Taste coding of heavy metal ion-induced avoidance in Drosophila

Summary Increasing pollution of heavy metals poses great risks to animals globally. Their survival likely relies on an ability to detect and avoid harmful heavy metal ions (HMIs). Currently, little is known about the neural mechanisms of HMI detection. Here, we show that Drosophila and related species of Drosophilidae actively avoid toxic HMIs at micromolar concentrations. The high sensitivity to HMIs is biologically relevant. Particularly, their sensitivity to cadmium is as high as that to the most bitter substance, denatonium. Detection of HMIs in food requires Gr66a+ gustatory neurons but is independent of bitter-taste receptors. In these neurons, the ionotropic receptors IR76b, IR25a, and IR7a are required for the perception of heavy metals. Furthermore, IR47a mediates the activation of a distinct group of non-Gr66a+ gustatory neurons elicited by HMIs. Together, our findings reveal a surprising taste quality represented by noxious metal ions.


INTRODUCTION
Heavy metals are commonly defined as metal elements with a relatively high molecular weight and density. About 70% of the elements in the periodic table are heavy metals. Weathering and volcanic eruptions contribute significantly to the natural occurrence of heavy metal pollution. However, most environmental contamination and human exposure to heavy metals are the results of anthropogenic activities such as mining, industrial production, and the use of metals and related compounds. Heavy metal pollution has severely threatened humans, animals, and plants. 1 Some heavy metals, such as iron (Fe), copper (Cu), cobalt (Co), and zinc (Zn), are essential nutrients and are required in physiological processes, 2 whereas other metals, such as cadmium (Cd), mercury (Hg), and lead (Pb), are highly toxic or carcinogenic to organisms. 3 For most individuals, diet is the largest source of exposure to heavy metals. Although related mechanisms are still not well-understood, bioaccumulation of heavy metals is known to interfere with normal functions, induce cancer, and damage organs, including the heart, intestines, kidneys, reproductive system, and nervous system. [3][4][5] Toxic heavy metals are also associated with the progression of neurodegenerative diseases, including Alzheimer's disease. 6 Assessing ingredients in food via sensory systems is critical to avoid ingesting toxins or harmful substances. 7 Many species, ranging from worms 8 to mammals, 9 develop taste sensations for heavy metal ions (HMIs). Humans have a complex taste profile of HMIs. Divalent and trivalent HMIs, such as Fe 2+ , Zn 2+ , and Cu 2+ , are commonly known to elicit bitter, salty, and astringent tastes. 10,11 The metallic taste of Fe 2+ is due to a retronasal smell. 10,12 While mercury salts have a metallic taste, lead acetate has a sweet taste. 13 A recent study suggested that a G protein-coupled receptor (GPCR) in humans, taste 2 receptor member 7 (TAS2R7), mediates the bitterness of multiple HMIs in vitro. 14 In mice, calcium (Ca 2+ ) and magnesium (Mg 2+ ) activate T1R3, a GPCR expressed in fungiform taste buds. 15,16 FeSO 4 or ZnSO 4 activate taste system through the T1R3-TRPM5 pathway at low concentrations, and through a member of the transient receptor potential (TRP) family, TRPV1, at high concentrations. 17 However, the ability and mechanism of sensing other HMIs, especially those with higher molecular weight, are little investigated in mammals, as well as in other organisms.
Drosophila has multiple taste modalities and associated behavioral responses. [18][19][20] It was demonstrated that adult flies, as well as larvae, tended to stay away from food containing a high concentration of several HMIs, although responsible receptors and neurons have not been reported. 21 In recent years, ionotropic receptors (IRs), 22 were reported to also mediate taste sensation, including the detection of amino acid iScience Article and fatty acid. [23][24][25] Flies were shown to detect sodium (Na + ), Ca 2+ and several essential elements via IRs. Low Na + is attractive, whereas high Na + is repulsive. 26 Loss of Ir76b reverted attraction to repulsion at low Na + . Additionally, a rejective response to Ca 2+ is mediated by IR25a, IR62a, and IR76b. 27 Very recently, the rejective response to Zn 2+ , an essential element, was shown to be mediated by IR25a, IR76b, and IR56b expressed in ppk23 + neurons. 28 However, gustatory avoidance of different essential HMIs seems to rely on different receptors because GR66a and GR33a were required for the aversive taste response to Cu 2+ , while IR76b, IR25a, and IR56b were not. 29 Toxic HMIs possess properties very distinct from those of Na + or Ca 2+ , which are abundant, easily accessible, and essential for normal cellular functions. Indeed, trace amounts of essential elements such as Zn 2+ or Cu 2+ are indispensable for the well-being of organisms, but venomous after excessive intake. It is not clear whether a similar IR-mediated mechanism applies to toxic HMIs.
Studies on metallic sensation in mammals and flies have generally utilized HMIs at concentrations in the millimolar range. However, a contaminated area could also give rise to far larger areas with lower but still potentially harmful concentrations; thus animals with higher sensitivity should have a higher chance of detection and avoidance, resulting in adaptive advantages. To address the questions above, we investigated the behavioral, cellular, and molecular basis for heavy metal taste in Drosophila. We found that flies strongly avoid toxic HMIs at the micromolar level, which is comparable to their sensitivity to common bitter agents. This high sensitivity helps to protect flies from HMIs-contaminated food. Both Gr66a + neurons and non-Gr66a + neurons in the labellum are activated by Cd 2+ . Furthermore, IRs, instead of GRs, in these neurons are required for the proper detection of HMIs. Our results demonstrate that Drosophila is able to sense a broad spectrum of HMIs with high sensitivity.

