The exploitation of sexual signals by predators: a meta-analysis

Abstract

Sexual signals are often central to reproduction, and their expression is thought to strike a balance between advertising to mates and avoiding detection by predatory eavesdroppers. Tests of the predicted predation costs have produced mixed results, however. Here we synthesized 187 effects from 78 experimental studies in a meta-analytic test of two questions; namely, whether predators, parasites and parasitoids express preferences for the sexual signals of prey, and whether sexual signals increase realized predation risk in the wild. We found that predators and parasitoids express strong and consistent preferences for signals in forced-choice contexts. We found a similarly strong overall increase in predation on sexual signallers in the wild, though here it was modality specific. Olfactory and acoustic signals increased the incidence of eavesdropping relative to visual signals, which experienced no greater risk than controls on average. Variation in outcome measures was universally high, suggesting that contexts in which sexual signalling may incur no cost, or even reduce the incidence of predation, are common. Our results reveal unexpected complexity in a central viability cost to sexual signalling, while also speaking to applied problems in invasion biology and pest management where signal exploitation holds promise for bio-inspired solutions.

1. Introduction

Sexual signals rank among the most elaborate and conspicuous innovations showcased by animals. Their ubiquity speaks to their importance in sexual reproduction where they advertise, among other things, the location [1], identity [2], availability [3] or quality [4] of prospective mates. These benefits are met by countervailing costs, however, with predation and parasitism standing among the most exhaustively studied [5–7]. Early work noted that the conspicuousness of sexual signals should also attract unwanted attention from predators, thereby establishing a fundamental trade-off [8,9]. This simple observation has since been borne out in a wealth of empirical tests [10–12], while more recent efforts have built on these foundations to highlight predation as a selective force in sexual systems more generally. In ecological terms, predation-induced shifts in signalling behaviour [13,14] and mate choice [15,16] are now well documented, while at evolutionary scales the need to balance between signalling to conspecifics and avoiding predators can culminate in genetic polymorphisms [17,18] and population divergence [19,20].

The interception of signals by unintended receivers is best known as ‘eavesdropping’ and is the route through which predation costs are imposed upon sexual signallers. Birds localize lizard prey using their colourful ornaments [21,22], flies attend the advertisement calls of frogs to extract a blood meal [23,24], and wasps use the sex pheromones of aphids as kairomones to identify hosts [25,26]. While the risk to sexual signallers is well articulated in general terms, questions remain as to whether and how such costs vary predictably across contexts. A signal's modality, in particular, may drive differential predation, with the expectation of modality-specific costs guided by knowledge of the structure of signals and signalling environments [27,28]. A general prediction from signal detection theory is that modalities characterized by shorter ranges, faster transmission and/or reduced longevity—as typifies visual and vibratory signals—should comparatively reduce the risk to signallers, while the increased range and longevity of olfactory and auditory signals may lead to a heightened risk of predation, on balance (table 1). Coevolution between signallers, receivers and eavesdroppers will of course modify the balance of risk and reward in a given system (e.g. [27,30,31]), but the inherent properties of modalities set the foundation for, and ultimately constrain, such processes. Recent work on multimodal signalling has broached this question indirectly [32–34], but most studies (understandably) still take a narrower taxonomic and modality-specific focus (as noted in [7]).

Table 1.

The fundamental properties of signals across five modalities which, in part, define the predicted magnitude of eavesdropping risk to signalling individuals. Adapted from [29].

property	visual	olfactory	acoustic	tactile
range	short/medium	medium/long	long	short
longevity	short	medium/long	short	short
transmission speed	fast	slow	fast	fast
specificity	medium	low	medium	high
complexity	low	high	high	medium
energetic cost	medium	low	high	low
localizability	high	low	medium/high	low
predicted risk	low	medium/high	medium/high	low

While central to theory in sexual selection and communication, signal exploitation holds interest across fields which have often progressed in isolation. For example, predicting the likelihood of invasiveness by introduced species is a longstanding conservation goal [35]. The responses of native predators to invasive prey, or vice-versa, is key to such ends, for which knowledge of the use of prey cues—including sexual signals—is vital [36,37]. In more applied terms, the promotion of natural predators is a central tenet of integrated pest management [38] for which the sexual signals of prey have proven a source of commercially viable bioinspiration in trap and lure design. Much effort has been expended on extracting and synthesizing pest sex pheromones for use in attracting predators amid managed crops (reviewed in [39]), though with mixed success (e.g. [40–43]). Underlying these disparate programmes is a shared interest in the predation costs to signalling, but the fragmentation of knowledge has constrained opportunities for fruitful, reciprocal exchange.

