In their natural habitat, Nematostella are hidden by burrowing within the sandy substrate and use an opportunistic predatory strategy to capture prey with their tentacles. In contrast, symbiotic Exaiptasia are found within open waters where they experience greater risk of predation and therefore must adopt a more defensive stance. Thus, we first asked whether differences in ecological pressure are reflected by stinging behavior. Consistent with our previous findings, we observed Nematostella stinging in response to simultaneously delivered prey extract and touch, reflecting stinging control adapted for predation (Figure 1) (Weir et al., 2020). Strikingly, Exaiptasia tentacles instead exhibited robust stinging even in the absence of prey chemicals (touch alone, Figure 1). Similar touch-evoked stinging was observed for Exaiptasia acontia, which are defensive nematocyte-enriched structures that are ejected and release toxins to repel predators (Lam et al., 2017). Considering the drastic differences in stinging behavior, we wondered if Exaiptasia’s indiscriminate stinging reflects a distinct control strategy.

To investigate whether different stinging behaviors might be suited for predation versus defense, we developed a normative theory aimed at predicting optimal stinging behavior as a function of nutritional state (see Supplementary Information for model details). We focused on stinging intensity, defined as the fraction of nematocysts discharged during a stinging event, and asked whether nutritional state would affect optimal predatory and defensive stinging. In the language of decision models, the intensity of stinging is an action, and it is chosen by the agent, or anemone. In our study, “choice” of stinging means modulation of behavior with nutritional state, rather than a cognitive process. Each choice has associated costs and benefits that depend on the environment. Because anemones sting many times over the course of their lives, an optimal behavior must account for overall costs and benefits after many events; therefore, this is a sequential decision-making problem.

We modeled the optimal stinging response to a given environment by using Markov decision processes (MDP). Each anemone was modeled as an “agent” that must hedge the intensity of its stinging response. The environment, including the identity of prey, predators, and the physiological state of the animals, defines the likelihood, costs, and benefits of successful stinging. Specifically, intense stinging responses are costly since each fired nematocyst needs to be regenerated. But they are also more likely to succeed because greater discharge of stinging barbs increases the likelihood of contact and envenomation. The cost per nematocyte was first assumed to be constant and equivalent for defensive and predatory stinging as nematocyst discharge requires regeneration in either case (Figure 2A, solid line, filled circles). We assumed that the benefits of successful predatory stinging depend on the capture and consumption of prey, which improves satiation (Figure 2B left). In contrast, stinging a predator for defense would not improve nutritional state, hence the benefits of stinging would not depend on starvation (Figure 2C left). We then used the model to predict optimal stinging that maximizes the sum of all future benefits while minimizing costs during starvation (i.e. maximizing the value function, see Materials and Methods). We then tested the prediction directly against experiments with behaving animals.

Using this approach, we found that optimal stinging strategies were completely different for predatory versus defensive behavior. Regardless of the specific environment (likelihood to succeed and specific costs and benefits), predatory stinging increased with starvation (Figure 2B, right solid lines, filled circles). To test our theory regarding predatory stinging, we carried out simulations in which agents discharged a random fraction of nematocytes between 0 and 1, regardless of starvation. Random stinging was unsustainable over numerous events and agents quickly reached maximal starvation state. Agents using optimal predatory stinging discharged more nematocysts when starved and less when satiated, leading to sustained stinging behavior and survival. This was true even if they fired the same fraction of nematocytes as the random agent (Figure 2D). In contrast, optimal stinging for defense stayed constant with starvation (Figure 2B, right solid lines, filled circles). Importantly, while the precise optimal response depended on the details of cost and reward that defined the MDP, the differences between increasing predatory stinging versus unchanging defensive stinging were consistent and largely independent of assumptions associated with each reward function (further described in Supplementary Information). These results reflect greater rewards to predatory anemones upon stinging during starvation, whereas defensive anemones sting at a similar rate regardless of nutritional status. Thus, our model predicts robust differences in predatory versus defensive stinging behavior.

We next sought to experimentally test whether pressure to predate regulates stinging in Nematostella and Exaiptasia. To do so, we fed both species of anemones copious amounts of prey (brine shrimp, Artemia nauplii) for 1-2 weeks and then deprived them of food for 5 days. Following manipulation of prey availability, Nematostella significantly increased stinging in response to starvation, while Exaiptasia stinging remained relatively constant despite complete deprivation of prey (Figure 2E, symbols with error bars). The behavior was remarkably consistent with our normative theory of optimal stinging strategies for predation versus defense (Figure 2E, filled circles). Furthermore, changes in Nematostella and Exaiptasia stinging were not due to changes in the abundance of nematocytes because tentacles from both animals were abundantly armed with nematocytes across feeding conditions (Figure 2S1A). The experimental behavior of Exaiptasia showed a slight decrease in stinging with starvation. To account for this decrease we revisited the theory and assumed that the cost per nematocyte slightly increased with starvation (Figure 2A, dashed lines and open circles). In this case, the optimal response slightly decreased for defensive stinging but increased for predatory stinging (Figure 2B right open circles and Figure 2C right, dashed lines, open circles). In fact, the fit between theory and data for both Nematostella and Exaiptasia improved when the cost increased slightly with starvation (Figure 2E, open circles). The more dramatic and less realistic increase of the cost with starvation is discussed in the Supplementary Information and may lead to a decrease in predatory stinging. Thus, we conclude that Nematostella controls stinging for opportunistic predation while Exaiptasia stinging is indiscriminate and serves a greater defensive role for this symbiotic anemone.