Robust aversive response to heavy metal ions
Metals broadly exist in the natural environment. In many locations, HMIs contaminate soil, water, and food with concentrations sufficient to cause physiological and mental damage. Thus, the ability to detect toxic ions before ingestion would be an admissible survival advantage for an animal. To systematically investigate the perception of HMIs by the common fruit fly, a food-choice assay was used to quantify their preference for food with various HMIs (Figures S1A and S1B). From the periodic table of elements, 17 representative elements were selected for a basic survey ( Figure 1A). These elements represent four classes: essential elements for humans or animals, including Na, potassium (K), Ca, and Mg; essential trace elements, including lithium (Li), chromium (Cr), manganese (Mn), nickel (Ni), Fe, Co, Cu, and Zn; toxic metals, including aluminum (Al), barium (Ba), Cd, and Pb; and a lanthanide rare-earth element, erbium (Er) (Figure 1A). Among these, Cr, Mn, Fe, Co, Ni, Cu, Zn, Cd, Ba, Pb, and Er are heavy metal elements. 30 At 0.1 mM, wild-type flies showed no bias toward the ions of four essential elements and most of the trace elements. However, ions of trace elements (Co 2+ ) induced significant avoidance behavior ( Figure 1B). Interestingly, flies strongly dislike food containing Ni 2+ , Cd 2+ , and Pb 2+ ( Figure 1B). The contributions from the common anionic groups in these compounds, Cl À or NO 3 À , are negligible. These results suggest that flies can detect and discriminate HMIs in food.
Previous reports evaluated responses of flies to HMIs at mM levels. 21,26,27,31 In our results, the avoidance response to most of the trace elements increased with concentration and peaked at 2 mM (Figures S1C-S1G). In our test, Figure 1. Wild-type flies avoid medium containing heavy metals in a two-choice assay (A) Periodic table of the elements highlighting the metal elements to be tested. Gray: essential metal, brown: trace metal, red: toxic HMIs. (B) Quantification of avoidance of Canton-S flies to food containing metal ions (100 mM) in a feeding-choice assay. Feeding preferences of flies was based on the food colors in abdomens and dyes were switched for each metal ion tested to eliminate the preference for dyes. N = 8-12. (F) Cluster analysis (K-means, K = 4) grouped metal ions and bitterants according to the evoked avoidance responses in feeding-choice and positionalchoice assays. Four clusters were labeled by gray, brown, blue, and red. The control group was presented with a black spot in the origin. The charges of metal ions are indicated by the sizes of spots. Box and whisker plots in B-D: the scatter points show all data points; the box includes the 25th to 75th percentile, and the line in the box shows the median of the dataset. Statistical analyses compared stimulus and control groups. No metal ions were added to the food in the control groups. One-way ANOVA followed by Tukey's post hoc test for multiple comparisons in (B) and (D). *p < 0.05, **p < 0.01, ***p < 0.001. See also Figures  iScience Article flies exhibited no preference to Na + at 2 mM, which is beyond the known effective range. 26 The 0.5 mM Ca 2+ was capable of inducing a repulsive reaction ( Figure S1C), consistent with the previous report. 27 In contrast, in the process to determine the lower limit of detection, flies could detect Cd 2+ even at 1 mM ( Figure 1C), suggesting that Drosophila exhibits a very high sensitivity to these ions.
Flies are known to avoid food containing alkaloids, which are the nitrogenous organic substances with bitter tastes in plants, through bitter-sensitive gustatory receptor neurons (GRNs). 20 At 0.1 mM, 5 of 7 bitterants (caffeine, CAF; L-canavanine, L-CANA; quinine, QUI; papaverine, PAP; strychnine nitrate salt, STR; lobeline hydrochloride, LOB; denatonium benzoate, DEN) induced obvious repulsive responses (Figure 1D), while at a lower concentration (1 mM), only DEN, the most bitter compound to humans and flies, elicited a significant repulsive response ( Figures 1D and S1H). Notably, evaluating the strength of avoidance response to Cd 2+ and DEN at a series of concentrations revealed that at a low concentration, the aversion elicited by Cd 2+ is comparable in strength to that of the bitterest compound ( Figure 1E).
To further investigate whether flies avoid areas containing metal ions, we used a modified positional twochoice assay to quantify the distributions of wild-type flies after feeding with relevant to the media contaminated with metal ions ( Figure S2A). The positional responses of sated flies divided the metal ions into three categories with neutral, attractive, and repulsive responses ( Figures S2B-S2D), demonstrating the different behavioral valences of these metal ions. Essential trace elements are beneficial for animals at very low levels, but harmful at high concentrations. The elicited behavior characterized here generally correlates with the putative biological effects of these metal ions.
Clustering the metal ions based on their effects in both food-choice and positional-choice assays revealed that flies could distinguish different groups of metal ions and subsequently take diverse actions ( Figure 1F). Interestingly, except for Al 3+ and Ba 2+ , the non-HMIs (Li + , Na + , K + , Ca 2+ , and Mg 2+ ) grouped near the control, while the HMIs (Cr 3+ , Mn 2+ , Fe 3+ , Ni 2+ , Cu 2+ , Zn 2+ , Cd 2+ , and Pb 2+ ), as well as two strong bitterants (DEN and QUI), were scattered away from the first group. The separation of HMIs from non-HMIs largely reflects the stimulating strength and behavioral valence of HMIs perceived by flies.
The behavioral response to a broad range of metal ions, and hence the ability to sense metal ions, is rather unexpected, given how well Drosophila sensory physiology is understood. Further experiments showed that male and female flies avoid substances containing HMIs similarly ( Figure S2E). Notably, the ability to detect HMIs is conserved among all four species of Drosophilidae tested ( Figure S2F), despite several million years of evolutionary distance between them. 32,33 Taken together, these behavioral results demonstrate that flies can detect HMIs. The high sensitivity to certain toxic HMIs, such as Cd 2+ , is as good as that to the bitterest compound known. Because Cd 2+ elicits a strong response, we chose to focus on Cd 2+ for further investigation to understand the mechanisms of HMIs detection.

