The energetic cost of filtration by demosponges and their behavioural response to ambient currents

Benthic suspension feeders affect marine and freshwater ecosystems by ingesting suspended particulates and dissolved nutrients from the overlying water column and releasing by-products for use by other organisms (Gili and Coma, 1998). Recycling of nutrients in this way provides an important link between the benthic and pelagic communities, known as benthic–pelagic coupling, yet the energetic cost of this type of feeding in invertebrates is still debated. Some work suggests the cost of pumping for filtration is minimal (Riisgård and Larsen, 1995, 2001), while other studies estimate that filtration can make up a quarter of the costs of metabolism (e.g. Hadas et al., 2008).

Many suspension feeders use ciliary or muscular pumps to induce water currents that draw water with food towards themselves. Bivalves, ascidians, polychaetes and sponges use filters to strain out particles from the water that are too small to be captured individually (Jørgensen, 1966). Where suspended particulates are low in concentration, this approach can be highly efficient to process large volumes of water (Jørgensen, 1966), and so it has been suggested that filter feeders evolved a low energetic cost of filtration to allow continuous feeding (Jørgensen, 1975). Yet, food availability varies widely on a temporal basis, with fluctuations occurring seasonally, daily and with the ebb and flood of the tide. It would therefore be advantageous for filter feeders to sense the variations in food availability and feed when concentrations are high. Two examples suggest this behaviour does occur. Bivalves respond to food availability by reducing filtration and respiration rates when food is scarce (Thompson and Bayne, 1972; Griffiths and King, 1979), and the demosponge Tethya crypta reduces pumping activity at night when ambient currents are lower (Reiswig, 1971); lower ambient currents often correspond with reduced food availability (Newell and Branch, 1980). As all filter feeders demonstrate some fluctuations in pumping rates in response to variations in environmental cues, filtration may be more costly than previously thought and the animals may be finely adapted to habitats that support the energetic cost to obtain food.

Models of the filter and pump system for a number of invertebrates have suggested that filter feeding is inexpensive, at less than 4% of total metabolism (Riisgård and Larsen, 1995). In contrast, direct measurements of the uptake of oxygen going from non-feeding to feeding has shown that in bivalves filtration accounts for up to 50% of total metabolism (Thompson and Bayne, 1972; Newell and Branch, 1980) and in sponges 25% of total respiration (Hadas et al., 2008). In addition, the resistance through the filter of sponges may be much higher than previously thought owing to the narrow dimensions of a difficult-to-preserve fine mucus mesh on the collar filter (Leys et al., 2011).

In support of the theory that the cost of filtration may be high, some animals have been found to reduce the energy expended by using ambient currents to draw water through the filter, for example, by dynamic pressure, the Bernoulli effect or viscous entrainment. In high-current environments, barnacles switch from active to passive feeding and orient their bodies toward the current (Trager et al., 1990). Cnidarians (Best, 1988), ascidians (Young and Braithwaite, 1980; Knott et al., 2004) and brachiopods (LaBarbera, 1977) also orient their bodies to the current, while other invertebrates may take advantage of current-induced flow through tubes (Vogel, 1977; Murdock and Vogel, 1978; von Dassow, 2005; Shiino, 2010). Sponges are often considered textbook examples of the use of current-induced flow in nature (Bidder, 1923; Vogel, 1974, 1977) but experiments to confirm this have been equivocal (Leys et al., 2011).

Even if models are correct and the cost of filtration is not great, a universal cost for all sponges is unlikely because structural differences in the sponge canal system owing to body form (Reiswig, 1975a), microbial content (Weisz et al., 2008) and tissue density (Turon et al., 1997) can cause wide differences in filtration rates between species. Temperature and food availability vary across habitats, causing differences in seawater viscosity and enzyme function that potentially lead to differences in the metabolic cost of filtering in different habitats. Estimates of the cost of pumping for a range of sponge species and habitats are therefore required to better understand sponge energy budgets.

Models can be informative about which structures contribute most to energetic costs; however, their accuracy depends on having correct dimensions for each region of the filter and canal system as well as volumetric flow rates. This approach used by Riisgård et al. (1993) was recently used for the filtration system of the glass sponge Aphrocallistes vastus (Leys et al., 2011), where the cost of pumping was found to be 28% of the total metabolism. Here, the resistance through the filter was found to be much higher than previous estimates for demosponges because of the small spaces of the glycocalyx mesh (Leys et al., 2011), a structure that is often not preserved with common fixation techniques and was not included in the model used by Riisgård et al. (1993). In addition, the volumetric flow rate in A. vastus was quite high compared with the ‘standard sponge’ studied by Riisgård et al. (1993), where volumetric flow rate was obtained indirectly using clearance rates of particles (flagellate cells) during incubation in a closed vessel. Closed vessels have been shown to cause reduced pumping behaviour of sponges (Yahel et al., 2005; Hadas et al., 2008), and the sponge may re-filter the water if sampling times are not well adjusted to pumping rates (Yahel et al., 2005). In contrast, Leys et al. (2011) used direct sampling of incurrent and excurrent water to determine filtration and pumping rates in situ.