We next probed the physiological basis underlying these significantly different stinging behaviors. We previously found that Nematostella stinging is triggered by a specialized Ca_V channel that exhibits strong inactivation at negative voltages to prevent responses to extraneous non-prey mechanical stimuli (Weir et al., 2020). Ca²⁺ influx triggers an increase in hydrostatic pressure inside the nematocyst capsule that forces the stinging thread to evert explosively at an acceleration of up to 5.41 × 10⁶g, placing it among the fastest biological processes in existence (Lubbock and Amos, 1981; Lubbock et al., 1981; Holstein and Tardent, 1984; Weber, 1990; Gitter et al., 1994; Tardent, 1995; Nüchter et al., 2006). Similar to Nematostella, Exaiptasia stinging required extracellular Ca²⁺ and was abolished by Cd²⁺, a Ca_V channel blocker (Figure 3A). Consistent with a Ca²⁺-dependent stinging mechanism, whole-cell patch clamp recordings from nematocytes revealed the presence of voltage-gated inward currents that were blocked by Cd²⁺, suggesting that Exaiptasia nematocytes also use Ca_V channels to control stinging (Figure 3B). Indeed, Ca_V currents in Exaiptasia nematocytes exhibited similar voltage-dependent activation properties compared with Nematostella nematocytes (Figure 3C). Thus, in agreement with previous findings, we conclude that Ca²⁺ influx via Ca_V channels is broadly important for stinging.

Ca_V channels respond to positive membrane potentials by opening to conduct Ca²⁺. However, sustained positive voltage drives Ca_Vs to transition to a non-conducting state (inactivation) that prevents re-activation until channels return to a resting state induced by negative membrane potentials. In most cells, voltage-gated ion channel inactivation prevents extended responses to repetitive or prolonged stimulation. Nematostella Ca_V, is unusual because it inactivates at very negative voltages to prevent responses from resting potential, resulting in nematocytes that cannot fire from rest (Weir et al., 2020). In contrast to Nematostella nematocytes in which half of all Ca_V channels (V_i1/2) were inactivated at ∼ -93mV, Exaiptasia nematocytes exhibited two distinct inactivation phenotypes: (1) nematocytes with low-voltage threshold (low-V) inactivation similar to that of Nematostella (low-V, V_i1/2 = ∼ -85mV); and (2) a distinct population with weak, high-voltage (high-V) threshold inactivation similar to its well-characterized mammalian orthologue (high-V, V_i1/2 = ∼ -48mV) (Figure 3D). While we did not observe a correlation with abundance or distinct cellular morphology (Östman, 2000; Kass-Simon and Scappaticci, 2002; Grajales and Rodríguez, 2014), we could clearly distinguish the two populations based on these electrophysiological features. Importantly, high-V nematocyte inactivation was minimal at resting voltages (∼ -70mV), so nearly all channels would be available to amplify depolarizing signals, such as those elicited by touch. Thus, these markedly different physiological properties correlate with distinct stinging behavior: Nematostella uses unusual low-voltage Ca_V inactivation to integrate sensory cues for tightly regulated predatory stinging. In contrast, Exaiptasia employs a population of nematocytes with weak Ca_V inactivation, consistent with direct activation from resting potentials and stinging to touch alone.

What is the molecular basis of distinct nematocyte physiology? Ca_V channels are made of at least three subunits: the pore-forming α and auxiliary β and α2δ subunits. Transcriptomics revealed that nematocyte-enriched tentacles of Exaiptasia expressed cacna1a, the pore-forming subunit homologous to that of the previously characterized Nematostella nematocyte Ca_V channel (Figure 4S1). We also analyzed the Ca_V β subunit, Ca_Vβ, which is required and sufficient for the unusual inactivation properties observed in Nematostella Ca_V (Weir et al., 2020) (Figure 4S1). From Exaiptasia, we identified two isoforms of Ca_Vβ: EdCa_Vβ1 and EdCa_Vβ2. Droplet digital PCR assays of mRNA abundance showed that both isoforms are expressed throughout Exaiptasia tissues, suggesting they could both be functionally important (Figure 4A, 4S1B). To localize Ca_Vβ, we used in situ hybridization to determine that distinct nematocyte populations expressed either EdCa_Vβ1 or EdCa_Vβ2 mRNA (but not both) in the same cell (Figure 4B).