Bitter-sensing neurons detect heavy metal ions
As flies primarily use their gustatory system to evaluate food before ingestion, we next investigated whether HMIs elicit ''bad tastes'' that result in food avoidance. In Drosophila, taste organs are distributed on the labellum, leg tarsi, wing margins, and pharynx. 7 We used the proboscis extension reflex (PER) to learn whether the labellum and foreleg could directly sense HMIs. In this assay, added Ni 2+ , Pb 2+ or Cd 2+ inhibited PER greatly, and increasing the content of ions resulted in a lower probability of proboscis extension (Figures 2A and S3A-S3C). Direct sensing of HMIs by either the labellum or the foreleg implicates the GRNs on these taste organs in acute HMIs detection. Remarkably, flies with surgically removed tarsi of the legs exhibited an identical level of aversive response to HMIs as wild-type flies ( Figure S3D), indicating neurons in the labellum alone are sufficient to mediate strong aversion. Therefore, we focused on the labellum to identify neurons essential for mediating aversive responses.
Next, we tested the behavioral response of flies without functional GRNs. In Poxn Dm22 mutants, poly-innervated chemoreceptors are transformed into mono-innervated mechanosensory receptors during development. 34 Accordingly, the sensilla on the labellum of Poxn Dm22 mutants appear longer and pointier than that of wild-type flies ( Figure S3E). Poxn Dm22 mutants exhibited significantly diminished avoidance of HMIs (Figure S3F), which suggests that GRNs are necessary for HMIs sensation.    23,24 and acetic acid and lactic acid (Gr66a + and Gr64f + neurons). [37][38][39] We set out to identify the GRNs responsible for detecting HMIs, using a selective inhibition approach by overexpressing Kir2.1, an inward-rectifying potassium ion channel, to hyperpolarize the targeted neurons. 40 Silencing neurons of the bitter-sensing pathway, including Gr66a + , Gr32a + , and Gr33a + neurons, significantly attenuated the avoidance of Cd 2+ , Pb 2+ , and Ni 2+ , whereas inhibition of other GRNs had minor or no effects ( Figure 2B). Our results here expand their ability to a drastically different class of chemicals, HMIs, thereby revealing a new mode of sensation of these neurons.

f > K i r 2 . 1 G r 5 a > K i r 2 . 1 I r 7 6 b > K i r 2 . 1 G r 6 6 a > k i r 2 . 1 G r 3 2 a > K i r 2 . 1 G r 3 3 a > K i r 2 . 1 K i r 2 . 1 / + G r 6 4 f > K i r 2 . 1 G r 5 a > K i r 2 . 1 I r 7 6 b > K i r 2 . 1 G r 6 6 a > k i r 2 . 1 G r 3 2 a > K i r 2 . 1 G r 3 3 a > K
To investigate whether HMIs trigger activity in GRNs, we performed calcium imaging on gustatory neurons of the labellum using the genetically coded calcium indicator, GCaMP6. 41,42 First, we tested the response of Gr66a + neurons to a bitter compound (DEN) as a positive control to demonstrate the reliability of our system ( Figures S4A-S4B). Gr66a-Gal4 labels all bitter-sensing neurons (approximately 20 neurons) per lobe of the labellum. 35 NaCl at a concentration of 50 mM, 20-fold lower than the low limit for inducing a behavioral response, 26 was used as a negative control. Although our behavioral paradigm was different, NaCl concentrations 2 mM or lower also failed to elicit clear aversive behaviors ( Figure S1C) and when quantifying the activation of individual Gr66a + neurons with calcium imaging, there was no significant change in fluorescent signals when exposed to 50 mM NaCl ( Figure S4C). As shown in Figure 2C, Cd 2+ , Ni 2+ , and Pb 2+ , as well as DEN, significantly increased the activity of Gr66a + neurons. However, responses to HMIs were distinct from those to DEN, but there was no noticeable difference among the three kinds of HMIs ( Figure 2C). Based on the strength of these evoked responses, cluster analysis further divided Gr66a + neurons into three categories with about 6%, 3.7%, and 7.6% of neurons exhibiting high responses to Cd 2+ , Ni 2+ , and Pb 2+ , respectively (Figures 2C, 2D and S4D). Portions of responding neurons ( Figure 2D) and the positive skew distribution of evoked signals ( Figure 2E) indicate strong heterogeneity of these Gr66a + neurons in terms of response to HMIs.
Together, these genetic and imaging results support that, in addition to bitterants, HMIs can activate Gr66a + neurons in the labellum.

