In rocky reefs and coralligenous habitats throughout the world, fish assemblages are important components of coastal biodiversity, contributing crucially to ecosystem functioning and representing important marine resources for humans (Ackerman and Bellwood 2000, Bussotti and Guidetti, 2009). Various underwater visual census (UVC) methods have often been used in ichthyological research, especially on hard littoral bottoms where most widespread methods for the quantitative assessment of fish assemblages, such as capture methods using trawls and other fishing gear, are hard to use (Harmelin-Vivien and Francour 1992, Lipej et al. 2003, Soldo and Glavičić 2013). However, it is confirmed that the standard UVC methods can miss a large number of cryptobenthic fish species (Ackerman and Bellwood 2000, Willis 2001, Smith-Vaniz et al. 2006, Kovačić et al. 2012). A cryptobenthic fish is a species (or a life history stage of a fish species) whose individuals exclusively or predominantly spend their lifetime in cryptobenthic microhabitats, i.e. in the restricted living spaces underneath the bottom surface of the substrate or bio-cover, with a physical barrier to open spaces (Kovačić et al. 2012). To gain a full understanding of how a fish assemblage is composed and structured on hard littoral bottoms, a study of small cryptobenthic and epibenthic species using destructive techniques is required. Quantitative studies on cryptobenthic fish communities are surprisingly rare worldwide and restricted to coastal reefs with depths of mainly 15 to 20 m (Prochazka 1998, Ackerman and Bellwood 2000, 2002, Willis 2001, Depczynski and Bellwood 2003, 2004, Smith-Vaniz et al. 2006, Beldade et al. 2006, Beldade and Gonçavles 2007, Kovačić et al. 2012). No quantitative research on cryptobenthic fish has been performed at or below 40 m depth (Kovačić et al. 2012), while only recently the methodology for the quantitative research at circalittoral depths was proposed by Glavičić and Kovačić (2016). Other quantitative studies of fish assemblages at deep rocky reefs, i.e. UVC using SCUBA diving with a rebreather, are very rare and limited to tropical seas (Brokovich et al. 2008). Studies that include destructive methods on fish assemblages have been performed at underwater reefs going deeper, i.e. below 40 m in tropical seas, but they used only qualitative methods (Feitoza et al. 2005, Pyle et al. 2008).

Destructive techniques, by definition, remove fishes from a particular area and have a harmful effect upon the fish under investigation (Ackerman and Bellwood 2000). The main destructive methods that have been used for the study of fish assemblages using SCUBA diving are ichthyocides and anaesthetics (Ackerman and Bellwood 2000, Kovačić et al. 2012). Ichthyocides, such as Rotenon, have a slow effect on fishes, e.g. Rotenone takes 10-15 min to asphyxiate them (Ackerman and Bellwood 2000). In the case of repeated samplings during SCUBA dives, sampling with Rotenone requires a lot of time, which is usually limited in deeper dives. Therefore, the use of the anaesthetic Quinaldine, following the procedure of Kovačić et al. (2012), was chosen for the present study as the most appropriate method with high space and time efficiency.

The aim of the present study was: 1) to continue the studies on unexplored Mediterranean littoral benthic habitats potentially harbouring cryptobenthic fishes, following proposals by Kovačić et al. (2012); 2) to provide the first quantitative assessment of cryptobenthic fish species diversity and abundance on hard bottoms below 20 m depth by studying the deep underwater vertical cliffs of the Mediterranean; and 3) to identify the environmental variables that best explain species distribution and abundance patterns in this habitat, and that are therefore expected to shape the cryptobenthic fish community structure in this habitat.

The present study is the first quantitative assessment of cryptobenthic fish species diversity and abundance at hard bottoms below 20 m of depth. It was performed on a Mediterranean underwater reef with vertical cliffs going down to about 45 m depth. The recorded fish assemblage is part of the Mediterranean infralittoral and shallow circalittoral ecozones, which are mostly occupied by coralligenous biocenoses (Table 1, quadrants with median of calcareous algae cover of 50%) and to some degree by biocenoses of infralittoral algae (Bakran-Petricioli 2007, 2011). The necessity of a large number of longer dives at greater depths in order to obtain quantitatively useful data could be the reason for the lack of this kind of study in general. These studies need an extended diving time for the use of ichthyocides or anaesthetics, and for the larger number of quadrants necessary for quantitative analysis. The use of the anaesthetic Quinaldine, which takes just 2-3 min to asphyxiate fishes, provided better use of diving time, and the SCUBA diving protocol depending not just on compressed air but on other gas mixtures provided longer diving time. The chosen procedure of Kovačić et al. (2012), with 1 m² squares, allowed for high space and time efficiency, while the quantity of Quinaldine used and the intensive examination of the substratum should have guaranteed a total census of present fishes.

