen identifies a critical role for mitochondria in C. elegans longevity. Nat Genet. 33, 4048.
  • Lund J, Tedesco P, Duke K, Wang J, Kim SK, Johnson TE (2002) Transcriptional profile of aging in C. Elegans. Curr. Biol. 12, 15661573.
  • McElwee J, Bubb K, Thomas JH (2003) Transcriptional outputs of the Caenorhabditis elegans forkhead protein DAF-16. Aging Cell 2, 111121.
  • Murphy CT, McCarroll SA, Bargmann CI, Fraser A, Kamath RS, Ahringer J, Li H, Kenyon C (2003) Genes that act downstream of DAF-16 to influence the lifespan of Caenorhabditis elegans. Nature 424, 277283.
  • Ogg S, Paradis S, Gottlieb S, Patterson GI, Lee L, Tissenbaum HA, Ruvkun G (1997) The Fork head transcription factor DAF-16 transduces insulin-like metabolic and longevity signals in C. Elegans. Nature 389, 994999.
  • Paradis S, Ruvkun G (1998) Caenorhabditis elegans Akt/PKB transduces insulin receptor-like signals from AGE-1 PI3 kinase to the DAF-16 transcription factor. Genes Dev 12, 24882498.
  • Pletcher SD, Macdonald SJ, Marguerie R, Certa U, Stearns SC, Goldstein DB, Partridge L (2002) Genome-wide transcript profiles in aging and calorically restricted Drosophila melanogaster. Curr. Biol. 12, 712723.
  • Smyth GK (2004) Linear Models and Empirical Bayes Methods for Assessing Differential Expression in Microarray Experiments. Statist Applications Genet Mol Biology 3 (2004), 1, Article 1.
  • Tissenbaum HA, Ruvkun G (1998) An insulin-like signaling pathway affects both longevity and reproduction in Caenorhabditis elegans. Genetics 148, 703717.
  • Wang J, Kim SK (2003) Global analysis of dauer gene expression in Caenorhabditis elegans. Development 130, 16211634.
  • Wolkow CA, Kimura KD, Lee MS, Ruvkun G (2000) Regulation of C elegans life-span by insulinlike signaling in the nervous system. Science 290, 147150.
  • Wood WB (1988) The Nematode Caenorhabditis elegans. Cold Spring Harbor Lab Press, Cold Spring Harbor, NY, USA.
  • Yang YH, Dudoit S, Luu P, Lin D, Peng V, Ngai J, Speed TP (2002) Normalization for cDNA microarray: A robust composite method addressing single and multiple slide systematic variation. Nucleic Acids Res. 30: e15.
  • Yang YH, Speed TS (2003) Design and analysis of comparative microarray experiments. In: Statistical Analysis of Gene Expression Microarray Data, TSpeed, ed. (Chapman & Hall/CRC, Boca Raton, FL, USA), pp. 3592.