Here we conducted the first quantitative synthesis of whether and to what extent sexual signals increase the risk of eavesdropping by predators and parasites. Our aim was to answer the following three related questions. (1) Do predators and parasites express preferences for organisms bearing sexual signals? (2) Do sexual signals increase the risk of predation and parasitism in the wild? (3) Are there biological or methodological moderators—such as signalling modality—which influence the magnitude of predator preferences or risks to signallers?

2. Methods

(a) . Systematic literature search

As our interest lay in identifying causal relationships, our broad aim was to retrieve experimental studies which manipulate the sexual signals of organisms, or models thereof, and quantify the outcome in terms of predation or parasitism. A preliminary search showed that the literature is dominated by two experimental paradigms which seek to answer closely related, but distinct, questions. One uses choice assays in which predators or parasites are presented with a forced binary decision between putative prey whose sexual signals have been manipulated by kind (i.e. entirely removed) or degree (e.g. [44–46]). The focal question such designs address is whether and to what extent eavesdroppers prefer organisms bearing sexual signals, and the resulting data are proportions from dichotomous choices. The other common approach manipulates the sexual signals of animals, or their signals in isolation, and exposes them to predation under natural or semi-natural conditions. Examples include the field deployment of clay or robotic models bearing colourful sexual ornaments [21,47], or sampling traps impregnated with isolated sex pheromones [48,49]. The question here being whether sexual signals increase the realized risk of eavesdropping, with the resulting data being quantitative measures of between-group differences in predation and/or parasitism. Given the conceptual and analytical differences between these approaches it is a distinction which we maintain throughout, and we refer to each as ‘eavesdropper preference’ and ‘eavesdropping risk’ assays, respectively, for convenience going forward. Note too that our focus is on sexual signals specifically, and so we excluded studies of other conspicuous signals like aposematic (warning) signals. See electronic supplementary material for further details on our systematic search and study screening.

(b) . Data extraction and effect size calculation

For eavesdropper preference assays we used the logit-transformed proportion, or log-odds [50], as the effect size describing the preference of eavesdroppers for organisms bearing sexual signals in binary choice assays. We back-converted all effects to raw proportions for reporting and display (below), and so values beyond 0.5 represent a greater number of choices for signalling individuals, and values less than 0.5 represent more choices for individuals with diminished or absent sexual signals (controls). For eavesdropping risk assays we estimated the standardized mean difference Hedges's g, and its variance [50], between treatment and control groups, with values above 0 therefore representing heightened predation on signalling individuals, and values below zero representing more attacks on control stimuli with reduced or no sexual signals [50]. In all cases these effects were estimated from raw or summary data presented in the main text or figures via the R package ‘metaDigitise’ v. 1.0.1 [51], or converted from available test statistics using the package ‘compute.es’ v. 0.2-5 [52].

We also recorded information from each study which we a priori hypothesized may moderate the strength of relationships between sexual signal expression and eavesdropper preferences or eavesdropping risk. This included signal modality, since differences in the active range, specificity, duration, localizability, exploitability and transmissibility may drive modality-specific costs to signalling (table 1). A general prediction being that modalities with reduced ranges, shorter durations, and/or brief transmission speed—as typifies visual and tactile signals—should comparatively reduce the threat to signallers. As a corollary, the increased range, duration, and transmissibility of olfactory and auditory signals should lead to heightened risks to signallers on balance. We note, however, that such functional differences will be reduced and/or eliminated in eavesdropper preference assays owing to the close quarters, forced-choice experimental designs, among myriad other differences imposed by the artificiality of laboratory-based settings. And so these predictions are unlikely to hold.

We also classified the focal eavesdropping receiver in each study as either predators, parasites or parasitoids. Here we predicted stronger effects among parasitoids owing to typically higher rates of prey (and hence, sexual signal) specialization as compared to predators and parasites, which tend toward generalism. However, we recognize a suite of factors which broadly covary with these designations such as taxonomy (with parasitoid hosts largely limited to invertebrates [53]), feeding guild, and trophic level, which will serve to temper the strength of our prediction.

