Marginal sinks or potential refuges? Costs and benefits for coral-obligate reef fishes at deep range margins

Abstract

Escalating climate-related disturbances and asymmetric habitat losses will increasingly result in species living in more marginal habitats. Marginal habitats may represent important refuges if individuals can acquire adequate resources to survive and reproduce. However, resources at range margins are often distributed more sparsely; therefore, increased effort to acquire resources can result in suboptimal performance and lead to marginal populations becoming non-self-sustaining sink-populations. Shifting resource availability is likely to be particularly problematic for dietary specialists. Here, we use extensive in situ behavioural observations and physiological condition measurements to examine the costs and benefits of resource-acquisition along a depth gradient in two obligate corallivore reef fishes with contrasting levels of dietary specialization. As expected, the space used to secure coral resources increased towards the lower depth margin. However, increased territory sizes resulted in equal or greater availability of resources within deeper territories. In addition, we observed decreased competition and no differences in foraging distance, pairing behaviour, body condition or fecundity at greater depths. Contrary to expectation, our results demonstrate that coral-obligate fishes can select high-quality coral patches on the deeper-reef to access equal or greater resources than their shallow-water counterparts, with no extra costs. This suggests depth offers a viable potential refuge for some at-risk coral-specialist fishes.

1. Introduction

Rapid climate change has increased ecosystem degradation and extinction risk across most of the Earth's biomes and taxonomic groups, with many species experiencing range contractions or displacements from core ranges [1–5]. Present-day populations of many extant species radiated from refuge populations that survived past global-scale environmental change (e.g. Quaternary glacial cycles) in cryptic refugia at range margins [6,7]. Population persistence in small refuges, often at range margins, is predicted to facilitate many species' long-term resilience to the asymmetric impacts of anthropogenic climate changes [8,9]. However, the potential for marginal habitats to act as refuges largely depends on individual performance at these ecological extremes. Species presence does not always indicate individual success or population viability; therefore, understanding the ecological factors that limit population viability and persistence at range margins is critical for predicting future trajectories.

Individuals occupying a species’s range margin face many potential costs. While species' realized niches are ideally centred on regions of optimal performance along gradients (e.g. [10,11]), source–sink dynamics, density-dependence and competition cause individuals to extend beyond a species’s ideal niche [12–14]. At range margins, habitats and populations often become more fragmented [15,16], occupancy decreases [17] and individuals may exhibit behavioural changes and/or experience physiological costs [18–20] in response to resources being more sparse. For example, summer range size of roe deer increases at higher altitudes in response to decreased food availability [21], while greater reliance on high-altitude habitats results in reduced body condition in red deer [22], and lower fecundity among birds [23] and insects [24].

Despite these costs, occupying range margins may also present advantages, such as reduced competition [25]. In addition, although population abundances may be lower in fragmented range-margin habitats, inhabited patches may not differ greatly from ideal conditions within range centres [26,27]. In such cases, margin-dwellers may incur little fitness cost [28,29]. Asymmetric disturbance impacts may also result in previously optimal habitats becoming suboptimal, while marginal habitats remain more stable. In such cases, marginal habitats may represent important refuges for species capable of maintaining viable populations. Determining the underlying mechanisms that support successful range-margin occupancy is therefore important for developing predictions about the generality of refuge potential for species distributed across environmental gradients, and also for species that shift distributions toward range margins in response to changing climate.

Assessing individual performance among marginal populations requires detailed knowledge of changes in key ecological strategies and demographic traits between the range core and range-margin habitats. However, range margins are typically under-sampled, and detailed ecological assessments of individual space use, resource access and physiological condition at range margins constituting potential refuges are rare among animals [30]. Coral reefs provide an ideal model ecosystem for assessing the responses of populations across environmental gradients because they are highly diverse ecosystems that exhibit strong ecological gradients over small spatial scales (e.g. [31–33]). Coral reefs are also highly vulnerable to global-scale degradation [34–36]; however, highly divergent responses often occur at smaller spatial scales (e.g. [37–39]). For example, while warm-water bleaching events are increasing in frequency and intensity, their effects often attenuate with depth [40–44]. Deeper water could provide a refuge for coral reef fishes vulnerable to shallow-water habitat loss (e.g. [45,46]), and fishes with broad depth ranges are considered at lower risk of extinction than species restricted to shallow depths [47]. Furthermore, recent studies have demonstrated large proportions (up to 85%) of coral-associated fish species occur at or below 20 m depth in some systems [45]. Despite broad depth ranges and no clear dispersal barriers, densities of coral specialists often decrease with depth, presumably in response to decreased resource quantity and/or quality [45]. Consequently, while it is clear that many coral-obligate reef fish species occur at greater depths than currently appreciated, the ecological and physiological costs of deeper residence on reefs remain unknown.

