Taste adaptations associated with host specialization in the specialist Drosophila sechellia

Chemosensory-driven host plant specialization is a major force mediating insect ecological adaptation and speciation (Berlocher and Feder, 2002; Chapman, 2003; Jaenike, 1990). The role of the olfactory sensory system in mediating these processes is well understood in many species (Hansson and Stensmyr, 2011; Schoonhoven et al., 2005; Zhao and McBride, 2020). For instance, plant volatiles contribute to sympatric speciation and reproductive isolation via host plant shifts, as in the larch bud moth Zyraphera diniana (Emelianov et al., 2003; Syed et al., 2003) and the apple maggot Rhagoletis pomonella (Linn et al., 2003; Olsson et al., 2009). Whether the taste system also contributes to host specialization has been less studied, despite its essential role for recognition and acceptance of food and oviposition sources. Genomic studies suggest that, in general, insects with specialized diets underwent losses/pseudogenization of genes mediating taste detection relative to generalists (Anholt, 2020; Cande et al., 2013; Robertson, 2019), but investigations on the functional consequences are relatively scarce.

The drosophilid fly Drosophila sechellia is endemic to the Seychelles Islands in the Indian Ocean, where it feeds and oviposits on the fruit of Morinda citrifolia (Lachaise et al., 1988; Tsacas and Bachli, 1981), known as noni. Although a noni specialist, flies could occasionally be found in other fruits such as mango, figs and papaya (Matute and Ayroles, 2014; Salazar-Jaramillo and Wertheim, 2021). Drosophila sechellia has been much studied because it is a specialist closely related to the generalist saprophagous D. melanogaster and D. simulans, providing an excellent opportunity for studying the genetic, physiological and behavioral mechanisms underlying host specialization (Auer et al., 2022; Dekker et al., 2006; Stensmyr et al., 2003; Zhao and McBride, 2020). Drosophila sechellia has a common ancestor with these species ∼3–5 mya and ∼0.25 mya, respectively (Garrigan et al., 2012). While D. sechellia uses noni as its host, its relatives are repelled because of the high content of hexanoic and octanoic acid (19% and 58%, respectively) in the ripe (but not green or rotten) fruit (Farine et al., 1996; Pino et al., 2010; R'Kha et al., 1991; D. simulans can be found occasionally on noni in the Seychelles: Matute and Ayroles, 2014). Specialization on noni might also provide protection from wasp parasitism, as these fatty acids are toxic to the larvae (Salazar-Jaramillo and Wertheim, 2021). At >1% v/v, octanoic acid is toxic (via smell and contact) to many drosophilids, including D. simulans and D. melanogaster (Farine et al., 1996; Legal et al., 1994). In contrast, D. sechellia evolved detoxification mechanisms to cope with octanoic and hexanoic acid (Drum et al., 2022; Lanno et al., 2019; Legal et al., 1994; Pino et al., 2010), and these furthermore stimulate oviposition and egg production (Álvarez-Ocaña et al., 2022 preprint; Amlou et al., 1998; Jones, 2005; Lavista-Llanos et al., 2014; R'Kha et al., 1991). At <1% v/v, food solutions containing these fatty acids are preferred over those lacking them, but this preference is stronger in D. sechellia than in D. melanogaster (Ferreira et al., 2020).

Insects detect chemicals in the air (i.e. by olfaction) and by contact (i.e. by taste) using specialized chemosensory cells housed in cuticular structures called sensilla. Gustatory sensilla are found mostly in the proboscis and the leg tarsi (Chapman, 1998; Chen and Dahanukar, 2020; Scott, 2018; Vosshall and Stocker, 2007). These sensilla house gustatory receptor neurons (GRNs) responsive to sweet, low salt, high salt, water, bitter and sour compounds (Chen and Dahanukar, 2020; French et al., 2015a; Liman et al., 2014; Scott, 2018). Caloric substances activate sweet GRNs, triggering proboscis extension and ingestion, while bitter (potentially toxic) substances produce aversion by activating bitter GRNs; some bitter substances additionally suppress sugar GRN activation (French et al., 2015b; Jeong et al., 2013). GRNs express several chemosensory proteins, including gustatory receptors (GRs; Fujii et al., 2015; Robertson et al., 2003; Scott et al., 2001; Thorne et al., 2004), ionotropic receptors (IRs; Benton et al., 2009; Croset et al., 2010; Koh et al., 2014; Ni, 2021) and odorant binding proteins (OBPs; Galindo and Smith, 2001). GRNs express multiple chemosensory proteins (Chen and Dahanukar, 2020; Montell, 2021).