Considering this expression profile, we wondered if the two Ca_Vβ isoforms could mediate low-V and high-V inactivation phenotypes in Exaiptasia nematocytes. To investigate this question, we heterologously expressed each β subunit isoform with other well-characterized Ca_V subunits (mammalian Ca_V α and α2δ) that express well in heterologous systems. Both channels exhibited functional Ca_V currents with similar activation thresholds (Figure 4C). However, EdCa_Vβ1- and EdCa_Vβ2-containing channels significantly differed in their inactivation properties. EdCa_Vβ1 inactivated at negative voltages, similar to channels containing Nematostella Ca_Vβ (NveCa_Vβ). In contrast, EdCa_Vβ2 mediated Ca_V currents with weak inactivation, more like channels containing rat Ca_Vβ2a (Figure 4C, D). Thus, EdCa_Vβ1 and EdCa_Vβ2 confer strong, low-voltage and weak, high-voltage steady-state inactivation, respectively, and are expressed in distinct nematocytes, consistent with low-V and high-V threshold inactivating nematocyte populations. Genomic alignment revealed that alternative splicing at the N-terminus gives rise to EdCa_Vβ1 and EdCa_Vβ2 isoforms, serving as a mechanism to dynamically tune nematocyte physiology and potentially stinging behavior in contrast to adaptation through gene duplication and divergence (Figure 4E). Furthermore, by expressing two functional variants, Exaiptasia could use distinct nematocyte populations for different behaviors, including a less pronounced role for predation.

We next asked how variation in Ca_Vβ structure mediates strong phenotypes by testing whether distinct protein domains confer low or high voltage-dependent inactivation. We first compared rat rCavβ2a and Nematostella NveCa_Vβ, which have significantly different voltage-dependent properties (Weir et al., 2020). Swapping the well-characterized SH3, HOOK, and GK domains had no effect on inactivation, but the NveCa_Vβ N-terminus was both required and sufficient for low voltage-dependent inactivation (Figure 5A, B). Indeed, swapping only the N-terminus of NveCa_Vβ was sufficient to shift rat rCavβ2a-conferred inactivation by ∼ -75mV (Figure 5A, B). This finding is consistent with the variation in EdCa_Vβ splice isoforms, in which differences in the N-terminus account for a ∼ 40mV difference in inactivation thresholds.

To explore evolutionary relationships of Ca_Vβ, we constructed a phylogenetic tree of sequences from various cnidarians including Nematostella vectensis (anemone, NveCa_Vβ), Exaiptasia diaphana (anemone, EdCa_Vβ1 and EdCa_Vβ2), Cyanea capillata (jellyfish, CcCa_Vβ), Physalia physalis (hydrozoan, PpCa_Vβ), Clytia hemisphaerica (jellyfish, ChCa_Vβ), Cassiopea xamachana (jellyfish, CxCa_Vβ), and the Rat β subunit (rCavβ2a) as an outgroup (Figure 5C). Sequence comparison across all amino acid positions revealed that the N-terminus exhibited the greatest sequence diversity (Figure 5D), consistent with previous findings showing extensive alternative splicing in this region in other organisms (Helton and Horne, 2002; Helton et al., 2002; Takahashi et al., 2003; Foell et al., 2004; Vendel et al., 2006; Ebert et al., 2008; Buraei and Yang, 2010; Siller et al., 2022). We found that all cnidarian Ca_Vβs conferred voltage-gated currents when co-expressed with Ca_Vα and α2δ subunits and had relatively low voltage thresholds for inactivation compared with rCavβ2a or EdCa_Vβ2 (Figure 5E, 5S1A, 5S1B). Importantly, swapping the N-termini of each cnidarian Ca_Vβ onto EdCa_Vβ2 was sufficient to shift voltage-dependent inactivation to more negative values (Figure 5E, 5S1C). Thus, alternative splicing at the N-terminus could serve as a broad molecular mechanism for tuning Ca_V function. Collectively, these findings substantiate the importance of Ca_Vβ in modulating inactivation and suggest a mechanism that could dynamically regulate a small region of only one subunit in the Ca_V protein complex to tune complex stinging behavior.

Starlet sea anemones (Nematostella vectensis) were provided by the Marine Biological Laboratory (Woods Hole, Massachusetts). Adult animals of both sexes were used and kept on a 14 hr light/10 hr dark cycle at 26°C in 1/3 natural sea water (NSW). Exaiptasia spp. were purchased through Carolina Biological Supply Company (Cat #162865). Adult animals of both sexes were used following being kept on either a 10 hr light/14 hr dark cycle at 26°C in natural sea water (NSW) or a 14 hr light/10 hr dark cycle at 26°C in natural sea water (NSW). Cassiopea spp. were purchased through Carolina Biological Supply Company (Cat #162936). Unless stated otherwise, all animals were fed freshly hatched brine shrimp (Artemia) twice a week.