The only published quantitative study (Kovačić et al. 2012 in the northern Adriatic Sea) that performed methodologically appropriate research on Mediterranean cryptobenthic fish communities was restricted to more shallow depths. Two other quantitative studies focusing on cryptobenthic fishes in the Mediterranean (La Mesa et al. 2004, 2006) used the visual census technique, a method which could not record "real” cryptobenthic fishes, i.e. fishes present in the cryptobenthic microhabitats, where they are never or rarely visible from the outside. Both studies were also depth-limited (to 23 m depth and 15 m depth, respectively). Two studies on cryptobenthic fish communities geographically closest to the Mediterranean were performed on the Atlantic coast of Portugal. Beldade et al. (2006) used only interference visual census with depth limits of 9.5 m, while Beldade and Gonçavles (2007) combined interference visual census with limited use of the anaesthetic Quinaldine with depth limits of 11.2 m. The qualitative studies of cryptobenthic fish communities in the Mediterranean were limited to the infralittoral zone and only one (Patzner 1999) reached 40 m depth, a depth similar to the present study. The qualitative studies of Patzner (1999) were performed in Ibiza, in the western Mediterranean Sea, using SCUBA diving and the anaesthetic Quinaldine, and it showed a high match in species composition with Adriatic researches. All 15 species recorded by Patzner (1999) were also recorded in the Adriatic researches (Kovačić et al. 2012, present research). Among them, 10 species were found in the present research and the 5 remaining shallow water species were recorded by Kovačić et al. (2012). Species biodiversity in Kovačić et al. (2012) was higher than in the present research (27 vs. 21 species). However, that study covered a depth down to 20 m, more diverse habitats and a higher total effort (76 quadrants vs. 50 quadrants). About half of the species were the same as in the present study. Most of the remaining half of the species diversity in Kovačić et al. (2012) belonged to shallow water species of the families Blenniidae, Gobiidae and Gobiesocidae and, to a lesser degree, species of these families that do not occur on vertical and steep bedrocks (e.g. Lepadogaster candolii and Gobius bucchichi).

The average density was surprisingly similar between two studies performed on different habitats and depths: the present research had an average total abundance of cryptobenthic and epibenthic fish of 6.64±0.36 ind. m^–2 (mean±SE), while in Kovačić et al. (2012) the median total abundance of cryptobenthic fish was 6 ind. m^–2 and of epibenthic fish 1 ind. m^–2. The cryptobenthic fish density is even higher on the tropical coral reef where Depczynski and Bellwood (2004) found a mean density of 11 ind. m^–2. Both Mediterranean studies showed an abundance of cryptobenthic fish several times higher than that of epibenthic fish (in Kovačić et al. (2012) cryptobenthic fish/epibenthic fish ratio was 7.8; in the present research 3.6). In the present research 13 out of 21 species were found strictly in the cryptobenthic position, compared with 13 out of 27 species found by Kovačić et al. (2012). Therefore, the studies of benthic fish assemblages on infralittoral and circalittoral bottoms rich with shelters should not overlook the influence and role of cryptobenthic fishes in the biological community and in the benthic ecosystem.

The present study used two methods: visual census to study epibenthic fish and anaesthetic to study cryptobenthic fish. The part of the bottom fish assemblage that was not covered by these methods includes hyperbenthic fish and large epibenthic fish, the second one due to the limited size of individual squares, and the relative contribution of cryptobenthic fish to total benthic fish diversity is certainly lower than 62% and 48% from these two studies.