GRNs protect flies from toxic heavy metals ions
The high sensitivity toward HMIs prompted us to determine whether the low concentrations that elicit behavioral aversion would be sufficient to generate physiological effects after prolonged exposure. Chronic exposure to either alkaloids or HMIs severely shortened the lifespan of flies ( Figure 3A), demonstrating the dire biotoxicity of HMIs to flies. Compared to Ni 2+ , Pb 2+ , DEN, and QUI, the Cd 2+ -treated group showed even higher toxicity with a median survival at only 15 days. Silencing Gr66a + neurons aggravated the decrease in lifespan when flies were cultivated on Cd 2+ -containing food ( Figure 3B). These results confirm that Gr66a + neurons also function in HMIs detection and further suggest that the detection sensitivity is physiologically relevant. iScience Article We next investigated whether chronic exposure to HMIs modifies the observed aversive response. Five days of pre-exposure to 20 mM Cd 2+ , Pb 2+ , or Ni 2+ yielded an adaptive effect where flies found the lower concentration of Cd 2+ more tolerable ( Figure S4E), implying a cross-adaptation among these ions. However, food with a concentration of Cd 2+ similar to or higher than that of pre-exposed Cd 2+ is still as aversive as before, which indicates that the sensitivity to higher concentrations is maintained despite pre-exposure.
We were also interested in whether the observed shortened lifespan is simply due to a refusal to feed on the contaminated food. We measured the amount of food that remained in the digestive tract after flies were maintained on food containing HMIs for 2 days. Compared with controls, flies treated with HMIs, as well as QUI and DEN, had a decreased amount of food intake but they did not stop feeding ( Figure S4F). We speculate that when limited in an environment with water and soil widely polluted, flies are likely left with no choice but to intake food containing HMIs.

Multiple IRs involved in heavy metal ion sensation in Gr66a + neurons
In Gr66a + GRNs, bitter-taste receptors, such as GR32a, GR33a, GR66a, GR89a, and GR93a, participate in the detection of multiple bitter alkaloids. 20,35 Therefore, we thought about whether these bitter-taste receptors also participate in detecting HMIs. Notably, all GR mutants tested showed a normal ability to avoid Ni 2+ , Pb 2+ , and Cd 2+ ( Figure S5A). Therefore, gustatory detection of alkaloids and HMIs in Drosophila uses distinct signal pathways.
TRP is a family of genes encoding cell surface cation channels. 43 Mice without TRPV1 or TRPM5 exhibited abnormal perceptions of CuSO 4 , FeSO 4, and ZnSO 4 . 17 In Drosophila, TRP channels are known for their vital roles in sensory perception including vision, taste, olfaction, thermosensation, and mechanosensation. 44 However, all mutants tested exhibited a normal avoidance of Cd 2+ , Pb 2+ , and Ni 2+ ( Figure S5B). Therefore, TRP channels, even those homologous to mammalian TRPV1 (TrpA1 and pain) and TRPM5 (Trpm), are not required for HMIs sensation.
We extended the scope of our screen to include ion channels, from which we identified IR76b, an ionotropic co-receptor broadly expressed in the gustatory system ( Figure S5C). Compared with wild-type flies, Ir76b mutants showed significantly decreased avoidance of Cd 2+ ( Figure 4A). Testing over a series of concentrations of Cd 2+ revealed that the aversive response of Ir76b 1 was much weaker than that of wild-type flies across the range, with a stronger difference at the lowest concentration ( Figure S5D). This suggests that B