To evaluate the importance of accurate measures of filter dimensions and volume of water processed when estimating the cost of filtration, we first examined the effect of changes in these values calculated in previous work. We then studied energetics of filtration in five species of demosponge from tropical and temperate habitats using in situ measurements of oxygen consumption and pumping rate, experimental tests of pumping at different ambient flow rates, and morphometric analysis of the canal and filter structures.

We first carried out a meta-analysis using data from the literature to determine the cost of pumping in four filter-feeding invertebrate groups used by Riisgård and Larsen (1995). Dimensions were calculated from electron micrographs of mucus filters (Fig. S1), where available, and volume flow rates that were obtained using ‘direct’ methods were used.

Field work was carried out at the Bamfield Marine Sciences Centre (BMSC) in Bamfield, British Columbia, Canada, and the Smithsonian Tropical Research Institute (STRI) on Isla Colon in Bocas del Toro, Panama. At BMSC, three species of demosponge, Neopetrosia problematica (n=7), Haliclona mollis (n=10) and Tethya californiana (n=8), were collected by SCUBA from Barkley Sound and kept in seawater tables at 8–9°C with unfiltered water from 30 m depth at flow rates of 3000 l min⁻¹.

Measurements on Cliona delitrix (n=8) and Callyspongia vaginalis (n=11) were conducted in situ by snorkeling at STRI Point (9°21.169′N, 82°15.528′W; Diaz, 2005), at approximately 2 m depth where seawater was 28–29°C. An aluminum frame (80/20 Inc., Columbia City, IN, USA) was placed over the sponge and instruments were attached using Loc-Line (Lockwood Products, OR, USA) and clamps (Fig. 1). The instruments were tethered via cables to a laptop computer on a boat anchored near the study site to monitor the data collection in real time.

Excurrent velocity was measured using a Vectrino II profiling acoustic Doppler velocimeter (ADV, Nortek, Rud, Norway), which provided 30 velocity estimates in 1 mm vertical bins, within a sampling diameter of approximately 6 mm. Because the sampling volume is some distance from the probe head, and because the excurrent ‘jet’ from a sponge osculum may be small, and may not travel very far from the osculum lip, positioning it over the sponge osculum is challenging. Fluorescein dye diluted in 0.2 µm filtered seawater was used to visualize the excurrent flow (Fig. S2), and a plastic cable tie was used to indicate the specific position of the sampling volume by blocking the signal. However, the profiling capabilities of the Vectrino II also helps because the ADV takes measurements over the full 30 mm vertical profile, and as a result the lip of the sponge osculum itself disrupts the profile, showing exactly where the sampling volume is above the sponge (Fig. S2). Using this method, it is possible to be sure that the velocity being recorded is just inside the sponge osculum, and this becomes important in preventing interference with ambient water velocities. For the three temperate sponge species the sample volume diameter (6 mm) was greater than the diameter of the oscula and therefore the profiler captured the average velocity across the full osculum. However, for the two warm-water sponges, the sampling volume was smaller than the osculum diameter. Because we sampled just inside the osculum, and because ambient flow could be seen to affect scalar (x,y,z) velocity (Denny, 1993) (Table S1), we therefore only used the Z velocity vector as a conservative measure of the average excurrent velocity of each osculum.

The ADV sensor was attached to Loc-Line for fine positioning above the sponge osculum to obtain the position of maximum excurrent velocity. Once the sensor was oriented correctly, excurrent velocity was recorded for 5 min at 25 Hz on low power (high power was strong enough to push the excurrent flow down, a process termed streaming). Data were binned using a 5 s median filter in MATLAB (vR2013b; MathWorks, Natick, MA, USA). Images of each sponge osculum were taken using a GoPro Hero2 for the duration of the recording and their diameters were measured in ImageJ (v. 1.43r; National Institutes of Health, Bethesda, MD, USA) to calculate volumetric flow.