Finally we also classified the experimental manipulation of signals as either discrete or continuous. Discrete manipulations are those in which the signal was entirely absent in the control group (e.g. a non-broadcasting speaker, or an unscented stimulus), while continuous manipulations are those in which signal variation was graded between treatment and control groups (e.g. differential volume of mating calls or brightness of visual signals). This is an attempt to control for experimental differences in the magnitude of manipulations across diverse modalities and contexts, and also stands as a test of the consequences of ecologically salient differences in signal variability. Discrete manipulations approximate gross differences between signalling and non-signalling individuals, as is common between sexes or life-history stages, while continuous manipulations represent situations of between-signaller variation, as is typical among individuals competing for mates. Here we predicted that predation risk should be heightened among discretely manipulated stimuli owing to the increased conspicuousness and salience of signallers relative to controls.

(c) . Statistical analyses

We constructed both multilevel meta-analytic and multilevel meta-regression models using the metafor package v. 3.0-2 [54] for R v. 4.0.1 [55]. To estimate an overall mean effect we ran intercept-only multilevel random-effects models, with study- and observation-level IDs included as random factors to control for the inclusion of multiple effects per study and to estimate residual variances, respectively. We also controlled for non-independence arising from the inclusion of effects estimated from shared control groups by fitting the off-diagonal covariances in the sampling covariance matrix assuming a correlation of r = 0.5 [56]. Signaller taxon and study were broadly confounded in our dataset, which argues for the inclusion of either a study- level or phylogenetic random effect to account for the possibility of phylogenetic non-independence, but not both. We favour the former for simplicity as several studies were conducted above the level of species, and because between-study variances are more commonly modelled in meta-analyses [57].

To examine the effects of the three moderators described above—modality, eavesdropper, and manipulation—we constructed separate univariate multilevel random-effects models with the random effects structure as above. We examined the Q_m statistic, an omnibus test of model coefficients, to determine whether moderators significantly influenced the mean effect size, and estimated the amount of variance explained by the fixed factors in each meta-regression model via the marginal R² [58]. Where an omnibus test was significant we performed pairwise post-hoc Tukey contrasts for each moderator level with Holm's method to control the family-wise error rate [59], using the R package multcomp v. 1.4-17 [60]. To estimate the heterogeneity of effect sizes we use both I², which we partitioned at the levels of study and observation ID [57], and we calculated 95% prediction intervals for meta-analytic means and all moderator levels [61]. In all models, we considered effect size estimates whose 95% confidence intervals did not overlap zero to be statistically significant.

(d) . Publication bias

We explored the possibility of small-study effects, including publication bias, via the visual inspection of funnel plots supported by Egger's regression [62] and trim-and-fill tests using the R₀ estimator [63]. In both cases we fitted study precision (the inverse of sampling variance) against meta-analytic residuals derived from our null models with the full random effects structure described above, to account for the multilevel structure of our data [57].

3. Results

(a) . Eavesdropper preference

We obtained 72 effects from 29 studies examining eavesdropper preferences for sexual signalling individuals (or the sexual signals of individuals) in forced-choice assays [26,30, 43,45,46,64–85]. Some 23 species were represented among signallers across five classes, with studies involving insects (k = 49) predominating over ray-finned fishes (k = 12), amphibians (k = 6), arachnids (k = 4), and mammals (k = 1). Predators (k = 34) and parasitoids (k = 38) were approximately equally represented among eavesdroppers, with fish (k = 10) and mammals (k = 9) the most common predators and wasps (k = 34) the most common parasitoid. We found no studies involving parasites. Studies of olfactory signals predominated (k = 31), with vibratory, auditory, and visual signals near equally common (k = 13, 13, 15, respectively).

Overall, we identified a moderate to strong preference of eavesdroppers for individuals bearing sexual signals (mean proportion = 0.712, 95% CI = 0.657–0.762, k = 72; figure 1). This held irrespective of the signal's modality (figure 1; see electronic supplementary material, tables S1 and S2 for full numerical results henceforth), as predicted, and neither the signal receiver nor the nature of the experimental manipulation modulated this result (figure 2). That is, we found a clear preference for signalling individuals (or their signals alone) across all MLMR models (figures 1 and 2). There was strong heterogeneity both overall (I² = 0.645, 95% CI = 0.562–0.717) and among subgroups (table 1), and relatively wide prediction intervals affirmed the existence of considerable variation in outcome measures (figures 1 and 2).

