Cell-to-cell variation in gene expression and the aging process

Abstract

There is tremendous variation in biological traits, and much of it is not accounted for by variation in DNA sequence, including human diseases and lifespan. Emerging evidence points to differences in the execution of the genetic program as a key source of variation, be it stochastic variation or programmed variation. Here we discuss variation in gene expression as an intrinsic property and how it could contribute to variation in traits, including the rate of aging. The review is divided into sections describing the historical context and evidence to date for nongenetic variation, the different approaches that may be used to detect nongenetic variation, and recent findings showing that the amount of variation in gene expression can be both genetically programmed and epigenetically controlled. Finally, we present evidence that changes in cell-to-cell variation in gene expression emerge as part of the aging process and may be linked to disease vulnerability as a function of age. These emerging concepts are likely to be important across the spectrum of biomedical research and may well underpin what we understand as biological aging.

Introduction

As we age, we accumulate molecular damage, and our cells adapt to these changes. That is generally a stochastic process. In addition to the accumulation of random molecular damage, there are also programs for, and events that result in, epigenetic diversification. These epigenetic diversifications occur independently in each organism, resulting in interindividual epigenetic diversity, as in [1]. These events and programs thus generate physiological differences that are not the result of differences in DNA sequences or environments, but instead are the results of these events and independently running programs. One of the results of these events and programs can be changes in the amount of cell-to-cell variation in gene expression as we age. Changes in heterogeneity of tissues can lead to failures of the physiological subsystems they comprise, resulting in age-related dysfunction and death (e.g., see Fig. 1 in [5], considering dysfunctional cells in organs and tipping points). In terms of lifespan, we can see that even isogenic individuals have differences in lifespan, shown in Fig. 1 [2–4]. The extent to which changes in cell-to-cell variation contribute to age-related loss of physiological capacities and death remains a mystery. Here we discuss what we know about age-related changes in cell-to-cell variation in gene expression and what the consequences of these changes might be in the context of the aging process.

Fig. 1.

Experiments demonstrating a significant nongenetic component to aging. a A histogram demonstrating that isogenic C. elegans hermaphrodites raised and aged in the same Petri dish have different lifespans. Raw data provided by Alex Mendenhall’s former advisor, Tom Johnson, and is also shown in a plot with a longevity mutant in Kirkwood and Finch (see reference [2]). b A step graph of lifespan demonstrating that isogenic male C57B6 mice have differences in lifespan. Figure adapted from Baumann et al., under Creative Commons License Agreement (see reference [3]). c A normalized distribution representing the data from Herskind et al. (see reference [4]), demonstrating that human monozygotic twins raised and aged in Denmark have differences in lifespan

In this review, we use the term nongenetic to encompass gene expression changes that can result from differences in epigenetic markings, the nature of mitotically inherited macromolecular machinery, or from differences attributable to chance events, which may be more consequential early in development. In any cell or organism, the traits that manifest are dependent on the biological program encoded by DNA, which is executed by the expression of genes in the context of a given environment (e.g., see [6]). In addition to DNA sequence and environmental differences, the execution of physiological processes comprises an added layer of inputs that also contributes to variation in traits. A few distinct and non-mutually exclusive mechanisms are likely players in nongenetic variation. The first could be differences in the embryonic inheritance of resources like yolk or ribosomes. A second could be stochastic noise in gene expression, which, for some genes, may be critical for ensuring the fitness of a population in a constantly changing environment. A third relates to differences in the inheritance of epigenetic information. Our goal is to present the relevance of these phenomena in all aspects of biomedical research, and to encourage others to join in investigation of this understudied biology that is rich for development of new ideas, new insights, and new avenues of research.

At the cell and whole organism levels of biological organization, heterogeneity (differences among isogenic individuals) can be detected as differences in discrete traits [7–11] or complex traits, like lifespan [12–16]. There is compelling evidence in each of the aforementioned cases that heterogeneity of traits can be correlated with or predicted by variation in the expression of genes that are not necessarily directly related to the specifics of the trait. For example, among isogenic model yeast, flies, and worms, differences in lifespan can be predicted by differences in expression of a limited set of genes [12–17], including cases where seemingly unrelated biological outcomes effectively serve as biomarkers of longevity. This suggests that changes in the degree of variation or heterogeneity in gene expression could be viewed as an index of a broader biological phenomenon.