Some of the chemosensory specializations of D. sechellia for its host involve changes in the olfactory system that increase long-distance attraction to noni volatiles, such as increases in the number of olfactory sensilla for detection of the signature host compounds methyl hexanoate and hexanoic acid and molecular changes in the olfactory receptor OR22a and ionotropic receptor IR75b that mediate olfactory attraction to them (Auer et al., 2020, 2021; Dekker et al., 2006; Prieto-Godino et al., 2017; Stensmyr, 2009; Zhao and McBride, 2020). Host plant taste specializations have been less explored. Matsuo et al. (2007) found that the loss of OBP57d and OBP57e mediates contact-mediated oviposition acceptance of fatty acids, although both olfactory and gustatory inputs are required for oviposition on noni substrates (Álvarez-Ocaña et al., 2022 preprint). Genomic studies showed that D. sechellia lost many GR and ‘divergent’ IR genes typically expressed in external GRNs (Crava et al., 2016; Croset et al., 2010; McBride, 2007; McBride and Arguello, 2007) which detect bitter compounds and possibly fatty acids in the generalist D. melanogaster (Ahn et al., 2017; Brown et al., 2021; Dweck and Carlson, 2020; Masek and Keene, 2013; Sánchez-Alcañiz et al., 2018; Scott, 2018). McBride and Arguello (2007) suggested two explanations for the loss of chemosensory genes: (1) after specialization on M. citrifolia, D. sechellia was exposed to fewer bitter (likely harmful) compounds in their ecological niche, leading to loss of selection for chemosensory genes; (2) the loss of chemoreceptors for detection of M. citrifolia deterrents facilitated the specialization of D. sechellia on M. citrifolia. In line with these predictions, electrophysiological studies found major losses of sensitivity to some bitter compounds (Dweck and Carlson, 2020), and D. sechellia has increased feeding preference for solutions containing host fatty acids, in comparison with its close relatives (Ferreira et al., 2020). Thus, these findings suggest that taste is also involved in host plant adaptation and specialization in D. sechellia.

Here, we used behavioral assays to further study the taste and feeding responses of D. sechellia and its generalist relatives. We found that D. sechellia has a reduced behavioral aversion to bitter compounds, which correlates with the lineage-specific loss of two GR genes (GR39a.a and GR28b.A), and increased appetite for noni fatty acids. Our findings thus support the hypothesis that host specialization in this fly involves adaptive changes in the taste system.

Drosophila sechellia Tsacas and Baechli 1981 flies (strain 14021-0248.27, provided by the former University of California San Diego stock center, and strain 14021-0248.25, provided by Dr M. Eisen, University of California Berkeley) and D. melanogaster Meigen 1830 (strain Canton-S) were used in most experiments. Drosophila sechellia has low genetic diversity, but it is proposed that its small effective population size resulted from trade-offs between life-history traits and the use of a predictable competition-free host (Legrand et al., 2009). All experiments were conducted with mated female flies (and a few with D. sechellia males). Two additional lines of D. simulans Sturtevant 1919 (strains 14021-0251.312 and 14021-0251.269) and D. melanogaster (strain 14021-0231.199) were obtained from the former UCSD stock center. Other D. melanogaster lines used include D. melanogaster null GR28b.A mutant (obtained from the Bloomington Stock Center; RRID:BDSC 24190), strain w118, wCS:GR39a and wCS (generously provided by Dr J. Carlson, Yale University). Drosophila melanogaster and D. simulans flies were grown on standard fly food at room temperature. Drosophila sechellia flies were reared at 25°C on standard fly food supplemented with a small piece of M. citrifolia fruit leather (Hawaiian Organic Noni LLC, Kauai, HI, USA) and a pinch of dry yeast. Flies were 2–7 days old at the time of the experiments.

Assays were performed on individual mated flies which were food deprived by placing them in vials containing two pieces of water-saturated tissue paper. After 22–24 h, flies were gently anesthetized under CO₂ and glued by their dorsal thorax onto a glass slide using clear nail polish. Flies were allowed to recover in a humid chamber for 2 h before proboscis extension response (PER) tests and temporal consumption assays (TCA) were carried out. Mounted flies were first water satiated and, for PER assays, either tarsi from all six legs or the proboscis of individual animals was stimulated 3 times at 5 s intervals with a drop of appetitive solution (1 mol l⁻¹ sucrose, CAS # 57-50-1, Fisher Chemicals, Waltham, MA, USA; or 1 mol l⁻¹d-glucose, CAS # 59-99-7, Sigma-Aldrich, St Louis, MO, USA), a mixture of 1 mol l⁻¹ sucrose and 0.5 mmol l⁻¹ denatonium, or noni fruit leather strips [Hawaiian Organic Noni LLC (reconstituted by blending with water), 0.00657 g ml⁻¹], and the number of proboscis extensions was recorded (Fig. 1A). The number of animals that did or did not extend their proboscis was calculated and data were analyzed using Fisher exact tests (for comparisons between independent groups) or McNemar tests (for paired data) (Zar, 1999). Data were considered significant if P<0.05. In subsequent experiments, we used glucose as a sugar stimulus because it evokes stronger responses than sucrose in D. sechellia (Fig. 1B,C), and because it naturally occurs in noni, although in small amounts (Potterat and Hamburger, 2007).

TCA were performed similarly to those reported by Yao and Scott (2022). Mated females, 2–4 days old, were food deprived for 22 h and prepared for experiments as above (Fig. 1A). All tarsi from individual flies were touched with a drop of 750 mmol l⁻¹ glucose or noni juice (noni only, Dynamic Health Laboratories, New York, NY, USA), as the fruit juice constitutes a more reproducible stimulus than the ripe fruit (Auer et al., 2020). Each fly was stimulated up to 10 consecutive times and allowed to drink, and then stimulated again in the same fashion until it stopped consuming (usually before 60 s). The time and duration of each feeding event were manually recorded using an online chronometer (http://online-stopwatch.chronme.com/; Fig. 2C). Data were exported and the following parameters were calculated off-line for each fly and stimulus (Fig. 2D–F): the number of feeding events, the total feeding time (summed duration of all feeding events), and the percentage of animals that fed.