Figure 1.

Forest plots depicting the meta-analytic mean effect of sexual signalling on the preference of predators eavesdroppers (a) and the risk of predation eavesdropping (b), as well as the mean effects within each signalling modality as estimated via moderator analyses. For preference assays the displayed effects are proportions back-transformed from logits, while Hedges's g was used to summarize the results of predation risk assays. Points are scaled by the precision of each estimate, and solid lines denote 95% confidence intervals while broken lines indicate 95% prediction intervals. (Online version in colour.)

Figure 2.

Forest plots denoting moderators of effect size estimates from studies of predator eavesdropper preferences for, or predation eavesdropping risk to, individuals bearing sexual signals. The effects are proportions back-transformed from logits for preference assays, and Hedges's g for predation risk assays. ‘Eavesdropper’ describes the guild of the unintended receiver(s), while ‘manipulation’ describes the magnitude of the difference in sexual signal expression between treatment and control group. Discrete manipulation represent cases of presence/absence between group, while continuous manipulations are those which introduce graded variation between treatments and controls. Points are scaled by the precision of each estimate, and solid lines denote 95% confidence intervals while broken lines indicate 95% prediction intervals. (Online version in colour.)

In terms of publication bias, inspection of the funnel plot revealed a weak asymmetry in the distribution of effects, with an under-representation of lower-powered, negative outcomes (electronic supplementary material, figure S2). This was affirmed by a significant intercept in our Egger's test (est = −1.215, z = 6.012, p < 0.001). A trim-and-fill analysis suggested the possible absence of five effects, whose addition slightly reduced the estimated meta-analytic mean (mean proportion = 0.642, 95% CI = 0.597–0.702, k = 72). The relative subtlety of both the asymmetry and adjusted estimate suggest that any underlying publication bias is exerting only a weak effect, however.

(b) . Eavesdropping risk

We collated 115 effects from 49 studies which quantified the eavesdropping risk to sexual signalling individuals [10,12,21,23–25,40–43,47–49,86–120]. Approximately 58 species were used as signallers, with insects (k = 55), reptiles (k = 21), and amphibians (k = 20) the best represented groups. A minority of studies drew on ‘generic’ representations of signals or signallers above the species level, and most manipulated model rather than live animals, thereby negating any effects of signaller behaviour (see discussion). The responses of predators (k = 66) were more often the focus than parasitoids (k = 40), and those involving parasites (k = 9) were uncommon. The available effects were unequally distributed across signalling modalities. Olfactory signals were again the most common (k = 57), followed by visual (k = 30) and auditory (k = 28), and we found no suitable effects from studies of vibratory signalling in the wild.

We found a strong positive mean effect overall, suggesting heightened eavesdropping risk for individuals bearing sexual signals (mean g = 0.958, 95% CI = 0.652–1.264, k = 115; figure 1). Here, unlike among eavesdropper preference assays, the effects varied by modality (figure 1). Consistent with our predictions, both olfactory and auditory signals increased eavesdropping risk to a greater extent than visual signals, and were strong and positive on-average (electronic supplementary material, table S2). We found no evidence for an effect of visual sexual signalling on the risk of predation, however, with the confidence interval from our MLMR including zero (figure 1). Indeed, the prediction interval for the visual modality was nearly symmetrical about the value of zero, suggesting that conspicuous visual sexual signals are almost as likely to decrease as increase the risk of eavesdropping by predators in the wild. Effects were strong and positive across eavesdropper types, though our prediction of significantly heightened risk from parasitoids, as compared to predators and parasites, was not supported (figure 2). The direction of the difference in mean effects between eavesdroppers was in the predicted direction, though the small sample of effects from parasites (k = 9) is limiting. The estimated mean effect was also positive across both types of experimental manipulation, though only weakly so among studies that induced continuous, graded variation in signal expression (figure 2). Counter to predictions, we found no difference in average effects between discrete and graded signal manipulations. Heterogeneity was high across all measures (table 2; I² = 0.865, 95% CI = 0.835–0.889), and prediction intervals wide and reaching beyond zero in all MLMR models (electronic supplementary material, table S2), revealing strong variability in outcomes.

Table 2.

