The question that wouldn't go away

For most of the history of biology, aging has been treated as a story about damage. Reactive oxygen species chew up DNA. Telomeres fray. Proteins misfold. Mitochondria sputter. The cell, in this telling, is a machine that simply runs down — and the best we can do is slow the rust.

But damage-only stories struggle with one stubborn fact: genetically identical cells, living in identical conditions, age in strikingly different ways. If aging were pure entropy, you'd expect a smooth distribution of outcomes. Instead, you see discrete, repeatable patterns. Some cells die one way; others die a very different way. The endpoints cluster. Something — some hidden logic — is steering cells toward a small menu of possible deaths.

In July 2020, a team led by Nan Hao, Jeff Hasty, Lorraine Pillus, and Lev Tsimring at UC San Diego published a paper in Science that puts a sharp edge on this mystery. Working in budding yeast (Saccharomyces cerevisiae), they tracked thousands of individual mother cells through their entire replicative lifespans, watched the molecular machinery of aging unfold in real time, and then — crucially — built a mathematical model that could predict how a cell would age and how to change its fate. The paper's title says it plainly: "A programmable fate decision landscape underlies single-cell aging in yeast."

The word programmable is doing a lot of work. It means aging is not just damage; it is a decision process, and decisions, unlike rust, can be edited.

Why yeast, and why single cells?

Budding yeast is the workhorse of aging research for reasons that are practical, not romantic. A mother yeast cell divides asymmetrically: it buds off smaller daughters, and the mother retains a "birth scar" that lets you count how many times she has divided. That count — the replicative lifespan (RLS) — is the operational definition of her age. Typical lab strains manage around 20–25 divisions before senescence.

This matters because the same asymmetric division happens in our own bodies. Stem cells in the gut, the skin, the blood, and the brain divide asymmetrically to produce one differentiated daughter and one stem cell that carries on. Yeast replicative aging is, in essence, a stripped-down model for how mitotic stem-cell lineages age — including ours.

The problem with yeast aging historically was that it was almost impossible to watch. Once a mother cell starts dividing, daughters pile up, fall away, and you lose track of who's who. The Hao/Hasty labs solved this with custom microfluidic chips — tiny traps that pin the mother cell in place under a microscope while fresh medium flows in and daughters are washed away. Combined with time-lapse fluorescence microscopy, this lets you sit with one cell from birth to death, watching her chromatin state, her mitochondria, and her gene expression change division by division. The 2020 paper is built on this kind of single-cell cinematography, at population scale.

Two aging paths, and only two

The first discovery was already provocative. When the team watched wild-type cells age, they didn't see a continuum. They saw two distinct phenotypic trajectories.

About 47.3% of cells — what the authors call Mode 1 — began producing elongated, oddly shaped daughters during the later divisions of life, and their nucleoli (the ribosome-factory region of the nucleus) became enlarged and fragmented. Their mitochondria, oddly, stayed normal — tubular and well-organized right up to the end.

The other 52.7% — Mode 2 — kept producing small, round daughters, kept a tidy nucleolus, but watched their mitochondria clump into aggregates before death.

So aging in yeast isn't one disease. It's two. And the two are mirror images: nucleolar failure with healthy mitochondria, or mitochondrial failure with a healthy nucleolus. Mode 1 cells also lived somewhat longer than Mode 2 cells, and Mode 2 cells stretched their cell cycle length more dramatically as they aged — they spent longer and longer stuck between divisions, like a runner slowing at the end of a marathon.

Two aging modes with distinct organellar failures raises an immediate question: what's the molecular switch? What flips a cell into Mode 1 versus Mode 2?

The two suspects: Sir2 and HAP

Sir2 — the silencer of the nucleolus

Mode 1's nucleolar enlargement is a classic symptom of ribosomal DNA (rDNA) instability. Yeast's rDNA locus — the genes that build ribosomes — is notoriously unstable, sitting on chromosome XII as a long tandem repeat. In aging mother cells, this locus spawns extrachromosomal rDNA circles (ERCs): little loops of DNA that pinch off, replicate themselves, and accumulate to toxic levels in the mother. This was worked out in the 1990s by David Sinclair and Lenny Guarente at MIT, who showed that ERC accumulation is, in a real sense, the clock of yeast aging — "set the clock early and the alarm rings early," as Guarente put it.

