The Landscape That Chooses

One of its most enduring explanations resembles a ball rolling downhill. In Conrad Hal Waddington’s developmental landscape, cells travel through branching valleys toward specialized identities. A nerve cell occupies one destination; a different cell type occupies another. The hills keep the possibilities from becoming an unrestricted wander.

But the ball is not simply traversing a preinstalled mountain range. The terrain can change as development proceeds. Cells respond to signals, regulatory networks change their activity, and routes that were available can disappear. The landscape does not literally choose. It represents the conditions under which particular outcomes become accessible and stable.

In an October 9 feature for Quanta Magazine, Philip Ball traces how this historical metaphor has become a productive research framework—and how measurements are exposing its limits. This is not one experiment announced this week. It is a fresh account of a long-running attempt to explain how cells acquire distinct identities, with consequences for how researchers might eventually steer them.

The interesting news is not that an old drawing was right. It is that scientists can increasingly specify which parts of the drawing hold, which need revision, and what evidence would distinguish them.

A valley is not an instruction

Waddington’s picture makes a useful distinction between an outcome and the conditions that keep producing it.

Imagine a shallow disturbance to a ball already rolling through a valley. It may veer, but the surrounding slopes return it toward the same route. In the developmental analogy, cells can differ somewhat in their molecular activity without becoming different cell types. Robustness does not require every component to remain perfectly still. It requires a system whose dynamics tolerate some variation.

Waddington called this channeling canalization. The landscape gives it a shape: a range of starting positions can lead toward a similar destination. That is a more informative picture than imagining each cell as a microscopic employee consulting an exhaustive instruction manual.

Genes remain central. The correction to the blueprint story is not that DNA has been demoted to decorative packaging. Gene regulation helps establish which states are possible and how transitions occur. But a list of genes is not, by itself, an account of the changing activity, interactions and surroundings through which a cell develops.

Ball reports that chemical signals between cells, along with mechanical interactions as tissues grow and fold, contribute to the developmental process. The relevant unit of explanation therefore cannot always stop at an isolated cell’s inventory. It may have to include what that cell receives, what it changes, and what its neighbors are doing.

A landscape can make these relationships easier to think about. It cannot substitute for identifying them. Calling something a valley is the beginning of a mechanistic question, not the answer delivered in a nicer font.

When two regulators make a fork

One concrete example comes from a 2007 paper by Sui Huang, Yan-Ping Guo, Gillian May and Tariq Enver in Developmental Biology. Its retrieved abstract describes a mathematical model and experimental measurements of a binary fate decision in blood-cell progenitors.

The researchers examined a regulatory circuit exemplified by GATA1 and PU.1, factors associated with erythroid and myelomonocytic development, respectively. Those are routes toward red blood cells and particular white-blood-cell lineages—not a complete account of every kind of blood cell.

The circuit combines self-stimulation with mutual inhibition. In the model, these interactions generate stable states corresponding to the two specialized fates, alongside an uncommitted, metastable state in which both regulators are expressed. The point is not simply that one molecular quantity becomes larger than another. The interactions help explain how a cell can remain uncommitted and then settle into one of distinct alternatives.

The abstract reports that measurements of GATA1 and PU.1 messenger RNA and broader transcriptome dynamics supported a two-stage commitment process. First the progenitor state becomes destabilized, leaving a poised state near the boundary between the lineage-specific attractors. Then the cell moves toward one of the now-accessible attractors.

An attractor is a state, or set of states, toward which a dynamical system tends to evolve under specified conditions. It is not a tiny magnet hidden inside a future blood cell. Here the term turns a picture of branching valleys into a model whose behavior can be compared with measurements.

The scope matters. This was a minimal model of a binary decision, not proof that every developmental event consists of two options. The authors suggested broader relevance to binary fate decisions. That is already substantial without appointing the fork as biology’s universal traffic policy.

The embryo declines the decision tree

The most revealing complication comes from a 2021 Cell paper linked in Ball’s feature: “A single-embryo, single-cell time-resolved model for mouse gastrulation,” by Markus Mittnenzweig and colleagues.

The team assembled data from 153 individually sampled mouse embryos spanning 36 hours of developmental diversification. They combined single-cell RNA measurements, embryo timing and computational modeling to infer how populations move between transcriptional states during gastrulation, an early period of body organization.

Some transitions did look like rapid bifurcations. Early specialized blood and node cells provided examples. For most lineages, however, the researchers reported more complex, branching-to-multiple-outcomes dynamics rather than a sequence of hierarchical binary choices.

This is not a minor artistic disagreement about how many tributaries to draw. A chain of yes-or-no decisions implies a particular organization of development: first split here, then split again within each branch. The study instead describes enduring progenitor populations that can progressively contribute to several fates, with combinations of regulatory factors involved.