There are a few additional technical reasons gerontologists are interested in cell-to-cell variation in gene expression. One reason is that gross anatomical pathology has shown that there are tremendous changes in tissue heterogeneity with age in terms of tissue morphology and resident cell population. Measurements of cell-to-cell variation in gene expression capture some of that heterogeneity. The reason that biologists often focus on gene expression is because it is a functional output of biological systems that incorporates genetic, epigenetic, and stochastic physiological differences. And, as gerontologists, we believe that differences in gene expression patterns among cells can be consequential for age-related disease [18]. So, now we have collectively begun to focus on gene expression in single cells. The focus on single cells is because the cell is the sole level of biological organization in which gene expression occurs, with many intricate processes spanning between accessing a gene and the eventual production of the encoded protein inside the cell. Extracting information on single cells’ gene expression levels, including the spatial patterns of heterogeneity whenever possible, will provide new insights into the biology of aging.

Finally, in quantifying cell-to-cell variation in gene expression, we hope to learn how tissue heterogeneity contributes to organs reaching tipping points of functionality and eventual failure during aging [5, 19]. It is important to note that, because many events during aging are driven by stochasticity (random probability), which organs become dysfunctional, and when, and how, is going to be unique to each individual—but still probabilistic. Taking all these considerations together, we hypothesize that loss of function of these organs arises not only from changes in average expression levels of groups of genes, but from changes in the range of expression that are reflective of a broader problem. It will be critical to not only capture the heterogeneity in gene expression, but also the spatial patterns in the intact tissues, which are often lost in experiments where we “grind and find”. To date, precise information on how cell-to-cell variation changes in intact tissues with age is mostly unknown.

As we discuss further below, but worth noting right away, the age-related changes in cell-to-cell variation in gene expression are not beyond intervention. Like most traits, we have found that the amount of variation in gene expression can be controlled. Recent advances in Caenorhabditis elegans [20] and cultured human cells [21] demonstrate that the amount of animal-to-animal or cell-to-cell variation in expression from one gene can be regulated in trans in metazoans. At this early stage, we do not know about the consequences of age-related changes in heterogeneity; however, being able to experimentally manipulate the heterogeneity also offers means to understand its consequences.

Here we will discuss what we know about age-related changes in cell-to-cell variation in gene expression and what it might mean. First, we discuss a brief history of nongenetic variation in traits. Then, we provide a brief overview of epigenetics, including how epigenetic diversification starts during development. Then we provide a conceptual overview of the study of cell-to-cell variation in gene expression, including physiological mechanisms by which cells can express more or less of a particular gene. Next, we detail the studies of age-related changes in cell-to-cell variation in gene expression. Finally, we discuss how age-related changes in cell-to-cell variation in gene expression might relate to age-related traits and diseases.

The landscape of nongenetic variation

A brief history of nongenetic variation in discrete and complex traits

Since at least 1925, biologists have known that isogenic animals in the same environment manifest differences in traits, from the penetrance and expressivity of mutations [22, 23] to longevity [24]. Variation in traits among isogenic animals was perhaps first identified in 1925 in Drosophila funeberis by Romaschoff when he reported that not all animals in a “pure bloodline” (inbred strain) exhibited the mutant phenotype for Abdomen abnormalis; this is called incomplete penetrance [25]. The severity of the phenotype was different as well, a phenomenon that came to be termed expressivity [22]. Also in 1925, Timofeeff-Resskovsky published that the penetrance and expressivity of another mutation, Radius incompletes, was altered in different genetic backgrounds [23]. Thus, the action of genes could affect the penetrance and expressivity of discreet traits conferred by other genes. In 1928, Pearl observed and quantified variation in the lifespan of isogenic Drosophila in various biological scenarios (see tables XXII-XXVII in [24]). The mechanisms underlying this variation remain mysterious, but more recent work indicates that the phenomenon is almost certainly relevant in mammals. Lifespans also vary among isogenic individuals for other species, both in model systems [26] and in human monozygotic twins [27] (see Fig. 1).

Epigenetics as a means to introduce differences in traits

The system of epigenetic molecular markings alters the DNA-based program of gene expression by changing which genes are accessible for expression or by changing the degree to which each gene is expressed. These silencing or activating instructions can be mitotically heritable and/or transmitted across generations. The epigenetic system serves at least three biological purposes: it provides a mechanism for cell fate determination through differentiation, it provides a mechanism for adaptive changes in gene expression that are heritable, and it provides a mechanism for physiological diversification, increasing the probability that some member of a population will be in a physiological state that happens to be fit for the current environment.