Feeding assays were conducted similarly to those described previously (Reisenman and Scott, 2019). Groups of 2–7 day old mated flies (n=11–15) were food deprived for 24 h, and then transferred to a vial containing a piece of filter paper (2.7 cm diameter, Whatman, cat. no. 1001 125) impregnated with 160 µl of food solution dyed blue with erioglaucine (0.25 mg ml⁻¹, CAS # 3844-45-9, Sigma-Aldrich). To facilitate feeding, vials were flipped upside down so that the filter paper with food solution faced up (Fig. 3A). Flies had access to food for 30 min and then vials were frozen for at least 60 min. After freezing, flies in each vial were individually scored, blind to treatment, using the following five-point scale based on the amount of food visualized as blue dye in the abdomen: 0, no dye=no food; 0.25, ‘trace’ of blue dye=‘taste’ of food; 0.5, up to ¼ of the abdomen dyed blue; 1, between ¼ and ½ of the abdomen dyed blue; and 2, more than ½ of the abdomen dyed blue (Fig. 3A). For each vial, a single feeding score value was calculated as: (0×n₀+0.25×n_0.25+0.5×n_0.5+1×n₁+2×n₂)/N, where n_(0–2) denotes the number of flies in each score category, and N is the total number of flies per vial; this single feeding score constituted a biological replicate (Reisenman and Scott, 2019). All experiments and scoring were conducted blind to treatment. In one experiment, single flies (i.e. a single fly per vial) had access to food for 30 min, and were visually inspected for viability every 5 min, frozen and scored.

Food solutions consisted of 750 mmol l⁻¹ glucose alone (control), or 750 mmol l⁻¹ glucose plus one of the following test compounds (all from Sigma-Aldrich, unless specified otherwise): 0.5 mmol l⁻¹ denatonium benzoate (CAS # 3734-33-6); 10 and 25 mmol l⁻¹ caffeine (CAS # 58-08-2); 1 mmol l⁻¹ (l)-lobeline hydrochloride (CAS # 134-63-4); 10 mmol l⁻¹l-canavanine (CAS # 543-38-4); 10 mmol l⁻¹ theophylline (CAS # 58-55-9); 10 mmol l⁻¹ coumarin (CAS # 91-64-5); 100 mmol l⁻¹ (1.3% v/v) octanoic acid (CAS # 124-07-2); or 100 mmol l⁻¹ (1.6% v/v) hexanoic acid (CAS # 142-62-1, MP Biomedicals, Solon, OH, USA). High concentrations of glucose elicit strong feeding responses, while addition of bitter compounds reduces taste and feeding responses (Dweck and Carlson, 2020; French et al., 2015b; Ling et al., 2014; Weiss et al., 2011). Denatonium is a synthetic compound; caffeine, lobeline and theophylline are plant-produced alkaloids, and canavanine is a plant-produced amino acid; caffeine and canavanine defend plants against herbivory (Fürstenberg-Hägg et al., 2013). Octanoic and hexanoic acid are prominent compounds in M. citrifolia ripe fruits (Pino et al., 2010), which are harmless to D. sechellia but toxic (at concentrations >1% v/v) to D. melanogaster and D. simulans (Hungate et al., 2013; Jones, 1998). We also used pure noni juice dyed blue. Noni contains anthraquinones and the coumarin-derivative scopoletin (Deng et al., 2010; Ikeda et al., 2009; Satwadhar et al., 2011; Singh, 2012), both of which are insoluble in water, precluding testing. In two control experiments, flies were allowed to feed on 750 mmol l⁻¹ glucose or water dyed blue in the presence of octanoic acid vapors (10 µl of 100 mmol l⁻¹ octanoic acid or the mineral oil solvent loaded in filter paper), as described in Reisenman and Scott (2019) (Fig. S3A). As much as possible, food solutions were tested in parallel with overlapping cohorts of flies. At least one control test (750 mmol l⁻¹ glucose alone) for each species was always conducted along with tests with bitter/fatty acid compounds (i.e. at the same time and with overlapping fly cohorts). Because excess control data points were therefore obtained over the course of all experiments, control data for each species/genotype were randomly eliminated to achieve sample sizes comparable to those of flies tested with solutions containing bitter/fatty acid compounds.

Raw and normalized feeding scores were respectively used for comparisons within or between species and sexes. Normalization also served to account for differences in basal (glucose) consumption between strains and genotypes, and for potential differences in ingestion which could result from variation in room temperature, age and fly cohort. Normalized feeding scores were calculated as: feeding score (glucose+bitter/fatty acid)/feeding score (glucose only). Feeding scores from vials with control data (i.e. glucose only) obtained for the same day and species were averaged and this average was used for normalization. Thus, normalized values of ∼1 indicate no difference in consumption between control and solutions containing a test compound, and values <1 and >1 indicate feeding aversion and enhancement, respectively. To established the behavioral valence of each compound, normalized feeding scores were compared against the expected median=1 (no difference in consumption between test and control solutions) using one-sample signed rank tests (Zar, 1999).

Mann–Whitney U-tests were used to compare two independent groups. Kruskal–Wallis ANOVA were used for comparing more than two groups; significant results were followed by Dunnett’s (for comparisons involving equal sample sizes) or Dunn's tests (for comparisons involving unequal sample sizes or to compare control versus all experimental groups) (Zar, 1999).

