Fasciola gigantica, F. hepatica and Fasciola intermediate forms: geometric morphometrics and an artificial neural network to help morphological identification

Introduction

Fasciola hepatica Linnaeus, 1758 and F. gigantica Cobbold, 1855 are well-known as parasites that cause fascioliasis (Mas-Coma, Bargues & Valero, 2005). The parasites belong to the family Fasciolidae Railliet, 1895, subfamily Fasciolinae Stiles & Hassall, 1898. The adult worm infects the liver of various species of mammals, particularly livestock that consume fresh plant material contaminated with infective-stage metacercaria (Mas-Coma, Bargues & Valero, 2018). Climatic changes seem to have exacerbated the problem by expanding the geographic distribution of such parasites (Mas-Coma, Valero & Bargues, 2008; Mas-Coma, Valero & Bargues, 2009a). Typically, F. hepatica is reported worldwide, while F. gigantica occurs in a narrower range across Asia and Africa. The geographic distribution of these two parasites can overlap, particularly in the equatorial zone. The hybrid forms that can result between F. hepatica and F. gigantica are often reported as Fasciola “intermediate forms” (Periago et al., 2008; Mas-Coma, Valero & Bargues, 2009a; Mas-Coma, Valero & Bargues, 2009b). Like other countries in the tropical and subtropical areas, Thailand has reported Fasciola infections in both humans and livestock (Mas-Coma, Valero & Bargues, 2009a; Buranasin & Harinasuta, 1970; Srihakim & Pholpark, 1991; Aroonroch et al., 2006; Kanoksil et al., 2006; Tippawangkosol et al., 2017; Wannasan et al., 2014).

Up until now, traditional morphometrics, which is based on linear measurements between anatomical landmarks, has been used to help in the identification of Fasciola species (Valero, Panova & Mas-Coma, 2005). Many factors have been noted which can modify the accuracy of phenotypical classification done in this way, including: (1) lymnaeid-host specificity, (2) bias arising from different types of definitive hosts, the technical bias for worm fixing, (3) the computerized image analysis system used, (4) the standardized methodology used for measurements, and (5) the effect of allometric growth (Valero, Panova & Mas-Coma, 2005; Periago et al., 2006). Size and some other variables or ratios that have been erroneously considered as a proxy for “shape” (body roundness (BR = BP2/4 πBA; BP = body perimeter; and BA = body area) could discriminate between adult F. hepatica, F. gigantica, and Fasciola intermediate forms collected from different countries in Asia (Periago et al., 2006), as well as being used to examine intraspecific diversity (Periago et al., 2008). A molecular identification technique was developed to discriminate between fasciolid adult worms based on ribosomal internal transcribed spacers (ITS) 1 and 2 (Lin et al., 2007; Lotfy et al., 2008). According to Amer et al. (2011), the identification of fasciolids requires at least ITS2 and mitochondrial cox1 or nad1 sequences. Itagaki et al. (2011) suggested that ITS1 sequences, which do not show variation in Europe or Oceania, are nonetheless useful molecular markers for discriminating F. hepatica, F. gigantica, and their hybrids occurring in Asia (Amer et al., 2011).

In Thailand, Fasciola intermediate forms were found in the livers of cattle in slaughterhouses and identified as hybrids because of the heterosequences of nuclear ribosomal ITS1 and ITS2 (Chaichanasak et al., 2012; Siribat et al., 2018). For molecular discrimination between F. gigantica, F. hepatica, and Fasciola intermediate forms, ITS1 has been noted as the most effective DNA marker (Chaichanasak et al., 2012). Fasciola intermediate forms present no qualitative morphological attributes that allow clear identification. Alongside molecular discrimination techniques, traditional morphometrics has been suggested as a relatively simple and effective approach for species identification (Periago et al., 2006; Valero, Marcos & Mas-Coma, 1996; Valero, Panova & Mas-Coma, 2005; Ashrafi et al., 2006; Afshan et al., 2014). However, the dimensions of Fasciola intermediate forms may overlap with those of both F. gigantica and F. hepatica (Periago et al., 2006; Srimuzipo et al., 2000).

