Adaptive radiation and burst speciation of hillstream cyprinid fish Garra in African river

Adaptive radiation of fishes was long thought to be possible only in lacustrine environments. Recently, several studies have shown that also riverine and stream environments provide the ecological opportunity for adaptive radiation. In this study, we report on a riverine adaptive radiation of six ecomorphs of cyprinid hillstream fishes of the genus Garra in a river located in the Ethiopian Highlands in East Africa. Garra are predominantly highly specialized algae-scrapers with a wide distribution ranging from Southeastern Asia to Western Africa. However, adaptive phenotypic diversification in mouth type, sucking disc morphology, gut length and body shape have been found among these new species in a single Ethiopian river. Moreover, we found two novel phenotypes of Garra (‘thick-lipped’ and ‘predatory’) that were not described before in this species-rich genus (>160 species). Mitochondrial and genome-wide data suggest monophyletic, intra-basin evolution of Garra phenotypic diversity with signatures of gene flow from other local populations. Although sympatric ecomorphs are genetically distinct and can be considered to being young species as suggested by genome-wide SNP data, mtDNA was unable to identify any genetic structure suggesting a recent and rapid speciation event. Furthermore, we found evidence for a hybrid origin of the novel ‘thick-lipped’ phenotype, as being the result of the hybridization of two other sympatrically occurring species. Here we highlight how, driven by ecological opportunity, an ancestral trophically highly specialized lineage is likely to have rapidly adaptively radiated in a riverine environment, and that this radiation was promoted by the evolution of novel feeding strategies.

Our goals were twofold: i) to investigate the morpho-ecological relationships of six Garra 100 sympatric ecomorphs from the Sore River, and ii) to test whether this assemblage has evolved 101 sympatrically. In detail, we aimed at elucidating the population structure and evolutionary history 102 of these ecomorphs using both mitochondrial DNA (mtDNA, cytochrome b) and genome-wide 103 nuclear loci obtained with a double digest restriction-site associated DNA (ddRAD) approach.

107
Study area 108 The Sore River is a headwater tributary of the Baro-Akobo-Sobat drainage in the White Nile basin, basins (Fig. 2, Table S1). Fish sampling was conducted under the umbrella of the Joint Ethiopian-   Table 2 (Table S1). 216 Akrokolioplax bicornis and Crossocheilus burmanicus were included as outgroups according to 217 previously published phylogenies (Yang et al., 2012). 218 measured using dsDNA HS Assay Kit for fluorometer Qubit 3 (Life Technologies, USA). ddRAD-245 library was constructed following the quaddRAD protocol (Franchini, Monné Parera, Kautt,& 246 Meyer, 2017) using restriction enzymes PstI and MspI. In total, 77 DNA samples of Garra 247 ecomorphs from the Sore River (see Table 2) and 11 DNA samples from five other species of 248 Ethiopian Garra from adjacent basins were sequenced by two independent runs of Illumina 249 HiSeq2500 and Illumina X Ten (2 x 150 bp paired-end reads). The raw sequencing data were 250 demultiplexed by the sequencing provider using outer Illumina TruSeq dual indexes.

252
Processing of RAD-seq data 253 The resulting reads were trimmed for remaining adapters and low quality reads Cutadapt were used for downstream processing and analyses. Prior to next steps, these R1 reads were 267 trimmed at their 3` ends to a uniform length of 130 bp to reduce the influence of sequencing error 268 (due to declined base quality at 3` end).

269
The de novo pipeline of STACKS was used to assemble loci and perform genotype calling. 270 We selected optimal parameters using the approach suggested by Paris, Stevens, & Catchen 271 (2017). Following the aforementioned procedure, we found that minimum stack depth (-m) of 5, 272 distance allowed between stacks (-M) of 3, and the maximum distance required to merge catalog 273 loci (-n) of 5 provided the best balance between data quality and quantity for our dataset (Fig. S1). proceeds to sampling iterations. Parameter 'rungs' was set to 10 (number of multiple MCMC 294 chains with different 'temperature' to run simultaneously). Both no admixture and admixture 295 models were run, and compared by plotting values of the posterior distribution and overall model 296 evidence in log space (log-evidence) (Fig. S2-S5). According to this comparison, the admixture 297 model is decisively supported over the no admixture model, and used here to report the results.