Exaiptasia diaphana were bleached through chemical methods (menthol-induced). Menthol (100mM in 100% ethanol; Sigma-Aldrich) was added to NSW at a final concentration of 0.2mM (Matthews et al., 2016). The anemones were incubated in the menthol/NSW treatment solution for a maximum of 8hr per day and outside of treatments anemones were incubated in NSW. For 2 weeks, anemones were treated 4 days per week and kept in the dark continuously starting from day 1 of treatment, aside from treatment changes. Animals were fed with Artemia approximately twice per week between bleaching treatments, enabling successful bleaching with minimal mortality. Their symbiotic status was assessed via fluorescence microscopy at the end of each week. For starvation experiments, animals were fed to excess for 1-2 weeks before the trial and withheld food entirely during the trial period and given water changes twice a week.

Nematostella nematocytes were isolated from tentacle tissue, which was harvested by anesthetizing animals in high magnesium solution containing (mM): 140 NaCl, 3.3 Glucose, 3.3 KCl, 3.3 HEPES, 40 MgCl₂. Cells were isolated from tentacles immediately prior to electrophysiology experiments by treatment with 0.05% Trypsin at 37°C for 15-20 min and mechanical dissociation in divalent free recording solution (mM): 140 NaCl, 3.3 Glucose, 3.3 KCl, 3.3 HEPES, pH 7.6. Dissociated cells were held on ice until use. Basitrichous isorhiza nematocytes were isolated from tentacles and identified by the presence of a capsule with high refractive index containing a barbed thread, oblong shape, and the presence of a cnidocil. Exaiptasia nematocytes were isolated from tentacle tissue immediately prior to electrophysiology experiments by incubation in a heat shock dissociation solution with (in mM): 430 NaCl, 10 KCl, 150 sucrose, 5 NaEGTA, 10 HEPES, 10 glucose, pH 7.6 at 45°C for 15-20 min and mechanical dissociation in the same solution. Dissociated cells were held on ice until use. Nematocytes were isolated from tentacles and identified by the presence of a capsule with high refractive index, oblong shape, and the presence of one or multiple apical cilia.

HEK293T cells (ATCC) were grown in Dulbecco’s Modified Eagle Medium (DMEM) (Gibco) supplemented with 10% fetal calf serum (Gibco) and 50 I.U./mL penicillin and 50 μg/mL streptomycin (Gibco) at 37°C, 5% CO2 using standard techniques. For transfection, HEK293 cells were washed with Opti-MEM Reduced Serum Media (Gibco) and transfected using lipofectamine 2000 (Invitrogen/Life Technologies Cat #11668019) according to the manufacturer’s protocol. 1 μg each of M. musculus (mouse) cacna1a and rat cacna2d1 and one of a wide variety of beta subunits (Nematostella vectensis cacnb2.1 (NveCa_Vβ), Rattus norvegicus (rat) cacnb2a (rCa_Vβ2a), Exaiptasia diaphana Ca_Vβs (EdCa_Vβ1, EdCa_Vβ2), Cyanea capillata Ca_Vβ (CcCa_Vβ), Physalia physalis Ca_Vβ (PpCa_Vβ), Clytia hemisphaerica Ca_Vβ (ChCa_Vβ), Cassiopea xamachana Ca_Vβ (CxCa_Vβ2)) were coexpressed with 0.5 μg eGFP. We also assayed an array of different EdCa_Vβ2 mutants with N-termini from different animals by coexpressing 0.5 μg eGFP, 1 μg of M. musculus (mouse) cacna1a and rat cacna2d1, and one of a variety of beta subunits (Nematostella vectensis cacnb2.1 mutant (NveCa_Vβ-N), R. norvegicus (rat) cacnb2a mutant (rCa_Vβ-N), Exaiptasia diaphana Ca_Vβ mutants (EdCa_Vβ1-N, EdCa_Vβ2-N), Cyanea capillata Ca_Vβ mutant (CcCa_Vβ-N), Physalia physalis Ca_Vβ mutant (PpCa_Vβ-N), Clytia hemisphaerica Ca_Vβ mutant (ChCa_Vβ-N), Cassiopea xamachana Ca_Vβ mutant (CxCa_Vβ2-N)). To enhance channel expression, cells were incubated with transfection mix containing plasmid DNA and Lipofectamine 2000 in Opti-MEM for 6 hr at 37°C. Cell were then re-plated on coverslips, incubated for 1-2 hr at 37 °C, and then incubated at 30°C for 2-6 days before experiments. Rat cacna2d1 (RRID: Addgene_26575) and cacna1a were gifts from D. Lipscombe (RRID: Addgene_26578) and cacnb2a was a gift from A. Dolphin (RRID: Addgene_107424).