Sponge volumes and surface area for each of the species were calculated by measuring the dimensions of the sponge from images taken of whole animals in situ, and using ImageJ (v. 1.43r). Volumes and surface areas were estimated as a cylinder (C. vaginalis), an ellipsoid (C. delitrix and H. mollis) and a sphere (T. californiana). When there was more than one osculum per sponge, sponge volume was calculated and divided by the number of oscula to obtain sponge volume per osculum. Because C. delitrix bores into coral, the ratio of sponge to coral skeleton in C. delitrix was estimated by dissolving the coral skeleton using 5% ethylenediamenetetracetic acid (EDTA) for a piece approximately 2 cm³, or 4.5 g, and scaling up to the whole specimen using the relative immersed volumes. Neopetrosia problematica is highly irregular in shape; therefore, a combination of triangles was used to estimate volume and surface area. The ratio of sponge volume to dry weight was calculated by drying three pieces of each species at 100°C to constant mass. This ratio was then used to standardize volume flow rate to dry weight.

Ambient and excurrent oxygen were measured using a two-channel FireStingO₂ optical oxygen metre (Pyro Science, Germany) with 250 µm diameter bare fibre minisensor probes. Before positioning the excurrent sensor in the sponge osculum, both probes were left in ambient water for at least 5 min to determine the difference in readings between the two sensors (here termed the offset value); this difference was subtracted from the difference between ambient and excurrent oxygen for all analyses. Oxygen sensors were calibrated to 21% (saturated water) and zero oxygen prior to experiments using air-bubbled water and a solution of sodium hydrosulfite. Oxygen readings were corrected using both internal and external temperature probes directly by instrument and recorded in mg l⁻¹. In addition, the ambient and excurrent sensors were positioned at the same height in the water column, such that any temperature fluctuations were accounted for between probes. Data were collected every 1 s and binned using a 5 s median filter using MATLAB (vR2013b). Oxygen removal per hour was calculated using the volume of water filtered per hour, and standardized to per gram dry weight.

To assess the effect of ambient current velocities on sponge excurrent flow, experiments were conducted on C. delitrix and C. vaginalis at STRI by manipulating ambient flow with an underwater aquarium pump (Eheim compact+3000). The pump was anchored near the sponges with a weight and the outflow was directed through a 50 cm long, 10 cm diameter PVC pipe at and over the sponges following the method of Genin et al. (1994). The aquarium pump had variable speeds that could generate flow at 5 to 40 cm s⁻¹ through the PVC pipe when positioned 30 cm from the sponge, as measured in a flow flume. Three flow speeds were used in experiments by setting the pump to low, medium and high speeds. Ambient current velocity was recorded using a Vectrino I point ADV (Nortek) with a sampling volume of 6 mm by 7 mm, positioned approximately 10 cm from the sponge and perpendicular to the pump outflow (Fig. 1F). The ADV could not be positioned right next to the sponge because of interference between the ADVs. Data were measured with a transmit length of 1.8 mm at 25 Hz on high power and binned using a 5 s median filter using MATLAB (vR2013b).

Paired excurrent velocities and oxygen removal were measured during experiments as described above. The profiling ADV (Nortek) was first positioned to ensure maximum excurrent velocity recordings from the osculum. Then the oxygen sensor was positioned inside the osculum, ensuring that it did not interfere with the ADV sampling volume as determined by a probe check analysis. Paired recordings were measured for 5 min at each flow rate (no additional flow, low, medium and high pump setting), and this experiment was repeated three times.

A GoPro Hero2 camera with underwater dive housing was positioned on the frame above both sponges to record osculum size during the experiment; images were captured every 30 s and a ruler was positioned in one of the images for calibration. Changes in osculum area were measured using a script developed for MATLAB (vR2013b). Volume filtered was calculated using excurrent velocity and area of the osculum. The ratio of sponge volume to dry weight was used to standardize volume flow rate and oxygen removal to per gram dry weight.

All statistical analyses were completed using SigmaStat in SigmaPlot v12.5. Data were tested for normality and linearity and, subsequently, variables were tested for association using a Spearman's rank order correlation test to allow for non-linearity.