Gr66a>GFP; Ir76b>DsRed
Gr66a>GFP; Ir25a>DsRed Ir76b is necessary for the normal detection of HMIs, while an additional aversive response, probably independent of Ir76b, is triggered as the concentration of Cd 2+ is elevated.
Another ionotropic co-receptor broadly expressed in the gustatory system is IR25a, which is often found to function together with IR76b. 45 Ir25a and Ir76b were co-expressed in labellar neurons. 23 The reduced avoidance response of Ir25a mutants and restored avoidance response of Ir25a genomic rescue flies toward Cd 2+ indicates that Ir25a is required to detect HMIs ( Figure 4B). As a negative control, the mutants of another co-receptor expressed in the olfactory system but not in the gustatory system, Ir8a, 45 exhibited normal avoidance responses ( Figure 4B). Furthermore, IR62a, 27 a critical tuning receptor for Ca 2+ -elicited aversive responses, was irrelevant to Cd 2+ sensation ( Figure 4B). These results suggest that in addition to the difference in detection limit, gustatory sensing of Ca 2+ and Cd 2+ involves different mechanisms.
IR76b and IR25a were present in a portion of Gr66a + neurons in the labellum ( Figure 4C). Knockdown with Ir76b RNAi in either Ir76b + or Gr66a + neurons rendered a reduced avoidance of Cd 2+ ( Figure 4E). Expressing Ir76b in Ir76b + neurons, or even in Gr66a + neurons restored the avoidance response of the Ir76b 1 mutant ( Figure S5E). Additionally, reducing the expression of Ir25a in Gr66a + neurons with RNAi also diminished avoidance performance ( Figure 4F). Notably, when the mutants of Ir76b and Ir25a were tested for the avoidance response to DEN, their performance was similar to that of wild-type controls ( Figure S5F), demonstrating that Ir76b and Ir25a specifically mediate HMIs sensation. Combined with the result that the bitterants receptors are not required for HMIs detection, behavioral avoidance of HMIs and alkaloids is likely based on distinct sensory pathways in Gr66a + neurons.
IR76b and IR25a are the broadly expressed co-receptors, believed to exert specific functions by forming heteromeric complexes with sparsely expressed tuning receptors. 22,[46][47][48][49][50] To identify the tuning receptors for HMIs detection, we screened members of the IR family expressed in Gr66a + neurons and found that knockdown of Ir7a in Gr66a + neurons reduced the Cd 2+ , Pb 2+ , and Ni 2+ avoidance response ( Figures 4D, 4G, and S5H). This indicates that these IRs participate in the detection of multiple HMIs. Ir7a was reported to be required for the avoidance of acetic acid in a subset of Gr66a + neurons, 38 it appears that Ir7a is at the joint pathways mediating the detection of acetic acid and HMIs, both of which are aversive cues.
To investigate the functional role of multiple IRs in HMIs detection, we directly measured the neuronal activity of Gr66a + neurons, with a normal or reduced expression level of Ir76b or Ir7a, when stimulated with HMIs. The response profiles revealed that a subpopulation of Gr66a + neurons was activated by Cd 2+ (Figure 4H). Notably, when the Ir76b or Ir7a expression was reduced by RNAi, neuronal activation was diminished ( Figure 4H). The results demonstrate that HMIs activate Gr66a + neurons, and this process requires Ir76b and Ir7a.
Taken together, IR76b, IR25a, and IR7a in Gr66a + neurons of the labellum play a central role in acute sensation and avoidance of Cd 2+ . iScience Article IR47a-mediated heavy metal ion sensation in non-Gr66a + neurons The reminiscent avoidance of Cd 2+ in Gr66a>Kir2.1 flies ( Figure 2B) implies that while all Gr66a + neurons are silenced, additional neurons help flies detect HMIs. Because IRs are considered another repertoire of gustation in addition to GRs, we addressed whether these unidentified neurons use IRs for HMIs sensing. We screened through Ir-Gal4 labeled neurons for their essential roles in the behavioral avoidance of Cd 2+ and identified Ir47a-Gal4 labeled neurons ( Figure S6A). Ir47a + neurons were located on the labellum and legs ( Figure S6C), consistent with a previous report. 45 GRNs in the fly labellum are grouped into five classes: A neurons (sweet), B neurons (bitter), C neurons (water), D neurons (expressing in ppk23 glut neurons), and E neurons (expressing IR94e). 51 In terms of behavioral valence, A, C, and E neurons mediate attraction responses, while B and D neurons mediate rejection responses. 51 We then investigated how Ir47a + neurons fit into these classes. Co-labeling experiments revealed that Ir47a + neurons were not coincident with attractive Gr5a + , ppk28 + , and Ir94e + neurons, nor with the repulsive Gr66a + neurons ( Figures 5A and S6D). Although both are required for aversion to Cd 2+ , the neurons labeled by Ir47a-Gal4 and Gr66a-Gal4 are two distinct populations. Silencing both populations together resulted in a partial reduction, rather than complete loss, of Cd 2+ avoidance ( Figure 5B), implicating additional pathways mediating Cd 2+ sensation. Instead, Ir47a + neurons extensively co-localized with ppk23 + neurons, the repulsive class D neurons ( Figure 5C). We further characterized the Ir47a + neurons for their repertoire of IR co-receptors with co-expression experiments ( Figure 5C). The expression patterns of Ir47a + and Ir76b + did not fully overlap in Figure 5C, this is likely due to the QF reporter of Ir76b used, which did not show a pattern fully overlapped with that of the Gal4 reporter of Ir76b ( Figure S6E). To circumvent the lack of faithful non-GAL4 Ir76b driver, we looked into the labellar neurons known for expressing Ir76b, ppk23 + neurons. 28,51 As shown in Figure 5C, the extensive co-expression of Ir47a + and ppk23 + neurons strongly suggest the ubiquitous expression of Ir76b in Ir47a + neurons.
We next analyzed single-cell transcriptome data of the Drosophila gustatory system to gain an additional molecular perspective of taste coding of GRNs. 52 Number of occurrences derived from this dataset served as a qualitative indication, rather quantitative representation, of co-expression frequencies. Expression profiles of Ir76b, Ir25a, Ir47a, and ppk23 in labellar GRNs of both male and female flies further supported the co-existence of these IRs and ppk23 in the labellum (Figures S6F and S6G).
We next used an RNAi approach to investigate whether the broadly expressed ionotropic co-receptor, IR76b and IR25a, and tuning receptor IR47a are required for ppk23 + neurons to sense HMIs. As shown in Figure 5D, the knockdown of IR76b, IR25a, or IR47a in ppk23 + neurons significantly reduced the avoidance response to Cd 2+ . Similarly, in Ir47a-Gal4 labeled neurons, decreased expression levels of Ir76b, Ir25a, or Ir47a also reduced the aversive response to Cd 2+ , indicating that all three IRs are needed in Ir47a + neurons for the proper detection of Cd 2+ ( Figure 5D).
Ir47a + neurons project into almost all L-type and s-type sensilla in the labellum. 48 To survey the excitability of the Ir47a + population induced by Cd 2+ , we visualized the activity states of all Ir47a + neurons with ex vivo calcium imaging. 42 Each Ir47a + neuron projects to different sensilla. All Ir47a + neurons responded strongly to Cd 2+ ( Figures 6A-6C), demonstrating that non-Gr66a + neurons mediate aversive reactions to HMIs in addition to Gr66a + neurons. Besides L-type and s-type sensilla, we occasionally observed that Ir47a + neurons projecting to I-type sensilla were activated by Cd 2+ (Figure 6B). We chose neurons projecting to L4 and s6 sensilla for further analysis. When removing Ir76b from these neurons, the responses of L4 and s6 sensilla to Cd 2+ were severely affected ( Figure 6C). Additionally, calcium imaging results indicate that consistent with the lack of Ir47a expression in Gr66a + neurons, Ir47a + neurons were not excited by DEN ( Figure 6D, blue mark). Moreover, the lack of neuronal responses of Ir47a + neurons to Cd 2+ when Ir76b, Ir25a, or Ir47a was knocked down by RNAi suggests that both IR76b, IR25a, and IR47a are required in Ir47a + neurons to detect Cd 2+ (Figures 6D and 6E).
To evaluate the detection spectrum of Ir47a + neurons, we analyzed whether other HMIs could elicit an aversive behavior in flies when Ir47a + neurons were silenced. In addition to Cd 2+ , Pb 2+ was also avoided when Ir47a + neurons were silenced by Kir2.1 ( Figure S6H). Decreasing the expression of Ir47a in Ir47a + neurons also reduced the avoidance responses to Pb 2+ and Ni 2+ , supporting a general role of Ir47a in the sensation of HMIs ( Figure S6I).
Taken together, taste perception provides flies with the capability to avoid food contaminated by HMIs. This is accomplished through two parallel pathways in the gustatory system, both of which are mediated