Despite the limited range of ecozones in this study, i.e. the lower infralittoral ecozone and the beginning of the circalittoral ecozone, one of three significant environmental variables that mostly contributed to variation in species occurrence (with an effect of 0.27 explained variance in species distribution) was the bottom depth. The presence of sand was another significant environmental variable that explained species distribution and abundance, with several gobiid species showing preferences for mixed bottoms and a few species showing an avoidance of sand. The third significant variable was the bottom of the cliff area, where some species occurred exclusively or mostly. It seems that the size and quantity of semicaves, caves and cavities, and even the biocover type as the second source of shelters, were without significant influence on species distribution and abundance, i.e. species were generalist in the choice of shelters. In a previous study of Kovačić et al. (2012), the three variables were highly significant for variation in species occurrence: depth, quantity of bedrock and presence of short thallus algae on boulders. Depth was, as in the present study, the variable contributing most strongly to variation in species occurrence in Kovačić et al. (2012). The differences in other significant and highly explanatory variables between the two studies could be the results of habitat differences, because not all the variables were present in both studies (e.g. the bottom of the cliff area in Kovačić et al. 2012). The depth dependence of cryptobenthic species composition for Mediterranean is also clearly visible in descriptive data in Patzner (1999: Fig. 1). The only other research that quantitatively studied, among other variables, the influence of depth on the cryptic fish community composition was also done in temperate waters, in South Africa, at depths of 0-20 m (Prochazka 1998) and showed depth to be the second most important environmental variable explaining cryptic fish community composition.

The indiscriminate use of the term cryptobenthic fishes in the past was criticized by Kovačić et al. (2012). We follow their definition of cryptobenthic fishes (provided here in the Introduction). The terms of cryptobenthic fish species and fish specimen in the cryptobenthic position should be clearly distinguished (see Material and methods). For the general conclusions on species character, even the large number of samples is not enough: the data need to be from various habitats, seasons and parts of day to understand general species relationships to the habitat and behaviour. In the present study, among species included in the quantitative data analysis, specimens of Z. zebrus, P. rouxi, O. balearica, C. zebratus, D. splechtnai and C. liechtensteini were found exclusively in the cryptobenthic positions. For the four of them the results match those in Kovačić et al. (2012) and represent one more evidence for the cryptobenthic character of these species. The exclusively cryptic behaviour of P. rouxi in the present study was surprising compared with previous data and general knowledge on the species (Kovačić et al. 2012), while D. splechtnai was not collected by Kovačić et al. (2012). The group of species present at the cliff base (G. kolombatovici, T. macrolepis, V. dollfusi) preferred epibenthic positions close to shelters. Another group along the entire cliff surface (G. auratus, G. vittatus and S. notata) could be termed epicryptobenthic according to Kovačić et al. (2012), considering their ambivalent bottom position records that include open and hidden places. The exclusively cryptic sample of Chromis chromis in the present study is too small for any conclusion but intensive use of cryptic spaces by this hyperbenthic species was known (at least for juvenile fish) at night-time (Harmelin 1987) and was recently reported even in daylight (Kovačić et al. 2012).

Some of the species found in the present study are considered to be "rare”, i.e. they have a limited number of total known records (summarized by Francour (2008) for D. splechtnai; Kovtun and Manilo (2013) for Steinitz’s goby, Gammogobius steinitzi Bath, 1971; Francour and Mangialajo (2007) for G. kolombatovici; Fesser (1980) for Speleogobius trigloides Zander and Jelinek, 1976; Ahnelt and Kovačić (1997) for T. macrolepis; and Kovačić (2008) for V. dollfusi). Most of them have been described or rediscovered during the last two decades (D. splechtnai, G. kolombatovici, T. macrolepis, V. dollfusi) and two species were described in the 1970s (G. steinitzi and S. trigloides). In addition, several species recorded in the present study as abundant (C. liechtensteini, O. balearica) or common (C. zebratus, Z. zebrus) had until recently been considered rare or even very rare species, but the recent study of Kovačić et al. (2012) revealed them to be a common part of the cryptobenthic fish assemblage.

Potential and possible directions of future research on Mediterranean cryptobenthic fish communities, using quantitative sampling by SCUBA diving and anaesthetics were reviewed by Kovačić et al. (2012). They could be summarized as studies of cryptobenthic fish communities at different times of day, seasons or different habitats to those used in Kovačić et al. (2012). The quantitative data on the influence of night and the cold season on the behaviour of Mediterranean benthic fishes and on their utilization of cryptic spaces is still unknown for any habitat. The present study, with a median of cover of calcareous algae in quadrants of 50% (Table 1), mostly studied the coralligenous biocenoses. This habitat was poorly included in the quadrants during the study by Kovačić et al. (2012), so the two studies are quite complementary by depths and bottom types. Other Mediterranean infralittoral benthic habitats well known for their rich cryptobenthic fish fauna, but without any quantitative data on this fauna, are Posidonia oceanica meadows and Cystoseira forests (Patzner 1999, Hofrichter and Patzner 2000). Future quantitative studies on cryptobenthic fishes should also target these two valuable and endangered Mediterranean habitats that have experienced a dramatic decline over the last decades in the Mediterranean (Boudouresque et al. 2003).