For scanning electron microscopy (SEM), sponges were cut into 1 mm³ pieces and fixed in a cocktail consisting of 1% OsO₄, 2% glutaraldehyde in 0.45 mol l⁻¹ sodium acetate buffer with 10% sucrose at 4°C for 6–12 h (Harris and Shaw, 1984). Modifications included addition of 10% Ruthenium Red to some samples to preserve mucus on the collar filter and adding 4% OsO₄ directly to the sponge tissue before fixation as above, to minimize contraction of canals. Samples were rinsed in distilled water, dehydrated to 70% ethanol and desilicified in 4% hydrofluoric acid (HF) at room temperature (RT) to dissolve spicules. Following desilicification, samples were dehydrated through a graded series to 100% ethanol, fractured in liquid nitrogen and critical point dried. Pieces were mounted on aluminum stubs using an eyebrow brush, gold coated and viewed in a field emission scanning electron microscope (SEM; JEOL 6301 F). Some pieces of C. delitrix and C. vaginalis were embedded in paraffin wax, sectioned at 12 or 30 µm and mounted on aluminum stubs. Prior to embedding, C. delitrix (which bores into coral and so is surrounded by the coral skeleton) was placed in 5% EDTA for 24 h to remove the coral skeleton. After sectioning, the wax was removed in toluene for 15 min, and the stubs were gold coated and viewed in a field emission SEM.

For histology, 1 cm³ pieces of sponge were fixed in 4% paraformaldehyde in filtered seawater for 24 h, rinsed twice in phosphate buffered saline (PBS), dehydrated in 70% ethanol and desilicified in 4% HF/70% ethanol at RT for 24–72 h. Cliona delitrix was decalcified in Cal-Ex Decalcifier (Fisher Scientific) for 24 h. Tissue was embedded in paraffin wax and sectioned at 5 µm (T. californiana), 12 µm (H. mollis and C. delitrix) and 30 µm (N. problematica and C. vaginalis) and sections were stained in Masson's trichrome. Images were captured with a QiCam using Northern Eclipse v.7 (Empix Imaging Inc., Mississauga, ON, Canada) on a Zeiss Axioskop2 Plus microscope.

To determine resistance through the aquiferous system, dimensions of the canals and flagellated chambers were measured from both SEM and histological sections using ImageJ (v. 1.43r). Care was taken to select regions of the canal system that were not contracted by looking at the tissue surrounding the canal system. It was not always possible to identify incurrent versus excurrent canals; in these instances it was assumed that the dimensions and path length were the same between excurrent and incurrent canals, following Riisgård et al. (1993).

The cross-sectional area of each region of the aquiferous system was calculated for a 100 mm³ (100 µl) piece after Reiswig (1975a) and Leys et al. (2011). Owing to differences in the shape of the sponge body, the dimensions for this 100 µl piece differed for each species. For N. problematica, H. mollis, T. californiana and C. delitrix, the inhalant surface was 4.5×4.5 mm and the wall 5 mm thick. The body wall of C. vaginalis was only 3 mm thick and therefore a larger inhalant surface was used (5.77×5.77 mm) to generate the same 100 µl volume for the piece.

The cost of filtration can be estimated by the power needed to move water through the sponge aquiferous system. The power is estimated by converting oxygen used in filtration to watts of energy. The movement of water experiences friction from the small dimensions of the canal and filter system and this resistance is called the head loss. Therefore, to determine the cost of filtration, we calculated the velocity of water through each part of the sponges' aquiferous system, and then estimated the resistance (head loss) at each part of the aquiferous system.

where A_i is the cross-sectional area of the region, A_osc is the cross-sectional area of the osculum and u_ex is the measured excurrent velocity from the osculum.

Two separate approaches were used to estimate the resistance through the canal system of sponges. The first uses an approach summarized by Riisgård and Larsen (1995), where a different equation to model each region of the canal system is applied based on the characteristics of the region. The different equations in this first approach reflect the estimated different architectures of different regions of the sponge. The second approach uses only one equation for the whole canal system (Leys et al., 2011), based on the Hagen–Poiseuille equation for fully developed laminar flow in a tube. This approach assumes that different equations do not capture the accurate differences between regions and therefore one equation is more straightforward and just as accurate. A detailed description of the equations is provided in the Appendix.

Using both dimensions of filter size gathered from electron micrographs and direct volume flow rate measurements resulted in an increase in the estimate of the cost of filtration for taxa studied previously (polychaetes, sponges, tunicates), in some instances to more than five times previously calculated values (Table S2).

Mean excurrent velocity, volumetric flow rate and oxygen removal for each species are provided in Table 1. Cliona delitrix had the fastest excurrent velocity; however, C. vaginalis had the highest volumetric flow rate (processed the most water) and used the most oxygen. Tethya californiana filtered the smallest volume of water per unit time and removed the least oxygen of all five species.

