Insect eyes are built from an ancient plan. The transcription factors and signaling pathways that pattern the retina and specify its photoreceptor types are recognizable across lineages separated by more than 300 million years, which means the Drosophila retina is a reasonable proxy for a much larger group than expected. Yet insect eyes differ dramatically in what they can see. Our comparative work shows that the variation comes from secondary patterning layered onto the conserved plan: some species shift the stochastic ratio, some add regions or receptor types, and some abandon stochastic patterning altogether (Gao et al., PNAS 2025).
That combination, a fixed blueprint and a well-defined set of modifications, lets us ask three questions that organize our evolutionary work. Where does novelty come from, and in particular, how do new types of neurons arise? When different animals face the same problem, do they converge on the same genes and the same genetic changes, or find different solutions? And how free is the retina to change, given that every change has to be accommodated by the brain it connects to?
Novelty: where new cell types come from
The clearest case of a new neuron type in our work is the house fly Love Spot. Male house flies have a dorso-frontal eye region specialized for detecting and chasing females in flight, and within it R7 photoreceptors normally dedicated to color vision are repurposed for target detection. We showed that these cells are first specified as canonical R7s and then adopt a chimeric identity, combining regulators of color and motion vision under the control of Doublesex and Spineless. New cell types can arise from new combinations of existing regulators (Donati et al., in review; preprint on bioRxiv). The lab established transgenesis in Musca, which makes these functional tests possible.
In the monarch butterfly, new R7 types arose by a different route. Nymphalid butterflies evolved red color vision, and single-cell profiling of the developing monarch retina resolves four R7 types rather than two, including long-wavelength types that make their ommatidia red-detecting. We are working out how these new types are specified during development through a second stochastic cell fate decision, and how that decision relates to the Spineless choice that precedes it.
Butterflies as a group made an earlier change: they added a second R7 photoreceptor per ommatidium, so that two independent Spineless decisions yield three ommatidial types instead of two (Perry et al., 2016). We found that the duplication was achieved by converting R3/4 cells to an R1/6 fate, which triggers R7 recruitment on both sides of the ommatidium, and we engineered a “butterfly-fly” in Drosophila that recapitulates it (Gao et al., Science Advances 2026).
A new project asks a related question within a single cell type. In males of the cabbage white Pieris rapae, violet-sensitive R7 photoreceptors in one ommatidial type carry a fluorescing pigment that retunes their spectral sensitivity; females lack it. Tiezheng Fan is taking the lead, building on single-cell data generated by Ke Gao, to find the genes that make the pigment and the regulators that create the difference between the sexes.
Convergence: same problem, same solution?
Red vision is a case in point. Nymphalid butterflies such as the monarch and swallowtails such as Papilio evolved the ability to see red independently, using different photoreceptors and circuits, and the monarch work asks which regulators were recruited to produce red-detecting cell types.
Across species from 23 fly families, we are tracing how the Rhodopsin family gained and lost members under a tight constraint: only a few photoreceptor types are available to express them. Different lineages found different solutions. Some regionalized the retina to make room for a new Rhodopsin without displacing another; others let one Rhodopsin replace another, sharpening one color comparison at the expense of a different one. We have pinpointed cis-regulatory and trans-regulatory changes behind two of these shifts.
The same pattern holds at the level of retinal organization. Several lineages have abandoned stochastic patterning for deterministic spatial maps, and they did so by different routes (Gao et al., PNAS 2025).
Constraint and flexibility: matching the eye to the brain
A change in the retina is only useful if the brain can use it. Our butterfly-fly showed how the developing brain accommodates an added photoreceptor. In the fly brain, R7 subtypes connect to Dm8 neurons that are born in excess and die if they fail to find a partner; in the butterfly-fly, the surplus Dm8 cells are rescued and form subtype-appropriate connections. Developmental overproduction maintains a reservoir of neurons that evolution can draw on (Gao et al., Science Advances 2026).
Developmental history also constrains what is possible. Love Spot R7s switch identity only after their axons have already targeted the color-processing medulla, so the axons must subsequently shorten to reach motion vision circuits in the lamina, a detour imposed by the order of events in development (Donati et al., bioRxiv 2026).
At the level of the molecules that wire neurons together, the system is more flexible than expected. In Drosophila, R7 subtypes are matched to their Dm8 partners by a defined pair of cell-adhesion molecules. Our cross-species single-cell data show that this code is not conserved: different molecular families carry the same matching function in house flies, mosquitoes, and butterflies, so three lineages wire the same homologous connection with three different solutions. We are now defining these codes lineage by lineage, testing their function, and asking how new codes arise alongside new photoreceptor types. In parallel, with Nikos Konstantinides (Paris), we are comparing optic lobe cell types across house flies, mosquitoes, butterflies, and a cricket outgroup.
People: Ke Gao, Antoine Donati, Takami Sho, Tiezheng Fan. Collaborators: Nikos Konstantinides (Institut Jacques Monod, Paris), Kentaro Arikawa, Gregor Belušič, Mathias Wernet, Andrew Tomlinson (Columbia).
Key papers: Gao et al., Science Advances 2026 (butterfly retina evolution); Gao et al., PNAS 2025 (deep conservation of the insect eye ground plan); Donati et al., bioRxiv 2026 (Love Spot); Perry et al., Nature 2016 (three-way stochastic choice in butterflies); Perry et al., Annual Review of Genetics 2017; Wernet, Perry, and Desplan, Trends in Genetics 2015.
Funding: NSF CAREER award (IOS-2339031); NIH NHGRI R01HG013634 (EDGE program); Hellman Fellows Program.