Differences in ontogenetic and diurnal microhabitat selection by sympatric live- bearing fish species with different reproductive modes

1. A pregnancy imposes a heavy reproductive burden on females. Some live- bearing species have evolved reproductive adaptations to reduce this burden, which may influence their ability to use specific microhabitats. 2. We investigate whether two such reproductive adaptations, placentation (embryo provisioning via a placenta) and superfetation (the ability to carry multiple broods at various developmental stages), influence microhabitat selection by five sympatric Costa Rican live- bearing fish species (family Poeciliidae). Theory predicts that placentation and superfetation should both reduce the reproductive burden of females during pregnancy, improve their body streamlining, and swimming performance, and consequently allow them to use more performance- demanding microhabitats. Here we apply underwater visual fish surveys to


| INTRODUC TI ON
Microhabitat selection affects community assembly and structure in response to a variety of biotic and abiotic gradients (e.g. competition, predation risk, parasites, resource availability; Morris, 2003;Rosenzweig, 1991). It enables the use of a different set of local resources by different species, minimising competition with cooccurring species (Bolnick et al., 2003(Bolnick et al., , 2007. Microhabitat selection results from optimisation processes involving individual fitness costs and benefits (Sibly & McFarland, 1976). For instance, in environments where patches vary in food availability, optimal foraging theory predicts that individuals should prefer habitats with high food density (MacArthur & Pianka, 1966). However, when these patches are simultaneously subject to high predation risk, individuals must balance the conflicting demands of food acquisition and predator avoidance (Sih, 1980). Different species may balance these conflicting demands (feeding vs. predator avoidance) in different ways, resulting in species-specific differences in microhabitat selection and affecting the spatial and temporal dynamics of species interactions (Morris, 2003). Microhabitat selection is therefore an important mechanism that allows for the competitive coexistence of species (Rosenzweig, 1981).
Microhabitat selection not only differs among species but can also vary within species. For instance, microhabitat use can change throughout an individual's life depending on their age, size, or developmental stage (i.e. ontogenetic microhabitat shifts). Sizedependent predation risk was shown to account for differences in habitat use between bluegill sunfish size classes (Lepomis macrochirus), with large bluegills preferring open-water habitats with high foraging returns and small bluegills choosing less profitable habitats close to vegetation to avoid predation (Mittelbach, 1981;Werner & Hall, 1988). Likewise, armoured catfish (Loricariidae) display sizespecific spatial distributions, with small catfish preferring shallow water to reduce predation risk by piscivorous predators while large catfish avoid shallow water where they are vulnerable to avian predators (Power, 1984). In coastal lagoons, many fish species display size-related shifts from settlement habitats (larvae) to nursery areas (juveniles) and subsequently to nearby coral reefs (adults). These ontogenetic shifts in (micro)habitat use may be related to ontogenetic changes in dietary preferences or susceptibility to predation (Werner, 1984;Cocheret de la Morinière et al., 2002; Cocheret de la Morinière, Pollux, Nagelkerken, Hemminga, et al., 2003; Cocheret de la Morinière, Pollux et al., 2007).
Many species furthermore display pronounced day-night shifts in habitat use (diurnal microhabitat shifts). Well-known examples include the diurnal vertical migration of freshwater and marine pelagic invertebrates from deeper water during the day to avoid visually orienting predators, to shallower surface waters at night to feed (Bollens & Frost, 1989;Bollens et al., 1992;Kaartvedt et al., 2007;Sainmont et al., 2013). Juvenile Atlantic salmon (Salmo salar L.) show a temperature-dependent shift in diel activity and habitat use, which is probably the result of maximising feeding efficiency in summer, and reducing predation risk in winter (Fraser et al., 1993). The Eurasian lynx (Lynx lynx) selects open habitats at night where prey is more abundant and shifts to dense understory cover and rugged terrain during the day to avoid human activity (Filla et al., 2017). Such diel differences in activity and microhabitat use are often the result of a complex trade-off between feeding and avoiding predators (Fraser et al., 2004;Metcalfe et al., 1999;van der Vinne et al., 2019).
Finally, microhabitat selection can also change in individuals depending on their reproductive state (reproductive microhabitat shifts). Gravid and pregnant females, for example, may temporarily select low performance-demanding microhabitats: for example, areas where they are less prone to predation or, in stream ecosystems, where they are less exposed to strong currents. The reason is that gravidity and pregnancy may negatively affect the locomotor performance of females Noren et al., 2011;Plaut, 2002;Seigel et al., 1987), increasing their susceptibility to predation (Pires et al., 2011;Pollux et al., 2009). This risk is compounded by the fact that predators often prefer gravid or pregnant females (Trexler et al., 1994), because they are large and (due to the presence of eggs or embryos) represent a high-quality resource that is rich in energy and nutrition. To reduce predation risk, gravid female common lizards (Lacerta vivipara), for instance, strongly rely on crypsis and remain motionless in the immediate vicinity of hiding places (Bauwens & Thoen, 1981). Moreover, gravid female three-spined stickle-backs (Gasterosteus aculeatus) remain closer to refuges than non-gravid females when inhabiting habitats with predators (Rodewald & Foster, 1998). Similarly, the Trinidadian guppy (Poecilia reticulata) uses habitats with lower water velocity late in pregnancy to offset some of the performance-related costs of pregnancy (Banet et al., 2016). Thus, when gravidity or pregnancy are associated with increased vulnerability to high performance-demanding conditions, this is likely to induce a shift in habitat use to less performancedemanding microhabitats.
Here, we study microhabitat selection (Allee et al., 1949)  that co-occur in rivers and streams in Costa Rica. We quantify ontogenetic and diurnal microhabitat preferences between sympatric live-bearing fish species living in environments characterised by large flow variation.

