The optical transparency of C. elegans allows non-invasive visual assessment of living worms without compromising subsequent measurement of longevity. We used a chromosomally-integrated transgenic strain (TJ375), containing the 400 bp hsp-16.2 promoter coupled to the gene encoding green fluorescent protein (GFP) and encoding no HSP-16.2 product itself (Fig. 1a). This reporter provides an accurate assessment of the total amount of native HSP-16.2 protein⁸ (see also Supplementary Fig. 1). No detectable GFP is observed in uninduced worms, but following a one or two-hour 35 °C pulse (Fig. 1b), GFP becomes readily apparent, peaking at 15–18 hours (Figs. 1c, d).

Heat-shocked populations displayed a wide and normally-distrubuted variation in individual GFP fluorescence (Figs. 1c–g), even though individuals were isogenic and grown in an environment designed to minimize environmental heterogeneity. Such heterogeneity was observed from the earliest times at which GFP expression was detectable and continued until the time GFP had completely dissipated (several days, data not shown). The degree of heterogeneity increased markedly with time (Figs. 1 c, d), and was both replicable and quantifiable (Figs. 1 g, h).

We asked if HSP-16::GFP expression might predict longevity. Our initial findings (Supplementary Figure 2), on individual, isogenic worms measured manually, suggested a significant correlation between GFP-expression level and subsequent longevity (r = 0.48; P = 0.002); so we extended our studies to large populations. Worms were sorted into high, intermediate, and low GFP-expression classes at various times after heat induction (Figs. 1 b, f) and were subsequently tested for resistance to a lethal thermal stress or kept for longevity analysis. We routinely observed significant differences in subsequent longevity and thermotolerance among worms that expressed GFP at high, average or low levels (Figs. 2, ​,3).3). When sorted after a two-hour induction at 35°C, worms that differentially expressed GFP showed large differences in life expectancy and thermotolerance. In a typical experiment, we found life expectancies of 16.4 days in the brightest worms, while the worms expressing the lowest levels of GFP after heat shock lived only about 3.2 days (Fig. 2a, Supplementary Table I). Similarly, worms with the highest GFP levels also showed higher thermotolerance (9.5 hours) than the average (6.7 hours) or than worms showing the lowest levels of thermotolerance (4.0 hours; Fig. 2d; P < .001).

We asked whether sorting at different times after the heat shock affected the observation of differential survival. We sorted worms at various times following induction and found differential life expectancy to be as much as 10 to 15 days (Fig. 2b, Supplementary Table II), averaging 8.0 days over all 19 experiments (Fig. 2c). Following two-hours of induction, lifespan averaged 15.1 days for the high and only 7.1 days for the low, more than a two-fold difference (Figs. 2 c, Supplementary Table I; P = 8.0 x 10⁻²⁹). Sorting earlier than 9 hours led to non-significant results. Worms expressing different levels of GFP, 9 to 36 hours after induction, also differed significantly with respect to subsequent thermotolerance (Figs. 2d, e and Supplementary Tables I, III). We found statistically significant differences in all but two of sixteen replicates. This differential thermotolerance was very robust, averaging about 3.4 hours (Fig. 2f, Supplementary Table I; P = 1.7 x 10⁻²⁸). Differential survival and thermotolerance were highest at about 18 hours after induction, about the time when variance is maximal.

In prior studies⁸, incubation of the TJ375 reporter strain for two hours at 35°C resulted in subsequent thermotolerance and increased longevity in a process termed hormesis. Under the two-hour heat induction used in the current experiments we also saw hormesis for both longevity and thermotolerance, but only in a subpopulation of worms. Our calculations indicate that 27% of the worms were damaged by this treatment regimen (D.W., S.L.R., T.E.J., J.W.V., manuscript in preparation). These results are different because we used new induction conditions to maintain a more uniform environment (abrupt vs slow temperature shift, see Methods). When we decreased induction time to one hour, we again found significant differences in survival between the brightest and dimmest worms in 7 of 9 experiments (Figs. 3a–e and Supplementary Tables I, II and IV). Average life span differed by as much as 14 days in one study (details in Fig. 3b legend). Furthermore, after one hour of heat we now observed a robust hormesis effect, consistent with previous observations^9,10.

To correct for possible interrelationships between the effect of heat on survival and its effect on fertility¹¹we crossed two temperature-sensitive (ts) fertility mutations, fer-15(b26) and spe-9(hc88), into the TJ375 reporter strain to form a new strain: TJ550. At the non-permissive temperature, the combination of both mutations completely blocked reproduction, but not germ cell formation or proliferation⁵. Differential survival was, however, still observed in this background with all six replicates showing significant differences in longevity between bright and dim worms (Fig. 2b, Supplementary Table II).

