What If Evolution Didn't Have to Take Forever?

The case for engineering biological convergence on a human timescale

In Madeira, there are elegant woody bushes in the wet valleys cooled by the waterfall breeze , that looked like the rhododendrons that fill waterfall valleys like in Pennsylvania. The lily-of-the-valley tree in Madeira seems to have taken on the same role as rhododendrons in Appalachia- both liking rich, acid soil, in the damp, shaded soil: an example of convergent evolution arriving at a similar shaped and appearing organism. This raises the question: could we make convergent evolution happen on purpose, and fast? 

The Wrong Question We Keep Asking

When scientists want to improve a biological molecule — an enzyme, a protein, an antibody — they use directed evolution. They mutate it, test it, keep the winners, repeat. It works extraordinarily well, and Frances Arnold won a Nobel Prize for it. But the trick only works because you have a single, clean readout: does this enzyme catalyze the reaction faster? Yes or no.

Now suppose what you want isn't a faster enzyme but a better healer — a plant whose roots secrete compounds that calm neuroinflammation, paired with a fungal partner that amplifies those compounds under stress, embedded in a microbial community that self-regulates when something goes wrong. You can't optimize that one variable at a time. The whole point is that the thing you want is the relationship, not the molecule.

How do you select for an entire ecosystem?

Folding Time

Can we fold time by making many ecological pressures coincide at once? 

In a mountain valley, Clethra and rhododendron converge on similar evergreen strategies not because evolution runs on a special geological clock, but because the relevant pressures — hydrology, shade, soil chemistry, fungal partnerships, pathogen load, disturbance — take a long time to stabilize and shape one another.  Because the constraints arrive gradually, the feedback loops between them take centuries to close.

But if you could make those constraints appear together, persistently, and with feedback? The search space collapses. What took a valley ten thousand years might take a laboratory greenhouse ten. Not because you've circumvented biology, but because you've compressed the geometry of the problem.

The Holobiont Insight

The key move is changing what to select on.

Evolution experiments typically select on individual organisms. But an organism is not the whole story — especially for plants and animals whose phenotype depends critically on their microbial passengers. The rhizosphere of a healthy plant is not backdrop; it is co-author. A plant's ability to tolerate drought, resist pathogens, or synthesize secondary metabolites is substantially a property of the plant-microbe ensemble.

In 2025, researchers demonstrated exactly this with bank voles selectively bred for an unusual dietary trait. What evolved wasn't just the vole — the microbial community co-evolved with it, producing distinct and stable gut bacterial consortia that tracked the selection pressure. The unit of evolution was the holobiont: host plus microbiome, selected as one.

This matters for timescale. Plant genomes change slowly — a generation might be a year or a decade. But microbial communities turn over in hours to days. A plant that improves its phenotype by recruiting better microbes can adapt at microbial speed without changing a single nucleotide in its own genome. That alone could compress centuries into growing seasons.

What CRISPR Is Good For Here

CRISPR enters this picture as a diversity scaffold. You wouldn't use it to encode the right phenotype — you can't know in advance what that is. Instead, you'd use it to pre-load the starting population with variation along axes you already know matter: leaf longevity, symbiosis competence, secondary metabolite pathways, stress tolerance, developmental plasticity. You're seeding the search, not dictating the destination.

Think of it as setting up orthogonal dimensions for selection to work in, rather than pointing at a target. Once those eigenvectors are in place, selection has something to grip immediately, instead of waiting for the right mutation to wander into existence.

The Missing Ingredient: Environmental Memory

With drected evolution, in ery generation, you start fresh. Clean plates, fresh media, identical conditions. This eliminates noise, which is scientifically tidy. But it also eliminates something essential — the way yesterday's winners shape today's playing field.

In a real ecosystem, soil chemistry drifts. Inhibitory compounds linger. Pathogens track their hosts. Mycorrhizal networks form persistent architectures that new seedlings either plug into or fail against. The environment has memory, and that memory is itself a selection pressure. It's what gives evolution something to lean against, what allows functional attractors to form and deepen over time.

The experimental systems that could accelerate convergent evolution would need to preserve this. Let secreted metabolites alter the substrate for the next cycle. Let the winners chemically modify the arena before the next generation competes. This isn't contamination to be eliminated — it's the mechanism.

Ecosystem as Problem-Solver

The developmental biologist Michael Levin is developing a taxonomy of complex systems as goal-directed, information-processing agents operating at multiple scales simultaneously. A single cell navigates toward chemical gradients. A tissue coordinates cell behavior toward a body plan. An organ detects damage and marshals repair. Each level has what Levin calls a "cognitive light cone" — a scale at which it can sense, model, and act.

If you establish the right boundary conditions, if you structure the constraint landscape carefully enough, biological systems at every scale will begin solving toward functional attractors. You don't have to specify the solution. You have to specify the problem correctly.

Applying this to compressed-timescale convergent evolution: the experimenter's job is not to engineer the therapeutic plant-microbe consortium directly. It is to create an environment whose pressures make a therapeutic ensemble the stable solution — and then get out of the way while selection closes in on it. This is closer to morphogenetic engineering than to protein optimization, and it requires a fundamentally different experimental posture: less control, more constraint design.

Who Is Doing This, and What's Missing

Some of the conceptual groundwork exists. Researchers studying rhizosphere eco-evolutionary dynamics have shown that rapid microbial evolution occurs at ecological timescales — within and between plant generations — and that this evolution feeds back into plant phenotype. A 2024 paper in Nature Communications proposed "Microbiome gene breeding," selecting crop varieties based on their genome's capacity to recruit beneficial microbial communities. A 2025 piece in Cell coined the term "hologenome breeding" explicitly.

On the experimental evolution side, there are extraordinary microbial systems — including Richard Lenski's 37-year E. coli experiment, now past 80,000 generations. And there are demonstrations of mutualism evolving in real time: one research team took a plant pathogen and, within a handful of selection cycles, evolved it into a nitrogen-fixing symbiont.

But the integrated version — multi-species consortia under high-dimensional recursive pressure, selected across seasonal cycles on an emergent therapeutic phenotype — does not exist yet as a deliberate program. The closest approximations are in agricultural science, focused on yield, not medicine.

The bottleneck is not biological. The tools are ready: CRISPR diversity loading, mesocosm design, multi-omics tracking, machine learning readout of complex phenotypes. Th goal is to run an experiment whose output is not a clean number but a relationship.

A Different Kind of Medicine

Imagine selecting not for a molecule but for a persistent association: a root-microbe-soil ensemble that reliably produces anti-inflammatory compounds under physiological stress, self-maintains under fluctuating conditions, and degrades gracefully when the stress resolves. 

You would not get tall trees quickly. Long-lived woody architectures are genuinely constrained by development time. But you absolutely could get convergence in evergreen leaf analogues, chemical defense strategies, water-handling traits, symbiotic dependency, and growth habit — and more importantly for medicine, in the secondary metabolite profile and the ecological conditions under which it's expressed.

What you would have made is not a drug. It is something stranger and more durable: a biological solution space that keeps finding the answer even as conditions shift.

Nature did this accidentally, slowly, in ten thousand mountain valleys. We could do it deliberately, faster, with our eyes open — if we're willing to let the ecosystem be the engineer.

The field currently calls this experimental evolution, or holobiont biology, or microbiome engineering. None of those names quite captures it. Perhaps the right word is what it always was: sculpted ecology.