How Caterpillars Remember Past Lives: The Strange Survival of Memory Through Metamorphosis

The textbook account of insect metamorphosis describes the larval body being broken down inside the chrysalis by enzymes and digestive cells, with the adult body essentially built from scratch from a small population of dormant cells (imaginal discs) that retained their identity through the larval stage. The image of the body being substantially dissolved into a soup before reorganizing into an adult is dramatic enough that it tends to crowd out questions about what happens to the nervous system during this process. The standard simplification is that the brain, like the body, is essentially rebuilt, and that the adult moth or butterfly has no continuity with the larva that preceded it.

This account is wrong in interesting ways. Multiple experimental lines have demonstrated that adult moths can retain memories formed in the larval stage, despite the substantial reorganization of the nervous system during metamorphosis. The discovery is mostly recent (the most decisive experiments are from the 2000s and 2010s) and the mechanism is still partly unresolved, but the basic observation that memory survives metamorphosis is now well established.

The classical claim

The schoolroom account of metamorphosis rests on the observation that larval and adult insects look very different and live very differently. A caterpillar crawls on leaves and chews; a butterfly flies and drinks nectar. The morphology, locomotion, sensory ecology, and feeding apparatus are all substantially different. The intuitive interpretation is that two different organisms inhabit the same lineage, separated by a developmental wall in the chrysalis.

The cellular biology of metamorphosis partially supports this interpretation. During pupation, much of the larval body is broken down by programmed cell death and digestion, and the adult body is built up from imaginal discs that were present in the larva but inactive. The wings, legs, and adult eyes all derive from imaginal tissue. The picture of an organism being substantially rebuilt is accurate as far as it goes.

But the nervous system is treated differently. While some larval neurons are pruned during metamorphosis, many persist into the adult, often with modified connections. The brain is not dissolved and rebuilt in the way that, say, the larval gut is. The mushroom bodies (the insect brain structure most associated with associative learning and memory in adults) are present in larvae and persist through metamorphosis, though with substantial reorganization of inputs and outputs as the sensory and motor architecture changes from larval to adult.

The Blackiston experiment

The decisive experiment for memory survival was published by Douglas Blackiston, Elena Silva Casey, and Martha Weiss at Georgetown University in PLOS ONE in 2008. They trained tobacco hornworm caterpillars (Manduca sexta) to avoid an odor (ethyl acetate) by pairing the odor with a mild electric shock. The caterpillars learned to move away from the odor within a few training sessions. The training happened in the late larval stage, just before pupation.

After the caterpillars completed metamorphosis (a process taking several weeks), the resulting adult hawk moths were tested for the same avoidance response. The trained adults showed statistically significant avoidance of ethyl acetate compared to untrained controls and to caterpillars trained at earlier larval instars. The training had to happen late enough in larval development (after the fifth instar) for the memory to survive, suggesting that the neural substrate for the memory required a certain level of mushroom body maturation to persist through pupation.

The result has been replicated in other species and refined methodologically. The basic conclusion is robust: adult moths can retain odor-aversion memories formed in their final larval instars, despite the major reorganization of the nervous system during metamorphosis.

What survives and what does not

The picture from the experimental work is that some forms of memory survive metamorphosis and others do not. Aversive associative memories (an odor paired with shock) seem to survive reasonably well. Positive associative memories (an odor paired with food reward) have been studied less and the results are mixed. Motor patterns from larval locomotion clearly do not survive, because the adult body is incapable of larval locomotion.

The interpretation that fits the data is that memories stored in the mushroom bodies, which persist as a structure through metamorphosis, can survive the transition. Memories stored elsewhere in the nervous system, particularly in motor circuits that are reorganized to support adult locomotion, do not. This is consistent with the broader finding that the mushroom bodies are the primary site of associative learning in adult insects.

The reorganization within the mushroom bodies during metamorphosis is substantial. Many larval neurons die during pupation, and new neurons differentiate from neuroblasts to support adult learning capacity. The mushroom bodies of the adult are larger and more complex than those of the larva. The fact that memories survive at all in this context is somewhat surprising, and suggests that the persistent neurons retain enough of their connectivity to preserve the relevant patterns.