RNA was prepared from tentacles, body, and acontia tissues of WT and bleached adult Exaiptasia using published methods (Stefanik et al., 2013). Each tissue was homogenized (Millipore Sigma Cat #Z359971) and RNA was extracted using TRIzol Reagent (Thermo Fisher Cat #15596026), then after skipping the salt precipitation steps, RNA was purified and concentrated with the RNA Clean & Concentrator-5 kit (Zymo Research). For ddPCR experiments, droplet generation (QX200™ Droplet Generator BioRad Cat #1864002) and transfer of droplets to ddPCR™ 96-Well Plates (Bio-Rad Cat #12001925) were performed according to manufacturer’s instructions (Instruction Manual, QX200™ Droplet Generator – Bio-Rad). Custom primers and probes and One-Step RT-ddPCR Advanced Kit for Probes (Bio-Rad Cat #1864021) reaction reagents and Droplet Generation Oil for Probes (Bio-Rad Cat #1863005) were sourced from Bio-Rad (see Key Resources Table for primer and probe sequences). The ddPCR plate was sealed with a Pierceable Foil Heat Seal (Bio-Rad Cat #1814040) and the PX1™ PCR Plate Sealer (Bio-Rad Cat #1814000). Plates were transferred to a Bio-Rad Thermalcycler C1000 (Bio-Rad Cat #1851197). The cycling protocol was the following: 45°C reverse transcription step for 60 minutes, 95°C enzyme activation step for 10 minutes followed by 40 cycles of a two-step cycling protocol (denaturation step of 95°C for 30 seconds and annealing/extension step of 58°C for 1 minute), 98°C enzyme deactivation step for 10 minutes, and holding at 12°C for an indefinite period before transfer to the QX200 Droplet Generator. The plates were read with the Bio-Rad QX200 Droplet Generator & Reader (Cat #1864003) and the RNA concentration per sample was processed using QuantaSoft (Bio-Rad Cat #1864011). Data were exported to Microsoft Excel and Prism (Graphpad) for further statistical analysis.

Most plasmids, including Nematostella vectensis cacnb2.1 (NveCa_Vβ), Exaiptasia diaphana Ca_Vβs (EdCa_Vβ1, EdCa_Vβ2), Cyanea capillata Ca_Vβ (CcCa_Vβ), Physalia physalis Ca_Vβ (PpCa_Vβ), Clytia hemisphaerica Ca_Vβ (ChCa_Vβ), Cassiopea xamachana Ca_Vβ (CxCa_Vβ2)), Nematostella vectensis cacnb2.1 mutant (NveCa_Vβ-N), R. norvegicus (rat) cacnb2a mutant (rCa_Vβ-N), Exaiptasia diaphana Ca_Vβ mutants (EdCa_Vβ1-N, EdCa_Vβ2-N), Cyanea capillata Ca_Vβ mutant (CcCa_Vβ-N), Physalia physalis Ca_Vβ mutant (PpCa_Vβ-N), Clytia hemisphaerica Ca_Vβ mutant (ChCa_Vβ-N), Cassiopea xamachana Ca_Vβ mutant (CxCa_Vβ2-N), were synthesized by Genscript (Piscataway, NJ). Sequence alignments were carried out using Clustal Omega. Wild type and Chimeric Ca_Vβ sequences are listed in Supplemental Table 1.

Cnidarian beta sequences were obtained from RNA sequencing or NCBI: Nematostella vectensis cacnb2.1 (NveCa_Vβ) sequence (Weir et al., 2020), Exaiptasia diaphana Ca_Vβs from RNA sequencing and confirmation from NCBI accession number KXJ28099.1 (EdCa_Vβ1) and NCBI accession number XP_020893045.1 (EdCa_Vβ2), Cyanea capillata Ca_Vβ (CcCa_Vβ) from NCBI accession number AAB87751.1 (Jeziorski et al., 1998), Physalia physalis Ca_Vβ (PpCa_Vβ) from NCBI accession number ABD59026 (Bouchard et al., 2006), Clytia hemisphaerica Ca_Vβ (ChCa_Vβ) from the MARIMBA database (Leclère et al., 2019), Cassiopea xamachana Ca_Vβ (CxCa_Vβ2) from RNA sequencing as TRINITY_DN5778_c3_g1_i5.p1.