Even when using many characters and multivariate statistical tools, the risk of using a morphometrics method is that linear distance measurements mainly provide information relating to size and provide limited information about shape. Previous studies inaccurately labeled as “shape” some variables which are actually another estimation of size, such as body roundness (BR = BP²/4πBA), body perimeter (BP), and body area (BA) (Periago et al., 2006; Mas-Coma, Valero & Bargues, 2009b; Valero et al., 2018). These characters are secondary constructions from linear measurements and produce variables redundant with size (Burnaby, 1966).

A more convenient approach has recently been applied which enables variations in shape to be estimated from linear measurements (Valero et al., 2018). However, there are various techniques available that can be used for this. A comprehensive review was recently published distinguishing two “schools” via which it is possible to approach the allometric concept: the Huxley–Jolicoeur and the Gould–Mosimann schools (Klingenberg, 2016). The former, initially developed for bivariate analyses, considers morphological form as a single unified feature (form as both size and shape), emphasizing the covariation among traits as a consequence of variation in size. The latter, initially developed for multivariate morphometric measurements, separates two estimates: an estimate of global size and a multivariate estimate of shape (Mosimann, 1970; Darroch & Mosimann, 1985). According to Klingenberg (2016), landmark-based geometric morphometrics represents a straightforward extension to this second method. Its concept may be naturally extended to the outline-based approach, which also separates a global estimator of size and a multivariate estimate of shape (Rohlf & Archie, 1984; Lestrel, 1989). Instead of using distances between anatomical landmarks, the modern methods use their coordinates in a common system of axes. One of the clear improvements introduced by the modern approach is that it enables a visual representation of shape (Rohlf & Marcus, 1993). Visualization of changes in shape are frequently presented using D’Arcy Thompson-like plots, or deformation grids (Bookstein, 1991). Deformation grids help to visualize the shifts in aligned coordinates from one shape to another. The grid is a visual aid and can even be amplified to improve the visibility of subtle changes. It can be omitted when these changes are clearly visible, as is frequently the case between species, without the need for any amplification artefacts.

The use of coordinates usually applies to animal species presenting a solid structure on their body. However, this method has also been applied to soft structures. For example, when combined with photogrammetric techniques, it can reveal variations in the shape of rodent sperm-cell heads and the diversity of these shape variations among species (Varea Sánchez, Bastir & Roldan, 2013).

In our study, we used recommended traditional morphometric techniques and, to preserve the parallelism of shape extraction from the data, we applied the Darroch & Mosimann technique (1985) to the linear measurements, which produces “log-shape ratios” as estimates of shape.

We compared the results obtained by traditional techniques with those obtained from two methods based on the coordinates of anatomical points: the landmark-based (Rohlf, 1990) and the outline-based (Kuhl & Giardina, 1982; Lestrel, 1997) approaches. These two approaches are conceptually different. The landmark-based approach uses anatomical points called “landmarks”. These points are small biological structures, as small as a point at the requisite scale. They are considered to be homologous in the sense that they are relocatable points from one individual to another. The outline-based approach also uses an ordered set of discrete point coordinates; these are superficially similar to landmark data, but they are conceptually quite different. They are of a different nature to true landmarks because comparability between them is not expected separately, but from the structure they describe. The outline-based method is generally restricted to closed contours (outlines) where landmarks are lacking. In this approach, the contour is described by points that we call “pseudolandmarks” (Dujardin, 2017). There is no need for these pseudolandmarks to be equally spaced, and there is no need to have the same number of them from one individual to another (Rohlf, 1990; Dujardin et al., 2014).

Pseudolandmarks should not be confused with “semilandmarks”. Semilandmarks are landmarks that lack one degree of freedom because they depend on other landmarks. They are generally used to compare open curves between other landmarks, and, most importantly, they are processed as true landmarks by Procrustes superposition, combined with a sliding procedure to minimize total bending energy or to minimize the Procrustes distances to the consensus (Dujardin, 2019).

Outline data are processed in a very different way, either by elliptic Fourier analysis (Kuhl & Giardina, 1982) or, less frequently, by eigenshape analysis (Lohmann, 1983). Briefly, the observed contour is decomposed in terms of sine and cosine curves of successive frequencies called harmonics, with each harmonic being described by four coefficients. These coefficients are normalized by the semi-grand axis of the first harmonic ellipse. The latter may be used as an estimate of global size, as well as the root-squared area of this same starting ellipse, the root-squared area of the digitized contour, or its perimeter. We explored the capacity of these different approaches to discriminate between F. hepatica, F. gigantica, and Fasciola intermediate forms.