Standing between the rDNA and chaos is Sir2, the founding member of the sirtuin family. Sir2 is an NAD⁺-dependent lysine deacetylase: it strips acetyl groups off histones, particularly histone H4 at lysine 16, which tightens up chromatin and silences transcription at the rDNA, at telomeres, and at the silent mating-type loci. By keeping the rDNA silent, Sir2 suppresses the illegitimate recombination that produces ERCs. Delete SIR2 and lifespan shortens; bump up its dosage and lifespan extends. This is one of the most replicated findings in yeast aging.

The Hao team had previously shown, in a 2017 PNAS paper, that rDNA silencing is lost sporadically and stochastically in young cells, then — for some cells — collapses into sustained loss late in life. To track this in real time, they used an rDNA-GFP reporter: a green fluorescent protein gene inserted into the silent spacer region of the rDNA. When silencing holds, GFP is off. When silencing fails, the cell glows.

Watching the reporter through life, they found that Mode 1 cells — the ones with nucleolar failure — underwent sustained loss of rDNA silencing late in aging. Mode 2 cells did not. And if you artificially crippled Sir2 with nicotinamide (NAM), a known Sir2 inhibitor, almost all cells were driven into Mode 1. Sir2 and rDNA silencing were steering the nucleolar-failure path.

HAP — heme's messenger to the mitochondria

Mode 2 pointed somewhere else entirely: to heme, the iron-containing porphyrin ring that is both a cofactor for cytochrome c oxidase (complex IV of the electron transport chain) and a signaling molecule in its own right. The connection to human aging came from work by Hani Atamna at UC Berkeley and colleagues, who showed in 2001–2002 that heme deficiency selectively disrupts assembly of mitochondrial complex IV in human fibroblasts and brain cells, and proposed that declining heme synthesis is a driver of the mitochondrial decay seen in the aging brain and in neurodegeneration.

In yeast, heme activates the HAP transcriptional complex (Hap2/3/4/5), which in turn drives expression of genes needed for mitochondrial biogenesis and respiration — including COX5a and CYC1, components of complex IV and cytochrome c. HAP is, in effect, heme's voice inside the nucleus, telling the cell to keep its mitochondria running.

To track heme, the team used two independent reporters in the same cell: a fluorescent HS1 heme sensor (where GFP signal goes up when free heme goes down) and a nuclear-anchored iRFP reporter whose fluorescence depends on biliverdin, a product of heme catabolism, and therefore tracks cellular heme abundance. Both reporters told the same story: heme dropped in Mode 2 cells, but not in Mode 1.

And just as NAM pushed cells into Mode 1, the heme-depleting drug succinylacetone pushed cells into Mode 2. The Mode 2 path was a heme/HAP collapse, with mitochondrial aggregation as the visible consequence.

The anti-correlation that broke the puzzle open

The breakthrough came when the team put both reporters — rDNA-GFP and nuclear iRFP — into the same cells and watched them age together. Newborn cells started in a tight cluster, with similar amounts of both signals. As they aged, they fanned out in two opposite directions:

• Mode 1 ended life with high GFP, high iRFP — i.e., low rDNA silencing, high heme.
• Mode 2 ended life with low GFP, low iRFP — i.e., high rDNA silencing, low heme.

The two endpoints were anti-correlated. When Sir2 lost, HAP won; when HAP lost, Sir2 won. That's not what you'd expect from two independent damage pathways running their own race. That's the signature of mutual inhibition: two circuits actively suppressing each other, with the loser dragging the winner up.

This is the moment the paper shifts from cataloging phenotypes to actually explaining them. If Sir2 and HAP mutually inhibit each other, then the cell isn't choosing between two independent forms of damage. It's being pulled, by its own circuitry, into one of two self-reinforcing attractor states.