Epigenetics first came to light as a way to explain how populations of isogenic cells diversified during development, forming distinct mitotically heritable physiological states comprising different cell types (embryogenesis). The field subsequently broadened into a study of heritable, non-DNA-based changes in gene expression (reviewed in [28]). The modern take on epigenetic regulation of gene expression is that changes are conveyed by three non-mutually exclusive mechanisms: (a) differences in methylation of the DNA backbone, (b) differences in the molecular markings of specific residues on histones (acetylation or methylation), or (c) differences in noncoding RNAs that act to guide chromatin modifying enzymes, regulate translation and/or initiate mRNA silencing (reviewed in [29]). Some epigenetic changes are “generation-autonomous” and are not passed on to following generations, while others are inherited in ways that can be intergenerational (between generations, e.g., P0->F1) or transgenerational (across multiple generations, beyond F1 progeny, e.g., F3) (reviewed in [30]). Epigenetic regulatory mechanisms have been associated in one shape or form with thousands of the approximately 20,000 protein coding genes in the human genome. Importantly, epigenetic markings can be acquired (e.g., from exercise [31]), or even arise by chance. Consequently, the potential to generate physiological diversity using epigenome modification strategies is vast.

Inheritance of epigenetic information

From the time of fertilization, or even during gametogenesis [32, 33], differences are detected in the partitioning of both resources (e.g., yolk protein [34]) and epigenetic information (e.g., whether or not a silencing small RNA is inherited). Klaus Gartner and colleagues showed that variation in complex traits in inbred mouse lines was innate, concluding that some difference or differentiating factors were present at the time of oogenesis [32]. Since then, several groups have posited that variable DNA methylation patterns contribute to both the emergence of variance in traits and the manifestation of disease [35, 36]. Interestingly, in mammals there is random loss of sperm DNA methylation marks with age, and those in the oocytes are better preserved (reviewed in [37]). Traditionally, the loss of gene silencing marks has been viewed as aberrant, but we would argue that nature could have selected for these random desilencing events as a way of bet-hedging in a world with uncertain future conditions. These chance differences in epigenetic markings and zygotic resources can echo throughout development and lead to differences in biological traits, including the rate of aging (discussed extensively in [38]). Consistent with these observations, experiments with laboratory animals aged in the same environment [2, 26] and analyses of human monozygotic twins [27, 39] both demonstrate that there is a large component of aging that is nongenetic (see Fig. 1). While we do not yet know the causes, increased variation in epigenetic markings has been observed between monozygotic twins with age [1, 40, 41]. This may be due to adaptation, programs to diversify gene expression, or may be purely stochastic dysregulation.

More recently, small RNAs have been shown to play a role in the inheritance and specificity of gene silencing (e.g., [42, 43]). In a study ahead of its time, the Fire lab first found that animals can inherit silenced genes for up to five generations, but that, without an exogenous silencing cue, the number of animals that inherited the silenced gene waned with each passing generation, until almost no individuals in the population silenced the particular gene by the F5 generation. What is interesting here is that some animals in each generation did get the silencing cue and others did not, despite coming from the same genes and environment. So, from that study, we learned that exogenously induced gene silencing is mediated by the inheritance of small RNAs, which requires the RNA binding/dicer protein encoded by dcr-1, and the exogenously induced silencing is variably inherited with decreasing efficiency each generation [42]. These RNA-based regulators work in concert with classic epigenetic mechanisms including chromatin modification by histone methyltransferases [44, 45].

A very new discovery important for epigenetic heritability in C. elegans is the addition of poly UG tails to germline RNA transcripts, which are sufficient to confer the ability to silence a gene [46]. What we know now is that the “pUGylation”, as it is termed, happens in two ways. First, an RNA can be pUGylated when an exogenous silencing cue is introduced; pUGylation is in fact sufficient to induce silencing [46]. Second, for reasons currently unknown, over 300 genes get naturally pUGylated without any exogenous cue [46]. Interestingly, hsp-17/HSPB5 was the only chaperone that was shown to be endogenously silenced [46], and it is also the gene that is most distinctly transcribed between animals expressing more or less of the hsp-16.2-promoter-based lifespan biomarker [47]. Based on the facts that hsp-17 is naturally endogenously pUGylated and that it is the most distinctly transcribed gene between isogenic animals expressing different amounts of a chaperone biomarker of lifespan and mutation penetrance, it seems reasonable consider that endogenous silencing is a plausible mechanism to generate nongenetic physiological diversity. It will be fascinating to explore the biology of pUGylation and endogenous gene silencing to establish whether it is a general mechanism of silencing conserved among species, and if other species, especially humans, are born with randomly silenced genes. The data showing random loss of epigenetic marks in early mammalian embryos is consistent with this idea [37].

