Fish larvae were collected from nearshore waters off Montemar (32°57′S, 71°33′W) and El Quisco (33°24′S, 71°43′W), central Chile. During the austral spring of 2011-2013, zooplankton samples were collected using a Bongo net (60 cm mouth diameter, 300 μm mesh size), equipped with a TSK flow meter (The Tsurumi-Seiki Co., Ldt. Tsurumi.Ku, Yokohama, Japan) to quantify the filtered seawater. Tows were done from surface to near-bottom depths (~20 m) during the dawn and night hours (1900 to 2300 h) at 1 nautical mile from the coastline. Five to 8 trawls at 1-2 knots were conducted during each oceanographic sampling at the same location. The seawater filtered by the net ranged from 13.3 to 437.4 m³ (mean ± standard deviation; 141.2±102.5 m³). All plankton samples (n=314) were initially fixed in 5% formalin buffered with sodium borate, and after 12 h they were preserved in 96% ethanol.

Additionally, juvenile fish were collected with hand nets in the intertidal zone off Las Cruces (33°29′S; 71°37′W) during April 2015. Individuals were immersed in sea water with benzocaine in excess for a bioethical death, and later preserved in 96% ethanol.

All fish larvae were sorted in the laboratory from the plankton samples. Identification of larval S. sanguineus followed Pérez (1981). Individuals were measured under an Olympus SZ-61 stereomicroscope with a Moticam 2500 video camera (5.0MPixel) connected to a PC with Moticam Image Plus 2.0 software. Later, larvae were divided into four size groups: <8 mm, 8-12 mm, 12-16 mm and >16 mm standard length (SL). Larvae measurements were not corrected for shrinkage. Juveniles were measured with a caliper after fixation in ethanol.

A set of 9 landmarks were digitized in lateral views of the 159 larval and juvenile fish (Fig. 1) following suggestions of Farré et al. (2016), using TpsDig (version 2.17). Shape information was extracted from the landmark coordinates with a generalized Procrustes analysis (GPA; Dryden and Mardia 1998), using MorphoJ (version 1.05f, Klingenberg 2011). GPA removes non-shape variation by scaling, translating and rotating all specimens to one consensus configuration (Rohlf and Slice 1990). Then the shape coordinates can be used to visualize and perform any type of statistical analysis (Kerschbaumer and Sturmbauer 2009). Residuals from this superimposition were analysed with the thin-plate spline interpolating function (Bookstein 1991). The digitizing procedure was repeated for 25 randomly selected cases, each specimen was slightly rotated and photographed twice and Procrustes analysis of variance (Procrustes ANOVA) was carried out to assess the relative amounts of variation among individuals and of measurement error using MorphoJ software (Klingenberg 2011).

To quantify the allometry, a regression of shape represented by Procrustes coordinates (landmark coordinates after Procrustes superimposition representing the shape of an object; Mitteroecker and Gunz 2009) on size, represented by centroid size (CS) was performed in MorphoJ software.

Principal component analysis was performed with the shape coordinates to identify the main axes of shape change and the specific changes in the larval body (reflected as movements of landmarks) that explain the variability of the data. For the visualization of the shape changes, a warping transformation grid was used, following Klingenberg (2013), in addition to classical transformation grids (Russo et al. 2007, 2009).

The same 159 larval and juvenile fish were utilized for gut content analysis. The gut of each larva was dissected from the body and opened length-wise with fine needles. Prey items were counted and identified to the lowest possible taxon under a Motic BA310 microscope at 10×-40× magnification. The maximum SL and width of each prey item (maximum prey width, MPW) were measured with the microscope and a Motic Moticam 2500 camera (resolution 5.0 MPixel) using the Motic Image Plus 2.0 software. The volume of each prey item (i.e. copepods, nauplii, etc,) was estimated using the three-dimensional shape that most closely resembled the item, following Cass-Calay (2003) and Sun and Liu (2003). The prosome length of copepodite prey was measured because the urosome was often missing.