The ecological, behavioural and physiological effects of coral decline on coral-dependent fishes in shallow water are well established. Low densities of preferred coral genera lead to increased space use and effort to protect resources, altering social dynamics and resulting in sub-lethal costs (e.g. [33,48–50]). Similar dynamics along depth gradients would reduce the refuge potential of deeper reefs. However, due to the difficulty of deep-water diving, there is a paucity of detailed ecological data among vulnerable taxa with wide depth ranges on coral reefs (but see [20]). Consequently, assessments of extinction risk and debate regarding the refuge potential of deep reefs are based on the unwarranted assumption that intraspecific ecology is static along depth gradients [47,51,52]. Further, mechanisms that support successful occupancy of range margins remain largely undetermined within much of the animal kingdom, limiting the capacity for predictive generalizations regarding the refuge potential of range margins more broadly [9]. Here, we examine changes in the ecology of two corallivorous butterflyfishes (Chaetodon baronessa and C. octofasciatus) with contrasting levels of dietary specialization along a depth gradient from 0–35 m to investigate the behavioural and physiological costs of living at greater depths. Specifically, we tested whether: (i) individuals' space use increased with depth; (ii) lower resource densities resulted in fewer secured coral resources in deeper territories; (iii) decreased resource availability led to behavioural costs related to accessing and securing resources at depth; and (iv) individual body condition, energy storage and fecundity declined with depth.

2. Material and methods

(a). Study site and focal species

We recorded the spatial parameters, conspecific-neighbour densities, rates of maintenance behaviours and coral resource densities within all territories of two obligate coral-feeding butterflyfish species between less than 1 m and 35 m depths within a 250 m wide section of Christine's Reef in Kimbe Bay, Papua New Guinea. The vertically continuous coral habitat along the entire depth gradient of the focal reef presents no physical barriers to a movement among depths and is representative of reefs in the region. The two focal species are both obligate corallivores with equivalent depth ranges (0–40 m), but contrasting density distributions within their depth range and contrasting levels of specialization within their coral-obligate diets. The depth range examined covers the full range of disturbance for known bleaching events. We define C. baronessa (n = 39 territorial pairs) as a ‘shallow-specialist’ because its distribution is strongly skewed toward shallow water, and it has a narrow dietary niche (niche breadth = 0.07) with high selectivity for Acropora corals (C.M. 2018, unpublished data). By contrast, Chaetodon octofasciatus (n = 21 territorial pairs) is defined as a ‘deep-generalist’ due to its broad dietary niche (niche breadth = 0.23) and infrequent occurrence in depths ≤5 m, but comparatively high abundance from 10 to 30 m (C.M. 2018, unpublished data).

(b). Territory size

Territorial butterflyfish patrol territory perimeters frequently, using habitual swim paths [49].Territorial pairs were identified using individual markings, external tags (Floyd T-bar) inserted in the dorsal muscle (see [53]), and their site fidelity, which was confirmed from many repeat observations over the course of the study. We demarcated territories by observing pairs for multiple 5–15 min periods (minimum of three initial observations), marking swim paths and territorial boundaries with flagging tape. Territories were confirmed via frequent re-visitations over multiple weeks, and the perimeter, minimum and maximum depths of territories were measured in situ.