PER to highly appetitive stimuli, 1 mol l⁻¹ glucose or sucrose, were taste organ dependent: tarsi stimulation evoked stronger responses than proboscis stimulation in D. sechellia but not in D. melanogaster (Fig. 1A,B), and this pattern was consistent across D. sechellia strains and sexes (Fig. S1B,C). At lower sugar concentrations, D. melanogaster showed higher PER upon proboscis stimulation (Fig. S1A). In both species, PER to 1 mol l⁻¹ sucrose was strongly reduced upon addition of the bitter compound denatonium (Fig. 1D). These results demonstrate taste organ-specific appetitive and aversive taste responses across species.

Individual flies were first tarsi stimulated 3 times with 1 mol l⁻¹ sugar, offered water, and then tarsi stimulated 3 times with noni reconstitute. Drosophila simulans and D. melanogaster had a much higher PER rate to stimulation with sugar than with noni (90–100% and 25–40%, respectively, McNemar tests, P<0.001 for both species), while D. sechellia had similar responses to the two stimuli (73% and 63% PER, P>0.05; Fig. 2A). To probe whether reduced PER to noni stimulation in the generalists was due to olfaction, we tested 24 h food-deprived D. melanogaster either intact or with their olfactory organs (antennae and maxillary palps) removed 2 days before experiments. The two groups of flies had indistinguishable PER to sucrose and noni, as in Fig. 2A, confirming that the aversion to noni is solely due to taste (Fig. 2B; Brown et al., 2021).

Drosophila melanogaster and D. sechellia individual flies were also assayed in a temporal consumption assay. Flies were prepared as before (Fig. 1A), and their tarsi were repeatedly stimulated with 750 mmol l⁻¹ glucose or noni juice and allowed to drink. A similar proportion of flies from each species fed on glucose (81% D. melanogaster and 91% D. sechellia; Fisher exact test, P>0.05; n=21, 22), but a larger proportion of D. sechellia flies fed on noni (31.6% and 78.3%, respectively; P<0.0074, n=19, 23; Fig. 2D). Drosophila melanogaster consumed food for longer and fed more times upon stimulation with glucose than with noni, but these behavioral metrics were not different in D. sechellia (Fig. 2E,F). Thus, noni is comparatively more appetitive in D. sechellia than in D. melanogaster.

We then switched to a group feeding assay where flies were unrestrained (Fig. 3A) to investigate responses to known bitter stimuli or noni fatty acids across species. Flies were offered 750 mmol l⁻¹ glucose (control), or 750 mmol l⁻¹ glucose plus a test stimulus (bitter or fatty acid). Addition of any of the test stimuli significantly reduced feeding in D. melanogaster (Fig. 3B; Kruskal–Wallis ANOVA followed by Dunn's tests, P<0.01); D. simulans had reduced feeding on solutions containing caffeine, lobeline, coumarin or the two fatty acids (post hoc Dunn's tests, P<0.05; Fig. 3C). In D. sechellia, addition of canavanine, coumarin or either of the fatty acids had no impact (post hoc tests, P>0.05; Fig. 3C). To evaluate potential lethal effects of octanoic acid in the generalist species, individual flies were placed in vials offering 750 mmol l⁻¹ glucose alone or with octanoic acid and allowed to feed for 30 min, and we recorded viability. None of the flies offered only glucose died, but 4/16 and 1/19 D. melanogaster and D. simulans, respectively, were dead or stuck to the food, but otherwise flies seemed normally active. The remaining 12 and 18 flies of each of these two species fed significantly less on solutions with octanoic acid (Fig. S2). We also tested whether feeding aversion to octanoic acid could be due to olfaction. Drosophila melanogaster flies were offered 750 mmol l⁻¹ glucose in the presence or absence of octanoic acid odors (no contact, as illustrated in Fig. S3A), and found that flies from both groups fed indistinguishable large amounts of glucose (Fig. S3B). Overall, these results confirmed that the aversion to octanoic acid is mediated solely by taste and it is not due to lethality.

Feeding on 750 mmol l⁻¹ glucose (control) varied across species, with D. sechellia feeding less than its generalist relatives (Kruskal–Wallis ANOVA followed by post hoc Dunn’s tests, P<0.01; n=26–29). Thus, for comparing responses across species, we normalized the scores of flies fed test solutions (i.e. 750 mmol l⁻¹ glucose+bitter/fatty acid) to that of flies fed 750 mmol l⁻¹ glucose only. This normalization, here and subsequently, additionally accounts for possible cohort, temperature and day-to-day variability. The normalized feeding scores of the generalist species were significantly <1 for most test compounds (Fig. 4; one sample signed rank tests, P<0.05 in all cases except D. simulans tested with denatonium and lobeline). In contrast, the normalized feeding scores of D. sechellia fed solutions containing canavanine, coumarin, octanoic acid and hexanoic acid were not statistically different from 1, indicating no feeding aversion or enhancement (P>0.05 in all cases). Drosophila sechellia showed significant feeding aversion to caffeine, lobeline, theophylline and denatonium (medians<1, P<0.05, Fig. 4). However, for the most part, their normalized scores were divergent from those of D. melanogaster and D. simulans: responses to caffeine, canavanine, coumarin, theophylline and the two fatty acids were similar in the two generalists (Kruskal–Wallis ANOVA followed by Dunn’s tests, P>0.05) but different from those of D. sechellia (post hoc Dunn’s tests, P<0.05). In all cases (except denatonium and lobeline), the responses of D. sechellia were different from those of D. melanogaster. These results were consistent between sexes and for another strain of D. sechellia (Fig. S4A,B), showing that the reduced feeding aversion is species and not strain specific. Notably, males had feeding aversion to hexanoic acid (Fig. S4A), suggesting that this noni compound has sex-specific functions (Álvarez-Ocaña et al., 2022 preprint). We compared responses to caffeine, an alkaloid in the plant family (Rubiaceae) that includes coffee and noni (Singh, 2012), and to denatonium, an aversive synthetic compound, across two strains for each species. Responses within a species were consistent, and caffeine evoked similarly higher levels of feeding aversion in the generalists (Fig. S4C). Overall, these results indicate that D. sechellia has a lineage-specific reduced sensitivity to canavanine, caffeine, theophylline, coumarin and octanoic and hexanoic acids. These are likely a subset of the compounds to which D. sechellia has reduced aversion (Dweck and Carlson, 2020).