For all sponge species except C. delitrix, more oxygen was removed when more water was filtered (Fig. 2). For one individual of each species measured over a 5 min period, oxygen removal increased with volume filtered except for C. delitrix (N. problematica: r_S=0.813, P<0.0001; H. mollis: r_S=0.869, P<0.0001; T. californiana: r_S=0.905, P<0.0001; C. delitrix: r_S=−0.180, P=0.169; and C. vaginalis: r_S=0.734, P<0.0001; Fig. 2A). Mean oxygen removal between individuals of each species was also positively correlated with the amount of water filtered, again except for C. delitrix (r_S=0.843, P<0.0001; Fig. 2B).

Callyspongia vaginalis responded to changes in ambient flow with changes to both excurrent velocity and oxygen used, but the sponge's response depended on the magnitude of ambient current (Fig. 3A,B). An increase in ambient flow from 0 to 5 cm s⁻¹ caused a reduction in both excurrent velocity and oxygen removal, but when ambient current rose from 5 to 15 cm s⁻¹, then excurrent velocity and oxygen removal both increased. At ambient currents greater than 15 cm s⁻¹, however, excurrent velocity and oxygen removal did not change (Fig. 3B). Overall, both excurrent velocity (r_S=0.141, P<0.001) and oxygen removal (r_S=−0.221, P<0.0001) of C. vaginalis were not correlated with ambient current (Fig. 3A,B). The same is shown by the filtration to respiration (F/R) ratio, which did not change with increasing ambient flow, remaining at approximately 0.42±0.03 l µmol⁻¹ O₂ (r_S=0.232, P<0.0001, Fig. 3C).

For C. delitrix, excurrent velocity was positively correlated with ambient flow (r_S=0.485, P<0.0001; Fig. 4A,B); however, the osculum constricted in response to the flow generated by the pump, reducing in area from 4 cm² to less than 2 cm² (Fig. 4A). The reduced osculum size meant excurrent velocity increased but volumetric flow (volume filtered) decreased with more ambient current (r_S=−0.407, P<0.0001; Fig. 4A,B). Oxygen removal (µmol h⁻¹) initially increased at ambient velocities of 0–5 cm s⁻¹, but overall oxygen removal was negatively correlated with ambient current (r_S=−0.456, P<0.0001; Fig. 4A,B). At ambient currents greater than 15 cm s⁻¹ the excurrent velocity increased because of the constriction of the osculum, and oxygen removal decreased correspondingly. As with C. vaginalis, in C. delitrix there was no significant change in F/R ratio with increasing ambient flow (0.40±0.18 l µmol⁻¹ O₂, r_S=0.307, P<0.0001; Fig. 4C).

Dimensions of each region of the aquiferous canal system for the five species of demosponges studied are given in Table 2. The path the water travels is illustrated in Fig. 5. Briefly, water flows in through minute holes (ostia) in the dermal membrane (a three-layered tissue) into a large subdermal space in four of the five species (except possibly for C. delitrix). From there, water enters into the largest incurrent canals, which branch into smaller and smaller canals leading to the choanocyte chambers (Fig. 6; Fig. S3). Callyspongia vaginalis is distinct from the other species in that the smallest incurrent canals empty into a common lacunar space that holds the choanocyte chambers as described by Johnston and Hildemann (1982) (Fig. 6A). Therefore, for C. vaginalis, openings between the choanocytes were considered the equivalent of prosopyles (Fig. 6B). In the other four species, water enters the choanocyte chambers from the smallest incurrent canal through one or more prosopyles. At the choanocyte chambers, water moves through the collar microvilli. In H. mollis, a set of cells forms a flat layer between all collars in the chamber (Fig. 6C). In both N. problematica and C. vaginalis, a dense mesh of glycocalyx occurs in the same region between each of the collars (Fig. 6B,D). Although a similar gasket has not yet been found in T. californiana and C. delitrix (Fig. 6E,F), this sort of structure may be more common in demosponges than has previously been appreciated because those made from mucus glycocalyx are difficult to preserve. A glycocalyx mesh was found between the microvilli of the choanocyte cells in each species studied (Fig. 6B), but in the case of T. californiana and C. delitrix it was only found in a few well-preserved choanocytes within a chamber. After passing through the glycocalyx mesh filter on the collar, the water flows into the chamber and from there out of the apopyle (exit of the sponge choanocyte chamber). In T. californiana, the apopyle consists of a sieve-plate (Fig. 6E); in other sponges, it is a circular aperture. From the apopyle, the water enters small excurrent canals that merge into increasingly larger canals before flowing out of the osculum.

