Stochastic processes dominate community assembly in cichlid communities in Lake Tanganyika

The African Great Lakes are characterized by an extraordinary diversity of endemic cichlid fish species. The cause of this diversity is still largely unknown. Most studies have tried to solve this question by focusing on macro-evolutionary processes, such as speciation. However, the ecological processes determining local cichlid diversity have so far been understudied, even though knowledge on these might be crucial for understanding larger scale biodiversity patterns. Using trait, environmental and abundance data of cichlid fishes along 36 transects, we have studied how differences in local environmental conditions influence cichlid community assembly in the littoral of Lake Tanganyika, Zambia. We investigated changes in average trait values and in trait-based community assembly processes along three key environmental gradients. Species diversity and local abundance decreased with increasing sand cover and diet-associated traits changed with depth. Analyses on within-community trait diversity patterns indicated that cichlid community assembly was mainly driven by stochastic processes, to a smaller extent by processes that limit the similarity among co-existing species and least by filtering processes that limit the range of species traits occurring in an environment. Despite, the low impact of habitat filtering processes, we find community dissimilarity to increase with increasing environmental difference. Our results suggest that local environmental conditions determine cichlid abundance, while the predominance of stochastic community assembly across all environments explains why the communities with the highest abundances contain most species.

ABSTRACT 23 The African Great Lakes are characterized by an extraordinary diversity of endemic cichlid 24 fish species. The cause of this diversity is still largely unknown. Most studies have tried to 25 solve this question by focusing on macro-evolutionary processes, such as speciation. 26 However, the ecological processes determining local cichlid diversity have so far been 27 understudied, even though knowledge on these might be crucial for understanding larger scale 28 biodiversity patterns. 29 Using trait, environmental and abundance data of cichlid fishes along 36 transects, we have 30 studied how differences in local environmental conditions influence cichlid community 31 assembly in the littoral of Lake Tanganyika, Zambia. We investigated changes in average trait 32 values and in trait-based community assembly processes along three key environmental 33 gradients. 34 Species diversity and local abundance decreased with increasing sand cover and diet-35 associated traits changed with depth. Analyses on within-community trait diversity patterns 36 indicated that cichlid community assembly was mainly driven by stochastic processes, to a 37 smaller extent by processes that limit the similarity among co-existing species and least by 38 filtering processes that limit the range of species traits occurring in an environment. Despite 39 the low impact of habitat filtering processes, we find community dissimilarity to increase with 40 increasing environmental difference. 41 Our results suggest that local environmental conditions determine cichlid abundance, while The stunning diversity of cichlid fish in the African Rift lakes has fascinated scientists during 47 the past 50 years (Brooks 1950;Fryer & Iles 1972;Coulter 1991  Community assembly is generally viewed from two alternative perspectives, either from a 60 niche-based perspective, or a neutral perspective. The niche-based hypothesis postulates that 61 species are adapted to their local environment and occupy a species' specific niche: a set of 62 conditions in which the species thrives, and outcompetes other species (Hutchinson 1959;63 Tilman 1982; Chesson 2000). The traits of a species reflect the adaptation of a species to its 64 own niche, and studying patterns in trait-distributions can inform us about any underlying 65 processes driving species-coexistence and community composition. In benign environments 66 that do not pose strong requirements on traits, the niche-based hypothesis predicts that species 67 are abundant, and that the presence or absence of species' traits is mainly shaped through 68 interactions with each other, rather than interactions with the abiotic environment. Due to 69 competitive exclusion of species with overlapping niches or due to exclusion of species with 70 shared specialist predators, niche-based assembly is expected to lead to an increase in trait 71 diversity of co-occurring species in benign environments (Macarthur & Levins 1967;72 Mayfield & Levine 2010). In harsh environments, species with traits that make them 73 intolerant to stress, herbivory and/or predation pressures might be excluded from a local 74 community, leading to a reduced observed trait diversity of co-occurring species (Weiher & 75 Keddy 1995; Cornwell et al. 2006).