298
The same protocol was followed for consecutive hierarchical rmaverick runs for the identified    proportions, mouth width, interorbital distance, and snout length). The same pattern was detected 343 for PC2 -nine of ten most loaded characters belonged to head proportions (mainly disc length, 344 mouth width, height of head at nape and at eyes etc. -see Table S2 for details).

345
After excluding ecomorph 5, the ecomorphs 1 and 2 became more distinguishable with low 346 overlapping (Fig. 3B). The PC1 explained 73.8% of variance, while PC2 8.1%. The most loaded 347 eigenvectors of both PC1 and PC2 were from head proportions with few more contributions of 348 some body proportion characters (see Table S3). The difference between ecomorphs 1 and 2 349 revealed in PC2 is explained by height of head at both nape and eyes, interorbital distance, head 350 width, body height as well as other characters (Table S3).

376
Mitochondrial data 377 Both BI and ML analyses of cytb revealed monophyly of the Garra from the Sore River (Fig. 4A).     Compared to mitochondrial data, the nuclear phylogenomic tree shows much better 425 segregation of Garra ecomorphs from the Sore River (Fig. 5A). Ecomorphs 3, 4, and 6 form 426 monophyletic clusters, while other ecomorphs are divided into two (nos. 1 and 5) or even three 427 (no. 2) clusters. We assign two distantly located branches of both ecomorph 1 (generalized) as 428 1a/1b as well as ecomorph 2 (stream-lined) as 2a/2b according to population genomics analyses 429 done below (Fig. 6-8). Ecomorphs 1 and 2 from one hand, and other ecomorphs from another hand 430 form two clusters within Sore River adaptive radiation according to SVDQ species tree (Fig. 5B).

497
The eighth genetic clusters possess from three (ecomorph 6) to 38 private alleles (ecomorph 4) 498 (Table 4). The ecomorph 6 has also the lowest heterozygosity (Ho = 0.00058) as well as nucleotide   Our study provides genetic support for the hypothesis of the evolution of an adaptive radiation in 506 a riverine environment. By analyzing trophic features and sucking disc variation, as well as trophic 507 ecology, we show morpho-ecological diversification of the cyprinid fish Garra dembeensis into 508 six distinct ecomorphs. First, diversification of two novel phenotypes (thick-lipped and predatory) 509 in the Sore River has evolved rapidly, an event that can be classified as burst of speciation sensu  shape. The latter has streamlined appearance and probably is adapted for life in more rapid flowing 538 water. Ecomorph 3 has shorter gut length (ca. 2-times longer than body length) and a mixed diet 539 with significant additions of benthic invertebrates. Ecomorph 5 has an extremely short gut, whose 540 length is as long as the fish body. Short gut is a strong marker for predatory/piscivory feeding predominantly periphyton in diet, but it is characterized by distinctly divergent mouth phenotype 548 compared to ecomorphs 1 and 2 (Fig. 3). The gut of thick-lipped phenotype (ecomorph 6) was not  Another novel phenotype for Garra detected in the Sore River is the "predatory" niche. A 570 conspicuously piscivory trophic strategy is rare among Cypriniformes, presumably because they 571 have a toothless jaw. Nevertheless, this feeding strategy is quite common among cyprinid fishes    Five endemic, and one introduced non-Garra species were recorded in the Sore River in the 683 study area (data of this study). This is an extremely low number compared to more than 110 fish 684 species (Golubtsov & Darkov, 2008, and our data) recorded in the Baro River at Gambella at 440m 685 altitude (our data) to which the drainage of the Sore River belongs with a distance of ~150km 686 between compared localities. The segment of the Sore River where Garra's diversification was 687 detected is rather rich in biotope complexity -pools are alternating pools slow currents, rift areas 688 and rapids (Fig. S6). The depauperated fauna was suggested to provide the ecological opportunities         Right -heat map of -branch metric for selected ecomorphs/lineages of the Garra Sore radiation.

1422
The used guide tree is shown along the x and y axes (in 'laddered' form along the y axis). The 1423 matrix shows the inferred -branch metric, reflecting excess allele sharing between the branch of 1424 the 'laddered' tree on the y axis (relative to its sister branch) and the branches defined on the x 1425 axis.