The results of meta-regression models examining moderators of effect sizes among eavesdropper preference and eavesdropping risk assays. Significance was determined via Q_m test for all fixed effects, marginal R² is the amount of variance explained by each fixed factor, and I² is an estimate of effect size heterogeneity. Each factor was tested using a separate multi-level mixed-effects model with a single fixed factor and two random factors (study ID and observation ID), while controlling for non-independence arising from shared controls within studies.

Context	moderator	d.f	Qm	p	R2	I2
eavesdropper preference	modality	4	44.248	<0.001	0.076	0.832
	eavesdropper	2	49.590	<0.001	0.052	0.806
	manipulation	2	46.694	<0.001	0.008	0.816
eavesdropping risk	modality	3	61.454	<0.001	0.161	0.927
	eavesdropper	3	45.141	<0.001	0.073	0.935
	manipulation	2	36.750	<0.001	0.013	0.940

Here too inspection of the funnel plot revealed some evidence of asymmetry, with apparent missing effects in the lower left (electronic supplementary material, figure S2). This was supported by Egger's regression (est = −0.926, z = 5.277, p < 0.001), and our trim-and-fill analysis suggested up to 17 unreported effects. Adjusting to explore the influence of these ‘missing’ effects reduced the meta-analytic mean (mean g = 0.759, 95% CI = 0.436–1.082, k = 115), though again the difference was relatively minor and the overall estimated effect remained moderate to strong. This suggests a weak influence of any publication bias, though we acknowledge its possible existence and inflationary influence on model estimates when interpreting our results.

4. Discussion

Signalling to potential mates is often central to reproduction, though it brings with it the risk of eavesdropping by predators. Here we examined this predicted cost of sexual signalling in a meta-analysis of experimental studies. We found evidence for strong preferences for sexual signals by predators and parasites in a forced-choice context, which held irrespective of the signalling modality, the type of eavesdropper, or the nature of the experimental manipulation (figures 1 and 2). Under more natural conditions we found a similarly moderate to strong average effect, suggesting a heightened risk to signallers in the wild. This varied in a modality-specific manner as predicted (table 1), with the greatest costs borne by olfactory and auditory signals (figure 1). Curiously, however, the incidence of predation on visual signallers was not only reduced relative to other modalities, but was on average indistinguishable from non-signalling controls. Further, we found substantial variation in outcomes across all modalities and contexts (electronic supplementary material, table S2), which suggests that circumstances under which signals incur no costs, or even reduce the burden of eavesdropping, should be common. The persistence of high heterogeneity across all models, however, also emphasizes the role of unmeasured differences in methodology and ecology between studies in shaping the distribution of effects. Contrary to our expectations, we found no statistical difference in average effects based on the type of eavesdropper or the nature of the experimental manipulations, though in both cases the weak estimated differences were in the predicted direction.

The clear preference for signallers (or their signals in isolation) expressed by eavesdroppers is unsurprising (figure 2), for two reasons. One is the salience of signals relative to controls. Selection, in general terms, favours increased conspicuousness in signals as a consequence of both the need for detectability amid environmental noise [27] and concurrent sexual selection favouring elaboration via runaway or indicator processes, or the exploitation of sensory biases [121]. All else being equal, this heightened salience of signals over controls will naturally attract the interest of viewers. The other reason is the perception of stimulus identity or category. Predators and parasites rely on informative cues—such as the colours, calls and odours which characterize sexual signals—as a guide to potential prey [12,21]. The preference for signals is therefore interpretable as a preference for likely prey, given that most studies focus on interactions between signallers and their known, ecologically relevant eavesdroppers (e.g. [43,74,83]). The absence of modality-, manipulation- and predator-specific variation in the average strength of preferences (figures 1 and 2) is similarly unsurprising under a forced-choice experimental paradigm since most of the functional differences between signals (table 1) are negated in a laboratory setting. This also emphasizes the need for caution when extrapolating laboratory-based results to realized predation in the wild, where the effects of signal ecology are rendered apparent (electronic supplementary material, table S2).