Proving the circuit with genetic perturbations

If Sir2 and HAP really are mutually inhibitory, then perturbing one should bias the cell toward the opposite fate. The team ran the experiment in four directions, and the results are almost embarrassingly clean:

• Delete SIR2: 83.1% of cells showed rising iRFP (rising heme, rising HAP activity) and aged with Mode 1 phenotypes. Without Sir2's brake, HAP ran unchecked.
• Delete HAP4 (the major HAP subunit): 89.8% of cells showed falling rDNA-GFP (rising Sir2 silencing) and aged with Mode 2 phenotypes.
• Overexpress HAP4: 95.9% of cells lost rDNA silencing and aged in Mode 1 — strong evidence that HAP actively inhibits Sir2.
• Overexpress SIR2 two-fold: more Mode 2 cells (Sir2 suppressed HAP), but — surprisingly — a new third phenotype appeared.

That third phenotype is the paper's most tantalizing discovery. A subset of cells with 2× Sir2 aged toward a state of both high Sir2 and high HAP — a state that, on the simple mutual-inhibition model, shouldn't be stable. The authors named it Mode 3.

Mode 3 cells were long-lived, with extended replicative lifespans. Like Mode 1 cells, they produced elongated daughters — but unlike Mode 1, their cell cycle length stayed roughly constant through life. They aged, in other words, more gracefully than either of the two wild-type modes.

However, 2× Sir2 alone wasn't a longevity hack. Because extra Sir2 also biased the landscape toward the short-lived high-Sir2/low-HAP Mode 2 well, the population-average lifespan only nudged up. The interesting state was reachable, but rare.

Waddington meets the Fokker-Planck equation

To make sense of all this, the team built a deliberately minimal mathematical model: two ordinary differential equations, one for Sir2 and one for HAP, with two key features:

1. Mutual inhibition — each suppresses the other (consistent with the genetic data, and with prior observations that HAP-component transcripts rise in sir2Δ cells, and that HAP helps regulate cofactors in Sir2 silencing complexes).
2. Positive autoregulation — each reinforces itself. Sir2 deacetylates H4-K16, creating high-affinity binding sites that recruit more Sir2. HAP activation drives TCA-cycle gene expression, which feeds back into more heme biosynthesis, which activates more HAP.

This is the textbook topology for multistability: a network that can settle into more than one stable steady state.

The deterministic version of the model produces a Sir2–HAP phase plane with seven fixed points, three of them stable — corresponding to Mode 1 (low Sir2, high HAP), Mode 2 (high Sir2, low HAP), and a shallower "low Sir2, low HAP" well that matches a subpopulation of Mode 2 cells seen in the data.

But real cells are noisy. So the team then solved the corresponding two-dimensional Fokker-Planck equation — the equation that describes how a probability distribution evolves over time under noisy dynamics — to compute a true aging landscape: a potential surface whose "wells" correspond to the stable fates, with well depth encoding how likely a cell is to fall in.

This is a deliberate echo of Conrad Waddington's famous "epigenetic landscape," the rolling hills developmental biologists use to picture how a pluripotent cell rolls downhill into one of several differentiated fates. What Hao's team is saying, in effect, is that aging has its own Waddington landscape — and the hills and valleys are sculpted by the Sir2–HAP circuit.

When the model's Sir2 parameter is doubled, a fourth stable well — the high-Sir2/high-HAP Mode 3 — appears on the landscape, exactly as observed. The model isn't just describing the data; it's predicting the existence of a state that the experimenters then found.

Engineering the long-lived cell

If Mode 3 is a high-Sir2/high-HAP state that the simple mutual-inhibition model says shouldn't be reachable, then the way to reach it is to overcome that mutual inhibition. The model said: if 2× Sir2 alone biases toward Mode 2, raise the basal HAP abundance too, and the landscape will tilt toward the high-HAP states — Mode 1 and Mode 3.

