Morphological Plasticity in a Wild Freshwater Fish, Systomus Sarana (Cyprinidae) from India: A Glimpse Through Advanced Morphometric Toolkits

Body morphology supposed to underpin a wide differences of animal performance that can be used to understand diversication of characters. Further, identifying sh population with unique shape due to variations in their morphometric characters enables better management of these subunits. Advanced statistical toolkits of morphometry called truss network system and geometric morphometrics have been increasingly used for detecting variations in morphological traits. Present study was carried out with the objective of determining whether there are morphological characteristics that separate freshwater sh Systomus sarana from different groups. heterogeneity between S. sarana populations. The study suggests that the S. sarana distributed across selected Indian rivers shows morphological plasticity. The high degree of classication accuracy of these two approaches advocates their extension to other problematic species and highlights their importance as exploratory tools in morphological based population studies.


Introduction
Morphologically similar populations thriving together in a region are not easily distinguishable. Therefore, it is essential to recognize characters that differentiate populations. Morphological characters are Interestingly, studies on intra-species morphological differentiation are essential in solving the problems related to species recognition, as it is agreed that insu cient information on intra-species geographic dissimilarities can lead to incorrect species identi cation (Ishihara 1987). Taxonomy is important to shery scientists for the delineation of sh resources, and aids in developing balanced conservation strategies (Sangster et al. 2014). The taxonomic signi cance of the variation observed in the present study has to be assessed concerning the available taxonomic information on the species. Worldwide taxonomy of the Puntius including other Cyprinidae species has been dubious (Kortmulder 1972 Sukham et al. 2015), and this has been the case for S. sarana in Asia. Hamilton in 1822 had described this species from the Ganga River and named Cyprinus sarana, and afterward, it was synonymies with Puntius sarana. Likewise, Puntius sarana have many synonyms assigned by various authors (Puntius sarana sarana, Puntius sarana subnasutus, Puntius sarana spilurus, Puntius subnasutus, Systomus immaculatus, Barbodes sarana, Barbodes sarana subnasutus, Puntius saberi, etc) these synonyms undoubtedly generate confusions in the identi cation of this species (Pethiyagoda 1991). So far ve sub-species of S. sarana were identi ed worldwide; P. sarana orphoides, P. sarana subnasutus and P. sarana sarana from India and P. sarana spilurus from Sri Lanka (Irfan and Gunawickrama 2011). S. sabnasutus is referred to as S. sarana, although very recently it is categorized as a different sub-species (Biswal et al. 2018). These subspecies further adding taxonomic complexity of this species. Due to a lack of proper systematic studies and having phenotypic resemblance among subspecies, Pethiyagoda (1991) has recommended extensive population studies on these species by accompanying intra-species/population delineation studies (Talwar andJhingran 1991, Irfan andGunawickrama 2011).
Our organism of interest, Systomus sarana belong to the subfamily Barbinae, is a taxonomically diverse and complicated group of freshwater sh for studying morphological differences due to their wide distribution (Talwar and Jhingran 1991) and ipping systematic status i.e. many species formerly placed in Puntius have been moved to other genera (Kottelat 2013; Pethiyagoda et al. 2012; Raghavan et al. 2013). It is characterized by a deep and moderately compressed body with a dorsal pro le elevated. The maximum length of sh is 42.0 cm TL (FishBase). Systomus sarana (then allocated to Puntius) is an ecologically important, pro table, and cultivable candidate sh species (Gopakumar et al. 1999;Chakraborty et al. 2003). In India, this species is distributed widely excluding peninsular India-south of Krishna River and is also found in Afghanistan, Bangladesh, Bhutan, Nepal, and Pakistan (Talwar and Jhingran 1991 Considering the above context, this study aims to nd the morphological divergence of S. sarana populations from Ganga (North), Godavari (South), Mahanadi (East), and Narmada (Central) based on morphometric measurements by utilizing the following toolkits, truss network system and geometric morphometrics. This allows quantitative analysis of morphological divergences and may provide insight into microevolution.

Study area
For the present study, four rivers have been selected viz. Ganga (2600 km), Narmada (1312 km), Godavari (1465 km), and Mahanadi (900 km). The Ganga River originates in the Garhwal Himalayas from the Gaumukh glacier in Uttrakhand, India, and drains into the Sunderbans delta in the Bay of Bengal. The Narmada River originates from the Amarkantak, located in the Shahdol district of Madhya Pradesh, India and drains into the Arabian Sea. The Godavari River is also known as Dakshina Ganga, originating from the Nasik district of Maharashtra, India, and drained into the Bay of Bengal. The Mahanadi River, a major river in east-central India, originated Dandakaranya in Raipur district of Chhattisgarh, India empties itself into the Bay of Bengal. All the rivers taken into account are east owing except the Narmada River, which is west-owing.