Taken together, these findings support the idea that there are intrinsic non-genetic mechanisms for physiological diversification. In longitudinal studies of aging in mice it is apparent that even in inbred strains there is considerable heterogeneity among individuals [48]. It is not at all clear why any particular mouse should land on one side or the other of average lifespan. In mice, metabolic and biometric traits in middle age are associated with different survival probabilities [49], with similar traits being shown to have clinical value in human studies [50]. The basis for the emergence of these traits and the heterogeneity with which they are exhibited is unclear, and currently, there are no predictors detectable in young genetically isogenic mice that might explain differences in future trajectories of aging. Figure 2 shows a schematic of the processes that may be contributing to nongenetic differences in the manifestation of traits during the aging process.

Fig. 2.

Contributions to nongenetic variation at different levels of biological organization. Epigenetic, environmental, and stochastic events contribute to differences in the epigenome. The epigenome affects how much and which genes are expressed, thereby contributing to differences in cell physiology, resulting in differences in tissue composition and function, and resulting in differences in the manifestation of traits

Implications of non-genetic variance at the population level

Organisms that do not have the ability to generate physiological diversity from a non-diverse gene pool are unlikely to succeed. There is ample evidence that geographically isolated species appear to have been founded by few individuals (island and founder effects), arguing in favor of non-genetic diversification programs. There have also been a multitude of mass extinctions over the course of geologic history; the implication is that this would have also left several species with limited diversity among some few survivors, also arguing for the benefit of some program(s) for physiological diversification. Such a program would ensure the fitness of a population that lacks genetic diversity, allowing generation of physiological diversity through intrinsic variation in the execution of the genetic program.

Generation of physiological diversity whether via epigenetic or via inherently stochastic processes retains a genetic component (see bet hedging discussion in [51]). After all, the amount of variation in any given biological process must also be encoded in the genome, at least to some extent. It follows from this that mutations which affect the amount of variation in gene expression could be isolated. We and others have shown that there is genetic control for interindividual variation in gene expression [20] and for cell-to-cell variation in protein expression [21, 52–54]. Genetic regulation of intrinsic variation in gene expression opens up possibilities for therapies designed to reduce inappropriate or pathological changes in cell-to-cell variation in gene expression. Indeed, and as we will discuss more later, evidence suggests that these epigenetic drifts affecting cell-to-cell variation in gene expression may well contribute to age-related diseases like cancer [55–57].

Cell-to-cell variation in gene expression

At its simplest level, cell-to-cell variation can be understood as genetically, chronologically identical cells expressing the exact same gene at different levels. That description is sufficient for isolated single cells, but there can be more to it than that in a multicellular organism. Within a tissue, normally functioning genetically identical cells exhibit spatial specialization that is dependent on where they are in the tissue and what cells are neighboring. The optimal comparison for measuring cell-to-cell variation in a complex tissue would be to quantify expression from equivalent cells in equivalent positions. It is this kind of change in variation that we are most concerned with when considering the physiology of the aging individual. Figure 3 shows a cartoon diagram showing different kinds of variation in gene expression. Based on our own unpublished clinical and laboratory observations, and what has been published to date on age-related changes in cell-to-cell variation in gene expression [58–62], we think the impact of aging on gene expression within a tissue will manifest as an increased heterogeneity among the cell populations comprising a given tissue. It remains unclear whether the age-related increase in diversity of expression serves a physiological role or is simply a reflection of a loss in integrity of gene expression regulation. For cell-to-cell variation in gene expression, what has been measured during aging to date in the studies we cite below is cell-to-cell differences in the abundance of particular transcripts, measured in dissociated cells. There are opportunities to investigate these changes in cell-to-cell variation in gene expression further by incorporating spatial data and relating these states to functional states of tissues or organs.

Fig. 3.

Different kinds of variation in gene expression at different levels of biological organization. a Two different Caenorhabditis elegans hermaphrodites with different whole animal expression levels for the same imaginary gene. The organ inside the cartoon is the intestine, comprised of twenty cells derived from a single cell in the 8-cell embryo. These kinds of whole animal expression level measures are often made on biological model nematodes or flies. b A cartoon diagram of two different intestines with two different amounts of cellular heterogeneity in the expression of the exact same gene. The intestine pictured on top is more heterogeneous than the one on bottom. c Two gray arrows pointing to two cells from two different intestines from two different animals expressing the same gene at different levels

Physiological mechanisms of cell-to-cell variation in gene expression

Distinct cellular processes can result in cell-to-cell variation in gene expression, and each can be tracked by their measurable physiological outcomes. These processes, discussed below, range from the silencing of individual alleles of a gene, to differences in a cell’s reception of a signal, and even to a cell’s general ability to execute the genetic program by translating genetic information into protein. Below we go over each of these types of variation in gene expression, including how they are experimentally measured, shown as a series of illustrations in Fig. 4.

Fig. 4.