Exaiptasia were anesthetized in 15% MgCl₂ NSW solution in a dish surrounded by an ice bath for 15 minutes. Tentacle, body, and acontia tissue were dissected and flash frozen in the presence of liquid nitrogen. Cassiopea xamachana were anesthetized in 10% MgCl₂ NSW solution in a dish surrounded by an ice bath for 15-20 minutes. Oral arms, bell, and cassiosome tissues were dissected and flash frozen in the presence of liquid nitrogen. All tissues were stored at -80°C until RNA extraction, library preparation, and RNA sequencing was performed by Genewiz (Azenta) using a HiSeq (2x150 bp) platform. Reads were examined for base quality distribution, kmer frequencies and adapter contamination by position in the read using fastqc (The Babraham Institute Bioinformatics Group), then where relevant, Rcorrector was used to remove erroneous k-mers (Song and Florea, 2015) and the FilterUncorrectablePEfastq python script from the Harvard Informatics group was used to discard read pairs. TrimGalore (The Babraham Institute) was then used to remove adapter contamination in reads and where relevant, Bowtie2 (Langmead and Salzberg, 2012) was used to remove reads originating from rRNA and Trinity was used to assemble reference transcriptomes de novo (Grabherr et al., 2011). Transdecoder was used to identify open reading frames (Haas, BJ) and Diamond used to annotate the transcriptome (Buchfink et al., 2015). Reads were pseudo-aligned and transcript abundance (TPM) was quantified using Kallisto (Bray et al., 2016) and our novel transcriptome assemblies as a reference and visualization and alignments were performed with Geneious Prime software and/or Clustal Omega (Madeira et al., 2022).

Recordings were carried out at room temperature using a MultiClamp 700B amplifier (Axon Instruments) and digitized using a Digidata 1550B (Axon Instruments) interface and pClamp software (Axon Instruments). Whole-cell recording data were filtered at 1kHz and sampled at 10kHz. Ca_V activation data were leak-subtracted online using a p/4 protocol, and all membrane potentials were corrected for liquid junction potentials.

For whole-cell nematocyte recordings, borosilicate glass pipettes were polished to 8-10MΩ for Nematostella and 4-6MΩ for Exaiptasia, respectively. The standard Nematostella medium was used as the extracellular solution and contained (in mM): 140 NaCl, 3.3 glucose, 3.3 KCl, 3.3 HEPES, 2 CaCl₂, 0.5 MgCl₂, pH 7.6, 260-280mOsm. The standard Exaiptasia medium was used as extracellular solution and contained (in mM): 430 NaCl, 10 KCl, 10 CaCl₂, 50 MgCl₂, 10 HEPES, pH 7.6, 800-900mOsm. The intracellular solution for both Nematostella and Exaiptasia contained (in mM): isolating inward currents (in mM): 500 cesium methanesulfonate, 4 MgCl₂, 10 CsEGTA, 10 HEPES, 30 sucrose, pH 7.6, 260-280mOsm for Nematostella and 800-900mOsm for Exaiptasia. For Nematostella nematocyte recordings, voltage-dependent inactivation was measured during a 200ms activating pulse of -20mV following a series of 1s pre-pulses ranging from -110mV to 30mV, holding at -110mV. Voltage-gated currents were measured through a series of 200ms voltage pulses in 10mV increments from -110mV to 70mV, holding at -110mV. For Exaiptasia nematocyte recordings, voltage-dependent inactivation was measured during a 200ms activating pulse of 0mV following a series of 1s pre-pulses ranging from -110mV to 30mV, holding at - 110mV. For both Nematostella and Exaiptasia, voltage-gated currents were measured through a series of 200ms steps 200ms voltage pulses in 10mV increments from -110mV to 70mV, holding at -110mV. For Cd²⁺ experiments, 500μM Cd²⁺ (dissolved in water) was applied locally and voltage-dependent activation was assessed through a single 200ms step to 0mV from a holding potential of -110mV.

For whole-cell recordings in HEK293 cells, pipettes were 3-6MΩ. The standard extracellular solution contained (in mM): 140 NaCl, 5 KCl, 10 HEPES, 2 CaCl₂, 2 MgCl₂, 10 Glucose, pH 7.4, 300-310mOsm. The intracellular solution contained (in mM): 5 NaCl, 140 cesium methanesulfonate, 1 MgCl₂, 10 EGTA, 10 HEPES, 10 sucrose, pH 7.2, 300-310mOsm. For Ca²⁺ currents in heterologously expressed channels, voltage-dependent inactivation was measured in one of two ways: (1) during an activating pulse of 0mV following a series of 1s pre-pulses ranging from -110mV to 50mV and holding potential of -80mV; or (2) during an activating pulse of 0mV following a series of 1s pre-pulses ranging from -110mV to 80mV and holding potential of -90mV. Voltage-gated Ca²⁺ currents were measured in response to 200ms voltage pulses in 10mV increments from -130mV to 80mV with -110mV holding potential. Voltage-dependent inactivation was quantified as I/I_max, with I_max occurring at the voltage pulse following a -110mV prepulse. In some instances, inactivation curves could not be fitted with a Boltzmann equation and were instead fitted with an exponential. G-V relationships were derived from I-V curves by calculating G: G=I_CaV/(V_m-E_rev) and fit with a Boltzmann equation. Data was processed and analyzed in Clampfit (pClamp 11 Software Suite, Molecular Devices) and Microsoft Excel and Prism (GraphPad).