Individual differences in hsp-16.2::GFP reporter expression may result from genetic variation – i.e. epigenetic changes may occur in isogenic individuals during propagation leading to differential inactivation or expression of one or more of the large number of repeats in the transgenic array present in the reporter strains¹². To address this question, we asked whether differences in levels of GFP expression were heritable. We sorted a population at 11 hours post heat shock into sub-populations containing a few hundred of the total initial population of 60,000 worms (Fig. 1i). Progeny were collected, allowed to grow to maturity, induced by heat shock and assessed for level of GFP expression using the identical protocol. We found that progeny of both the high- and low-expressing parents showed almost identical average levels of GFP expression (Fig. 1i). Progeny of both high- and low-expression parental classes had almost the same mean expression level (298.0 vs 288.6 GFP units) and both displayed almost identical variation in levels of GFP expression (p = 0.5, χ² test for distributional difference), essentially recapitulating that of the parental population. Thus, the precise level of GFP expression is not heritable. While it is possible that further experimentation may reveal discrete causal factors determining variance of GFP expression, the results shown here were obtained from an isogenic population, maintained in a uniform environment during their propagation. Non-heritability of GFP expression level suggests the presence of a large underlying stochastic component specifying level of GFP expression in individual worms, similar to that observed in bacteria^13,14.

Finally, we also asked whether level of GFP fluorescence is a predictor of longevity when GFP is tagged to promoters of non stress-inducible genes (myo-2 and mtl-2). It is not. Since GFP fluorescence is dependent upon redox activation we also utilized a promoter tag of a gene normally activated in response to oxidative stress (gst-4) and again found no relationship between GFP levels and subsequent longevity (Fig. 3f, Supplementary Fig. 3 and Supplementary Table I).

HSP-16.2 expression level in young adults is a robust predictor of remaining life expectancy. This variation is not heritable.

For C. elegans, mutational analysis has long been the preferred approach for understanding gene action and biological function¹⁵, no less so for aging and life span. Despite the success of the genetic approach in explaining life-span extension between distinct genotypes in C. elegans, most life-span variation is not under genetic control. Even under rigidly controlled laboratory conditions, 60% of the variation in longevity in F₂ intercrosses in nematodes is not genetic¹⁶. Similar findings are true in all species that have been studied^1,3; in humans, only about 25% of the variation in life span (even after excluding early deaths due to childhood disease and accident), is due to measurable genetic effects^1,2,17leaving the vast majority of variation in life span as unexplained or “environmental”, some of which results from chance or stochastic events within individuals^1,2,18.

Stochastic variation arises from fundamental thermodynamic and statistical mechanical considerations. A large fraction of individual variation in life span must stem from the fundamental fact that life results from an integrated series of metabolic reactions which themselves are under fundamental physical constraints on the specificity and rigidity with which they, too, can be regulated¹⁹. At the molecular level, two points are germane to the present study. First, when the number of molecules regulating a biological process becomes countably small, “chance” distributions come into play such that some regulatory molecules can vary several-fold between individual cells²⁰. Second, the Maxwell-Boltzmann (M-W) equation specifies the distribution of kinetic energies among molecules and requires kinetic energy to be a distributed function. Strehler and Mildivan²¹ utilized this equation to develop a general theory explaining mortality kinetics. Several sources of variation at the molecular level could conceivably alter GFP (HSP-16.2) expression level while simultaneously affecting more global processes. These include intracellular differences and fluctuations in the rates of molecular processes such as transcription, ribosome loading and translation (as postulated by Kirkwood et al²²). Chance variation in the number of HSF effecter molecules present within each cell at the time of heat shock also could have dramatic phenotypic consequences. Variation in the frequency of mitochondrial genomic rearrangements, as previously observed in isogenic populations of C. elegans ^23–25, could have an effect. There is an increasing literature describing variation among isogenic individuals at the molecular level, typically in microbial or yeast cultures where such effects can be visualized^13,14,26. Clearly, significant variation among genetically identical individuals is a fact of nature and inherent molecular variability implies that biochemical and molecular genetic processes must exhibit inherent variability.

From the earliest studies of C. elegans aging populations²⁷, it has been apparent that individual age at death varies greatly in isogenic populations - spanning several weeks between those dying on the first day of adult life (excluding larval and embryonic death) and individuals who died last. In the case of long-lived Age mutants, this span can be several months. Stochastic variation provides a means by which one can start to understand this huge variation in lifespan. Genetic regulatory systems can be viewed in terms of robustness or sensitivity toward chance environmental fluctuations, maintaining expression of either a single phenotype or leading to the expression of a distributed phenotype. When multiple phenotypes are useful, such as for sampling changing environments, genetic systems that have a built in capacity to reveal variance, or indeed amplify it, could be selected²⁸. Such systems might be of particular use to self-fertilizing organisms like C. elegans.

Supplementary Figure 1

Supplementary Figure 2

Supplementary Figure 3

Supplementary Table I

Supplementary Table II

Supplementary Table III

Supplementary Table IV