(c). Within-territory resource density and ‘total secured resources'

We recorded the density of all coral resources (total-coral) and the preferred resource (genus ‘Acropora’) within each territory, from 1 m² photographed benthic quadrats at approximately 2 m intervals around each perimeter. The benthic component directly under each of 25 randomly allocated points was recorded for each quadrat in coral point count (CPCe) [54]. Corals were recorded to genus. We calculated ‘total secured resources' of total-corals (TSR_T) and Acropora corals (TSR_A) within territories by multiplying the mean density of each resource by the area encompassed within a 1 m internal border around the perimeter length. To examine the distribution of resources across larger spatial scales, we also recorded total-coral cover and Acropora coral cover using 60 random points in 4–6 replicate transects at depths of less than 1, 5, 10, 20 and 30 m, on 10 reefs throughout Kimbe Bay [45]. We used more intensive benthic sampling on the focal reef from 90 to 120 replicate photo-quadrats per depth (approx. 1 m²) at less than 1, 5, 10, 15, 20, 25 and 30 m.

(d). Behaviour

We recorded rates of territorial interactions, movement and pairing within the same focal territories. Observations were recorded simultaneously during four to six replicate 3 min observation periods for each pair, with focal fish followed at a distance of approximately 2–4 m. Movement paths were marked every 0.5–1 m with weighted flagging tape markers, and total distance measured. Pairing status and water depth were recorded every 15 s (paired, ≤ 2 m from partner). Observations were not recorded if focal fish showed flight or aggressive display responses. Conspecific density was calculated as the number of directly adjoining territories divided by the length of the focal perimeter.

(e). Body condition

We determined the condition of fish residing at different depths with five commonly used metrics. After collection of behavioural data, fish from most territories were harvested by spear in January 2016, with no obvious sampling biases (C. baronessa = 35 females, 31 males—90% of 39 territories, C. octofasciatus = 16 females and 15 males—75% of 21 territories). A further 10 male and 21 female C. octofasciatus were collected in November 2016. All fish were gutted, weighed and measured. Gonads and livers were removed, weighed and stored in 4% calcium buffered formalin. The five physiological condition metrics were: total length (TL); Fulton's K (K = 100 × W_gutted/TL³); histosomatic index (HSI = 100 × W_liver/W_gutted); gonadosomatic index in females (GSI = 100 × W_gonad/W_gutted); and proportion of vacuolated hepatocyte cells in males, as a measure of energy storage [50]. The proportion of points intersecting vacuoles was recorded in CPCe from 50 random stratified points in each of three digital frames (400× magnification) from each of three stained (Mayer's Haemotoxylin and Eosin) 5 µm sections per liver.

(f). Analyses

Differences in territory sizes among depths were examined using linear models (lm) of log (perimeter) length against median-territory-depth (med.ter.depth). The density of both total-corals, and Acropora was modelled against med.ter.depth using binomial comparisons of the number of points identifying (i) coral and non-coral substrata, and (ii) Acropora and non-Acropora substrata within territories. Models were performed in glmer, from the r package lme4 using quadrat as a random factor. TSR_T and TSR_A were modelled against med.ter.depth using lm. Conspecific density was modelled against med.ter.depth using glm with a Poisson error-wise family. Rates of territorial interactions were modelled against med.ter.depth using a Poisson error-wise family in glmer with territory (ID) and observation (Obs) included as random factors. The log of distance moved was modelled against med.ter.depth in lmer, with ID and Obs included as random factors. Pairing ratios were modelled against med.ter.depth in glmer, using a binomial comparison of paired and not-paired observation counts, with ID and Obs included as random factors. TL, Fulton’s K, HSI and GSI were modelled for variation among med.ter.depth using lm. Variation in hepatocyte vacuolation rates with med.ter.depth was tested using a binomial comparison of vacuolated and non-vacuolated cell counts in glmer, with ID and Obs included as random factors. For models fit in lmer and glmer; r.squaredGLMM was used to obtain pseudo-R-square estimates; dispersion_glmer was used to test for over-dispersion; deviance-based tests of fit were undertaken; confint was used to obtain confidence intervals; and glht was used to obtain probability estimates of effects. For C. octofasciatus, analyses of body condition metrics first incorporated collection date as an interaction term. No interactions were present (no 95% CI crossed zero, all p > 0.10), so the term was excluded from final models (condition.metric_i ∼ med.ter.depth). All analyses were therefore consistent between each species and were undertaken in R v. 3.3.2 [55].