Cross-sectional area, velocity of water and head loss for each region of the canal system calculated using the models of Riisgård and Larsen (1995) and Leys et al. (2011) are given in Table 3. In each species, the cross-sectional area increases as the water enters the choanocyte chambers (Table 3, Fig. 7). Velocity through each region (u_i) was calculated from total cross-sectional area of each region (i), A_i, and excurrent velocity out of the sponge osculum, u_ex, using Eqn 1. The effective velocity u_i through the collar slit of the two warm-water species C. vaginalis and C. delitrix was 0.018 and 0.017 mm s⁻¹, respectively, which is two to 10 times higher than the effective velocity in the temperate species (N. problematica=0.009 mm s⁻¹, H. mollis=0.006 mm s⁻¹, T. californiana=0.002 mm s⁻¹), a difference arising from the higher excurrent velocities of C. delitrix and C. vaginalis. The total head loss through the canal system – the sum of head loss through each region – is also five to 17 times higher for the tropical species (Table 3).

To determine the cost of filtration, the power required to pump water across the sponge, P_p, was calculated using the total head loss, ΔH, and volume flow rate, Q, and divided by the total measured oxygen removal, R_tot (Riisgård and Larsen, 1995), using the conversion 1 µl O₂ h⁻¹=5.333 µW (Riisgård, 1988) and the mean oxygen removal and volume flow rates reported in Table 1. Using the model by Riisgård and Larsen (1995), the estimates of the cost of filtration were quite variable, ranging from 1% for T. californiana to 15% for C. delitrix. The simplified model developed by Leys et al. (2011) gives the same relationship for cost of filtration between the five species of demosponge, although the range is slightly lower, from 1% for T. californiana and 12% for C. delitrix. Both models estimate a very similar cost of pumping for each species. Both of the warm-water species C. vaginalis and C. delitrix had the highest cost of pumping regardless of the model used (Table 3).

Our study aimed to assess the energetic cost of filtration in demosponges and to determine whether sponges reduce their cost of filtration by taking advantage of ambient currents. Of the five species of demosponges studied, the tropical species C. delitrix and C. vaginalis filtered the most water and extracted the most oxygen per gram of tissue. Because volumetric flow rate drives the model describing the energy required to overcome the resistance through the sponge aquiferous system, both tropical sponges have higher estimates of the cost of filtration than the three temperate demosponges N. problematica, H. mollis and T. californiana. Our data also show that the F/R ratio is constant for ambient flow velocity and therefore the cost of filtration depends only on the amount of water filtered. This suggests that the cost of filtration for demosponges varies by species and that reducing the volume of water filtered would reduce the cost of filtration in regions and seasons with lower food availability.

Interestingly, we found that demosponges can control the water flow through their bodies by responding behaviourally to changes in ambient flow. Both C. delitrix and C. vaginalis reduce the amount of water filtered at very high ambient velocities, which may be a mechanism to protect themselves from damage during storms. This reduction in the amount of water filtered at very high ambient velocities also suggests that neither species uses current-induced flow, despite the cost of filtration being higher than previously estimated.

Animals allocate energy to a variety of processes including growth, reproduction, feeding and digestion. For filter feeders, the energy allocated to feeding has been considered to be low, at 0.1–4% of total metabolism based on theoretical models (Jørgensen, 1955, 1986, 1988; Riisgård et al., 1993, 1988, 1989). In those studies, the cost of filtration was assumed to be equivalent to the energy lost owing to frictional resistance as water flows through the filter and canals. By changing filter size and volume flow rate using numbers gleaned from the literature, the estimate for the cost of filtration increased by up to five times previous estimates. This suggests that both filter dimensions and volume flow rates contribute substantially to the cost of pumping in filter-feeding invertebrates. It also suggests that accurate measurements of filter dimensions and volume flow rates are important when modelling the cost of filtration, and focused our attention on these for this study.

Although there were differences in gross morphology among the five species of sponge we studied, we found similar morphologies of the filter for each of the species studied. The cost of pumping was variable, ranging from 1% (T. californiana) to 12–15% (C. delitrix) of the total oxygen consumed, assuming oxygen consumed is used for all metabolism including filtration. Previous estimates on the energetic cost of pumping in demosponges also show a large range, from 0.4% for Haliclona urceolus (Riisgård et al., 1993) to 25% for Neogombata magnifica (Hadas et al., 2008). Although this variability may reflect the use of theoretical models (Riisgård et al., 1993) compared with direct measurements (Hadas et al., 2008), our results suggest that the main variability in ‘cost’ arises from differences in the volume of water processed.