Experimentally, analytically tractable mechanisms of cell-to-cell variation in gene expression. a A simplified analytical framework equation derived from Colman-Lerner et al. [54]. The variation in gene expression can be experimentally dissected into different bins using fluorescent reporter proteins. The bins attribute differences in gene expression to differences in allele expression, termed γ; differences in signaling through particular pathways, termed P; and differences in the general ability to express genes into proteins, termed G, which includes production maintenance and turnover. b Two cartoon diagrams depicting the two types of experiments conducted to dissect and quantify different sources of variation in gene expression. Type I experiments quantify signals from differently colored alleles of the same gene. Type I experiments quantify allele expression bias as uncorrelated variation. Type II experiments quantify signals from two distinctly colored and distinctly regulated reporter genes. Type II experiments quantify signaling noise as uncorrelated variation after subtracting intrinsic noise. Correlated variation in type II experiments is attributable to differences in the expression and protein maintenance machinery. c Cartoons illustrating the potential results of type I experiments in a metazoan tissue. In a scenario with low intrinsic noise, the red and green alleles would merge to make the stereotyped yellow pattern seen in worm on the top left of this panel. The gene used to model this pattern was hsp-16.2, which has a relatively higher expression level in the four anteriormost cells in ring one of the intestine, and relatively flat/even expression profile in the rest of the cells. On the bottom left of the panel a worm shows the random pattern of red, green, and yellow, reflecting a random array of monoallelic, biased and biallelic expression in a tissue with high cell autonomous intrinsic noise. On the right of the panel, the three cartoon worms depict the effects of high cell nonautonomous intrinsic noise, in which all the cells in the tissue would bias together toward one allele or the other. d Cartoons demonstrating potential results of type II experiments analyzing the uncorrelated variation beyond the limits of intrinsic noise, thereby attributable to signaling noise. Two smaller nematode cartoons at the top of the panel show the stereotyped patterns of expression formEGFP and mCherry based reporter genes controlled by the vit-2 and hsp-16.2 promoters, respectively. On the top of the left side of the panel, a worm with a merged pattern of the two promoters is shown, indicating what a population of animals low signaling noise would look like. Little or no intrinsic noise is assumed for these cartoons; high intrinsic noise would require the use of three or four distinctly colored fluorescent reporters to account for allelic variation for one or both reporters. The worm on the lower right shows what high cell autonomous signaling noise would look like. On the right side, three cartoon worms demonstrate what it would look like with relatively low (bottom worm) to relatively high, whole tissue, cell nonautonomous signaling noise to the hsp-16.2 promoter. e Cartoons demonstrating potential results of type II experiments analyzing the correlated variation of two distinctly regulated reporter genes. Two smaller nematode cartoons at the top of the panel show the stereotyped patterns of expression for mEGFP and mCherry based reporter genes controlled by the mtl-2 and hsp-90 promoters, respectively. On the top of the left side of the panel, a worm with a merged pattern of the two promoters is shown, indicating what animals would look like with minimal noise in all three components. The worm on the lower right shows what high cell autonomous gene expression capacity differences would look like; the cells are different brightnesses, but all have the proper ratio of red-to-green, indicating low intrinsic and signaling noise, but high cell autonomous differences in expression capacity. On the right side, three cartoon worms demonstrate what it would look like with relatively low (bottom worm) to relatively high, whole tissue, cell nonautonomous differences in gene expression capacity; this was the major mechanism of variation in gene expression in young adult animals reported in Burnaevskiy et al. (see reference [9])

Allele bias/monoallelic expression

Cells can express one or both alleles of a gene at a given locus, and differences in allelic expression can lead to differences in cellular function, from biochemical activity to synthetic capacity. In some cases, individual alleles may confer distinct biochemical activities such that expression of both alleles may provide functional changes beyond a simple increase in abundance of a single gene product. Allele bias taken to the extreme, monoallelic expression, can result in escape from dominant mutations or condemnation to recessive traits. Thus, differences in allele bias can account for differences in the effective dosages of genes, incomplete penetrance of traits, and missing heritability.