K E Y W O R D S
habitat selection, matrotrophy, placenta, Poeciliidae, superfetation differences in diurnal, ontogenetic, and reproductive microhabitat use among these species and their piscivorous predator (Gobiomorus maculatus). These five co-occurring poeciliid fish species are of similar size and generally prefer similar environmental conditions (Bussing, 2002;Meyer, 2015). However, they differ in the absence/presence of two reproductive adaptations (Table 1). The first reproductive adaptation is the placenta: P. turrubarensis, P. gillii, and B. roseni are lecithotrophic (or yolk-feeding), committing all nutrients required for development during pregnancy to the eggs prior to fertilisation, while P. retropinna and P. paucimaculata are placentotrophic (i.e. mother-feeding) provisioning nutrients to the developing embryos throughout pregnancy via a placenta (Furness et al., 2019;Pollux et al., 2014). The second reproductive adaptation is superfetation: P. retropinna, P. paucimaculata, and P. turrubarensis have superfetation, which means that they are able to carry multiple broods at different developmental stages, while P. gillii and B. roseni lack superfetation (Pollux et al., 2009).
Theory predicts that (different combinations of) these two reproductive adaptations may be associated with differences in microhabitat use. The reason is that the presence of the placenta and superfetation are both thought to reduce a female's reproductive burden during pregnancy (Bassar et al., 2014;Furness et al., 2021;Hagmayer et al., 2020;Pollux et al., 2009;Reznick et al., 2007). A reduced reproductive burden has further been associated with enhanced body streamlining and improved locomotor performance of females during pregnancy (Fleuren et al., 2018Pires et al., 2011;Pollux et al., 2009;Quicazan-Rubio et al., 2019;Thibault & Schultz, 1978;Zúñiga-Vega et al., 2010), potentially allowing the use of different stream (micro)habitats (Banet et al., 2016). Here, we test a key prediction of this hypothesis, namely that the two reproductive adaptations should be associated with a species' microhabitat use. Specifically, (1) we predict that reproductive adults of species with a placenta and/ or superfetation will inhabit relatively deeper, faster-flowing sections in the middle of the river compared to species that lack both reproductive adaptations. (2) Furthermore, we know from preliminary nocturnal observations (Hagmayer, Furness, & Pollux, per-sonal observations) that poeciliid species become inactive at night and tend to move to the shallows to rest. We therefore predict that if adults of placental species with superfetation indeed inhabit deeper, faster-flowing parts of the river during the day (see prediction 1), then they should show a far more pronounced diurnal shift at dusk towards shallower, slow-flowing microhabitats compared to species that lack both reproductive adaptations. (3) Finally, we expect that juveniles and immatures of all five species tend to avoid the faster-flowing sections of the river, because their swimming abilities are still limited (Dial et al., 2016;Lankheet et al., 2016). If true, then these younger ontogenetic stages (juveniles and to a lesser extent immatures) should display a similar habitat selection regardless of the species, with all preferring shallow, slow-velocity areas near the riverbank. By comparing microhabitat use of these five sympatric live-bearing fish species, our study provides new insights into the potential effects of reproductive adaptations on microhabitat selection and local diurnal and ontogenetic migration.
Furthermore, although adult female B. roseni are generally smaller than the females of the other study species, female body size does not differ between P. retropinna, P. paucimaculata, P. turrubarensis, and P. gillii (Methods 1.1 in Supporting Information; Table S1; Figure   S1). However, the study species differ in the absence/presence (or degree) of two reproductive adaptations: placentation and superfetation (Table 1). The degree of placentation is quantified as the ratio of offspring mass at birth to egg mass at fertilisation, also referred to as the Matrotrophy Index (MI; Pollux et al., 2014;Reznick et al., 2002). Some live-bearing species, known as lecithotrophs (yolk-feeding), allocate all resources to eggs prior to fertilisation in the form of large fully-yolked eggs. Embryos subsequently lose dry mass over the course of gestation due to metabolic processes. Such species have an MI less than 1. Other live-bearing species, known as matrotrophs (mother-feeding), allocate nutrients to the developing offspring post-fertilisation throughout pregnancy. Such species TA B L E 1 Summary of the reproductive modes among the study species have an MI greater than 1, indicating that embryos gain dry mass during pregnancy. Placentotrophy represents one specific type of matrotrophy that is achieved through a follicular placenta, roughly an analogue to the mammalian placenta (Pollux et al., 2009). The degree of superfetation is the number of broods at various developmental stages that are carried by a female (Turner, 1937). Females with superfetation tend to produce smaller broods, but do so more often (Reznick & Miles, 1989).
Poeciliopsis retropinna and P. paucimaculata are both characterised by superfetation and post-fertilisation maternal provisioning (i.e. matrotrophy). P. retropinna females carry up to four broods at various developmental stages and offspring increase in dry mass more than 100-fold during pregnancy (MI = 117; Reznick et al., 2002 (Tables 2 and 3). Underwater visibility was high at all study sites. We recorded the occupancy (i.e. presence or absence) of juveniles, immatures, and adults of all study species including G. maculatus during daytime in each quadrat of each transect. Specifically, fish occurring in deep-water transects in the middle of the river were identified by means of underwater visual census while snorkelling (Pollux et al., 2007). The snorkeller began at the downstream end of the transect and slowly worked his way upstream, metre by metre, while recording the occupancy (i.e. presence or absence) of each species and size class after completing each meter mark ( Figure S2). In very shallow transects (too shallow to snorkel), fish were instead identified from above while standing or sitting on the shore. Poeciliid fish can be closely approached by a snorkeller without being disturbed (e.g. without scaring them away or altering feeding behaviour). All individuals were classified into three categories based on their ontogenetic stage: adults, defined as large (potentially pregnant) females and mature males (fully developed gonopodium present); immatures, defined as small (nonpregnant) females and males that did not have fully developed gonopodia; and juveniles, defined as fish <2 cm.