Adult Exaiptasia were paralyzed in anesthetic solution (15% MgCl₂), rinsed in PBS, then embedded in Tissue-Tek O.C.T. Compound (Sakura Cat #4583) in cryomolds (Sakura Tissue-Tek® Cryomold®, Intermediate, Cat #4566) and flash frozen on dry ice and stored at -80°C. Cryostat sections (18-20μm) were adhered to Fisherbrand™ Superfrost™ Plus Microscope Slides (Fisher Scientific Cat #12-550-15) and flash frozen on dry ice and stored at -80°C until used for BaseScope. The BaseScope Duplex Detection Reagent Kit (Advanced Cell Diagnostics Cat #323800) and the manufacturer’s manual (BaseScope Duplex Detection Reagent User Manual, ACDBio) was followed to hybridize custom probes or positive control (Cat #700101) or negative control probes (Cat #700141) to targets in tissue cryosections and amplify signals. Samples were imaged on an Olympus BX41 Phase Contrast & Darkfield Microscope (Olympus Cat #BX41-PH-B) and images were acquired using the Olympus CellSens software and Olympus DP25 5MP Color Firewire Camera.

Discharge of nematocysts was assessed based on well-established assays (Gitter et al., 1994; Watson and Hessinger, 1994; Weir et al., 2020). For assaying discharge, 5 mm round coverslips were coated with a solution of 25% gelatin (w/v) dissolved in NSW (for Exaiptasia) or 1/3 NSW (for Nematostella) and allowed to cure 3-4 hr prior to use. Coverslips were presented to the animal’s tentacles for 5 seconds and then immediately imaged at 20X magnification using a transmitted light source. To assay behavioral responses to prey-derived chemicals, freshly hatched brine shrimp were flash frozen and ground to a powder with a mortar and pestle (Fisherbrand), then filtered through a 0.22μm syringe filter (VWR Cat #28145-501) and osmolarity adjusted for the specific anemone species. Coverslips were submerged in prey extract for 10 seconds then immediately presented to the animal. Nematocytes visualized on coverslips were only those that embedded in the gelatin after discharge. For experiments using pharmacological agents such as CdCl₂, coverslips were submerged in a solution of 1M (Exaiptasia) or 10mM (Nematostella) Cd²⁺ in milli-Q water for 10 seconds then immediately presented to the animal. After performing the experiments, the animals were given several water changes to remove Cd²⁺. Experiments carried out in the absence of extracellular Ca²⁺ were nominally Ca²⁺ free and did not include use of extracellular chelators. The region of the highest density of discharged nematocytes on the coverslip was imaged at 20X. Images were acquired with MetaMorph Microscopy Automation and Image Analysis Software (Molecular Devices) and the number of discharged nematocysts was counted by eye. Images were processed in Fiji (ImageJ) (Schindelin et al., 2012). Exaiptasia and Nematostella tentacles were examined by cutting a small portion of exposed tentacles and sandwiched between glass coverslips and then imaged at 20X with the MetaMorph software.

To infer exon boundaries and isoforms, we aligned EdCa_Vβ1 (NCBI accession number LJWW01000015.1) and EdCa_Vβ2 (NCBI accession number XM_021037386.2) to the Exaiptasia diaphana reference genome (BioProject PRJNA261862) (Baumgarten et al., 2015) using GMAP version 2015-07-23 (Wu and Watanabe, 2005). For phylogenetic analyses, we aligned nucleotide sequences with MAFFT v.7 (Katoh and Standley, 2013). We used ModelFinder (Kalyaanamoorthy et al., 2017) to assess the best model of substitution for phylogenetic inference. We estimated a maximum likelihood gene tree in IQ-TREE v2.0 (Minh et al., 2020). Support for clades was calculated using ultrafast bootstrap approximation UFBoot2 (Hoang et al., 2018). Percentage of identity for amino acids was calculated in overlapping windows.

To model Nematostella, we assume the agent stings for predation. We thus introduce the state of starvation, s, that ranges from 0 to 1; at s = 0 the agent is least starved and at s = 1 the agent is most starved. At each time step, the agent decides to perform an action (sting), a, representing the intensity of the attack; a ranges from 0 to 1, and is experimentally compared to the fraction of nematocytes fired in the behavioral assay. Each action has a cost that is proportional to the fraction of nematocysts that are fired,c(a) = c₀a where c₀ is the cost of discharging all nematocysts at once, or cost of full discharge, and we first consider c₀ constant. Each stinging event has a probability p(a) of achieving successful predation, where p(a) increases with a (more intense attacks are more costly and more likely to succeed). A successful attack leads to the transition to the next state s′ where the agent is more satiated s → s′ = s − 1 whereas a failed attack leads to higher starvation state s → s′ = s + 1. The most starved state is absorbing, which is equivalent to a point of no return. A reward r(s′) is assigned to the state of starvation reached upon attack, indicating its desirability (r(s′) is a decreasing function of s′), and the most starved state entails a starvation penalty r(1) < 0. Without loss of generality, all costs and rewards are normalized to the penalty of starvation, hence penalty of starvation is r(1) = −1. Our goal is to choose actions that maximize the expected sum of all future net rewards (reward - cost) for each state, which is called the value function. As customary in infinite horizon problems, we ensure convergence of the value function by introducing an effective horizon, i.e. by discounting exponentially rewards that are further in the future with a discount rate γ < 1.