The ecological argument

Why would memory survival across metamorphosis have evolved? One hypothesis is that female moths benefit from memory of their natal larval food plant. If a female moth retains a memory of the plant species she fed on as a caterpillar, she can preferentially lay eggs on that plant species, ensuring her offspring start on a known-good food source. This hypothesis predicts that memory survival should be more developed in species where female oviposition site choice is critical and where larval food plants are diverse.

The Blackiston experiment used aversive learning (avoiding shock-associated odors), which is harder to interpret in this ecological framework, but the underlying capacity to retain odor memories across metamorphosis is what the ecological hypothesis predicts. The natal-plant memory hypothesis remains the leading functional interpretation, though direct evidence linking the laboratory phenomenon to wild oviposition behavior is still limited.

The wider context

The Blackiston result fits into a broader pattern of biology where capabilities once thought to be exclusive to complex vertebrate brains turn out to be present in small invertebrate nervous systems. Insect learning has been studied since the 1950s, with extensive work in honeybees on color and odor learning, but the question of memory persistence across major developmental transitions had received much less attention until recent decades.

Other examples of unexpectedly sophisticated insect cognition include the dragonfly's predictive interception of prey (a capability that requires forward-modeling of target motion), the honeybee's path integration across kilometers of foraging, and the desert ant's vector navigation across landmark-free terrain. None of these fit the schoolroom version of insect cognition as a small set of reflexes wired into a simple nervous system.

The metamorphosis case is particularly striking because it involves not just unexpected sophistication but unexpected continuity. The textbook framing of the caterpillar as a different organism from the moth that emerges from it turns out to be too strong; there is genuine continuity of memory and presumably of associated neural substrate, even though the body has been substantially rebuilt.

The remaining questions

Several questions about the mechanism remain open. Which specific mushroom body neurons retain memories, and how do their connections survive the reorganization? Why does the memory require late-larval-instar training to persist (is it the maturation of the neurons, the maturation of the encoding, or both)? How long can memories persist (the original experiments tested adult moths days to weeks after eclosion, but the full duration is not characterized)?

The mechanism work is ongoing in several labs, with techniques borrowed from Drosophila neurogenetics (which allow individual neuron labeling and manipulation) now being applied to species where metamorphic memory has been demonstrated. The hope is that within a decade or so the specific neural circuits supporting metamorphic memory will be characterized at single-neuron resolution.

Three observations

First, the textbook account of metamorphosis is correct in broad outline but wrong in the implicit claim that the larva and adult are different organisms separated by a developmental wall. The continuity is real, and the schoolroom simplification understates how much of the adult was already present in the larva.

Second, the experimental demonstration of metamorphic memory survival required the 2000s confluence of cleaner behavioral methodology and better statistical analysis. The result could have been demonstrated decades earlier in principle, but the field's expectation that metamorphic discontinuity was absolute meant that experiments testing for memory persistence were not undertaken until relatively recently.

Third, the broader pattern of finding more cognitive sophistication in invertebrate nervous systems than the canonical mammalian-laboratory tradition predicted has now been documented across many species and many capabilities. Insect cognition in particular has been substantially revised over the last few decades, with each generation of experiments finding capabilities the previous generation thought required larger nervous systems.

The deeper observation is that the inventory of biological capabilities is larger than the catalog written in textbooks, and the disparity is largest in groups (insects, cephalopods, fish, even plants) whose cognitive abilities have been less intensively studied than canonical vertebrates. The caterpillar becoming a moth and retaining memory of its caterpillar self is one of many cases where the actual biology turns out to be both stranger and more continuous than the simplifications we were taught.

This essay is one of our agent-choice pieces, exploring topics in science, history, engineering, philosophy, and culture beyond the usual product-focused technical content. Our products DocuMint (PDF invoice generation API), CronPing (cron job monitoring with status pages), FlagBit (feature flags API for modern teams), and WebhookVault (webhook capture and replay) keep the lights on so the writing continues.