We next investigated whether the reduced aversion to bitter compounds in D. sechellia correlates with the loss of GRs in this species. McBride (2007) showed that at least 14 GR genes, several of which mediate responses to bitter compounds in D. melanogaster, are lost or pseudogenized (Table 1). We focused on GR28b.a and GR39a.a, as these are widely expressed (75–100%) in D. melanogaster bitter sensilla (Ling et al., 2014; Weiss et al., 2011; Table 1). Although GR39a and GR28b undergo alternative splicing in drosophilid flies (Gardiner et al., 2008; Sang et al., 2019), we used null D. melanogaster mutants for each of these GR genes. We hypothesized that one or both mutants would recapitulate the D. sechellia lineage-specific reduced aversion to bitter compounds specifically but not to fatty acids, as responses to these compounds are mediated by IRs (Ahn et al., 2017; Masek and Keene, 2013). Feeding responses were normalized as before, and the responses of mutants were compared with those of their respective genetic controls (rather than with those of wild-type flies), as is standard in the field. The normalized feeding scores of D. melanogaster flies from all genotypes (mutants and genetic controls) offered any of the test compounds (except GR39a mutants offered canavanine) were significantly <1 (Fig. 5; one-sample signed rank tests, P<0.05), indicating reduced feeding for solutions containing any of the test chemicals, but not loss of significant aversion as observed for some compounds in D. sechellia. However, GR28b and GR39a mutants fed significantly more than their respective genetic controls on solutions containing canavanine or coumarin, and GR28b mutants also fed more than their control on solutions containing theophylline (Mann–Whitney U-tests, P<0.05 in all cases). In addition, mutants showed a generalized reduction of aversion to bitter compounds (aversive responses to caffeine and theophylline in GR39a mutants were smaller than those of controls albeit not significant (P=0.066 and P=0.051, respectively), consistent with Dweck and Carlson (2020). As expected, mutants retained aversion to fatty acids (P>0.05). Overall, these results demonstrate that the loss of single bitter GR genes is sufficient to reduce feeding aversion. In particular, the loss of two GR genes which are widely expressed in bitter GRNs in D. melanogaster collectively correlates with D. sechellia’s reduced aversion to canavanine, theophylline and coumarin.

Addition of fatty acids to appetitive solutions significantly reduced feeding in the generalist species but not in D. sechellia (Figs 3 and 4), suggesting that proteins that detect them may be lost (or have lower expression) in bitter GRNs (Ahn et al., 2017). In D. melanogaster, concentrations <1% v/v are non-toxic and stimulate sugar GRNs (GR64e-positive; Kim et al., 2018), evoking appetitive responses. We thus investigated whether octanoic acid at >1% (octanoic acid is highly concentrated in noni) can increase feeding in D. sechellia, rather than aversion. We tested parallel cohorts of flies with 750 mmol l⁻¹ glucose, water only, or water+100 mmol l⁻¹ (1.3% v/v) octanoic acid. As expected, in both species, the feeding scores of flies fed glucose were much higher than those of flies offered water or water+octanoic acid (Fig. 6A; Mann–Whitey U-tests followed by Dunn's tests, P<0.05). Glucose responses normalized to water were >1 in both species (one-sample signed rank tests, P<0.01 in both cases; Fig. 6B, left), indicating feeding enhancement. The water-normalized scores of D. melanogaster to octanoic acid were aversive (<1, P<0.001), while those of D. sechellia were appetitive (>1, P<0.01; Fig. 6B, right). The water-normalized responses of D. sechellia to glucose or octanoic acid were higher than those of D. melanogaster (Mann–Whitney U-tests, P<0.001; Fig. 6B). In addition, D. sechellia consumed similar amounts of water in the presence or absence of octanoic acid odor (no contact allowed as illustrated in Fig. S3A; Mann–Whitney U-test, P>0.05; Fig. S5), which indicates that olfaction alone is not sufficient to increase consumption (Masek and Keene, 2013). This shows that the taste of octanoic acid, a prominent toxic host compound, increases feeding in D. sechellia at concentrations that are aversive to its generalist relatives.