76
In contrast, the neutral hypothesis, which treats all individuals from all species equally, 77 explains community composition from a stochastic point of view, where local abundance of a 78 species is the outcome of stochastic birth, death and migration over time (Hubbell 2001;79 Rosindell et al. 201179 Rosindell et al. , 2012. Local community composition is assumed to be a dynamic 80 equilibrium between random immigration from the species pool, and local ecological drift.

81
The neutral hypothesis acknowledges that there might be benign and harsh environments, but 82 that these environments affect all individuals equally. As a consequence, benign environments 83 have more individuals than stressful environments, but both benign and stressful areas contain 84 individuals that form a (dispersal-limited) random subset of the species pool. The null 85 expectation is then that areas with high abundances also have high diversity, as a result of 86 random sampling.

87
Here we investigate how niche-based and neutral-based community assembly shapes 88 biodiversity in cichlid fishes across three environmental gradients: depth, sand cover and 89 topographical complexity. If community assembly is highly niche-driven, we firstly expect 90 that decreasing algae densities with depth provide less scope for algae-feeding cichlids and 91 that hence, communities in deep areas are predominantly assembled by filtering processes.

92
Secondly, because many cichlids are highly dependent on rocks for either food (as substrate 93 for algae growth) or for shelter (Konings 2005), we expect that an increase in sand cover, 94 which brings a reduction of available rock surface, will increase habitat filtering and decrease diversity. However, males of several cichlid species use sand to construct bowers to attract 96 females and are thus highly dependent on sand, which could counteract filtering processes and 97 possibly drive limiting similarity processes (Sefc 2011). Thirdly, we expect that areas with 98 high topographical complexity induce higher levels of limiting similarity and higher diversity, 99 because a more complex habitat provides more shelter and a larger surface area for algae 100 growth. Furthermore, complex habitats are often associated with a reduction in territoriality  Alternatively, if neutral-based processes dominate, we expect resource-rich areas, such as 104 shallow, rocky and complex habitats that allow for abundant algae growth, to show highest 105 diversity, as these habitats can sustain the highest number of individuals. Furthermore, we 106 expect these habitats to have the highest species diversity, and we expect that the trait 107 compositions of such communities to reflect a random subset of the larger species pool.

110
Abundance data 111 Abundance and composition data of cichlids was collected in Lake Tanganyika  were collated for 11 traits: standard length, total length, weight, stable isotope ratios of carbon 140 and nitrogen, lower pharyngeal jaw height, lower pharyngeal jaw width, gut length, lower 141 pharyngeal jaw shape and body shape. In order to obtain a set of traits that does not strongly 142 correlate with each other, to avoid the overemphasis of the importance of some traits over 143 others, we only used a subset of these traits. Firstly we checked the Pearson Rank correlation 144 between traits and removed one of the traits that correlated strongly. Total length and weight 145 were found to correlate strongly with standard length (standard length vs total length: R 2 = 146 0.98, standard length vs weight: R 2 = 0.75), and we decided to only use standard length. 147 Furthermore, lower pharyngeal jaw (LPJ) width and lower pharyngeal jaw height were found 148 to correlate with each other (R 2 = 0.67) and to correlate with standard length (lpj height vs 149 standard length: R 2 = 0.44, lpj width vs standard length: R 2 = 0.37). Partly to avoid 150 overemphasis of pharyngeal jaw traits, and partly because of the high correlation coefficients 151 we decided to omit both these traits. Lower pharyngeal jaw shape and body shape were 152 assessed using landmark-based geometric morphometric methods. xy coordinates of the 153 landmarks were combined into PCA components for both body shape and lower pharyngeal 154 jaw shape. From these PCA components we included the first PCA component for both body 155 and lower pharyngeal jaw shape, which contained 37% and 49% of the variation for body and 156 pharyngeal jaw shape respectively. The inclusion of more PCA components would allow us 157 to cover a larger part of variance in shapes, but would overemphasize the importance of shape 158 relative to the other traits.