An early example of monoallelic expression comes from mouse studies of neuronal odorant receptors, first reported 25 years ago [63]. Later, studies found that an interleukin, IL-2, was also expressed in a monoallelic fashion in mice [64]. Meanwhile, in prokaryotes, differences in the expression of two different copies of the same fluorescent reporter gene were reported [52]. Though not alleles, this was an important step forward in measuring differences in the expression of two different copies of the same gene at the protein level; they termed these differences intrinsic noise. In unicellular eukaryotes, in 2004, actual alleles were observed to be expressed in a biased and sometimes monoallelic fashion, and the amount of random allele bias was shown to be controlled in cis and trans primarily by TATA boxes and the SWI/SNF chromatin remodeling complex, respectively [53]. Then, in 2007, once methods to quantify monoallelic expression globally were developed using SNP arrays, the phenomenon was reported to be widespread in cultured human cells (300 out of 4000 assayable genes) [65]. Using a clever chromatin signature approach to identify both open and closed chromatin marks, another study reported monoallelic expression for over 40% of approximately 10,000 technically ascertainable genes in the human genome [66, 67]. Recently, using the same fluorescence microscopy approach developed in microbes, methods were developed for assessing allelic expression in living intact C. elegans tissue, where differences in allele expression, a.k.a., intrinsic noise, could now be quantified in real time [9, 68]. Investigation of how monoallelic expression may be controlled and how it may contribute to heritable human diseases is an active area of current research [69].

Signaling noise

Cells can perceive signals differently, either due to chance differences in perception, or due to differences in signaling capacity for a particular signaling pathway, which may be stochastic (chance based) or epigenetic in origin. In yeast, the ability to send signals through the pheromone response pathway in response to the same external stimulus is not equivalent among individual cells despite the fact that they are clones [54]. This unexpected discovery was made possible by adapting an expanded analytical framework [52], and required the observation of distinct pathway genes for reference (analogous to triangulation). Precise quantification of signals allowed distinctions to be drawn between allele bias and signaling noise. Differences in perception and signaling sensitivity in a metazoan may be especially consequential, in particular in cases where one cell controls other cells, such as neuronal control of muscle contraction.

Capacity for protein synthesis

With the development of the expanded analytical framework [54], general protein expression capacity was identified as another source of cell-to-cell variation. The analytical framework allowed scientists to reclassify what was formerly known as extrinsic noise [52] as gene or protein expression capacity by measuring the correlated variation between two distinctly regulated genes at the protein level. In terms of cellular physiology, this cell-to-cell variation led to a difference in cellular ability to translate genetic information into proteins, resulting in global differences in proteome between cells. Since then, general differences in protein expression capacity have been observed in cultured human cells [70, 71] and in vivo in C. elegans [9]. In HeLa cells transducing apoptosis signals, mitotically heritable differences in post-translational signal transduction capacity have been identified [72] and may be linked to differences in protein expression capacity.

The breadth of sources of variance at the cellular, and indeed, whole animal levels is remarkable. The precise mechanisms involved in generating variance are beginning to be uncovered and are almost certain to be highly conserved among species. It will be fascinating to explore how these features might be beneficial in cellular physiology, how they are balanced to ensure homeostasis, and what happens when the integrity of these mechanisms drifts, as might be expected in the context of aging.

Nongenetic variation as it relates to aging

One of the consistent observations in aging research, independent of which species is under investigation, is an increase in variation detected across traits as a function of age [50, 58, 73, 74]. But, we do not yet understand how age-related increases in trait variation relate to age-related changes in cell-to-cell variation in gene expression. The field of cell-to-cell variation is still growing, and currently, we know much less about the causes and consequences of cell-to-cell variation in gene expression in metazoan systems compared to what we know about cell-to-cell variation in microbes, notwithstanding some key developments involving individual mammalian cells. The challenge becomes even greater when considering that, as organisms age, their tissues become more fibrotic making sampling more difficult, and cells can become more sensitive to shearing during physical disruption, impacting the integrity of isolation. Furthermore, there is an increase in time and money required to maintain aging animals, especially mammals. Age-related changes to cell-to-cell variation in gene expression within tissues could be tremendously consequential. They could take the form of increased heterogeneity or even loss in heterogeneity among cell populations, changing tissue function and whole organism physiology. We would argue that taking cell-to-cell variation to be a de facto age-related trait could yield valuable new insights into the aging process and the consequences of age in terms of functional loss.

Cell-to-cell variation in gene expression with age

Several lines of evidence point to changes in the amount of cell-to-cell variation in gene expression as a function of age. In mouse cardiomyocytes, a targeted approach using qPCR-based analysis revealed increased cell-to-cell variation in gene expression for several genes [58]. Subsequent studies using unbiased using RNA-seq also observed the same general trend of increased cell-to-cell variation in gene expression [59–61]. Later, a three-tissue single cell RNA-seq study found that there is both increased and decreased variation in gene expression with age that is both gene and cell type dependent [62]. Figure 5b graphically summarizes the observations of age-related changes in cell-to-cell variation in gene expression made to date.

Fig. 5.

