Characterizing fish populations and species is highly fundamental for fisheries management. The more species or populations identified, the more attention is needed for management of fisheries. In the Philippines, the fishery sector is a major industry providing food and livelihood to millions of people. However, the country is among the top most affected by climate change in the world (Kreft et al. 2015) which could have great impact on fishery production.
Among fish species, sardines are among the fish species with the highest value of production in the Philippines (PSA 2016). Sardines, with Sardinella lemuru as the dominating species, are most abundant especially in Zamboanga region where there is upwelling ((Villanoy et al. 2011). The Philippines is among the top producers of sardines in the world (FAO 2014). S. lemuru production, however, are shown to decline from previous years (PSA 2017). This shows that this high catch species could also be prone to vulnerabilities. Many procedures have risen to address this issue such as molecular tools and tagging to characterize fish populations for sustainable management but there is a need for a less time-consuming and cost-effective method that would be able to characterize fish populations. For this reason, geometric morphometrics is employed.
Geometric morphometrics is a method that is preferred for stock differentiation as it can process bulk samples at rapid speed and low cost. It only requires images of the samples which can be easily processed in image analysis programs. The images can readily be subjected for reprocessing and reanalysis of data. This method is preferred as its results can be visualized immediately as opposed to tables produced from traditional linear morphometrics (Cadrin and Friedland 1999). Spatial localization of morphological variation can also be detected easily and be interpreted biologically (Webster and Sheets 2010).
Otoliths are calcareous structures found in the fish that aid in balance and hearing (Popper et al. 2005). Otoliths have been preferred for geometric morphometrics because they are stock specific (Ihssen et al. 1981), not susceptible to modification (Green et al. 2009), not affected by short-term variabilities in the condition of fish or methods of preservation (Campana and Casselman 1993), do not deteriorate over time and do not require special storage methods. Otolith shape is shown to be less vulnerable to short term variability than fish body shape (Stransky 2014). Since its development, otolith shape studies to detect fish species or populations have been increasing in recent years (Javor et al 2011; Renán et al 2010; Begg and Brown 2010; Watkinson & Gillis 2003). For otolith shape analysis, two main methods are employed – outline and landmark-based analyses. In this study, the two different approaches on otolith shape geometric morphometrics will be used and compared.
The aim of the study is to utilize geometric morphometrics on fish otolith to discriminate sardine populations and species in Northern Mindanao for effective management. Specifically, the objectives of the study are:
1. To determine if otolith outline and landmark-based geometric morphometrics are useful in characterizing different species of sardines and different populations of S. lemuru
2. To detect any spatial or temporal patterns in population structure of S. lemuru
2. To determine which geometric morphometrics method in otolith is most effective in discriminating sardine species and populations
Review of Literature
The tropical sardine Sardinella lemuru are small, highly migratory pelagic fish species that form shoal in coastal waters of less than 200m depth (Willette et al. 2011). They are sporadic throughout the Philippines but they are known to be abundant in productive waters or upwelling areas such as in the North Zamboanga Peninsula, Visayas and Eastern Mindanao (Viillanoy et al 2011; Willette et al 2011; De Guzman et al 2012). It is said that sardines have two recruitment peaks but spawning season are more pronounced during October to December. S. lemuru are abundant from January to February during Northeast monsoon (NEM) season ( Villanoy et al. 2011). Sardine catches are highly seasonal and is correlated to high chlorophyll especially during NEM when upwelling occurs.
Half of sardine catches are coming from Region 9 or Zamboanga Peninsula especially in Sindangan bay where there is upwelling and Butuan Bay in Bohol Sea System where Agusan River, which is the largest river, has discharges. (Villanoy et al 2011; Willette et al 2011). Sardines said to be overexploited from heavy fishing pressure as manifested in smaller standard length of captured sardine fishes than the presumed standard length at first maturity (Guanco et al. 2009). This is an affirmation of the major impact of climate variability to sardine catches. Deteriorating sardine stocks was also attributed to sea temperature rise, thermal stratification of the water column and presence of predators (Piontkovski et al. 2014).