3. Results

(a). Territory size

Mean territory sizes did not differ between the two species overall (figure 1a). However, territory area increased approximately three-fold along the depth gradient for the shallow-specialist C. baronessa (table 1, figure 1b), but did not change for the deep-generalist C. octofasciatus (figure 1c).

Figure 1.

(a) Interspecific similarities in territory size between a shallow-specialist (Chaetodon baronessa—red) and deep-generalist (C. octofasciatus—blue) obligate coral-feeding butterflyfish species, and intraspecific variation in territory size along a depth gradient for (b) the shallow-specialist species and (c) the deep-generalist. (d–f) Within-territory resource densities, showing (d) the interspecific similarities, (e) intraspecific variation along the depth gradient for the shallow-specialist and (f) depth variation for the deep-generalist. (g–i) Total secured resources within territories, showing (g) interspecific similarities, (h) intraspecific variation along the depth gradient for the shallow-specialist and (i) depth variation for the deep-generalist. Lines above bars in (a,d,g), represent statistically similar means. In regression plots, each variable is modelled against the median depth of territories; solid lines and straight dotted lines represent best fits, bands represent 95% confidence intervals and each data point represents a territory.

Table 1.

Result summaries for models of depth-related variation in territory size, within-territory densities and total availability of two types of coral resources, and number of directly neighbouring territories of a shallow-specialist and deep-generalist obligate coral-feeding butterflyfish species.

species	metric	response variable	R2	estimate ± s.e.	intercept ± s.e.	F(df)	p-value
Chaetodon baronessa (specialist)	territory size	perimeter	0.42	1.45 ± 0.14	23.39 ± 4.14	27.09(1,38)	<0.001
	resource density	total-coral	0.14	0.52 ± 0.11	55.96 ± 3.37	5.37(1,32)	0.027
		Acropora	0.05	0.41 ± 0.17	21.26 ± 4.90	1.56(1,32)	0.22
	total secured resources (TSR)	total-coral	0.46	0.96 ± 0.10	13.56 ± 2.78	26.66(1,32)	<0.001
		Acropora	0.44	0.42 ± 0.09	4.75 ± 1.28	24.76(1,32)	<0.001
	neighbour density	conspD	0.40	−0.004 ± 0.0009	0.14 ± 0.01	20.20(1,31)	<0.001
Chaetodon octofasciatus (generalist)	territory size	perimeter	0.02	−0.25 ± 0.39	45.87 ± 8.00	0.43(1, 19)	0.52
	resource density	total-coral	0.26	−0.56 ± 0.24	34.81 ± 4.86	5.50(1,16)	0.032
		Acropora	0.16	−0.32 ± 0.19	12.98 ± 3.79	2.99(1,16)	0.10
	total secured resources (TSR)	total-coral	0.06	−0.50 ± 0.50	70.52 ± 10.20	1.01(1,16)	0.33
		Acropora	0.04	−0.46 ± 0.54	27.00 ± 11.08	0.74(1,16)	0.40
	neighbour density	conspD	0.03	−0.0009 ± 0.002	0.07 ± 0.03	0.55(1,19)	0.47

(b). Resource availability

Total hard coral cover throughout the bay did not decline with depth (mean approx. 55% at all depths), but Acropora cover declined from approximately 25% at less than 1 m to approximately 2% at 30 m (electronic supplementary material, figure S1). Within the focal reef, total hard coral cover peaked at 15 m (approx. 72%) and was lowest at 30 m (approx. 48%) (F_1,5 = 10.79, p < 0.001). Acropora cover on the focal reef was highest at less than 5 m (approx. 23%) and lowest at 30 m (approx. 7%) (F_1,5 = 10.91, p < 0.001).

Within-territory resource densities did not decrease with depth for either species (table 1). Mean densities of both total-coral and Acropora resources did not vary between territories of the two species overall (figure 1d) and did not decline with depth in the territories of either species (figure 1e,f). In fact, the density of total-coral resources increased approximately 15% along the depth gradient for the shallow-specialist (figure 1e). As a result of increasing territory size and stable or increasing resource densities with depth, TSR_T and TSR_A within territories of the shallow-specialist both increased approximately three-fold along the depth gradient (figure 1h). By contrast, TSR_T declined by almost half between the shallowest and deepest territories of the deep-generalist species, while TSR_A was consistent along the gradient (figure 1i).