The feeding incidence (FI) was calculated as a percentage of larvae with any gut content, in relation to the total number of larvae examined (Sassa and Kawaguchi 2004). Comparisons of FI among size ranges were carried out with contingence tables.

The diet was described using the percent frequency of occurrence (%F) of a diet item in larvae with food in their guts, the percent of the total number (%N) of diet items that were examined and the percent of volume (%V) of each item out of the total volume of prey items. An index of relative importance (IRI) was calculated as follows: IRI=(%N+%V)×%F. To readily allow comparisons among prey items, the IRI was the standardized to %IRI for each prey item (Cortés 1997).

To establish the variability in the feeding success of S. sanguineus, three measures of feeding success were calculated: the number of prey items per gut (PIPG, number), total prey volume per gut (TPVG, mm³) and MPW (µm) (Reiss et al. 2002, Landaeta et al. 2011, 2015). Because these indicators of feeding success may vary with larval size, one-way ANCOVA was utilized for inter-size comparisons, using SL as covariate.

Pearre’s trophic niche breadth (Pearre 1986) was adopted to analyse the relationship between prey size and predator size. This model uses the standard deviation (SD) of the log₁₀-transformed prey size as a measure of trophic niche breadth. In this analysis, clingfish larvae and juveniles were classified according to body length at 0.5-mm intervals. Only fish with more than two prey items in the gut were used for further analysis. The mean and SD of the log₁₀-transformed prey width was calculated for each available size class of clingfish. The relationship between SL and the corresponding mean and SD of the log₁₀-transformed prey size was examined using linear regression analysis to determine any shifts in niche breadth with SL.

To establish the differences in the composition of prey among size ranges, a multivariate approach was utilized. The prey composition (number of prey) of each larva was utilized for the generation of Bray-Curtis similarity matrices. A permutational multivariate analysis of variance (PERMANOVA, Anderson 2001), which is analogous to a univariate analysis of variance, was utilized to determine the differences in prey composition among size ranges. Finally, similarity percentage analyses (SIMPER) were carried out to explain aspects of the similarity between diet and larval size ranges.

To determine the covariation pattern between shape and diet along the early development of S. sanguineus, a partial least squares (PLS) analysis was performed. PLS is a method for exploring patterns of covariation between two blocks of variables and can be used to analyse the relationship between form and function (Zelditch et al. 2012). Additionally, the number of variables can greatly exceed the number of cases, making PLS useful for discriminating between groups when there are far more shape variables than specimens. This kind of analysis has been used before to study the covariation between shape and diet changes (Klingenberg and Ekau 1996, Russo et al. 2008). PLS analysis were performed, using one block as the shape coordinates (after Procrustes analysis) and other block of diet composition (prey number or prey volume). These analyses were done to determine coupling or independence of shape and diet changes in early stages of clingfish.

Procrustes ANOVA indicates that measurement error was 1.91% (Table 1). The relationship between CS and SL was well expressed by a linear model (R²=0.979, P<0.01), and CS is therefore a good proxy of the individual length. The first three principal components of shape explained most of the variance (79.54%) of the shape changes during early ontogeny of S. sanguineus (Fig. 2 and Table 2). PC1 (47.62% of the total Procrustes form variance) expresses a lengthening of the visceral cavity (given by a noticeable displacement in the horizontal axis of the LM3, corresponding to the anus) during the formation of the sucker, as well as a shortening of the tail and a lengthening of the head. The factor loadings for the landmarks representing these shape changes were the greatest in magnitude in this PC (anus, 0.626; posterior body extremity, –0.474; posterior extremity of supraoccipital bone, 0.456; Fig. 3A). PC2 (24.68%) shows a dorsal displacement of the tail as well as changes in the body height, with an increase in the height at the level of the anus and a flattening of the head (factor loadings of posterior body extremity, 0.454; anus, –0.422; insertion of operculum, –0.442; Fig. 3B). The PC3 accumulated low variance (7.24%), showing a lengthening of the head caused by an increase in the distance between the premaxilla tip and the opercle insertion as well as a lengthening of the premaxilla (factor loadings of insertion of operculum, 0.725; tip of the premaxilla, –0.427; Fig. 3C).

