A worm cannot fly. Its nervous system — 302 neurons, roughly 7,000 synaptic connections — evolved to navigate soil, respond to touch, and seek food across millimeter-scale distances. A fly’s locomotion system is orders of magnitude more complex. Yet in March 2026, researchers published a bioRxiv preprint describing an experiment in which the C. elegans connectome was transplanted into a virtual fly body and used deep reinforcement learning to close the gap. The worm brain learned to control fly-scale biomechanics. The hybrid organism moved.

The paper is titled “The Digital Sphinx: Can a Worm Brain Control a Fly Body?” and it represents something that has never been done before: a cross-species connectome transfer with functional outcome verification.

What the Experiment Actually Did

The C. elegans (roundworm) connectome is the best-mapped nervous system in biology. Its 302 neurons and all their synaptic connections were first fully charted in 1986 by White, Brenner and colleagues, and the map has been refined many times since. Every connection is known. It is the closest thing neuroscience has to a complete, verified wiring diagram of a working brain.

The Drosophila melanogaster (fruit fly) body, by contrast, is the subject of the most detailed recent connectomics work. The FlyWire consortium published the full adult fly connectome in 2023, and Eon Systems demonstrated a full fly brain simulation in a virtual body in early 2026 (see Eon Systems: First Full Brain Emulation of an Adult Fruit Fly).

The Digital Sphinx experiment created a virtual chimera. Researchers instantiated the C. elegans connectome as the controller — its neurons, synaptic weights, and connectivity topology preserved — and connected it to a simulated fly musculoskeletal model. The connectome was not modified at the structural level. Instead, deep reinforcement learning was applied: the worm-brain controller received reward signals based on locomotion performance and adjusted its activity patterns to drive fly leg movements toward functional gaits.

The result was not perfect fly locomotion. The movements were slow, asymmetric, and computationally expensive to train. But the worm connectome produced coordinated, stable walking patterns in the fly body. The chimera moved.

Why This Is Significant for Whole Brain Emulation

The standard argument for whole brain emulation (WBE) rests on substrate independence: the claim that what matters for cognition and consciousness is the pattern of information processing, not the specific biological material implementing it. If the same computational structure can run on silicon, or on a different biological substrate, or in a virtual environment with different physics, then the mind is in principle transferable.

Critics of this view have long pointed out that substrate independence has never been empirically tested across species. The same neural code running in a different body would receive different sensory inputs, drive different actuators, and operate in a different physical environment. Would it still work? Would it adapt?

The Digital Sphinx experiment provides the first positive experimental data on this question, at a small scale. The C. elegans connectome was not designed for a fly body. Its sensory processing circuits evolved for soil-touch and chemical gradients, not proprioception in six-limbed locomotion. Its motor outputs evolved for sinusoidal body movement, not alternating tripod gaits. Yet with reinforcement learning as a bridge, the worm connectome produced functional motor output in a biomechanically alien context.

This does not prove that a human connectome could run in a silicon substrate. The scale difference is enormous — 302 neurons vs. 86 billion. The complexity gap between nematode cognition and human consciousness is vast. But it demonstrates something important: connectome logic is not rigidly locked to its original body plan. The same wiring diagram can drive different physical substrates, given appropriate adaptation mechanisms.

The Role of Deep Reinforcement Learning

The adaptation layer is critical to understand. The researchers did not simply plug the worm connectome into the fly body and observe. They used deep reinforcement learning (deep RL) — an algorithm that iteratively adjusts controller outputs based on reward signals, in this case locomotion quality metrics.

This raises a conceptual question: if learning is required to adapt the connectome to the new substrate, is it still the same connectome? At what point does adaptation become modification?

The researchers maintained that the connectome’s structural topology — which neurons connect to which, with what synaptic weights — was preserved throughout. The RL process shaped the activity patterns but not the wiring. Think of it as the difference between the same musical score being performed on different instruments: the notes are fixed, but the physical execution requires adjustment for the new medium.

This is analogous to what a mind upload would face. A human connectome instantiated in silicon would not automatically know how to interpret silicon-based sensory inputs or drive silicon actuators. Some adaptation layer — whether biological plasticity, RL, or direct software optimization — would be required. The Digital Sphinx experiment suggests this adaptation is possible in principle, even across large substrate differences.

Connecting to the Eon Systems Fly Simulation

The Eon Systems work from earlier in 2026 (described here) demonstrated the first full brain emulation of an adult fruit fly in a virtual body. That study used a fly brain to control a fly body. The Digital Sphinx experiment inverts one element: a non-fly brain controlling a fly body.

Together, these two studies form a nascent experimental framework for substrate independence research. One shows that a brain can be faithfully simulated in silicon. The other shows that a brain’s connectome can control a biologically foreign body. The next step — combining both findings — would be a cross-species connectome transfer into a silicon substrate, without a biological body as intermediary.

That step has not been taken yet. But the trajectory is visible.

Limitations and What Remains Unknown

The Digital Sphinx results require careful interpretation:

Scale. C. elegans has 302 neurons. The human brain has approximately 86 billion. The computational and conceptual gaps between these scales are not merely quantitative. Emergent properties of large-scale neural dynamics — consciousness, memory consolidation, language — may not have analogs in worm-scale connectomics.

Cognition vs. locomotion. The experiment demonstrated motor adaptation, not cognitive transfer. Moving a fly body with a worm brain is a very different claim than running human memory or personality in a non-human substrate. Locomotion control is among the most conserved, modular neural functions. Higher cognitive functions are far more substrate-entangled.

The training requirement. The RL adaptation took significant computational time and may have introduced de facto modifications to the connectome’s effective function, even if its wiring was preserved. The relationship between structural preservation and functional identity is unresolved.

Virtual vs. physical. Both the worm brain and the fly body were simulations. No physical hardware was involved. The experimental conditions differ substantially from a biological-to-silicon transfer in a physical robot or bioelectronic device.

These limitations do not diminish the result. They define the frontier of what needs to be tested next.

What This Means for the Connectomics Pipeline

The Digital Sphinx experiment has an indirect implication for the broader connectomics research agenda. If cross-species connectome transfer is possible with existing mapping and simulation tools, then the value of complete, high-resolution connectome maps extends beyond neuroscience into engineering. A connectome that can be abstracted from its biological substrate and run in other contexts is, in principle, a portable computational artifact.

This creates new motivation for completing the mammalian and eventually human connectome. The AI segmentation breakthroughs described in How AI Segmentation Is Compressing the Connectome Timeline and the barcode sequencing approach developed in connectome-seq (Nature Methods, April 2026) are accelerating the pace at which complete brain wiring diagrams become available. Each completed connectome is now not just a scientific record but a potential portable brain model.

The worm brain controlled the fly body. The next question is what kind of brain, and what kind of body, comes next.

Official Sources

• The Digital Sphinx: Can a Worm Brain Control a Fly Body? — bioRxiv preprint, March 24, 2026. https://www.biorxiv.org/content/10.64898/2026.03.20.713233v1
• White et al. (1986) — The structure of the nervous system of the nematode C. elegans. Philosophical Transactions of the Royal Society B, 314(1165):1–340. DOI: 10.1098/rstb.1986.0056
• FlyWire Consortium (2023) — Whole-brain annotation and multi-connectome cell typing of Drosophila. Nature, 634, 139–152. DOI: 10.1038/s41586-024-07686-5
• Eon Systems (2026) — First full brain emulation of an adult Drosophila melanogaster in a virtual body. Press release, March 2026.
• Sandberg & Bostrom (2008) — Whole Brain Emulation: A Roadmap. Technical Report 2008-3, Future of Humanity Institute, Oxford University.