| Underwater visual census
Immediately following the daytime census, we measured the (1) water depth to the nearest cm by using an aluminium metre stick and (2) water velocity to the nearest 0.01 m/s with a Höntzsch Vane Wheel FA current meter (type ZS30 GFE md20 T/100-2/p10, Höntzsch Instruments) three times separately in the centre of each 1 × 1 m quadrat. In quadrats where the water depth exceeded 60 cm, the water velocity was calculated as the average between the velocities measured at 0.2 times the water depth and 0.8 times the water depth in the centre of the quadrat (Hauer & Lamberti, 2007). This ultimately yielded a mean water velocity for each 1 × 1 m quadrat, calculated as the average of the three (or six if water depth exceeded 60 cm) repeated measurements in its center.
Finally, to quantify diurnal (day-night) shifts in microhabitat use, each transect was censused a second time that same night (as described above; Figure S2). For this, we used a 2000 Lumen ThorFire S1 underwater lamp to identify fish occurring in deep-water transects while snorkelling and headlamps for identifying fish in the

| Data analysis
The habitat preference by the different species was quantified using the Bayesian programming environment JAGS (Plummer, 2003) in R v 3.5 (R Core Team, 2020).
For this, we quantified (1) the occupancy probability of a given age-class of a species as a function of water depth and velocity, as well as (2) the preferred mean water depth and velocity for a given age-class of a species.
For (1), the occupancy (i.e. presence or absence) of a given ageclass of a species per quadrat was fitted in a Bernoulli generalised linear mixed model as a function of the three-way interaction between water depth, age, and day-night cycle, and the three-way interaction between water velocity, age, and day-night cycle. Additional variables included the second order polynomials of water depth and velocity.
The model estimates species-specific slopes on each of the parameters and quadrat-, transect-, and site-specific random intercepts to account for pseudo-replication and between-transect/site variation, respectively. In the case of G. maculatus, information about age is not available. Thus, occupancy per quadrat was fitted as a function of the two-way interaction between water depth and day-night cycle, and the two-way interaction between water velocity and day-night cycle.
The quadrat, transect, and site identity were fitted as additional intercepts (see above), and the second order polynomials of water depth and velocity as additional slopes.

| Ontogenetic microhabitat use during daytime
During the day, adults of all species occupied deeper, faster-flowing water than immatures, and immatures occupied deeper, fasterflowing water than juveniles (Table S2; Figure 1: left panels, Figure 2).  Figure 1: left panels).

| Ontogenetic microhabitat use disappears at night
At night, the ontogenetic microhabitat use observed during daytime disappears. At night fall, all age-classes, regardless of the TA B L E 3 Study locations, mean and range of water depth (m) and velocity (m/s), number of quadrats (n q ), and percentage of occupied quadrats at day and night at each site species, either remain in or move towards the shallow river shore where they congregate in the slow-flowing waters presumably to sleep (Table S2; Figure 1: right panels, Figure 2). As a result, adults of species that have a placenta and superfetation (P. retropinna and P. paucimaculata) show a far more pronounced diurnal (daynight) migration towards shallower, slow-flowing microhabitats compared to species that lack both reproductive adaptations (P.

| Ontogenetic (age-related) microhabitat use during daytime