(c). Neighbour density and maintenance effort

There was no apparent increase in neighbour density or maintenance effort with depth (table 2). For the shallow-specialist, the number of directly neighbouring conspecific territories decreased with depth (table 2, figure 2a). Correspondingly, mean neighbour densities declined almost five-fold between shallow and deep territories (figure 2b). The rate of territorial interactions for the shallow-specialist also declined by over two-thirds along the gradient (figure 2c). Depth explained a small proportion of variation in movement rates of the shallow-specialist, with mean rates declining by approximately a third from the shallowest to deepest territories (figure 2d). Pairing behaviour in the shallow specialist did not vary with depth. By contrast, there was no depth-related change in neighbour density (figure 2a), territorial interaction rates (figure 2b), or movement rates for the deep-generalist (figure 2c), and pairing rates of the deep-generalist also did not decline with depth (figure 2d).

Table 2.

Result summaries for models of depth-related variation in territory maintenance effort of a shallow-specialist, and deep-generalist obligate coral-feeding butterflyfish species. Each variable is modelled against the median depth of territories.

						conf. int.(95%)
species	relationship	response variable			estimate	lower	upper	z	p-value
Chaetodon baronessa (specialist)	distance moved	log(distance)	0.04	0.26	−0.02	−0.04	1.55−3	−1.85	0.098
	interactions	interactions	0.14	0.22	0.92a	0.87a	0.97a	−3.29	0.001
	pairing	Cbind(p,np)	<0.00	<0.00	0.49	0.47	0.51	−1.05	0.315
Chaetodon octofasciatus (generalist)	distance moved	log(distance)	0.06	0.35	0.02	−0.04	0.04	1.61	0.134
	interactions	interactions	0.09	0.17	0.91a	0.80a	1.01a	−1.79	0.073
	pairing	Cbind(p,np)	<0.00	0.31	0.56	0.45	0.56	0.25	0.802

^aEstimates and confidence intervals based on log link for the Poisson error family are factorial. In this case, the evidence does not support an effect where confidence intervals cross 1.

Figure 2.

Depth-related variation in competitor density and territorial maintenance effort. (a) The density of directly neighbouring conspecifics around territory perimeters. (b) The number of territorial interactions (insert shows shallow-specialist only), (c) the distance moved and (d) mean paring ratios. Solid lines represent the best fit for generalized linear models and bands are 95% confidence intervals.

(d). Body condition

No aspect of physiological condition declined significantly with depth in either fish species (table 3, figure 3). Neither female nor male C. baronessa total lengths (TL) declined with depth (figure 3a). Similarly, neither relative body mass nor hepatosomatic index declined with depth in either sex (figure 3b,c). The fecundity (GSI) of female C. baronessa did not decline with depth (figure 3d), and neither did energy storage (hepatocyte vacuolation) among males (figure 3e). However, there was some indication that a small proportion of variation in body mass of male C. baronessa, and the fecundity and HSI of females show depth-related trends (all; R² < 0.13, probability = 0.1 ; table 3). For C. octofasciatus, none of total lengths, relative body mass nor hepatosomatic index declined with depth for either sex (figure 3f–h). The GSI of female C. octofasciatus did not decline with depth (figure 3i), and neither did energy storage among males (figure 3j).

Table 3.

Result summaries for models of depth-related variation in body condition of a shallow-specialist and deep-generalist obligate coral-feeding butterflyfish species. Each variable is modelled against the median depth of territories. Italicized p-values show metrics whose depth-related variations were approaching statistical significance.