The moderate to strong increase in eavesdropping risk we identified is consistent with the results from preference assays, though the concordance largely ends there. Of particular significance is the modality-specific nature of costs, with olfactory and auditory signals attracting heightened risk relative to visual signals (figure 2). This accords with signalling theory which, in general terms, describes how differences in the propagation of signals through natural environments underlies differential predation risk [29,122] (table 1). The on-average absence, and sometimes reduction, of risk associated with visual sexual signals is curious, but affirms recent work showing that predators impose no cost when their exposure to prey signals is infrequent, owing to neophobia and/or dietary conservatism [47]. Such effects are both widespread and common among predators [123], which therefore stands as a general working hypothesis for the modality-specific differences in outcomes seen in the current evidence base (figure 1). We also briefly note that this result touches on the ‘paradox of aposematism’ inasmuch as it suggests that, contrary to expectation, conspicuous but undefended prey may simply bear no additional predation cost in the wild (figure 1), thereby allowing the subsequent evolution of defences [124].

It should be remembered that most studies in our sample necessarily discounted the influence of behaviour by both signallers and predators. In that sense our estimates can be understood as representing the baseline risk of predation absent any adaptations for actively enhancing and/or subverting the privacy of communication. Such innovations are well documented (e.g. [125,126]). Predicting their form and occurrence, however, is an enduring challenge for which knowledge of the signal features that drive differential risk offers some guide (table 1). Visual signallers should seek to minimize their localizability, for example, through the coupling of highly directional signals with precision displays, as seen among iridescent insects (e.g. [127,128]). While auditory signallers may temper the reach of calls by shifting frequencies, or even modalities (from higher- to lower-risk; figure 1), under threat of predation, as seen among forest-dwelling katydids [129–131]. Fully appraising such possibilities demands deeper knowledge of the structure and diversity of sensory environments, signals, and receivers, which remain valuable avenues for future work [132].

Our findings also reach into applied domains, as reflected in the breadth of fields captured in our evidence survey (electronic supplementary material, table S1). Predicting and responding to biotic invasions is a pressing conservation challenge, for one, with recent work emphasizing the importance of cue recognition in mediating interactions between invasive and native species [36,37]. That visual signals attract minimal cost (figure 1), for example, suggests conspicuous ornaments will present little impediment to introduced species becoming invasive. This is consistent with general evidence for neophobia among predators, as well as the importance of search-image formation in guiding visual foragers [123]. By contrast, the heightened risk associated with olfactory signals is laid bare in work showing the rapid exploitation of such information by invasive predators (e.g. [36]). This dynamic also presents opportunities, however, with recent studies deliberately familiarising invasive predators with unrewarding odours and, in doing so, improving outcomes for vulnerable prey [133,134].

Similar efforts to ‘weaponize’ eavesdropping are ongoing in pest management, where the encouragement of natural predators is central to contemporary control methods [38]. The strong effects associated with parasitoids and parasites (albeit with a very limited sample of the latter; figure 2), and the heightened influence of olfactory and auditory signals over visual (figure 1), suggest them as profitable targets for future strategies. This aligns with a substantial body of work centered on developing artificial, bio-inspired kairomones from the sex pheromones of key pests such as aphids and moths [39,135]. The substantial variation in effects we found however, as captured in wide prediction intervals, suggests artificial lures may benefit from redundancy across modalities for improved efficacy (given the unreliability of any single modality, on average; electronic supplementary material, table S2), and supports more general calls for multimodal solutions to control problems [136].

The study of sexual communication has driven general advances in theory and its application [121,135]. That signals capture the interest of eavesdroppers and increase the risk of predation, but in a variable and modality-specific manner, anchors our understanding of a central cost to sexual communication, and sex more generally [8,121]. Much remains to be learned, however. Signalling in vibratory and electric modalities is sorely understudied in all respects and warrants further general attention. As does the extent to which plasticity, including behaviour, can dynamically balance the demands of signalling with the foundational costs of predation described here. Multimodal systems present promising, albeit underused, sources for progress on all fronts, including open problems of signal evolution amid sustained eavesdropping (e.g. [137]). These are exciting areas for progress on questions of broad significance.

Data Accessibility

All data and code necessary to reproduce our analyses are publicly available via GitHub (https://github.com/EaSElab-18/ms_metarisk) and are persistently archived via Zenodo at https://doi.org/10.5281/zenodo.6534204 [138].

Electronic supplementary material is available online [139].

Authors' Contributions

T.E.W.: conceptualization, data curation, formal analysis, methodology, visualization, writing—original draft, writing—review and editing; T.L.: conceptualization, data curation, validation, writing—review and editing; K.D.L.U.: conceptualization, data curation, validation, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interests declaration

We declare we have no competing interests.

Funding

This work was supported by the Hermon Slade Foundation (HSF20082).