DISCUSSION
We found that Drosophila exhibit a robust avoidance response to HMIs, especially Cd 2+ . The high sensitivity to Cd 2+ is comparable to that of denatonium. HMIs activate Gr66a + neurons, but the canonical iScience Article bitter-taste receptors are dispensable for this process. From several screens, we identified IR76b, IR25a, and IR7a are required for sensing HMIs in Gr66a + neurons, and a group of GR66a-independent neurons, which require IR76b, IR25a, and IR47a are also mediating HMIs-induced aversion. These findings offer a vital framework for understanding the biological detection of toxic HMIs.
The unexpected sensitivity and breadth of the tuning of a gustatory response elicited by a class of agents with similar chemical natures suggest an additional taste quality or taste category. The uniformity of underlying molecules for sensing HMIs, the IRs, further strengthens the qualitative attribute. In this respect, Drosophila is not alone. Throughout the history of the psychophysics of human taste, the metallic taste has been proposed as a basic taste, along with sweet, bitter, sour, and salty. 53 It is difficult to correlate the complex taste experience in humans with taste-evoked behavior in animals. Nevertheless, the response of the human bitter receptor TAS2R7 to multiple HMIs offers interesting parallelism to our finding in Drosophila.
It is also equally important for animals not to reject all HMIs outright because some are essential for survival. Particularly, trace elements are required for a variety of biological processes, 2 and thus they are beneficial at a low level. However, at high levels, they interfere with physiological functions. The complex characteristics and biological functions of HMIs likely contribute to the heterogeneity of behavioral responses in Drosophila. Although our results were consistent with the recent findings of feeding avoidance induced by essential trace elements at millimole concentrations, 28,29 we found that some essential trace elements at lower concentrations elicit the opposite behavioral response. As such, in the positional choices assay with 0.1 mM HMIs, flies are attracted to Cr 2+ , Fe 3+ , Zn 2+ , and Al 3+ and are not repelled by Li + , Na + , K + , Ca 2+ , Mg 2+ , and Ba 2+ . The aversion to Cd 2+ and Pb 2+ even at much lower concentrations is consistent with their toxic nature in a broad concentration range. It is not immediately clear why Ni 2+ , which is an essential trace element, elicits a strong aversive response, though this may represent a Drosophila-specific trait. It is worth noting that previous investigations conducted in worms, rodents, and humans covered only Ca, Mg, Fe, Cu, and Zn, at comparatively higher concentrations. The taste sensitivity and taste quality of Cd and other toxic HMIs in organisms beyond Drosophila remain to be determined. Furthermore, the protective role of a taste system needs to be established by comparing the detection limit and the minimum toxic level of HMIs.
The gustatory system of Drosophila has served as a major model system for investigations on taste. For each taste modality, different types of taste-responding neurons harbor a combination of cell surface receptors for chemical stimulants. 19,20 Our finding that Gr66a + neurons are also activated by HMIs to trigger avoidance behavior expands their response profiles and confirms the general role of Gr66a + neurons in warning animals of toxic agents. 54,55 However, Gr66a + neurons use distinct sensing mechanisms for HMIs and bitter agents. These findings are consistent with the emerging concept that bitter-taste neurons are heterogeneous in terms of receptor repertoires and functions. 56,57 Moreover, the perceptions of alkaloids and HMIs are likely still distinguishable because of the additional contributions from GR66a-independent neurons sensitive to HMIs. Two classes of GRNs in the labellum, B (bitter) and D (ppk23 glut neurons) neurons, mediate rejection responses. 51 Our findings here expand the response profiles of ppk23 + neurons to include toxic HMIs. Specifically, ppk23 + neurons in the labellum were involved in the sensation of high salt, Ca 2+ , and Zn 2+ using IR76b and IR25a as co-receptors, while specificity was determined by tuning iScience Article IRs. 27,28,51 Based on their neurotransmitters, ppk23 + neurons can be further divided into ppk23 chat neurons and ppk23 glut neurons. While ppk23 glut neurons are negative for GR66a, ppk23 chat neurons express GR66a. 51 Based on our co-labeling experiment, we conclude that Ir47a + neurons do not show any expression of GR66a. This suggests that these sensing neurons belong to the ppk23 glut subset, rather than the ppk23 chat subset, of class D neurons. The fact that Drosophila, and potentially other Drosophilidae, utilizes both B and D classes of GRNs in Cd 2+ sensation signifies the critical importance of active avoidance of HMIs for survival.