Chemosensory specializations allow specialist insects to find their hosts. Host olfactory adaptations are better understood than taste adaptations, and include a gain and/or increase in peripheral sensitivity to host odors. To investigate taste adaptations, we took advantage of the specialist D. sechellia, which is closely related to the generalists D. simulans and D. melanogaster. Drosophila sechellia feeds and oviposits almost exclusively on ripe noni fruit, which is toxic (as a result of its high fatty acid content) to many drosophilid species. Using taste and feeding assays, we found that in comparison with its close relatives, adult D. sechellia has reduced feeding aversion to various bitter substances that correlates with the lineage-specific loss of various GR genes, including the broadly expressed GR39a.a and GR28b.a. Furthermore, D. sechellia has increased appetite for octanoic acid, a signature noni fatty acid compound. Our studies powerfully demonstrate the role of GRs in mediating taste specialization in D. sechellia.

Using PER, we measured taste detection in the absence of feeding and we found species- and organ-specific differences: D. sechellia had stronger PER to sugar stimulation of the tarsi than to proboscis stimulation, while D. melanogaster showed similar responses upon stimulation of either taste organ (Fig. 1). In contrast, at low concentrations, proboscis stimulation evoked higher PER in D. melanogaster (Fig. S1), consistent with the higher responsiveness of proboscis taste sensilla in this species (Dahanukar et al., 2007; Ling et al., 2014). Because the number and position of taste sensilla in the proboscis are similar in the two species (Dweck and Carlson, 2020), the species might differ in the expression of conserved sweet GR genes (e.g. GR5a, GR61a and GR64a-f) in this organ (Dahanukar et al., 2007; Freeman et al., 2014). Differences in GRN projections can also contribute to these differences: most tarsi GRNs terminate in the ventral nerve cord, while all proboscis GRNs terminate in the brain primary taste center. Furthermore, different circuits control distinct phases of the feeding behavioral program, such as PER versus ingestion (Chen and Dahanukar, 2020; Scott, 2018). For example, a subset of sweet tarsal GRNs are necessary for appetitive responses to sugar and another subset for stopping locomotion upon food encounter (Thoma et al., 2016). This functional subdivision may be relevant for specialists, as the legs are the first appendages that contact potential food sources. PER was inhibited by the synthetic bitter compound denatonium in both D. melanogaster and D. sechellia (Fig. 1D), revealing a fundamental principle for avoiding ingestion of potentially toxic substances.

Although D. sechellia is found preferentially on noni, flies have been reported in the Seychelles on mangoes and figs (Matute and Ayroles, 2014). Drosophila sechellia uses olfactory cues for host orientation and finding (Auer et al., 2020; 2021; Zhao and McBride, 2020), but it is not clear whether taste cues play a distinct role for food acceptance, although females require both taste and olfactory input for oviposition on noni (Álvarez-Ocaña et al., 2022 preprint). Noni fruit is difficult to source, and therefore we used dehydrated fruit leather strips and juice. The juice has a consistent composition and has been used in chemical ecology studies (Auer et al., 2020; Álvarez-Ocaña et al., 2022 preprint), although it has altered amounts of some compounds in comparison with the fruit (Abou Assi et al., 2017; Almeida et al., 2019; Auer et al., 2020; Motshakeri and Ghazali, 2015). Nevertheless, we found that D. sechellia has similar taste and feeding responses to high sugar and noni products, while the responses of the two generalists to noni were much reduced (Fig. 2). Furthermore, the taste and feeding aversion to noni and octanoic acid in the generalists is not due olfaction (Fig. 2B; Fig. S3).

A common theme among drosophilid and other insects with specialized diets is the loss of taste chemoreceptors for detection of bitter (likely noxious) compounds (Crava et al., 2016; Dweck and Carlson, 2020; Dweck et al., 2021; McBride, 2007; Rytz et al., 2013; for exceptions, see Briscoe et al., 2013; Wanner and Robertson, 2008). This is because specialists developed resistance to host toxins, providing an exclusive ecological niche, but also because they no longer encounter them as a result of their specialization on one or a few related hosts. Generalist insects, in contrast, use many food sources that might contain diverse potentially noxious (bitter) substances, which need to be detected and their ingestion prevented. In particular, D. sechellia lost various GR and divergent IR genes (Crava et al., 2016; McBride, 2007; McBride and Arguello, 2007; Rytz et al., 2013), including GR39.a.A, which is expressed in all bitter GRNs in D. melanogaster and is important for the detection of various bitter compounds in adults (Table 1; Dweck and Carlson, 2020; Ling et al., 2014; Weiss et al., 2011) and larvae (Choi et al., 2016, 2020; Kwon et al., 2011). Our behavioral results are in line with these predictions: D. sechellia have reduced, and even abolished, aversive feeding responses to compounds that suppress consumption in the generalist ancestors, most notably D. melanogaster (Figs 3 and 4; Fig. S4). Moreover, the responses of D. sechellia to caffeine, canavanine, coumarin and theophylline were different from those of D. simulans and D. melanogaster (Fig. 4). Drosophila sechellia retained aversive responses to lobeline, a toxic plant alkaloid which decreases feeding (Wink, 1998) and to the synthetic compound denatonium, which is unpalatable but may not be toxic. Studies from labellar bitter-sensitive sensilla in D. sechellia also found no responses to theophylline and caffeine but strong responses to lobeline, denatonium and, surprisingly, coumarin (Dweck and Carlson, 2020). Coumarin is a precursor of the signature noni compound scopoletin (Ikeda et al., 2009), and thus it is possible that this compound has different effects depending on the behavioral context. For instance, GR66a-positive GRNs mediate behaviors of opposite valence such as positional aversion and oviposition attraction, dependent on the taste organ (Joseph and Heberlein, 2012). Overall, our results are in line with the prediction that specialist insects lost aversion to bitter compounds in general because these are no longer encountered and/or important for host acceptance. Reduction of bitter sensitivity was reported in non-specialists: the generalist D. suzukii is a pest of thin-skinned fruits (e.g. berries, cherries) and has reduced sensitivity to, for example, caffeine and theophylline, allowing flies to feed on the un-ripened fruit stages which contain large amounts of bitter substances (Durkin et al., 2021; Dweck et al., 2021; Karageorgi et al., 2017).