We found that adults generally tend to prefer faster-flowing and deeper water during the day than immatures, and that immatures F I G U R E 1 (a) Mean water depth and (b) velocity (±95% posterior density confidence interval) occupied by a given age-class of a species during day and night. Dotted line corresponds to a linear fit throughout the posterior samples of a given species at day or night, respectively. J: juvenile, I: immature, A: adult. red/orange: placental species with superfetation; green: non-placental species with superfetation; blue: nonplacental species without superfetation prefer faster-flowing and deeper water than juveniles. This is probably due to the positive association between fish size (and thus, indirectly, age or ontogenetic stage; Reznick et al., 1996) and swimming capability in Teleost fish (Gibb et al., 2006). Newborn poeciliid fish have relatively poor swimming abilities (Dial et al., 2016;Lankheet et al., 2016). They are physically incapable of maintaining their position in fast-flowing water and thus seek refuge in low-velocity areas of the river near the shore. An added benefit for these small juveniles is that here they can inhabit very shallow water (sometimes less than 1 cm deep) where they are out of reach of piscivorous predators. The subsequent growth of fish is associated with physical ability and better control of the locomotor system (Fuiman & Higgs, 1997;Gibb et al., 2006;Lankheet et al., 2016), leading to an increase in swimming ability with increasing body length (Gibb et al., 2006). This allows larger and physically stronger adults to inhabit the relatively faster-flowing parts of the river.
The above may explain why only adult individuals are physically able to swim in the fast flow. However, it does not explain why some would want to move to the fast-flowing parts of the river in the first place, considering that there they are exposed to high drag forces (Quicazan-Rubio et al., 2019) making it energetically expensive for them to swim. We tentatively propose that some adults may move to deeper water to avoid avian predation. Together with various other piscivorous birds (e.g. herons), the southern Pacific region of Costa Rica is inhabited by four species of kingfisher (Garrigues & Dean, 2014), which mainly feed on fish (Fry et al., 1999). Piscivorous birds are very effective predators in shallow shore water or close to the surface (Kramer et al., 1983;Whitfield & Cyrus, 1978), where they show a preference for larger fish (Power, 1984;Trexler et al., 1994). This means that large adult poeciliids are more vulnerable to avian predation in shallow water than small juveniles. At the same time, piscivorous fish are presumably feeding in deeper water to prevent avian predation (Power, 1984). Indeed, we found that the piscivorous predator G. maculatus inhabits relatively deeper water during the day (i.e. when most piscivorous birds are active in the shallows).
It has furthermore been shown that larger prey fish are able to perform faster escape responses, making them less vulnerable to predation by piscivorous fish (Gibb et al., 2006). Thus, one could argue that the mortality risk of live-bearing fish during daytime is highest for adults in shallow habitats from piscivorous birds, and highest for juveniles in deep habitats from piscivorous fish (Power, 1984;Schlosser, 1988). If true, then it would be advantageous for juveniles to inhabit shallow waters near the shore, while it would be beneficial for larger adults to move to deeper water as they outgrow the vulnerability to piscivorous fish (Power, 1984).
Collectively, these findings suggest that the observed ontogenetic microhabitat preference in live-bearing fish during daytime may be an adaptive response to predation risk.