Both B and D classes of GRNs drive high salt avoidance as well. 51 In the B class GRNs, the molecular mechanism of high salt sensation is unclear, although IR76b is required for this function. 51 This suggests that high salt and HMI activate different taste signal pathways in this class. On the other hand, in the D class, IR 7c (IR7c) functioning with co-receptors IR76b and IR25a was found to detect high salt. 58,59 Although both high salt and HMI require co-receptors IR76b and IR25a in the D class, the distinct tuning receptors would allow flies to distinguish between these stimuli. Furthermore, within the D class, the responsive neurons for high salt and HMI are likely different with high salt activating the Ir7c + ppk23 glut neurons, whereas Ir47a + neurons respond strongly to Cd 2+ . The presence of additional HMI-sensitive ppk23 glut neurons might also lead to distinct tastes of high salt and HMIs. Even if IR47a is also involved in sensation to high salt (R500 mM), one needs to explain why such a low sensitivity to Na + is not a nonspecific response as the sensitivity to Cd 2+ is below1 mM.  29 Although Zn and Cd locate in the same group but in different periods (IV and V, repectively), Zn 2+ detection is mediated by IR25a, IR76b, and IR56b in ppk23 + neurons. 28 From the initial screen of seventeen heavy metal elements, we chose to study toxic heavy metals (primarily Cd) as Drosophila exhibits extremely high sensitivity to them. Our findings that removing IR47a from Ir47a + neurons and IR7a from Gr66a + neurons decreases behavioral aversion to Pb 2+ and Ni 2+ suggest that, similar perception mechanisms are shared between certain heavy metals, likely for these with very high atomic numbers and extremely toxic nature.
Different from GRs, IRs belong to a subfamily of ionotropic glutamate receptors, which are ligand-gated ion channels. 22 IRs have been shown to play diverse roles in olfaction, 22,60-62 taste, 23,26 thermosensation, 63,64 and hygrosensation. 65 The co-receptors, IR76b and IR25a, are broadly expressed and facilitate the sensation of both attractive cues (low salt, fatty acid, polyamines, and carbonation) and aversive cues (high salt, Ca 2+ , and Zn 2+ ). The tuning IRs are responsible for specificity and valence. Our results suggest that in coordination with IR76b and IR25a, IR7a, and IR47a act as tuning IRs to convey the aversive taste of HMIs in Gr66a + neurons and GR66a-independent neurons, respectively. Further analysis of the structural-functional relationship of these IR complexes will help reveal novel gating mechanisms leading to HMIs sensation.
Given that sensing and avoiding water and food contaminated by HMIs provides a strong survival benefit to animals, it would not be surprising that multiple organs (proboscis, legs, and pharynx), neurons (Gr66a + and Ir47a + neurons), and receptors (IR7a, IR47a, and likely more) work in parallel and at different levels to ensure that fruit flies accomplish the task with sufficient redundancy. We believe that some of the weak phenotypes shown in this and other studies 27,28 demonstrate the robustness of such HMIs detection systems against perturbation. Our Ir>Kir2.1 suppression screen hinted that in addition to Gr66a + and Ir47a + neurons, other neurons are likely involved in HMIs-induced avoidance. Nevertheless, the roles of labellar Ir7a + and Ir47a + neurons are important, as suppressing both neuronal subtypes together resulted in a $70% loss in avoidance performance. Redundancy across molecular, cellular, and organ levels is presumably organized into a fail-safe scheme for avoiding naturally occurring toxicity in the complex natural world. iScience Article could be addressed as follows. First, besides silencing the target neurons, the RNAi technique was also utilized as a primary method to disrupt the functions of these neurons by reducing the expression levels of key GRs and IRs. Conducting mutant, rescue and ectopic experiments of IR47a and IR7a could provide further evidence for the involvement of IR repertoire in the perception of HMIs. Secondly, based on their responses to Cd 2+ , Gr66a + neurons are divided into three clusters: strong responders, weak responders, and non-responders, and Ir7a, which is expressed only in a subset of Gr66a + neurons, could help to divide the Gr66a + population for further understanding the molecular nature of their diverse response. Thirdly, our imaging results suggest that all Ir47a + neurons responded to Cd 2+ . Ir47a + neurons constitute about 90% ppk23 + population in the labellum. This adds another dimension to the response profile of ppk23 + neurons, which are well-known for high salt sensation. Our results, together with the recent discovery that Ca 2+ and Zn 2+ are detected by the ppk23 + population with different IR repertoires, indicate a need for in-depth dissection of the ppk23 glut population for their specific roles and sensitivities in the perception of Na + , Ca 2+ , Zn 2+ . and Cd 2+ . Lastly, as both GRs and IRs are involved in HMI detection, the combination codes of these receptors possibly determine the specificity toward individual metal ions. Further investigation via calcium imaging of GRN with multi-RNAi knockdown could provide a broader understanding of the response profile of different ions at both the molecular (GRs/IRs repertoire) and neuronal (subsets of GRNs) levels.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