The optimal value of a state, V^*(s) and the corresponding optimal action a^*(s) are obtained by solving the Bellman Optimality equation (Bellman, 2003). With the boundary condition V^*(1) = 0. We solve these equations numerically with the value iteration algorithm (Sutton and Barto, 2018) and analytically under the assumption that a^*(s) varies slowly with s (details of the asymptotic method can be found in Supplementary Information). The asymptotic solution reproduces well the numerical results (compare lines and full circles in Figure 2B right, where we used c₀ = 1 and p = p_M(2 − a²) and p_M = 0.8 and showcased two different functional forms for r(s), r(s) = 10 tan^-1 (1 − s); . We also explored how well the asymptotic result can capture the trend of the numerical result by varying the parameters in these three different forms for the reward (SI). The approximate solution is used to demonstrate that the stinging response increases with starvation under broad conditions, i.e. as long as r(s) and p(a) are concave functions and c(a) is convex (see Supplementary Information). To exemplify the importance of acting optimally to save resources, we considered two agents, one acting optimally and one acting randomly i.e. shooting with a number of nematocysts uniformly distributed between a_min and a_max.. Both agents start at the same starvation state (s = 0.9 in Figure 2D) and use on average the same number of nematocysts, but the random agent reaches starvation typically in tens of steps, whereas the optimal agent converges to a steady state (around s = 0.3 in the figure) and hardly ever reaches severe starvation.

To model Exaiptasia, we assume the agent stings for defense, thus the associated Markov process models transitions between the states of safety, which we indicate with L, and danger, which we indicate with D. The state of starvation is not affected by stinging and instead is dictated by a separate process that relies on symbionts and which we do not model. Similar to the previous model, the agent chooses an action, a, representing the intensity of the attack. Each attack has a likelihood to succeed p(a) and an associated cost c(a) = c₀a where c₀ is the cost of full discharge of all nematocysts at once. A successful attack allows the animal to remain in state L and receive a unit reward; a failed attack leads to state D and penalty −1.F is an absorbing state hence V^*(F) = 0. The optimal value and action in state L follow: Zeroing the derivative of the argument on the r.h.s. of equation (2) leads to Here, p(a) is the probability of success of action a, hence it goes from p(0) = 0 to p(1) = p_M < 1. Assuming a specific form p = p_M a(2 − a) leads to the specific solution where and . If c₀ is constant, there is a non-trivial optimal action as long as c₀ < 2p_M(2 − γ) and clearly the optimal action does not depend on starvation (constant solution in Figure 2E, right, with c₀ = 1;γ = 0.99 and p_M = 0.8).

Stinging predators does not improve nutritional state, thus transitions among different starvation states are not modelled for defensive stinging, but instead rely heavily on symbionts. However, to capture subtle effects of starvation on defensive stinging, we note that the cost of discharging nematocysts may still depend parametrically on whatever state of starvation the agent happens to be in. In this case, c₀ = c₀(s) and under the assumption that the cost increases with starvation, we find that optimal defensive stinging always decrease with starvation, for any functional form of p (see Supplementary Information for more details). We applied the same cost to predatory stinging and found that it still increases for moderate increase of c₀(s) with s. The result is exemplified in Figure 2C using a functional form for c₀(s) that is obtained by fitting the experimental data as described in the Supplementary Information. Hence for defensive stinging, the optimal strategy is either independent of starvation or may decrease with starvation.

We then considered more intense increase of c₀(s) with s for both models. In contrast to defensive stinging, we find that predatory stinging increases with starvation for mild increase c₀(s) and decreases with starvation if the cost increase is too steep (see Supplementary Information discussion and Supplementary Figure 3 for a comparison with 4 different cost functions).

Data were analyzed with Clampfit (Axon Instruments), Prism (GraphPad), or QuantaSoft (BioRad Laboratories) and are represented as mean ± sem. n represents independent experiments for the number of cells/patches or behavioral trials. Data were considered significant if p < 0.05 using paired or unpaired two-tailed Student’s t-tests or one-or two-way ANOVAs. All significance tests were justified considering the experimental design and we assumed normal distribution and variance, as is common for similar experiments. Sample sizes were chosen based on the number of independent experiments required for statistical significance and technical feasibility.