So the team overexpressed both Sir2 (2×) and Hap4. The results matched the model almost eerily:

• Only 5.8% of cells went Mode 2 (down from ~70% under 2× Sir2 alone).
• 44.7% of cells went Mode 3 — the long-lived state.
• Lifespan extended more than overexpressing either Sir2 or Hap4 alone — a synergistic, not additive, effect.
• Cell cycle length stayed relatively normal throughout aging.

The combined perturbation, in essence, partially neutralizes the mutual inhibition between Sir2 and HAP, opening up the high-Sir2/high-HAP corner of the phase plane as a stable place for the cell to live. The two factors that each, alone, would push the cell toward one short-lived fate or the other together carve out a new, longer-lived one.

The team then tested the same logic on a different long-lived mutant. fob1Δ deletes the Fob1 protein — the rDNA replication-fork barrier that's required for the recombination events that spawn ERCs. fob1Δ extends lifespan by reducing ERC accumulation, mechanistically a cousin of Sir2 overexpression. Combining fob1Δ with Hap4 overexpression produced the same synergistic lifespan extension. The principle generalized.

And it offers a tidy explanation for a long-standing puzzle in the aging field: why caloric restriction (CR) and Sir2 — two interventions that were each thought to act through independent longevity pathways — sometimes seem to synergize. CR induces Hap4 expression (it shifts yeast metabolism toward respiration). Sir2 stabilizes rDNA. The Hao model says they're not independent at all: they're the two arms of the same circuit, and combining them reshapes the same aging landscape.

What this means beyond yeast

Three claims from the paper deserve to be flagged as genuinely important, with appropriate humility about translation:

1. Aging is a fate-decision process, not just damage accumulation. This is the conceptual headline. The Sir2–HAP circuit shows that even in a single cell, with no environmental variation, the aging trajectory is chosen — early, by the dynamics of an internal gene-regulatory network, and stably, in a way that's robust to intracellular noise. That doesn't mean damage is irrelevant; it means damage is filtered through decision circuitry, and the circuitry can be edited.

2. The same circuit may exist, in some form, in humans. Sirtuins are conserved from yeast to humans — mammals have seven (SIRT1–7), with SIRT1 being the closest functional cousin of yeast Sir2 in terms of chromatin-silencing roles, and SIRT3 and SIRT6 being heavily implicated in mitochondrial and genomic maintenance. Heme regulation of mitochondrial biogenesis is also conserved, in part through nuclear receptor pathways (including the heme-binding transcription factor Rev-erbα and Bach1). The mammalian equivalents of the yeast HAP complex include the nuclear respiratory factors (NRF1/2) and PGC-1α, which coordinate mitochondrial biogenesis. The exact topology will differ, but the design principle — chromatin-stability machinery and mitochondrial-biogenesis machinery talking to each other, with feedback — is plausible in mammalian stem cells.

3. Combination therapy beats single-target therapy. The most actionable lesson is the synergy result. Overexpressing Sir2 alone, or Hap4 alone, gave modest lifespan gains. Overexpressing both together gave a much larger, synergistic gain — because they neutralized the mutual inhibition that limited each on its own. This is a strong argument against the "find one magic longevity gene" framing and in favor of combination interventions that target both halves of the chromatin–mitochondrial axis at once.

Caveats worth stating plainly: this is yeast. Mammalian aging is more layered — senescence, immune clearance, tissue-level inflammation, vascular decline all matter in ways that don't show up in a microfluidic trap. Sirtuin biology in mammals has had its share of walked-back claims (resveratrol as a direct SIRT1 activator is the famous example; clinical translation of NAD⁺ precursors like NR and NMN is still maturing, with some human trials showing metabolic benefits but no proven lifespan extension in people). And the Mode 3 state in yeast was reached by genetic engineering — not by a drug — so "today" there is no pill that does this.

What the yeast work does is hand the field a blueprint: a circuit, a model, a perturbation strategy, and a predicted optimal intervention. The blueprint now has to be tested in mammalian stem cells.