Geometric morphometrics has been proven to be useful for different disciplines, especially for its utility to distinguish close species and populations (Green et al 2013; Claude 2008). Geometric morphometrics has clear advantage over other fish stock identification methods because the life history and phenotypic interpretation can be derived from stock differences (Duarte-Neto et al. 2008).
Otolith shape analyses has also gained popularity as the calcareous structures are not reversed by acclimation (Ihssen et al. 1981), not prone to degradation through time and can be used readily for reanalysis through geometric morphometrics. Otolith shape are stock specific and can detect variations among species and populations (Cadrin 2014). Stransky et al (2008) were able to see strong geographical separation in orange roughy when otolith shape analysis was used while weak separation using genetic methods. In the same way, otolith shape analysis was shown to have high discriminatory power in mackerel populations compared to other stock identification methods (Smith et al. 2002). This makes the otolith shape analysis a promising tool for species and population discrimination. Shape studies can become good alternative and can be used with other stock identification tools such as molecular techniques, microchemical analysis and tagging to enhance the precision of differentiating stocks and achieve more comprehensive results. Otolith shape is affected by the growth rate (Campana and Casselman 1993) and feeding history (Gagliano & McCormick 2004) as influenced by environmental effects.
Otolith outline-based analysis deals with closely-packed points that form the perimeter shape (Cadrin 2014). On the other hand, landmark-based analysis utilizes landmark points in the otolith which are homologous morphometric points found in forms of samples (Bookstein 1990). This kind of analysis also look at linear distances and geometric relationships among landmark points. Outline-based analysis has the ability to compare structures that does not contain or do not have enough landmarks and could provide similar or even better scores than the landmark-based analysis (Dujardin et al. 2014). Most studies using geometric morphometrics use the landmark approach but some studies showed promising results based on the outline-based morphometrics (Loy et al 2000; Baylac and Frier 2005, cited in Dujardin et al 2014). Still, some results have shown that landmark and outline-based methods are equivalent and produced similar patterns on samples (Changbunjong et al 2016; Jensen et al 2002).
Sardine fish specimens that will be used in this study will be obtained from the samples collected during spawning peaks from years 2014-2016. About 1400 S. lemuru otoliths from four sampling locations in North Zamboanga Peninsula (NZP) and four from Bohol Sea System (BSS) (Figure 1) will undergo shape analysis. S. lemuru samples from Manila Bay will be used as outgroup. For species discrimination, about 30 otoliths each from S. gibbosa and S. fimbriata, S. tawilis and Amblygaster sirm will be tested. The left sagittal otolith of the samples will be utilized. If the left sagittal otolith is not available or broken, the sample will be omitted from the analyses. The samples will be stored dry in tubes.
Shape variations will be tested using the following groupings: between sites; Mindanao and outgroup samples; NZP and BSS samples; samples from each year of sampling; and species.
Fig. 1. Collection sites of S. lemuru samples from North Zamboanga Peninsula (labeled in italicized letters) – PAT, Patawag; SIN, Sindangan; DIP, Dipolog; DAP, Dapitan; and from Bohol Sea System (labeled in bold letters) – ILG, Iligan; MAC, Macajalar; GIN, Gingoog; BUT, Butuan.
To capture image for analyses, the otoliths will be placed on a dark background with mesial surface up and the rostrum pointing to the left. A Dino-Lite digital camera (AM4113T-FVW Dino-Lite Premier; AnMo Electronics Corp, Taiwan) will be used to image the otolith samples with proper calibration. Transmitted light in a dark field condenser and outside light source will be used to produce good quality images. The photo editing features of DinoCapture 2.0 v 1.5.22A will be used to increase the contrast. A scale for size will be placed in the images.
