The needles start out straight, packed in a tidy array no wider than a fingertip. Press the patch onto warm skin and they begin to move. Within two minutes they have hooked themselves into a coil, like dozens of tiny fingers tightening their grip, pulling the edges of a wound together from the inside. Nobody is turning a screw or pulling a thread. The needles are simply responding to the heat of the body, doing what they were printed to do.
That printed-in behaviour is the work of a team led by Hyun-Do Jung, an associate professor at Hanyang University in Seoul, and the inspiration comes from an unlikely source: a carnivorous plant. Drosera capensis, the Cape sundew, traps insects by curling its sticky tentacles around them, holding them fast, and then breaking them down with a chemical assault. Jung’s group took those three tricks — coordinated movement, adhesion, and a built-in defence against microbes — and folded all of them into a single wound-healing device.
Borrowing a Predator’s Playbook
The movement is the cleverest part, and it relies on what materials scientists call a shape-memory polymer. The needles are built from two acrylates, mixed and cured under ultraviolet light using a 4D-printing method (the fourth dimension being time, since the structure changes shape afterwards). Heated to 70°C and then straightened, the polymer holds that flattened, temporary shape until something warms it again. At room temperature it barely stirs. At 37°C, the temperature of human tissue, it recovers its original curl, and reaches full bend within about 120 seconds. The wound closure happens, in other words, the moment the patch meets living skin.
Getting that timing right is harder than it sounds. The shape recovery depends on a fiddly tangle of variables: how much crosslinker goes into the mix, how long the resin sits under the lamp, how warm the surroundings are. Working through every combination by hand would take an age.
So the team handed the problem to machine learning. They trained three different algorithms to predict how the printed material would behave, and one, a method called Gaussian process regression, came out clearly ahead, predicting the recovery angle with better than 99 per cent accuracy and, usefully, flagging how confident it was each time. From that the researchers settled on an optimal recipe and a printing window that balanced a brisk shape change against the structural stiffness the needles need to actually penetrate skin.
For Jung, this marriage of biology and computation is the whole point. “This study goes beyond conventional biomimicry by using artificial intelligence to translate nature-inspired principles into a functional biomedical device. The key point of this research is not only that it is inspired by nature, but that AI helps convert biological inspiration into a predictable, programmable, and clinically relevant wound-healing technology,” he said.
Closing the Wound, Then Healing It
Closure alone, though, would only be half a treatment. A wound that is held shut but left exposed can still fester, and for people with diabetes — whose wounds heal slowly, stay inflamed, and turn septic far too readily — infection is often the thing that turns a small injury into a serious one. Here the sundew’s other two talents come in. The needles were coated with adhesive DNA nanoparticles, assembled using a sticky chemistry borrowed from mussels, which release slowly into the wound and coax the cells that build new blood vessels and connective tissue into action. They mop up the reactive oxygen molecules that keep chronic wounds stuck in their inflamed state, too. Then a vanishingly thin layer of zinc, driven into the surface by an ion-implantation technique, supplies the antibacterial punch. In dish tests it cut colonies of Escherichia coli by more than 80 per cent and hit Staphylococcus aureus hard as well. The zinc has a second job, as it happens: by densifying the polymer surface it slows the DNA release, stretching it out over a fortnight or more.
Put a diabetic mouse’s wound under this patch and the difference shows. Wounds treated with the full system closed faster than those given saline, with near-complete skin regrowth by day ten, denser and better-organised collagen, and a flush of new blood vessels. Markers of inflammation fell. Some features of healthy skin, including hair follicles and the small glands that keep skin supple, came back at up to six times the density seen in untreated wounds.
None of this means a sundew patch is heading for your bathroom cabinet next year. The work was done in mice, not people, and a fair gap usually separates the two. The machine-learning models leaned on a fairly small pile of experimental data, which the authors are upfront about. And the needles, as designed, do not dissolve once their work is done — a wrinkle the team flags as worth ironing out, since a patch you have to remove is less appealing than one that simply vanishes.
Still, the broader idea has legs. The same AI-guided, shape-shifting strategy need not stop at skin. Jung sees it travelling further: “Beyond wound healing, the AI-guided 4D-printing strategy could also be extended to soft biomedical robots or tissue-interfacing devices that require programmable motion, controlled shape transformation, and stable contact with biological tissues,” he said. Stents that mould themselves to a vessel, scaffolds that fill a bone defect, soft robots that creep through the body and hold their grip — all could draw on the same trick of letting a material decide, on its own, when and how to move.
Which leaves a rather pleasing thought. A plant that spent millions of years perfecting how to catch and dissolve a fly may end up teaching our machines how to knit us back together.
DOI / Source: 10.1002/adma.202523665
Frequently Asked Questions
How can a patch close a wound without any stitching or pulling?
The needles are printed from a shape-memory polymer that holds a flat, temporary form until it warms up. At body temperature they recover a pre-programmed curl, hooking into the tissue and drawing the wound edges together on their own, reaching full bend in roughly two minutes. It is the body’s own heat, not any external force, that triggers the movement.
Why does this matter especially for people with diabetes?
Diabetic wounds tend to heal slowly, stay inflamed, and become infected easily, which is what turns minor injuries into chronic, dangerous ones. This patch tackles all three problems at once: it closes the wound, releases molecules that spur new blood vessels and tissue, and carries a zinc layer that kills bacteria. In diabetic mice it produced faster closure and far more regrowth than standard treatment.
What does the artificial intelligence actually do here?
Shape recovery depends on a tangle of manufacturing variables that would take enormous trial and error to optimise by hand. The team trained machine-learning models to predict how the printed material would behave, and the best one forecast the recovery angle with better than 99 per cent accuracy while estimating its own uncertainty. That let the researchers pick an ideal recipe without exhaustively testing every combination.
Could the same approach be used for anything besides skin wounds?
In principle, yes. The researchers suggest the strategy could extend to self-shaping stents, bone scaffolds, and soft robots that need to move in controlled ways and grip tissue reliably. The common thread is a material that decides for itself when and how to change shape inside the body.
How close is this to being used on actual patients?
Not very close yet. The work was carried out in mice, the AI models were trained on a relatively small dataset, and the needles do not currently dissolve after use, which the team acknowledges needs fixing. Several rounds of refinement and human trials would be needed before anything like it reaches a clinic.