Bottom line: if you wanted to act on this today

So what can a non-yeast, non-lab-bound person actually do with a paper like this? Translated conservatively into today's evidence, the Sir2–HAP axis maps onto two levers that human biology gives us some real control over: (a) supporting NAD⁺-dependent sirtuin activity, and (b) supporting mitochondrial biogenesis and heme sufficiency. The yeast paper's central insight — that you want both levers pulled at once, because they're two arms of one circuit — is the practical takeaway.

1. Don't pull one lever; pull both. The single most important practical lesson from the Li et al. study is the synergy result. If you're already doing something that supports mitochondrial biogenesis (e.g., zone-2 cardio, which raises PGC-1α), also do something that supports NAD⁺/sirtuin activity (e.g., time-restricted eating, which depletes NADH and raises the NAD⁺/NADH ratio), and vice versa. One lever alone is the yeast equivalent of 2× Sir2 by itself — modest gains, with the circuit fighting back.
2. Keep NAD⁺ availability up. Sirtuins are NAD⁺-dependent; their activity tracks NAD⁺ levels, which decline with age. In humans, the most consistent NAD⁺-raising levers with reasonable evidence are: caloric restriction or time-restricted eating (which lowers NADH and shifts the NAD⁺/NADH ratio), and supplementation with NAD⁺ precursors such as nicotinamide riboside (NR) or nicotinamide mononucleotide (NMN). Human trials of NR and NMN have shown NAD⁺ elevation and some metabolic benefits (e.g., blood pressure, insulin sensitivity in prediabetic adults); they have not shown proven lifespan extension in humans. Treat them as plausible-but-unproven.
3. Support mitochondrial biogenesis directly. The HAP equivalent in mammals is the PGC-1α / NRF1 / NRF2 / TFAM axis. The most evidence-backed way to upregulate it is regular endurance and zone-2 aerobic exercise. Cold exposure and certain phytochemicals (e.g., urolithins from pomegranate/gut-microbiome conversion, which activate mitophagy) have emerging human data but are not slam-dunks.
4. Don't smoke out your heme. Heme synthesis depends on adequate iron, vitamin B6 (as pyridoxal phosphate), and glycine. Iron deficiency is the most common nutritional deficiency worldwide and directly compromises complex IV assembly (the Atamna mechanism from the early 2000s, which the yeast paper leans on). If you're a menstruating athlete, a vegetarian, or have unexplained fatigue, getting ferritin and transferrin saturation checked — and correcting deficiency — is a high-yield, cheap, and undeniably real lever.
5. Caloric restriction is the one intervention that hits both levers at once. CR raises NAD⁺/NADH (favoring sirtuins) and induces the HAP/PGC-1α program (favoring mitochondrial biogenesis). This is exactly why the yeast paper offers CR as a unifying explanation for Sir2/CR synergy. In humans, CR (or its milder cousin, time-restricted eating) has the broadest evidence base of any lifespan-affecting intervention, with proven metabolic and cardiometabolic benefits — though sustained CR is hard, and its long-term effects on human lifespan remain unproven.
6. Avoid the Sir2 inhibitors we know about. Nicotinamide at high doses is a Sir2 inhibitor in yeast (that's how the Hao team induced Mode 1 experimentally). High-dose niacin/nicotinamide supplementation in humans is not directly comparable, but it's a reminder that "more is better" doesn't hold for sirtuin modulation — the system needs balance, not maxing out. Chronic high-dose nicotinamide without NAD⁺ precursor logic may not be doing what supplement marketing implies.
7. Sleep and circadian rhythm. SIRT1 activity is circadian — it oscillates with NAMPT-driven NAD⁺ biosynthesis, which is itself under circadian control. Chronic sleep disruption desynchronizes this loop. This is one of the cleaner mammalian examples of the chromatin–mitochondrial circuit being time-of-day modulated, and it gives a concrete mechanism for why chronically bad sleep ages people.
8. Read the original before believing any summary. Including this one. Aging biology is full of hyped extrapolations from yeast to humans, and the gulf is real. The Li et al. paper is rigorous and worth reading directly; the supplementary materials (23 figures, 6 tables, 2 movies, and the model code at github.com/yaj030/aging_science2020) are unusually generous.