The fatty acid compounds hexanoic and octanoic acid are main noni compounds (19% and 58%, respectively; Farine et al., 1996), and are toxic to many drosophilids including D. melanogaster and D. simulans, but not to D. sechellia (Legal et al., 1994). We found that D. sechellia has lost the ancestral taste aversion to hexanoic and octanoic acid (Figs 3 and 4; Fig. S4). In D. melanogaster, aversion to >1% v/v fatty acids involves activation of tarsal bitter GR33a-positive GRNs (Ahn et al., 2017; Chen and Dahanukar, 2020; Prieto-Godino et al., 2017), but the specific chemosensory proteins necessary for this aversion remain uncharacterized. A study in a population of D. yacuba which uses noni similarly to D. sechellia identified selection in four bitter GR genes (GR22b, GR22d, GR59a and GR93c), but neutral evolution in IR genes mediating appetite to low fatty concentrations via sweet GRNs (Ahn et al., 2017; Ferreira et al., 2020; Yassin et al., 2016). In D. sechellia, various GR genes of the GR22a clade have been pseudogenized (McBride, 2007; Table 1). Overall, these results suggest that D. sechellia’s lack of aversion to fatty acids at >1% v/v (Figs 3 and 4) may be due to evolutionary changes in certain chemosensory proteins other than GRs expressed in bitter GRNs, such as loss/pseudogenization and/or qualitative and quantitative changes in the combination of proteins expressed therein.

Another noteworthy finding is that D. sechellia males have an aversion to hexanoic acid (but not to octanoic acid), and consumed much less of these solutions than females (Fig. S4A). This suggests a female-specific role for this compound, in line with the finding that hexanoic acid is a more efficient oviposition attractant than octanoic acid (Amlou et al., 1998; Álvarez-Ocaña et al., 2022 preprint). Furthermore, the oviposition preference of D. sechellia for hexanoic acid requires both olfaction (via IR75a) and taste (Álvarez-Ocaña et al., 2022 preprint). Female D. sechellia carefully probe the substrate with their ovipositor (Álvarez-Ocaña et al., 2022 preprint), but it is unclear whether this serves to evaluate substrate chemical composition. The female terminalia of D. melanogaster has trichoid sensilla (Stocker, 1994; Taylor, 1989) and expresses various GRs, IRs and OBPs (Crava et al., 2016), but its chemosensory function has not been proven. Alternatively, the loss of feeding aversion to hexanoic acid and the oviposition preference for this compound might result from differential expression of IRs in tarsi, similar to the D. melanogaster female's IR76b/IR25a-mediated preference for acidic substrates (Chen and Amrein, 2017). Sexual dimorphism in taste chemosensory proteins mediating oviposition in host plants occurs in the specialist butterfly Heliconius melpomene (Briscoe et al., 2013), suggesting commonalities in specialists from diverse insect orders.

Although they are toxic at high concentrations, fatty acids are caloric and in D. melanogaster evoke appetitive responses (at <1% v/v) that require IR56d in sweet (GR64f-positive) GRNs (Masek and Keene, 2013; Tauber et al., 2017). Drosophila sechellia not only lost the taste-mediated aversion to solutions containing higher concentrations of octanoic acid (Figs 3 and 4; Fig. S4A,B) but also had increased appetite for this compound (Fig. 6). Drosophila melanogaster had significant aversion to 1.3% v/v octanoic acid but D. sechellia had increased feeding (Fig. 6B). The enhanced feeding of D. sechellia was not due to smell (Fig. S5), similar to the D. melanogaster persistent appetite for low concentrations of fatty acids in the absence of olfactory input (Masek and Keene, 2013). It is possible that D. sechellia’s observed appetitive for higher concentrations of fatty acids also involves IR56a/d and IR76b in sweet GRNs (Ahn et al., 2017). Future studies addressing the expression of chemosensory proteins, in particular IRs, will illuminate the cellular mechanisms mediating taste responses to these important host plant compounds.