Reported age-related changes in cell-to-cell variation in gene expression. A cartoon diagram depicts a plane of cells in an imaginary tissue with an optimal amount of heterogeneity in the expression of some gene in youth; shades of blue depict different gene expression levels in individual cells, represented as circles. Then, with age, two more cartoons depict the same imaginary tissue with abnormal age-related increases and decreases in heterogeneity. Single cell transcriptomic studies we cited to date have quantified isolated cells; actual positional information of cells in relation to age-related changes in cell-to-cell variation in gene expression remains largely unknown

To date, assessment of age-related changes in cell-to-cell-variation in gene expression has been largely limited to quantifying transcripts. We currently do not know much about how cell-to-cell variation in protein expression levels or activities changes with age beyond targeted approaches. The gap in our knowledge of heterogeneity among individual cells at the protein level is important because there can be little correlation between protein and RNA levels [75–77]. Layers of regulatory mechanisms exist between transcript and production of the eventual encoded protein, and another battery of mechanisms influence protein activity and stability, some even occurring at the time of translation though interactions with the transcript. Initial studies of cell-to-cell variation in protein expression of living cells, which we have not exhaustively listed, were conducted in bacteria [52] and yeast [53, 54, 78]. There have also been notable studies of mammalian cell-to-cell variation in protein expression using microscopy [21, 79, 80] or flow cytometry [81–83]. Finally, using live-imaging techniques, we have endeavored to measure cell-to-cell variation in protein expression levels in an intact metazoan, C. elegans [9], which should provide a blueprint for expansion to mammalian systems.

At this point we know that increased cell-to-cell variation in transcription is a hallmark of aging for some genes in some tissues [58–62]. How these genes are singled out is not known, the mechanisms specifically recruited to differentially regulate these genes are not known, and the functional consequences of changes in this suite of genes are also not known. It will be important next to determine how cell-to-cell variation in gene expression plays out at the protein level, and the extent to which age-related changes in mRNA transcripts and changes in non-coding RNAs involved in RNA-based regulatory processes are involved.

Cell-to-cell variation in the proteome with age

A number of lines of evidence would suggest that integrity of pathways associated with protein synthesis, protein stability, and complex assembly are all vulnerable to age. We know that translational capacity and protein turnover change with age, with evidence in support of this idea coming from studies of nematodes and killifish [84, 85]. Major changes in the stoichiometry of macromolecular complexes have been reported [84]. Interestingly, inhibiting protein turnover with proteasome poisons can increase levels of protein encoded by diverse unrelated genes [84], suggesting age-related changes in cell-to-cell variation can work through global differences in protein production, maintenance and/or turnover. And in support of the notion that large swaths of the proteome may covary as a result of some other difference in the protein production, maintenance, or turnover machinery, the majority of proteome turnover is controlled by insulin signaling [86–88], which we know affects aging. Additionally, we know that there are age-related changes in epigenetic histone marks that could contribute to changes in gene activation or silencing, reviewed in [89–91]. In the model we propose, genes, or just one allele of a gene, become stochastically silenced in some cells of some tissues during the aging process, and this leads to the age-related change in specific functions related to that gene, such as a change in macromolecular complex stoichiometry. Figure 4 shows a conceptual overview of different kinds of cell-to-cell variation in gene expression in tissue and how it might change in terms of the aforementioned mechanisms, and how this would have consequence for cell autonomous or whole tissue level outcomes.

Age induced cell-to-cell variation and physiological outcomes of age

Increases in variance among a range of age-related parameters have been reported in aging studies including population studies and studies of isolated cells from aged individuals [50, 73, 74, 92, 93]. Although age-related heterogeneity is reported in numerous biomedical, pre-clinical, and clinical studies, it is seldom the focus of study. A recent study of longitudinal aging in nonhuman primates reported increase in variance among biometric, systemic biomarker, and functional traits as a function of age and found that the prolongevity intervention of caloric restriction decreased animal-to-animal variation in all parameters measured despite equivalence in age between control and CR fed groups [94]. This finding is consistent with increased variance being a hallmark of aging, not only because it is observed across diverse phenotypes, but also in that the variance is compressed across diverse phenotypes in an intervention with proven ability to prolong health and improve survival [95]. Although conjecture at this point, it would be interesting to know to what extent differences in cell-to-cell variation in the tissues give rise to age-related heterogeneity in traits. It is currently not known if caloric restriction, other prolongevity treatments, or long-lived genotypes impact cell-to-cell variation in gene expression, and if they did, whether this would correlate with suppression of age-related phenotypes. Considering spatial heterogeneity of expression patterns in tissues, the coordination among cells must be tightly controlled if the distinct populations within a tissue are to function harmoniously. It may be that increased cell-to-cell variation in the expression of genes, for example, of particular alleles of immune-related genes like IL-2 [64], may also underlie age-related pathological changes. Data collected as part of longitudinal studies may be able to address this, with bulk profiling methods offering some insight, but single cell and single nucleus studies being more informative. A causal role for cell-to-cell variation in healthspan and lifespan is not difficult to imagine, and it seems equally likely that data already exist that would allow for first steps in exploring the concept.