The deepest thing this paper gives you is not a pill or a protocol. It's a reframing. Aging is not only damage. It is, in part, a decision your cells are making — and decisions can be influenced. The yeast in this study were coaxed into a longer-lived fate by editing two genes at once. The same logic, applied to mammalian cells, is one of the most concrete research programs in aging biology today.

Sources and further reading

• Main article (open access via UC eScholarship): Li Y, Jiang Y, Paxman J, O'Laughlin R, Klepin S, Zhu Y, Pillus L, Tsimring LS, Hasty J, Hao N. "A programmable fate decision landscape underlies single-cell aging in yeast." Science 369(6501): 325–329 (2020-07-17). DOI: 10.1126/science.aax9552. Author manuscript (PMC): PMC7437498. UC eScholarship permalink: escholarship.org/uc/item/22k6g3c1. Companion code: github.com/yaj030/aging_science2020 (DOI 10.5281/zenodo.3770529).
• Hao Lab research page (UCSD) — microfluidic single-cell aging platform: haolab.ucsd.edu/research.html
• Li Y et al. (2017) — earlier work on multigenerational rDNA silencing dynamics during aging: "Multigenerational silencing dynamics control cell aging." PNAS 114(46): 11253–11258. PMID 29073021.
• Jin M et al. (2019) — divergent aging of isogenic yeast cells: "Divergent Aging of Isogenic Yeast Cells Revealed through Single-Cell Phenotypic Dynamics." Cell Systems 8(3): 242–253. PMID 30852250.
• Sinclair DA & Guarente L (1997) — the original ERC mechanism of yeast aging: "Extrachromosomal rDNA circles — a cause of aging in yeast." Cell 91(7): 1033–1042. PMID 9428525. (Companion MIT news: news.mit.edu/1997/aging2)
• Kobayashi T & Horiuchi T (1998) — Fob1 and rDNA replication fork barrier: Requirement of Fob1 for rDNA recombination and ERC formation. SGD locus page: SGD FOB1.
• Wierman MB & Smith JS (2013) — review of yeast sirtuins and aging: "Yeast Sirtuins and the Regulation of Aging." FEMS Yeast Research 14(1): 73–88. PMCID PMC4365911.
• Atamna H et al. (2002) — heme deficiency in mitochondrial and neuronal decay of aging: "Heme deficiency may be a factor in the mitochondrial and neuronal decay of aging." PNAS 99(23): 14807–14812. PMID 12417755. pnas.org/doi/10.1073/pnas.192585799
• Hughes AL & Gottschling DE (2012) — vacuolar pH, mitochondrial decline, lifespan: "An early age increase in vacuolar pH limits mitochondrial function and lifespan in yeast." Nature 492: 261–265. PMID 23172144.
• Kaeberlein M, Kirkland KT, Fields S, Kennedy BK (2004) — Sir2-independent CR lifespan extension: "Sir2-independent life span extension by calorie restriction in yeast." PLoS Biology 2(9): e296. PMID 15328540.
• Kane AE & Sinclair DA (2018) — sirtuins and NAD⁺ in human trials: "Sirtuins and NAD⁺ in the Development and Treatment of Metabolic and Cardiovascular Diseases." Circulation Research 123(7): 868–885. PMCID PMC6206880.
• Imai S & Guarente L (2016) — NAD⁺ and sirtuins in aging/longevity: "It takes two to tango: NAD⁺ and sirtuins in aging/longevity control." npj Aging and Mechanisms of Disease 2: 16017. nature.com/articles/npjamd201617
• Funding for the main study: NIH National Institute on Aging AG056440 (Hao, Hasty, Pillus, Tsimring); NSF MCB-1616127 (Hao); NSF MCB-1716841 (Pillus).