The researchers also examined embryos containing cells with Foxc1/Foxc2 mutations, using the temporal framework to investigate regulatory effects. That matters because an atlas alone tells researchers what states they sampled. Perturbations help them ask how changing a regulator changes the process.

Their landscape did not disappear. It became less like a neatly pruned family tree and more like a terrain containing broad basins as well as sharp canyons. The familiar picture remained useful only by surrendering some of its tidiness.

There is a particularly important restraint in the paper: the flows are inferred. These are not continuous films of every measured cell becoming its successor. The model incorporates assumptions about proliferation rates, and the authors identify sampling and temporal-resolution limits. They also note that dividing a continuous differentiation space into named cell types requires thresholds.

The study therefore offers evidence against a universal binary-choice account in the development it examined. It does not establish a complete map of every embryo, nor does it make computational reconstruction equivalent to direct observation.

That boundary is where a useful model earns trust. It tells us both what became visible and how the visibility was constructed.

A beautiful map can smuggle in a mountain

Single-cell measurements make developmental cartography possible at a scale Waddington could not access. They also introduce a new temptation: treating a persuasive visualization as though the microscope had photographed the landscape itself.

It has not. RNA measurements capture one layer of cellular activity. They do not provide a complete inventory of proteins, physical forces, spatial relationships or future behavior. Ball quotes cell biologist Julie Theriot cautioning both about the limits of RNA-based characterization and about the assumptions embedded when high-dimensional data are compressed into simpler maps.

That compression is necessary for many kinds of analysis and display. A cell’s molecular state involves far more variables than fit comfortably on a page. The problem is not reduction itself. It is forgetting what the reduction preserved, what it discarded, and which structures the method was predisposed to reveal.

Timing presents a related difficulty. Researchers often arrange cells along pseudotime: an inferred progression based on differences in their measured states. This can suggest a developmental sequence, but it is not automatically a stopwatch reading or a recorded ancestry.

The 2021 mouse study addresses this problem by linking cells to individually staged embryos and building a temporal flow model. Its discussion also explains why pseudotime approaches that impose rigid trees struggle with continuous commitment and with distinct populations converging on similar programs.

Two dots close together on a plot may share aspects of their molecular state. That fact alone does not establish that one became the other. A branch in the display may represent a meaningful transition, but the evidence for the transition must do more work than the line drawn between the dots.

The map is an argument. Attractive contour lines do not exempt it from cross-examination.

Changing the route is not yet repairing the tissue

Why spend so much effort improving the terrain? Because understanding a transition could help researchers intervene at a point when the desired outcome is reachable.

Ball reports James Briscoe’s argument that knowing the topology of a developmental landscape and its branching points could, in principle, guide methods for steering cell populations toward particular states. Regenerative medicine and stem-cell engineering are obvious motivations. A better account of the route could replace some trial and error with more deliberate experiments.

The practical distinction is between pushing a system toward a label and changing the conditions that support the state behind that label. If the relevant regulatory configuration is unstable, briefly producing a marker is not the same achievement as establishing a durable, functional cell identity.

Ball also describes interest in viewing cancer through attractor landscapes and exploring whether cells could be redirected toward less harmful states. That is a research perspective, not evidence from these papers of a broadly effective treatment. A stable state can be disastrous for the organism. Stability describes behavior; it does not issue a certificate of health.

Nor does a diagram of possible transitions establish a safe route through them. Showing that a model permits a destination, demonstrating a cellular transition, and repairing damaged tissue are different evidentiary tasks. The landscape helps formulate those tasks. It does not complete them on the researchers’ behalf.

Keep the picture open

The coherenceist lesson here is narrower, and more useful, than declaring that everything is connected. Explain an outcome through the relationships and constraints that actually sustain it. Then keep the explanation open to evidence that those relationships differ from the ones the picture led you to expect.

The contrast between the blood-cell circuit and the mouse-embryo atlas makes this concrete. A two-way fork can illuminate a particular decision without becoming a universal rule. A broader measurement can reveal multiple routes without rendering all simpler models worthless. The work is to match the model to the question, not to crown a single image sovereign over development.

There is an ethical distinction worth retaining, too. Coherence is not automatically the good. An attractor can be stable without serving the organism; an elegant scientific account can be internally consistent while leaving out consequential observations. In both cases, admiring the order is not enough. We have to ask what it includes and what its apparent success fails to show.

Waddington’s landscape survives because it can be revised, not because biology has finally agreed to resemble a mid-century drawing. The cells are neither unconstrained wanderers nor actors reciting an immutable script. They develop through changing possibilities that researchers are learning to measure.

Somewhere in that untidy process, a body makes the nerves with which it will eventually wonder how bodies are made. No supervisor appears. We do, however, get better maps—and the rather lovely capacity to redraw them together.