species	condition (metric)	sex	R2	estimate ± s.e.	intercept ± s.e.	test statistic	p-value
Chaetodon baronessa (specialist)	body length (total length)	F	0.02	0.08 ± 0.10	91.34 ± 1.62	F = 0.58(1,30)	0.454
		M	0.06	0.13 ± 0.09	91.43 ± 1.50	F = 1.85(1,28)	0.184
	relative body mass (Fulton's K)	F	0.04	−3.46−6 ± 3.23−6	3.05−3 ± 5.18−5	F = 1.14(1,30)	0.294
		M	0.12	−1.11−5 ± 5.67−6	3.37−3 ± 9.33−5	F = 3.81(1,28)	0.061
	relative liver mass (LSI)	F	0.11	−0.01 ± 3.73−3	1.38 ± 0.06	F = 3.83(1,30)	0.060
		M	0.06	−0.01 ± 0.01	1.16 ± 0.10	F = 1.72(1,28)	0.201
	fecundity (GSI)	F	0.10	−0.025 ± 0.01	2.56 ± 0.22	F = 3.44(1,30)	0.074
	energy storage (hypatocyte vacuolation)	M	—	0.01 ± 0.01	−1.30 ± 0.20	z = 0.79(1,19)	0.429
Chaetodon octofasciatus (generalist)	body length (total length)	F	3.96−7	−3.16−4 ± 0.09	71.90 ± 1.40	F = 1.39−0.5(1,35)	0.997
		M	0.08	−1.18 ± 0.13	70.69 ± 2.23	F = 1.93(1,23)	0.178
	relative body mass (Fulton's K)	F	0.02	3.26−3 ± 7.67−5	−34.34−6 ± 4.65−6	F = 0.27(1,35)	0.606
		M	7.16−4	−9.96−7 ± 7.53−6	3.26−3 ± 1.37−4	F = 0.02(1,23)	0.899
	relative liver mass (LSI)	F	0.01	−3.13−3 ± 0.01	1.03 ± 0.09	F = 0.31(1,34)	0.581
		M	1.37−4	−3.45−4 ± 0.01	1.00 ± 0.11	F = 3.02−03(1,23)	0.957
	fecundity (GSI)	F	5.41−4	−2.96−3 ± 0.02	2.16 ± 0.35	F = 0.02(1,34)	0.893
	Energy storage (hypatocyte vacuolation)	M	—	−0.03 ± 0.02	−1.62 ± 0.46	z = −1.17(1,23)	0.243

Figure 3.

Relationships between body condition and depth. Metrics are: (a,f) total length; (b,g) relative body mass; (c,h) hepatosomatic index; (d,i) reproductive potential; (e,j) energy storage. Data points = individual fish. Closed circles = females. Open circles = males.

4. Discussion

Predictions of the impacts changing environments will have on biodiversity require information on changes in individual performance across a species range and a greater understanding of the mechanisms and ecological dynamics that support successful occupancy at range margins [9,17].To date, most studies that invoke the potential of increased resilience or refuge effects in marginal habitats, including ‘refuge models’ based on broad environmental parameters, largely assume intraspecific ecology recorded at range cores is static along sometimes very steep environmental gradients [9,47,51,52]. Moreover, few studies account for costs that can limit individual success and population viability at range margins [9]. Our results demonstrate that spatial ecology and resource maintenance behaviours can vary significantly between range-core and range-margin habitats, even over short, steep environmental gradients such as water depth, as well as between species with different expressions of similar ecological traits. Despite covariation between physiological condition and environmental gradients being commonplace in a multitude of ecological systems [56–59], our results demonstrate that responses to differential environmental and habitat conditions at range margins (here, deeper water) do not necessarily result in substantial ecological or physiological costs. The relative costs, compensatory mechanisms and risk-mitigating benefits of living at range margins are therefore likely to be species- and ecosystem-specific. Nonetheless, our detailed ecological assessment indicates that deeper reefs likely represent a valuable potential refuge for at least some vulnerable coral-obligate reef fishes. Similar investigations across other species and ecosystems would facilitate predictions of the importance of refuges to mediate species extinctions across a broad range of ecosystems.

Contrary to expectations, neither highly specialized C. baronessa nor deep-generalist C. octofasciatus residing near their deeper water range margins (to 35 m) experienced any physiological costs. Moreover, at the range margin the shallow-specialist C. baronessa had greater resource access and experienced lower competitive pressures than their range-core counterparts. While the size of C. baronessa territories increased with depth, this was not accompanied by a decrease in resource densities, indicative of individuals establishing territories on high-quality resource patches. Total quantities of preferred resources (Acropora) were therefore up to three times greater in deeper territories, with little or no cost in terms of defence effort, movement rates or time paired. In addition, the deep residence had little effect on a range of physiological condition factors, including fecundity. While the territory sizes of C. octofasciatus did not increase at greater depths and the species did experience some decline in within-territory resource densities, there was no indication this resulted in declines in total secured resources (TSR) or increased sub-lethal costs.