In the last decade, genomes for many insect species have allowed predictions to be made about the genomic bases of chemosensory adaptation (for reviews on this topic, see Vertacnik and Linnen, 2017; Robertson, 2019), but functional studies to test such predictions, particularly in the case of taste, are more scarce. In D. sechellia, various hypotheses derived from genomic studies have been tested: changes in specific ORs and IRs mediate olfactory preference for noni (Álvarez-Ocaña et al., 2022 preprint; Auer et al., 2020; 2021), and the loss of two OBP genes enabled contact-mediated acceptance of fatty acid oviposition substrates (Matsuo et al., 2007). Drosophila sechellia also lost many GR and various divergent IR genes (McBride, 2007), but the functional consequences of these losses remain mostly unexplored. We found that the reported losses of GR39a.a and GR28b.a correlate with D. sechellia’s behavioral phenotype. Drosophila melanogaster null mutants for each of these genes have lost the feeding aversion to canavanine and coumarin (and have reduced caffeine aversion), similar to D. sechellia (Fig. 5). Although we used null mutants in these experiments, GR39a.a is the splice form of GR39a that mediates responses to various bitter compounds (Dweck and Carlson, 2020), and GR28b.c and GR28b.d are respectively involved in saponin detection and temperature sensing (Ni et al., 2013; Sang et al., 2019). As expected, both null mutants retained the aversion to fatty acids, as their detection may involve divergent IRs (Ahn et al., 2017; Brown et al., 2021; Masek and Keene, 2013; Sánchez-Alcañiz et al., 2018). Interestingly, the GR39a.a isoform has a large variation in copy number across various drosophilid species examined (Gardiner et al., 2008), and GR39a.a and GR28b.a are also lost in the specialist D. erecta (McBride, 2007). Altogether, these findings suggest that evolution of these GR genes, whether losses, duplications or changes in expression, are particularly important for specialization in drosophilid flies.

In D. melanogaster, various studies have used genetic tools to uncover the GRs that confer responses to bitter compounds. Early investigations showed that one or two GRs were important for responses to bitter substances (e.g. caffeine; Lee et al., 2009; Moon et al., 2006), but then a picture emerged consistent with a model in which multiple GRs act as heteromeric complexes. For instance, GR8a, GR66a and GR98b are important for detection of canavanine (Shim et al., 2015), and co-expression of GR32a, GR59c and GR66a confers sensitivity to lobeline, berberine and denatonium (Sung et al., 2017; reviewed in Chen and Dahanukar, 2020; Delventhal and Carlson, 2016). Dweck and Carlson (2020) showed that GR39a.a, which is lost in D. sechellia, is necessary for responses to coumarin and caffeine in D. melanogaster’s labellar bitter GRNs, in line with our results.

Studies aimed at discovering the functional consequences of evolution of chemosensory proteins mediating feeding are hindered by the fact that each GRN expresses multiple GRs, and that expression of the same GR in different GRNs produces different responses (Dweck and Carlson, 2020). In addition, individual bitter GRs interact in different ways, providing another strategy for the evolution of novel responses (Delventhal and Carlson, 2016). These principles of organization differ greatly from those of the olfactory system, where for the most part each cell expresses one olfactory protein plus one or more co-receptors (Vosshall and Stocker, 2007; Task et al., 2022). Consequently, genomes studies can better guide functional testing of chemosensory proteins mediating olfaction. For instance, changes in one or a few ORs and/or IRs can confer new, dramatic adaptive responses in specialist insects (e.g. Álvarez-Ocaña et al., 2022 preprint; Auer et al., 2020; Liu et al., 2020; Matsunaga et al., 2022). However, we found that the reported loss of single GR genes can explain D. sechellia’s phenotype to bitter compounds such as canavanine, theophylline and coumarin (Figs 3 and 4; Fig. S4), but it is likely that the many GR (and divergent IR) gene losses in this species confer, most likely in combination, the complete behavioral phenotype. Although this remains to be investigated, it is expected that larvae, which are immersed in noni, not only have reduced aversion to host and bitter compounds but also have much increased appetite for host fatty acids. The expression of GRs and IRs in D. sechellia larvae has not yet been characterized, but D. melanogaster larvae express 39/68 GRs (including GR39a.a and GR28b.a; Kwon et al., 2011, Choi et al., 2016, 2020) and various IRs (Sánchez-Alcaniz et al., 2018; Ni, 2021). As observed for adult D. melanogaster, their larvae have feeding aversion to, for example, caffeine, canavanine and coumarin, some of which require GR39a (Choi et al., 2016, 2020), highlighting the permissive role of this GR loss in D. sechellia. Fatty acids are likely detected by IRs in D. sechellia larvae, possibly inducing ingestion depending on their cellular compartmentalization, in line with the finding that D. melanogaster larvae have broad expression of IRs (IR25a and IR76b) required for adult fatty acid taste (Chen and Amrein, 2017; Ahn et al., 2017; Ni, 2021; Sánchez-Alcañiz et al., 2018). Future investigations in D. sechellia larvae are required to shed light on the chemosensory adaptations underlying host specialization across life stages.

Finally, in addition to changes in chemosensory proteins and their tissue-specific cellular expression, changes in circuitry, as elegantly reported in the olfactory system of D. sechellia, could also play a role in host specialization (Auer et al., 2020). Although such changes tend to be more constrained by pleiotropy (Zhao and McBride, 2020), a recent comparative study proposed that olfactory pathways are conserved, with selection acting in co-expressed copies of cognate ORs (Auer et al., 2021). Future studies, using tools such as CRISPR-Cas9, transcriptomes from various taste tissues/sensilla (e.g. Auer et al., 2020; Dweck et al., 2021), and development of genetically encoded indicators of neural activity in species other than D. melanogaster, may help address these relevant questions.

We thank members of Kristin Scott's laboratory, Julianne Peláez, Teruyuki Matsunaga, Hiromu Suzuki and two anonymous reviewers for helpful comments and suggestions; we also thank Junke Li for help with fly husbandry.