The regression between Procrustes coordinates and CS was significant (P<0.001) and allometry accounted for 31.69% of the shape change during the transition of a pelagic larval to a benthic juvenile (Fig. 4). The allometric growth occurred early in the ontogeny of S. sanguineus mainly between 4 and 12 mm SL. After that size, the shape change was comparatively low (Fig. 4).

The FI was high throughout the SL analysed, ranging between 80% and 94.12% (Table 3). No significant differences in the FI were detected among size groups (χ²=0.974, P=0.807), suggesting that the foraging was independent of the larval fish size.

A total of 43 prey items were identified in the gut contents of larval and juvenile S. sanguineus, showing a specialization for each size group (Table 4). In the smaller larvae (<8 mm SL), the diet was dominated by copepod nauplii (42.45% %IRI), an item with high values of %N (29.36%), %V (15.31%) and %F (62.20%). Other important prey items in this size range were cirriped nauplii (16.79% %IRI) and dinoflagellate cysts (13.85%). Larvae between 8 and 12 mm SL also preyed mainly on copepod nauplii (31.85%) and cirriped nauplii (23.95%); however, in this size range, individuals reduced the ingestion of phytoplankton down to 8.34% %IRI, with a %V of 2.04% (Table 4). In this size group, fish eggs of at least two pelagic species, anchoveta Engraulis ringens and mote sculpin Normanichthys crockery, appeared as volumetrically important items (~43%). Specimens between 12 and 16 mm SL showed a preference for adult calanoid copepods, such as Paracalanus indicus (39.44% %IRI), gasteropod larvae (14.91%) and copepod nauplii (9.42%). The frequency of phytoplankton in the gut contents was low (11.76% %F) and almost irrelevant as a prey item (0.48% %IRI; Table 4). On the other hand, juvenile, completely transformed fishes (>16 mm SL) had prey items largely digested (61.98% %IRI) and red (25.49%) and brown (3.88%) macroalgae (Fig. 5), showing an important change in the prey items, varying from pelagic to benthic prey.

PIPG did not vary significantly among size groups (one-way ANCOVA, F=0.27, P=0.89, Fig. 6A), but TVPG did (F=11.61, P=6.65×10^–7, Fig. 6B). The largest differences in TVPG occurred between individuals <8 mm SL and >16 mm SL (F=31.39, P=2.61×10^–7). Finally, one-way ANCOVA indicated no significant differences in the maximum prey among size groups (F=2.58, P=0.051, Fig. 6C).

The regression between SL and SD of log-transformed MPW was positive and significant (R²=0.418, P<0.05), and the trophic niche breadth was 0.201±0.07 (Fig. 7). This result means that the trophic niche increases with larval length and the individuals of S. sanguineus are capable of ingesting wider prey as they grow.

One-way PERMANOVA detected significant differences in the diet composition among size groups (pseudo-F=4.31, P=0.0001) and SIMPER analysis indicated that the largest dissimilarity (99.31%) in the diet composition occurred between small (<8 mm SL) and transformed juveniles (>16 mm SL), and that the dissimilarities between contiguous size groups were relatively high (>80%, Table 5). The prey items that explain the differences among size groups were copepod nauplii, cirriped nauplii, Paracalanus indicus and Rodophyta (Table 5).

PLS analysis indicated significant, but low covariance (RV=0.045, P=0.0104) between morphometrics changes and variations in diet (expressed by number of prey per item). The PLS 1 explained 90.31% of the covariance. The regression between shape changes of clingfish (Block 1) and diet composition (Block 2) was low but significant (R²=0.243, P=0.032) (Fig. 8A). Likewise, the PLS analysis between the morphometrics changes and variations in the prey volume showed significant but low covariance (RV=0.096, P<0.001). The PLS 1 explained 99.89% of the covariance and the regression between shape changes (Block 1) and prey volume (Block 2) was significant but low (R²=0.386, P<0.001) (Fig. 8B). These results suggest that major changes in the shape of clingfish are decoupled with higher variations in feeding habits during habitat transition.