Data and code availability
All data have been deposited at Mendeley Data, and are publicly available as of the date of publication. DOI is listed in the key resources table. This paper does not generate original code. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Fly stocks and chemicals
Flies were raised on standard food at 25 C or 29 C (RNAi lines), with a humidity of 60%. Details of Drosophila strains and chemicals used are in the key resources table. Canton-S was used as wild-type strain.

METHOD DETAILS
Two-choice feeding assay Two-choice feeding assay was modified from a standard protocol. 67 Thirty 3-5-day-old female flies were first kept in a vial containing only filter paper soaked with distilled-water for 24 hr. The starved flies were temporarily anesthetized on ice for 15-30 s and transferred to a 35 mm diameter petri dish. There was a partition in the middle of the dish, and 1% agarose and 100 mM sucrose were added to both sides of the dish, with one side containing HMIs. Each side of the dish was labeled by either a blue dye (0.1 mg/ mL Brilliant Blue FCF, Sigma) or a red dye (0.2 mg/mL Sulforhodamine B, Sigma). Flies ate in the dark for 90 min, and the preferences of flies was calculated by observing the colors in their abdomens. NR, NB and NP denoted the number of fruit flies with red, blue and purple abdomens respectively. A fly that ate both blue and red dyes had a purple abdomen. To quantitatively analyze the feeding preferences, we defined the Avoidance Index (AI) as AI = (NR -NB) / (NR + NB + NP) (ions in the side of blue dye) or AI = (NB -NR) / (NR + NB + NP) (ions in the side of red dye). An AI of 1 indicates that all flies avoid food containing metal ions, while an AI of 0 indicates that there is no preference for the two kinds of food. To eliminate the preference for dyes, metal ions were tested with both dyes.

Two-choice positional assay
Similar to the two-choice feeding test, 3-5-day-old flies were pre-starved for 24 hr, and quickly transferred to a 60 mm-diameter dish for testing after brief anesthesia by freezing. Both sides of the dish contained 1% agarose and 100 mM sucrose. Different concentrations of metal ions were added to one side of the dish. We recorded the location of flies for 2 hr with 15 sec/frame and analyzed the distribution of flies with a custom MATLAB script. The distribution at the end of 2 hr was used as the indicator of positional choice. Positional AI was defined as Positional AI = (NC -NE) / (NE + NC). NE or NC indicates the number of flies on the side without or with metal ions. The walls and lid of the dish were pretreated with Sigmacote to restrict the flies to stay on the agarose surface at the bottom.

Proboscis extension response (PER)
The PER assay was modified from a standard protocol. 68 Flies that were starved for 24 hr were temporarily anesthetized on ice and fixed on a cover slide from the back with light-curable glue and recovered in a wet box for 2 hr. Before the test, flies were fed with distilled water to eliminate the influence of thirst. First, 100 mM sucrose was given to confirm that the animals were able to extend a proboscis. Flies that were not able to extend a proboscis were discarded. Next, filter paper containing sucrose solution and HMIs was given to the labellum or foreleg of flies. We touched the proboscis or foreleg with filter paper and then withdrew it quickly. Each solution was tested three times, and the labellum or foreleg was wiped with distilled water between each test. The proboscis extension probability of three tests was calculated to represent the perception of flies to the test solution.

Calcium image
Female flies were raised at 25 C for 2-3 days. A fly was immobilized by inserting an electrode into the thorax and extending it towards the labellum, 69 and then it was transferred to a fixed platform to observe the GRNs in the labellum using a confocal microscope. 42  iScience Article ions was circulated via a pump at a speed of 3.8 mL/min. Firstly, the fixed flies were recorded in Milli-Q water for 6-8 min. The stimulus of metal ions or alkaloid were presented for 3-4 min and washed away with water until the fluorescence decreased to the basal level. GCaMP and tdTomato were expressed in target neurons simultaneously. Images were acquired as timelapse 3D (XYZ) stacks using a Leica SP8 confocal microscope with a 403 water-immersion objective (NA = 0.8), then were aligned and processed in MATLAB.
For analysis, the fluorescence intensities were calculated separately for individual neurons. First, the red (mtdTomato) and green (GCaMP) channels of raw data were z-projected (maximum intensity projection). The projected image from a stack was calculated as a frame. The ROI (region of interest) region enclosing each neuron was manually defined based on tdTomato expression. Fluorescence intensities of the red and green channels in each ROI were calculated, and the corresponding normalized fluorescence intensity was derived from F = F green /F red . The mean fluorescence intensity of 15 frames before the stimulus was designated as

Immunohistochemistry and imaging
The heads of 5-7-day-old flies were removed and treated with a transparent solution containing 80% glycerol and 20% PBST (PBS with 0.5% Triton) for 1.5-2 hr and mounted with Vectashield solution (Vector Labs Inc.). The dissected brains removed labellum and legs of flies were fixed in 4% PFA (Paraformaldehyde) for 3-4 hr, then washed in PBST 3 times for 15 min each and mounted with a mounting solution. Images were acquired with a Leica SP8 with a 203 objective at a resolution of 1024 3 1024 and processed with ImageJ.

Food-intake assay
Food-intake assays were modified from the previous description. 66 Briefly, twenty 5-7-day-old female flies were tested in a vial filled with 1% agarose, sucrose (100 mM), blue food dye (Erioglaucine Disodium Salt), and various HMIs for 48 hr and frozen at -80 C to stop feeding. Flies were transferred to 1.5 mL Eppendorf tubes containing 500 mL of degrading buffer (1% PBST) and centrifuged at 13,000 rpm for 30 min. The supernatant was placed in a 96-well plate and its absorbance at 630 nm was measured to calculate feeding amounts.

Survival assay
Survival assays were referred to the previous description. 39 A total of 20 newly enclosed female flies were collected and fed in a medium containing 1% agarose, 100 mM sucrose, and 20 mM HMIs or alkaloids. The food was changed every 2 days and the number of surviving flies was counted every day.

Data statistics and analysis
Data analysis and result plots in the experiments were mainly completed in MATLAB (2018a, MathWorks) and GraphPad Prime 8 (GraphPad Software, Inc). K-means clustering algorithm was used for data partition. All error bars represent the standard error of the means (SEM). Differences between groups were analyzed by Student's t-test (two-sided) or one-way ANOVA with Tukey's post hoc test. P > 0.05 was considered nonsignificant. *P < 0.05, **P < 0.01, and ***P < 0.001 were considered significant.