Sample Collection
A total of 154 specimens of S. sarana were collected from Kanpur site of the river Ganga, Adilabad site of river Godavari, Haushangabad site of river Narmada, and Nadigaon site of Mahanadi river in two years duration (2016 to 2018). The specimens were caught before the breeding season and after the spawning period to avoid a bias toward size difference. The sh samples were collected with the help of hired local shermen. The identi cation of the sh was based on standard taxonomic keys of Talwar and Jhingaran (1991) and Jayaram (2010). Samples collection details and geographical coordinates of sites have been mentioned in Table 1. and Fig. 1.

Digitization Of Samples And Morphometric Measurements
The freshly caught sampled specimens (only undamaged) were placed with the left side up on a waterresistant paper and the body posture and ns were teased into a natural position to make the landmark points visible. Each individual was labeled with a speci c code for identi cation and archiving purposes.
Images of the specimens were taken by a camera (Canon IXUS145), set on a tripod stand directly above the specimen and the camera lens was adjusted and each image included a scale to normalize the individual sizes and additional scaling was applied in tpsDig making use of the millimeter gridiron in the graph paper.

Landmark-based Truss Analysis
Fourteen homologous anatomical landmarks (Winans and Nishioka 1987) were selected for the analysis (Fig. 2). A box-truss network was developed to give 91 morphometric variables through interconnection among these landmarks. Software including tpsUtil, tpsDig (Rohlf 2006), and software PAST (Hammer et al. 2001) was employed for generating truss data from the digital images. Since the standard length (SL) of sh specimens were different, it was necessary to remove dissimilarities due to size variations (Reist 1985). The truss measurements were standardized to account for size variation through the method described by Elliott et al. (1995) to eliminate the size component from the shape measurements: Madj = M*(Ls/Lo)b, Where M denotes original measurement, Madj is the size-adjusted measurement, Lo is the SL of the sh, and Ls is the overall mean SL for all sh from all samples in each analysis. Parameter b was calculated for each character from the observed data as the slope of the regression of log M on log Lo. SL (character code 1-6) was excluded from the nal analysis because SL was used as a basis for transformation (Mamuris et al. 1998) and thus 90 morphometric variables were retained for further analysis. The transformed data were validated for e ciency by testing the signi cance of the correlation between standard length and the transformed variables. The SL was excluded from the nal analysis. Univariate ANOVA was performed for each morphometric character to assess the signi cant variation among the four populations (Gomez-Rodriguez 2010). The transformed data representing characters that showed signi cant variation between populations were analyzed using PCA. This analysis was applied to determine the linear combinations of variables that responsible for a large amount of the variation in the data and to identify in uential variables (Johnson and Wichern 1998). PCA plot was formed by using components that con rmed high variance. In PCA, Jolliffe's rule with eigenvalues of at least 0.7 was applied to retain principal components (Dunteman 1989) and factor loading greater than 0.30 is considered signi cant, 0.40 more important, and 0.50 or greater very signi cant (Nimalathasan 2009). In the present study, only those factors were considered as signi cant that having loadings above 0.50. The Wilks' k was used to compare the differences. Further, a stepwise procedure was employed to lessen the number of variables to meet the requirement of a reduced set of characters for the DFA. Standardized canonical discriminant function coe cients and coe cients in the structure matrix were used as the criteria to identify the discriminating variables between two populations. DFA was used to assign individuals to their original group and to compute the percentage of correctly classi ed (PCC). Cross-validation (leave-one-out method) employing PCC was done to approximate the expected actual error rates of the classi cation functions. Statistical analyses were performed with the computer software programs MS-Excel (vers.2007), SPSS 16.0, and PAST 1.47.

Landmark-based Geometric Morphometric Analysis
Shape coordinates were superimposed to successfully eliminate the size effect, which was apparent from Procrustes analysis (Procrustes sums of squares: 0.363 and Tangent sums of squares: 0.361). Also, partial least square (PLS) revealed a non-signi cant covariance between superimposed shape and log centroid size (R = 0.54; P > 0.001), resulting in overlap among populations (Fig. 4). The deformed wireframe of average shape also showed variations between individuals and between populations (Fig. 5). Relative warp (RW) analysis illustrated deformation in shape (Fig. 6) from the reference that corresponds to selected positions in the ordination. The deformed wireframe was drawn on the shape among four populations to interpret shape changes that support the RW analysis.
The PCA extracted 24 components with a 100.00% variance. The rst two principal components (PCs) account for 40.22% of the total variance (22.48% for PC1, 17.74% for PC2). Overlap among the specimens obtained from four rivers is evident in the PCA plot of PC1 and PC2 (Fig. 7). A low level of variance and a high level of overlapping in the PCA demands further veri cation through CVA and DFA to determine shape variations. The CVA based upon 14 landmarks showed four groups with slight overlap among populations (Fig. 8). The larger part (82.98%) of the total variance (100.00%) was explained along the rst two canonical variates (CVs): CV1 and CV2 explained 55.47% and 27.50% of the total variance, respectively, while CV3 explained only 17.016% of the total variance. CVA extracted Mahalanobis and Procrustes distances among four groups found to be highly signi cant (p < 0.0001) ( Tables 6, 7).
Classi cation results of CVA indicated that all the specimens of each group were allotted to their respective groups with a slight misclassi cation rate. The classi cation of individuals into their crossvalidated groups showed a low level of mixing between the populations (Table 8; Fig. 9). These results go well together with those depicted by the deformed wireframe of average shape.