Whether a single equation was used for all structures of the aquiferous system (Leys et al., 2011) or different equations were used to describe potentially different aspects (aperture, duct, channel and lattice network openings) of the passages (Riisgård and Larsen, 1995), the estimate of cost was very similar. Small differences in cost found between the two models reflect differences in the dimensions and the velocity of water at the aperture or structure. The greatest head loss occurred at the glycocalyx mesh filter, which had the smallest dimensions and therefore offered the greatest resistance (Table 3). And yet while the mesh size was smaller in T. californiana than in C. delitrix, the head loss was far greater across the collar filter of C. delitrix because the velocity of water at its filter was substantially higher. Velocity at each aperture was governed by the exit velocity of each species, and so the two tropical species, having the largest volumetric flow rates (processing the most water), had the highest cost of filtration (Table 3). In addition, when individuals of each species filtered more water, they also consumed more oxygen (Fig. 2), demonstrating that the energetic cost of pumping depends on how much water is being pumped at any one time. This implies that sponges could save energy in times of low food availability by reducing the volume of water filtered.

That filtering more water costs more energy is not necessarily intuitive because sponges might use ambient flow to enhance the movement of water through their canal system. Our data demonstrate that neither C. delitrix nor C. vaginalis uses higher ambient flow to reduce the cost of filtration. Instead, both sponges actively modify their behaviour to increase or reduce the volume filtered. However, how a sponge changes its pumping rate and volume filtered is not known. One mechanism might be to increase the rate of flagellar beat in the choanocyte chambers or activate more pumping units. This would increase the velocity of water through the canal system and therefore increase the volume of water filtered. Dilation of the canals including the osculum diameter might also occur in addition to the increase in flagellar beat, which would increase the volume of water filtered. According to the Hagen–Poiseuille equation for flow through a pipe, the energy required to move water through a pipe is proportional to the square of the flow rate; therefore, an increase in water actively pumped through the canal system at any one time would increase the resistance and thus also the energetic cost of pumping. Similarly, because head loss is inversely proportional to the canal diameter to the power of four, constriction of the canals would result in a reduced volume flow rate, an increase in head loss, or a combination of both. Sponges likely control the rate of flagellar beating to save energy when the sponge contracts. We did not look at the energy spent on the flagellar beating nor the drag on the flagellum when estimating the cost of pumping, which is a limitation in the model and probably makes our estimates more conservative.

It is important to consider why the two tropical species of demosponge had higher volume flow rates and therefore higher estimates for the cost of filtration in this study. Differences in pumping rates between species can be caused by structural differences in the sponge canal system (Reiswig, 1975a), microbial content (Weisz et al., 2008) and tissue density (Turon et al., 1997). In addition, Riisgård et al. (1993) found that volume flow rate in Haliclona urceolus increased up to 10 times with a change in temperature from 6 to 15°C. The higher volume flow rate per gram weight in the two tropical species in this study therefore may be a result of the higher temperature of water and lower viscosity. However, because of the high energetic cost of filtering more water, it would only be adaptive to have such high volume flow rates if there is enough food available – either as picoplankton to be filtered from the water or from endosymbionts – to support the high energetic needs. We did not measure carbon flux in these species but this is a planned next step.

Sponges are found in almost every marine and freshwater habitat and, with the exception of carnivorous sponges, all feed on ultraplankton and dissolved organic carbon. Increasing cross-sectional area through the canal system slows the velocity of water, enabling food capture either in the small incurrent canals or at the filter. Slight differences in filtration ability and size of plankton captured between different species of sponges are known (Reiswig, 1975b; Turon et al., 1997; Yahel et al., 2007). There is little information, however, on the relationship between filtration ability, diet and microarchitecture of the aquiferous canal system. Among the five species studied here, slight variations in the architecture of the aquiferous canal system were found, such as the lacunar space in C. vaginalis and the sieve-plate apopyle of T. californiana. We also found wide differences in the velocities of water measured out of the osculum, but similar velocities of water at the filter (Table 3). This suggests that despite wide differences in volume flow rates and adaptations to a variety of habitats and ecological niches, the canal system of sponges is designed to slow the velocity of water down to a certain speed that enables food capture at the filter. The small differences that were found in both velocities at the filter and filter dimensions may reflect small differences in preferred plankton size or filtration ability. A smaller mesh size may allow the sponge to capture smaller food particles but, because of the higher cost of filtration, may also restrict the volume of water processed. Our model predicted that both tropical species C. vaginalis and C. delitrix had faster velocities of water at the collar filter. Callyspongia vaginalis lacks the small incurrent canals leading into the choanocyte chambers, suggesting that it does not rely on pinacocyte capture of plankton in the canals.