| Diurnal (day-night) shifts in microhabitat use
Many fish exhibit diurnal shifts in microhabitat use (Helfman, 1986;Lowe-McConnell, 1975), which have been attributed to shifts in foraging activity (Piet & Guruge, 1997), the use of shallow water as refuge from predation (Arrington & Winemiller, 2003;Copp & Jurajda, 1993), or the use of slow-flowing areas to reduce energy expenditure while resting at night (Matheney IV & Rabeni, 1995;Sempeski & Gaudin, 1995). In our study, we found that at dusk all fish, regardless of their ontogenetic stage (juvenile, immature, or adult), tend to move to shallow waters near the shore, where they sleep lying on the bottom in low-velocity areas. Occasionally, large adult individuals (particularly P. retropinna and P. paucimaculata) can be found sleeping while wedged into crevices or behind stones in deeper and faster flowing stretches of the river. Interestingly, G. maculatus, the most common piscivorous fish species in our study sites, is primarily a nocturnal sit-and-wait bottom predator (Swing, 1992). The finding that this piscivorous predator also moves towards shallow water at night, suggests that the observed diurnal microhabitat shifts in poeciliid fish towards shallow water are not related to predator avoidance. Instead, it is more likely that the day-night shifts are driven by a preference for low-velocity areas in the river to avoid being washed away while resting at night.

| Differences in diurnal and ontogenetic microhabitat use among poeciliid species with different reproductive adaptations
All five studied poeciliid species occur sympatrically in freshwater streams in Costa Rica (Bussing, 2002), yet show remarkable differences in ontogenetic and diurnal microhabitat use. This appears to be correlated with the absence/presence of the two reproductive adaptations. During daytime, adult P. retropinna and P. paucimaculata This raises the question why the adults of P. retropinna and P.
paucimaculata, and to a somewhat lesser extent of P. turrubarensis are found in fast-flowing water during daytime? It has been argued that the evolution of placentation and superfetation both reduce a female's reproductive burden during pregnancy, yet achieve this in fundamentally different ways (Pires et al., 2011;Pollux et al., 2009;Thibault & Schultz, 1978). The evolution of the placenta is associated with a shift in the timing of maternal provisioning from pre-to post-fertilisation. Non-placental live-bearers (e.g. P. gillii, B. roseni, and P. turrubarensis) typically produce large fully yolked eggs, committing all the nutrients required for embryo development to the egg prior to fertilisation. Placental species (P. retropinna and P. turrubarensis), by contrast, produce relatively small eggs and instead provide most nutrients to their offspring throughout pregnancy via a placenta (Pollux et al., 2009;Wourms, 1981). The shift in the timing of maternal provisioning from pre-to post-fertilisation reduces a female's reproductive burden (Bassar et al., 2014;Fleuren et al., 2018;Reznick et al., 2007). The evolution of superfetation furthermore correlates with the more frequent production of smaller broods (Reznick & Miles, 1989). By spreading reproduction more evenly over time, superfetation is thought to reduce a female's peak reproductive allotment during gestation without reducing maternal fecundity (Pollux et al., 2009). Thus, placentation and superfetation are both thought to reduce a female's reproductive burden during pregnancy (Bassar et al., 2014;Hagmayer et al., 2020;Pires et al., 2011;Pollux et al., 2009;Reznick et al., 2007;Thibault & Schultz, 1978). This is likely to cause a more slender body shape (Fleuren et al., 2018Zúñiga-Vega et al., 2007), reduced body drag (Quicazan-Rubio et al., 2019), as well as improved sustained swimming performance (Plaut, 2002) and fast-start escape response Ghalambor et al., 2004). Thus, one might argue that placentation and superfetation are reproductive adaptations that facilitate the use of high performance-demanding microhabitats (e.g. high-flow areas) in the river. Our study provides the first empirical evidence in support of one aspect of this hypothesis, namely the idea that placentation and superfetation are reproductive adaptations that can drive differences in ontogenetic, diurnal, and reproductive microhabitat use between sympatric live-bearing species.