Conclusion

We previously speculated that genetic/epigenetic programs to increase physiological diversity may constitute an antagonistic pleiotropy mechanism of aging [96]. We believe that the capacity for cell-to-cell variation confers both adaptive strengths for optimal youthful functional capacities of tissues and detrimental age-related changes in heterogeneity; however, we still have a lot to learn about the mechanisms and functional consequences, particularly in the context of aging. To learn the mechanisms and functional consequences will require two factors. First, it will require direct observation of age-related changes in cell-to-cell variation in gene expression inside intact somatic tissues. In this, the strengths and translatability of various organisms from C. elegans to mice to monkeys might be leveraged. Second, it will require the correlation of cell-to-cell variation in gene expression in a tissue to functional measures of tissue performance. Technical improvements in advanced confocal submicron-resolution endoscopes now offer means for directly observing gene expression inside living mammals [97–100]. The ability to track functional outcomes longitudinally will not only inform of correlations but also predictive qualities of these assessments in terms of the pace of aging. These are achievable goals in both vertebrate and invertebrate systems. Technological advances in the last decade have opened up new strategies that are well suited for this line of investigation, from molecular profiling, to genetically delivered reporter systems, to advanced microscopy and in vivo imaging.

Single-cell transcriptome sequencing has been successful in measuring the impact of age on cell-type distribution in tissues, age-related changes in gene expression at the individual cell level, and changes in cell-to-cell variation in the transcriptome as a function of age [58–62]. As promising as these data are, more can be done to further test the idea that changes in cell-to-cell variation in gene expression are a causal factor of aging. It will be critical to determine how changes in cell-to-cell variation relate to aging outcomes. Experiments will need to address whether the increases or decreases in cell-to-cell variation correlate with differences in age-related outcomes or measures of physiological function, and if the extent of age-related cell-to-cell variation in gene expression is reflective of differences in the pace of aging among isogenic individuals. It is not yet clear which tissues are more vulnerable to changes in cell-to-cell variation, or within tissues, which cell types are more likely to exhibit this trait.

The loss of spatial context within a tissue is currently a limitation of transcriptional profiling of isolated single cells or nuclei. There are a few ways this can be addressed without the need for endoscopy or transparent organisms. First, for reanalyzing existing data, for some tissues, known patterns of gene expression can be exploited to place cells in a stereotyped model of the tissue. For example, there is the mouse gene expression database [101]; both expression levels, and the ratios of gene pairs can function as markers of cell position, because there happen to be genes for which expression varies by position in the tissue. Thus, by knowing the expected transcription level, or the ratio of a pair of particular transcripts, for a cell in a particular position in a tissue, researchers may be able to place individual cells from their scRNA-seq datasets into particular places in a model tissue to infer how spatial changes in gene expression may be happening with age, computationally. Second, steady improvements in resolution of proteomic techniques including the development of single-cell capabilities [102], and enhanced mass spectrometry imaging of the proteome in tissues [103], could together allow mapping of individual cells contributing to tissue heterogeneity and cell-to-cell variation as a function or age. Finally, the relatively new spatial transcriptomics methodologies offer the potential to acquire the same transcript information as RNA-seq approaches while capturing the spatial data for each cell in the tissue [104]. The time is ripe for a full exploration of this necessary property of cellular diversity in multicellular organisms, how it arises, how it is controlled, and what mechanisms might be engaged to counter age-related drift.

Funding

Support for Dr. Kaeberlein and Dr. Mendenhall was provided by the Nathan Shock Center for Excellence in the Basic Biology of Aging center grant from the National Institutes for Health National Institute on Aging, P30AG013280. Support for Dr. Mendenhall was also provided by a grant from the National Institutes of Health, National Cancer Institute, R01CA219460, and by the University of Washington EDGE Center of the National Institutes of Health, funded by NIEHS, P30ES007033. Support for Dr. Martin was provided by National Institutes of Health, National Cancer Institute, R01CA210916, and the National Institutes for Health, National Institute on Aging R01 AG06397. Support for Dr. Anderson was provided by a grant from the National Institutes of Health, National Institute on Aging, AG040178, and a grant from the Department of Veterans Affairs, VHA ORD BX003846, the Simons Foundation, and the William S. Middleton Memorial Veterans Hospital at Madison, Wisconsin.