One way to reduce the cost of pumping in sponges would be to take advantage of ambient currents by using passive flow. In both C. delitrix and C. vaginalis, excurrent volume flow rates were not correlated to the ambient currents, suggesting that these species do not use passive flow. Rather, both species decreased their volume flow rate when ambient velocities reached a certain level, which may be to reduce damage caused by high currents or resuspension of sediments during storms (Gerrodette and Flechsig, 1979; Tompkins-MacDonald and Leys, 2008; Bannister et al., 2012).

We suggest that the discrepancy in our results and those of Vogel, who first described the apparent use of induced current by sponges (Vogel, 1974, 1977), are due to his use of excurrent velocity rather than volume flow rate and the use of individual points in time rather than a full time series. The increased excurrent velocity recorded by Vogel (1977) in the species Amphimedon viridis (referred to as Haliclona viridis), Ircinia variabilis (referred to as Ircinia fasciculata) and Aplysina fistularis (referred to as Verongia fistularis) may have been behavioural responses, as shown here in C. delitrix in which contraction of the osculum caused increased excurrent (exit) velocity, but reduced volumetric flow rate as ambient currents increased. A full time series with measurements of the osculum diameter throughout would indicate whether Vogel's findings were actually a result of current-induced flow or rather a behavioural response to the increased current.

Other sponges have also been reported to show reduced excurrent flow with increased ambient flow. Savarese et al. (1997) noted that two individuals of the globose freshwater sponge Baikalospongia bacillifera in Lake Baikal monitored over a diel cycle showed negative correlations between ambient and excurrent flow rates. In addition, they noted periodic cessation of pumping on the order of minutes to hours that did not correlate with ambient flow. Reiswig (1971) also found periodic oscular closures and therefore cessations of pumping in Tethya crypta that were negatively correlated to wave action. Here, we found that both C. vaginalis and C. delitrix reduced the volume of water filtered when ambient currents reached a certain threshold. Therefore, some demosponges show a large amount of control of filtration despite fluctuations in the ambient water velocity.

The over-arching finding of this work is that the energetic costs of sponge pumping are closely coupled to volumetric flow (how much water is filtered). We found a considerable variation in flow rates and corresponding costs, with tropical sponges having higher volumetric flow rates and costs. The difference may be related to food availability and temperature. We also found that despite the variation in costs, the similarity in architecture and design of the aquiferous canal system is such that the velocity of water slows at the filter, although each species accomplishes this in slightly different ways, thus making them all well suited to feed on ultraplankton and dissolved organic carbon.

For the three temperate demosponges, our estimates for the cost of pumping are comparable to those found by Riisgård et al. (1993) for what he termed the ‘standard sponge’. However, the two tropical demosponges we studied had much higher volume flow rates with a much higher cost of pumping: the cost of pumping for C. delitrix was more comparable to the cost of pumping for other sponges found by both Leys et al. (2011) and Hadas et al. (2008). It is likely that the habitat and ecological niche of sponges have led to adaptations in body form and physiology over time, and together these play a large role in differences in cost of pumping. Our third main finding was that despite the higher costs for the two tropical sponges, ambient flow is not a positive driver of flow through the sponge and is therefore not used to reduce the cost of filtration. Instead, both of these sponges responded to increased ambient currents by reducing the volume filtered. In the shallow sandy habitat of both C. delitrix and C. vaginalis, increased ambient flow results in resuspension of particles. Demosponges have sensory cilia in their osculum, which allow them to sense and respond to changes in their environment (Ludeman et al., 2014). Reiswig (1975a) showed that several species of tropical demosponge responded to storm events by reducing filtration, and so it is likely that the demosponges we studied here respond to changes in ambient currents to reduce damage to their filter that would be caused by intake of sediments resuspended during high ambient currents.

where u_ex is the velocity of water leaving the osculum. The total head loss through the system is equal to the sum of the head losses at each region of the canal system.

As a comparison to the model by Riisgård and Larsen (1995), we also used the simplified model developed by Leys et al. (2011) which assumes head loss through each region of the aquiferous canal system can be calculated based on the Hagen–Poiseuille equation for fully developed laminar flow in a tube (Eqn A2).

Here, Q is the volume flow rate through the system and ΔH is the total pressure drop, or head loss, through the system.

We thank the directors and staff at the Bamfield Marine Sciences Center, Bamfield, British Columbia, and the Smithsonian Tropical Research Station, Bocas del Toro, Panama, for use of facilities; N. Lauzon and N. Trieu for assistance with field work and histology, respectively; and A. Kahn and G. Yahel for valuable comments on an earlier version.