Landmark-based truss analysis
After the allometric transformation, there was no signi cant correlation (p > 0.05) found between standardized truss measurements with the standard length (SL), indicating that the size effect had been effectively removed from the data. Hence, all the measurements were utilized for further calculations. Further, the morphometric characters did not differ signi cantly (p > 0.05) between both sexes, therefore the data for both sexes were pooled for all subsequent analyses. By applying ANOVA (one way) on 90 morphometric characters, only 63 showed a signi cant difference in their mean values (p < 0.05).
Signi cant variables were subjected to principal component analysis (PCA) and DFA. PCA plot does not allow one to draw a conclusion about homogenous grouping based on visuals. By applying PCA, a total of 13 principal components were extracted explaining 93.311% of the total variance among populations.
Principal component 1 (PC1) and PC2 contribute 24.412% and 19.028% of total variance respectively ( Table 2) Table 3.  (Table 4). These variables or morphometric truss measurements were found to be the most important characters in distinguishing the selected populations. The linear discriminant analysis produced an average percentage correct classi cation (PCC) of 90.3% for morphometric characters indicating a high rate of correct classi cation of individuals into their original populations (Table 5). The percentage of correct classi cation ranged from 86.1% (Godavari) to 93.9% (Narmada). It was highest for the population of river Narmada followed by river Mahanadi (92.9%), river Ganga (86.2%), and lowest for river Godavari (86.1%). The results attained from the PCC cross-validation test were analogous to the results. Additionally, the plot of the discriminant variables showed a pattern that re ects successful discrimination among populations of four rivers (Fig. 3).

Discussion
Several statistical methods have been employed to study morphological divergences among wild populations of S. sarana collected from different geographic regimes. This is the rst study on the population delineation of S. sarana using truss network analysis with geometric morphometrics. The results revealed that heterogeneity exists among examined populations of S. sarana procured from the speci c sites of rivers (Ganga, Godavari, Narmada, and Mahanadi). Signi cant variations were detected for most of the analyses. The PCA loadings (truss analysis) of principal components revealed distinctness between populations. Though, there was a slight overlap found in the characters which were examined among the four groups. This separation was corroborated by DFA (truss analysis), showed signi cant morphological heterogeneity among populations, the level of differentiation between most of them as evidenced by a slight overlap of statistical data on derived plots.
Using geometric morphometrics, CVA plot obtained, have shown a slight level of overlaps among groups with a high percentage of correct classi cation suggesting differentiation among the examined populations. The PCA (geometric analysis) and DFA (geometric analysis) further con rmed the morphological heterogeneity among populations of S. sarana. The higher misclassi cation (DFA) observed for the Ganga with Mahanadi River and least with the Narmada. The biological variations of morphometric characters based on DFA are majorly associated with head morphology, covering lateral body lengths and caudal peduncle regions. Shape differences have been visualized with the deformation grids using geometric morphometrics. Geometric morphometry-based deformations grids (wireframes and relative warps) of average shapes between populations correspond to the high values of statistical distance between them and con rm the distinctness of populations in their immediate anatomical context.
Overall, the variations among the four groups in this study were largely owing to the dissimilarities of morphometric characters broadly associated to head, and body characteristics. However, the shape differences observed in this study presents little practical use in terms of discriminating sh populations in the eld. The visualization of the body shape differences, associated with other groups of correlated Earlier efforts have been made to differentiate S. sarana populations using traditional morphometry (Siddik, et al. 2016

Conclusion
To summarize, we quanti ed the morphological variation of populations of S. sarana from four major rivers of India. The basic characteristics of discrimination are overall body shape majorly associated with head morphology, covering lateral body lengths and caudal peduncle regions. Body morphology shows variation and could separate most populations, observed morphological variations provide good evidence for intraspecies heterogeneity between S. sarana populations. The study suggests that the S. sarana distributed across selected Indian rivers shows morphological plasticity. The high degree of classi cation accuracy of these two approaches advocates their extension to other problematic species and highlights their importance as exploratory tools in morphological based population studies.

Declarations
Ethics approval and consent to participate Fish specimens were obtained from the wild, directly from the commercial catches. The collection sites of sh specimens collected were fell outside Protected Areas (PAs). Fish were captured by gill nets. Fish if alive were euthanized with MS222 (Sigma) anesthesia and transported to the laboratory on ice to avoid damage to its morphological characters. The Research