A number of general ecological paradigms collectively suggest that environmental and ecological conditions at range margins should result in increased costs, particularly for specialist territorial species that are tertiary consumers (e.g. [15,60]). Resource fragmentation limits successes of a broad range of vertebrate communities, particularly at range margins [61,62]. Territory sizes are therefore assumed to increase toward range margins (e.g. [21]) and resultantly increase costs associated with foraging and resource protection. Cost–benefit models across a broad range of taxa from terrestrial and marine environments [63–67] suggest that the benefits of increased resource access in larger territories are offset by behavioural costs of defending resources. However, these relationships were not apparent in the highly specialized territorial reef fishes at deep range margins. Therefore, our results do not concur with avian [23,30], mammalian [21,22], freshwater fish [57] and insect [24] studies conducted along elevation gradients. Similarly, our results are not consistent with investigations of butterflyfishes across horizontal resource gradients within their core shallow-water range [33,49,50]. We therefore found no evidence that the costs of deep residence would mitigate the benefits of deep refuges in this system. Instead, deeper-reef habitats may offer refuges to populations of reef fishes vulnerable to coral loss in shallow waters even where the density of key (coral) resources decline along the depth gradient.

Our results suggest that selective placement of territories in pockets of comparatively high resource density and lower intraspecific competitive pressures are key mechanisms facilitating successful occupancy of (deeper) marginal habitats. Decreased resource availability may therefore reduce population densities at range margins, but not necessarily at the cost of resource access for the fewer individuals occurring there. This supports the hypothesis that low population to resource ratios may be an important characteristic of successes at range margins [68]. In addition, micro-environmental conditions or alternative energy-acquisition strategies that favour the maintenance of resources in sufficient quantity and quality may also be important factors influencing refuge potential for secondary and tertiary consumers at range margins.

One important caveat is that this study cannot directly address the potential ramifications of future increases in resource competition or predation if assemblages shift downwards towards deep marginal habitats and away from degrading shallow-water resources. Assessing the degree of potential costs arising from this currently hypothetical situation will require long-term depth-stratified data spanning periods of shallow-water disturbances. To our knowledge, no such data have been presented to date.

5. Conclusion

Increasing asymmetric habitat disturbances in response to rapid and ongoing climate change will increase the importance of marginal habitats for providing local-scale refuges for biodiversity. While generalist species are commonly considered better equipped to take advantage of marginal habitats, we show deeper reefs are a potential refuge for highly specialized coral-obligate reef fishes. In addition, we highlight the importance of detailed ecological investigations across a species range to improve understanding of refuge potential for populations occupying marginal habitats. Few detailed field-based investigations have examined ecological factors that enhance or limit refuge potentials at range margins for animals (but see [29]); therefore, our results are among the first detailed investigations of the condition and ecological responses among populations of vulnerable marine taxa dwelling at deep range margins. We show that the costs and benefits of occurring at range margins are likely to be species- and ecosystem-specific, and highlight the importance of examining a range of ecological and physiological factors to properly assess the costs and benefits of living at range margins. Given the rapid increase in scientific interest in refuges and their likely importance for the persistence of many species, we recommend future studies account for differences in species physiological and ecological performance across their entire range.

Ethics

All methods were undertaken within the guidelines permitted by the James Cook University Animal Ethics Committee under approval number A2149.

Data accessibility

Territorial, behavioural, resource availability and body condition data are available at Dryad Digital Repository: http://dx.doi.org/10.5061/dryad.c5v81v1 [69].

Authors' contributions

C.M. conceived of the study, designed the study, coordinated the study, collected data and drafted the manuscript; G.P.J. participated in the design of the study, provided funding and helped draft the manuscript; T.B. participated in the design of the study and drafted the manuscript. All authors gave final approval for publication.

Competing interests

The authors have no competing interests.

Funding

Funding was provided through an Australian Research Council grant, via the Centre of Excellence for Coral Reef Studies.