For the outline-based analysis, the ‘shapeR’ package (Libungan and Palsson 2015) available from the Comprehensive R Archive Network (CRAN) will be used. ShapeR package automatically detects otolith outline using Normalized Elliptic Fourier Transform (EFT) analysis (Kuhl and Giardina 1982) and Discrete Wavelet Transformation (WT) analysis. EFT utilizes Fourier harmonics which are trigonometric sine and cosine waves to describe outline shape (Rohlf and Archie 1984). Both outline methods will be used and compared in this study. EFT describes overall shape while WT can describe localized differences in the otolith (Bárðarson 2015).
The procedures that will be used for the outline-based geometric morphometrics analysis will be based from the methods by Libungan and Pálsson (2015). The otolith outlines will be obtained using detect.outline function of the package which automatically traces the contour of the the otolith. The outlines produced will be checked using the write.outline.w.org argument to place the extracted outline on top of the original image. Otolith images with low quality outlines will be fixed.To produce robust results for Fourier and Wavelet analysis, the smoothout function of the package will be utilized to smooth the outlines by reducing pixel noise (Haines and Crampton 2000).
Otolith shape coefficients will be obtained using the package and the shape coefficients will be adjusted for fish length to prevent allometric effects. Bonferroni adjustment will be conducted to reduce errors. Mean statistics including the perimeter, maximum or Feret length and width, and area of the otolith samples will be obtained. To visualize the variation in the area of the otolith, the mean and standard deviation of Wavelet coefficients will be plotted against the angle from ‘gplots’ package (Warnes et al. 2016).
TpsDig2 (v. 2.30) software will be used to digitize landmark coordinates from the otolith images. Twelve landmarks will be used for the analysis (Figure 2). The landmarks are present in all S. lemuru otolith samples and adhere to the guidelines of landmark selection from Zelditch et al (2004). The standard landmarks that will be used in this study based on homology. Additional landmarks will be used along the sulcus acusticus of the otolith as they are found to vary in some fish species ( Tuset et al. 2016). The order of selecting landmarks will be the same for all samples. Digitized images will be converted to .nts file using TpsUtil.
Fig. 2. Landmarks from S. lemuru otolith contour (1 – rostrum; 2- excisura major; 3 – antirostrum; 4 – postrostrum; 5 – collum) and sulcus acusticus (6-9 where the inferior crista change curvature; 10 – most distal point of the cauda; intersection between inferior and superior crista; 11-12 endpoints of crista superior).
Landmark data will be read using ‘geomorph’ package (Adams et al. 2017) in R statistical computing environment for further geomorphometric analyses. To eliminate size variation among samples, generalized least squares Procustes superimposition will be performed using the function gpagen. Procustes superimposition will remove the effect of size, rotation, and translation in the otolith images so that only shape differences will be accounted (Rohlf and Slice 1990). To visualize otolith shape variation patterns, plotTangentSpace function will be used to perform principal component analysis and will project the objects to tangent space.
ANOVA-like permutation test will be used for partition of variation among groups. Populations that will be found significantly different from ANOVA will undergo Canonical Analysis of Principal coordinates (CAP) from ‘vegan’ package using the Fourier and Wavelet coefficients to assess shape variation.
For statistical assessment of patterns of shape variation and covariation from landmark data, the function procD.lm will be used from ‘geomorph’. This function performs Procustes-ANOVA which is equivalent to non-permutational MANOVA on shape data (Anderson 2001). The size independent variables will undergo ordination – MANOVA to identify variation among individual samples, Hotelling’s pairwise comparisons (post-hoc test), and Canonical Variates Analysis (CVA) using ‘morpho’ package for testing differences between groups. Finally, shape differences that will be derived from the scores produced from the multivariate analyses will be graphed visually through thin-plate spline deformation using plotRefToTarget function.
Linear Discriminant Analysis (LDA) will be applied to the shape coefficients that will be produced using the ‘ipred’ package. Through this analysis, the classification success rate of the geometric morphometric method will be tested.