159
The final trait set thus consisted of 6 traits: standard length, stable isotope ratios of carbon and   values were calculated following Nakagawa and Schielzeth (2013). 198 We repeated this approach using unweighted trait means, thus treating every species equally 199 and disregarding differences in abundance. Using unweighted trait means has two benefits

218
Species compositions 219 We recorded on average 22 species per transect, and an accumulated total number of 53 220 species (Table S3)     Relationships between traits and habitat components 259 We found that the three habitat gradients explained only a low proportion of variation (R 2 260 around 0.5) of community-weighted means (CWM) of cichlid traits (  We have investigated whether community assembly in cichlid communities at the littoral of 278 Lake Tanganyika, Zambia is mostly driven by niche-based processes, or by neutral-based 279 processes. In line with the neutral hypothesis, we found that while abundance and richness of Although habitat filtering only contributed to community assembly to a limited extent, 289 environmental conditions were not without influence on local community patterns. We found 290 that an increase in sand cover was correlated with a lower total abundance in the local 291 community. This could be an effect of a reduction in food resources, as there is less substrate 292 for algae growth (a common food source for many cichlids). However, individuals of most 293 species showed similar trends, which explains why a) both in areas with high and low sand 294 cover local communities resembled a random subset of the species pool, and b) diversity is 295 highest in low sand areas: this is mostly the result of random sampling from the species pool.

296
Sand cover thus drives local community patterns through neutral based processes, rather than 297 niche-based processes.

299
Although we found no changes in the relative contribution of the different community 300 assembly processes across the different environmental gradients, we did find that species 301 compositions shifted over these gradients. With an increasing difference between the local 302 conditions between two sites, their species composition became more dissimilar. This seems 303 to suggest that habitat filtering does impose an effect on community composition. One 304 explanation of our lack of finding habitat filtering effects in the STEPCAM analysis could lie 305 in our choice of traits. In our trait-based analyses we focused on traits associated with shifts in 306 diet and survivability (total length, gut length and pharyngeal jaw shape). We expected these 307 traits to differ most across different habitats, as they are thought to be related to different shown by our stepwise community assembly models, and confirmed by patterns in 317 community weighted trait means over environmental gradients. We found significant trait-318 environment relationships for stable nitrogen and carbon isotopic signatures, lower 319 pharyngeal jaw shape and shape of the body. The first three of these traits are associated with 320 food uptake and diet and mainly change with depth (associated with light availability and 321 hence algal density), thus confirming our expectation that there is a diet shift along our 322 environmental gradients.

324
Here we have presented a first analysis of cichlid diversity from a community assembly 325 perspective. We have focused on disentangling niche-based and neutral processes and found 326 that the majority of community assembly in cichlid communities at the littoral of Lake 327 Tanganyika, Zambia is driven by stochastic effects, rather than trait-based effects.

328
Interestingly, the dominance of stochastic processes in shaping local diversity is in line with 329 previous findings focusing on factors influencing cichlid diversity from a more macro-  Furthermore we suggest that our findings are in line with the previously hypothesized stages 338 of adaptive radiation in cichlid fish (Kocher 2004). Kocher hypothesized that cichlids have 339 first adapted to the local habitat (habitat filtering), after which diversification proceeded 340 through diversification of the feeding apparatus (limiting similarity), and lastly diversification 341 proceeded through diversification of the color pattern due to sexual selection. Our findings 342 resonate this order, as we find that adaptations to the local habitat (habitat filtering) are 343 currently least important in shaping diversity, followed secondly by trophic adaptations 344 driven by competition for food (limiting similarity). Although the current contribution of 345 community assembly processes is not necessarily a strong predictor of the evolutionary path 346 that cichlids have followed, the resemblance is striking at least.

347
Here we studied cichlid biodiversity from a micro-ecological perspective, rather than from a 348 (macro-) evolutionary perspective. Ultimately, a combination of these alternative approaches 349 might be most effective in uncovering why the diversity of cichlids in the East African Rift 350 lakes is so extremely high.