Many organisms, including cosmopolitan drosophilids, show circadian plasticity, varying their activity with changing dawn–dusk intervals[1]. How this behaviour evolves is unclear. Here we compare Drosophila melanogaster with Drosophila sechellia, an equatorial, ecological specialist that experiences minimal photoperiod variation, to investigate the mechanistic basis of circadian plasticity evolution[2]. D. sechellia has lost the ability to delay its evening activity peak time under long photoperiods. Screening of circadian mutants in D. melanogaster/D. sechellia hybrids identifies a contribution of the neuropeptide pigment-dispersing factor (Pdf) to this loss. Pdf exhibits species-specific temporal expression, due in part to cis-regulatory divergence. RNA interference and rescue experiments in D. melanogaster using species-specific Pdf regulatory sequences demonstrate that modulation of this neuropeptide’s expression impacts the degree of behavioural plasticity. The Pdf regulatory region exhibits signals of selection in D. sechellia and across populations of D. melanogaster from different latitudes. We provide evidence that plasticity confers a selective advantage for D. melanogaster at elevated latitude, whereas D. sechellia probably suffers fitness costs through reduced copulation success outside its range. Our findings highlight this neuropeptide gene as a hotspot locus for circadian plasticity evolution that might have contributed to both D. melanogaster’s global distribution and D. sechellia’s specialization.

Confocal images were obtained using the Zeiss microscopy ZEN 2.3 SP1 software. Drosophila Activity Monitor data were collected using the DAM System software, version 311X. All image analysis was performed in Fiji running v2.9.0 of ImageJ. RNA spots were detected using the RS-FISH macro, and Scholl analysis was performed using the neuroanatomy plug-in. Sequences were analyzed and visualised using SnapGene (www.snapgene.com) running MUSCLE (v3.8.1551), and Jalview (v2.11.2), and using the R package Phangorn (v2.11.2). Sequence alignments were perfomed using Samtools (v1.19.2), and tests of neutrality were performed using the DNA sequence polymorphism software (v6). Drosophila Activity Monitor data were analysed using Rethomics in R (v3.6.3). All R code used for data analysis are available at: github.com/mshahandeh/circ_plasticity. No statistical methods were used to predetermine sample size. For behavioural experiments, we aimed for a sample size of 25-30 individuals, as significant differences were easily detected at these sizes. In the case of our hybrid screen, where hybrid flies were harder to make, we aimed for 15 individuals per strain but due to the strong reproductive isolation between species, some genotypes were difficult to cross to D. sechellia, resulting in a lower sample size. For image analyses, we obtained images from 5 brains from each strain per treatment because this allowed for parallel processing of multiple genotypes and time points. Additionally, quantifications of these stainings were largely consistent, and significant differences were apparent at this sample size. All behavioural experiments were collected over at least two technical replicates with corresponding controls to ensure reproducibility, with the exception of those shown in Fig. 1c, where we replicated only two photoperiod treatments: 12:12 h LD and 16:8 h LD, because we continued the use of these specific photoperiods throughout the work. All brain dissections and stainings for a given experiment were performed in parallel to ensure comparability between time points and fixed microscope settings were used within each experiment. Two time-points of each histological experiments (Pdf smFISH, Pdf immunofluorescence, and Scholl analysis of s-LNv axonal projections) were replicated once in order to ensure the overall patterns of expression were the same. We did not randomise within experiments, and instead ran all genotypes in parallel for all experiments, except for the hybrid screening, where we measured the behaviour of hybrids as and when crosses produced offspring, with at least one replicate run in parallel with corresponding controls (data collection for hybrid screening took >2 years). Quantifications for stainings and collection of behavioural data were performed blind to treatment.