| Other potential causes of microhabitat use
Although our findings are consistent with the idea that placentation and superfetation may shape microhabitat selection in our study species, we must entertain the notion that other factors, besides these two reproductive adaptations, could also potentially influence the observed interspecific differences in microhabitat use. For instance, larger-bodied species generally have a higher swimming capability (Gibb et al., 2006) and are therefore more likely to inhabit faster-flowing water. If co-occurring sympatric species that share similar dietary preferences have different body sizes, then spatial niche segregation (e.g. with the larger-bodied species using fasterflowing parts of the river) may reduce interspecific resource competition (Lanza, 1983 shape (e.g. increase in abdominal distention and wetted surface area), significantly increasing body drag and negatively influencing swimming performance (Fleuren et al., 2018Plaut, 2002;Quicazan-Rubio et al., 2019). By contrast, males are generally more streamlined than pregnant females (Quicazan-Rubio, 2019).
If microhabitat selection was solely a function of body shape, then we should expect to find spatial segregation of the sexes within streams, with encumbered pregnant females occurring in shallow low-flow areas and the more streamlined males in the deeper faster-flowing parts of the river. We did not separately quantify microhabitat use of the sexes; however, our snorkelling observations revealed that within each species the adult males closely co-occur with adult females. So why then do males occupy similar microhabitats as females, given that they are not affected by any reproductive burden? We propose that females may be the main driver of the observed interspecific microhabitat selection and males simply follow females. Indeed, in many poeciliid species, it has been shown that males obsessively follow females in a persistent attempt to mate with them (Magurran, 2011).

| CON CLUS ION
We report large differences in adult daytime microhabitat use between five sympatric poeciliid species, presumably associated with two reproductive adaptations (placentation and superfetation) that both improve body streamlining and swimming performance during pregnancy (Fleuren et al., 2018. The presence of these adaptations may explain, at least in part, why adult P. retropinna and P. paucimaculata, and to a lesser extent P. turrubarensis, inhabit deeper, faster-flowing areas during daytime, while B. roseni and P. gillii are more confined to the shallow waters near the shore. The finding that, at night, all fish (regardless of species or age-class) move to shallow, low-velocity areas to rest, lends additional support to this idea. Collectively our results suggest that a female's reproductive strategy (i.e. placentation and superfetation) may be a hitherto unrecognised biological feature that may help to understand microhabitat preferences between sympatric live-bearing fish species. Our study can be seen as a first step on which future, ideally experimental, studies can build to assess the costs of locomotion as a function of reproductive mode and pregnancy state. Future studies should furthermore focus on comparing microhabitat use in more live-bearing fish (e.g. from the family Poeciliidae, Anablepidae, Goodeidae, or Zenarchopteridae), but also other aquatic live-bearing animals (e.g. amphibians, reptiles, and mammals), to assess the generality of these findings.

ACK N OWLED G EM ENTS
We