To ensure correct evaluation of the methods, the samples collected from Thailand were also identified using molecular analysis. In addition to the Thailand samples, nine F. hepatica strains from the UK were provided and used to give reference data: they were subjected to morphometric analyses only.

As far as we know, morphometrics applied to the Fasciola identification always relied on traditional morphometric techniques. In this study, we tried to answer the following question: could modern morphometric techniques bring any improvement over traditional ones in terms of accuracy and/or simplicity?

Materials & methods

Outline data (OD)

The contour of each specimen was manually digitized (see Fig. 1A), and the resulting coordinates were subjected to an elliptic Fourier analysis. The “outline data” (OD) used as the input for validated classification were the normalized elliptic Fourier coefficients (NEF). Their numbers were restricted to an average set giving a power of at least 99% (n = 30). Their first eight principal components were used directly as input for the validated classification tests. An estimate of the allometric residue contained in shape variables was performed by regressing the principal components of the NEF on the square root of the area delimited by the contours. The non-parametric estimation of the statistical significance of this multivariate regression followed the procedure of Good (2000). On the other hand, the influence of the global size on interspecific discrimination itself was estimated after regressing the first discriminant factor on size. For the artificial neural network technique (ANN, see below), the OD were also combined with different numbers (4, 6, 8, 10, and 12) of LN.

Results

Global size

The global size of the parasites, either the body perimeter as computed from OD, the square-root area within the contour (Fig. 2A; Table 4A) or the centroid size as computed from LD (Fig. 2B, Table 4B) differed significantly between the intermediate forms and the other two species. These other two species, F. hepatica and F. gigantica, did not show statistically significant differences in global sizes (Table 5). Fasciola intermediate forms had the smallest global size, both in terms of centroid size and the perimeter or square-root area of the contour.

Table 4:

Global sizes of Fasciola spp.

Species	n	Mean	Min	Max	SD
4a. Body perimeter and Square-root area of Fasciola spp.
	Body Perimeter
Fg	17	65.44	45.22	88.47	13.96
Fh	9	62.25	55.55	75.40	6.88
Fif	10	38.67	29.05	56.83	9.14
	Square-root area
Fg	17	13.5	9.04	17.89	2.54
Fh	9	14.3	10.64	14.68	1.76
Fif	10	9.8	5.20	10.40	2.33
4b. Centroid size of Fasciola spp.
Fg	17	17.44	11.83	28.33	4.98
Fh	9	17.16	14.48	22.00	2.52
Fif	10	9.45	6.93	13.36	2.31

DOI: 10.7717/peerj.8597/table-4

Notes:

Fg

Fasciola gigantica

Fif

Fasciola intermediate forms

Fh

Fasciola hepatica

n

number of specimen

Min

minimum

Max

maximum

SD

Standard deviation

Perimeter and square-root area as computed by the outline-based technique, and centroid size by the Generalized Procrustes Analysis (landmark-based technique).

Table 5:

Non-parametric comparisons of global size estimations, 1,000 permutations (P-values).

Species	Centroid size	BP / Square-root area
F. gigantica / F. hepatica	P=0.909	P = 0.650∕P = 0.555
F. gigantica / F. intermediate forms	P < 0.001	P < 0.001/P = 0.003
F. intermediate forms / F. hepatica	P < 0.001	P < 0.004/P < 0.001

DOI: 10.7717/peerj.8597/table-5

Notes:

P

percentage of values equal or higher than the observed differences of mean values between species after 1,000 random permutations. It is the empirical risk of error saying that the differences are significant

BP

body perimeter

Outline-based analyses

Shape variation among Fasciola spp. according to the outline-based analysis

Direct visualization, without amplification, of contour differences suggested Fasciola intermediate forms possess a more globular contour (Fig. 3A). The factor maps of the discriminant analyses (Fig. 3B) showed non-overlapping distributions. Accordingly, satisfactory validated classification scores were obtained, ranging from 89% to 94% (Table 3, column “OD”).

Landmark-based analyses

The mean specimens aligned according to the five anatomical landmarks used here showed some separation between the three forms, with a seemingly more critical difference for Fasciola intermediate forms (Fig. 4A), as for the OD (Fig. 3). The factor map of the discriminant analysis showed that Fasciola intermediate forms could diverge from F. gigantica and F. hepatica (Fig. 4B). However, some overlapping was observed, and the validated classification scores based on landmark data were not satisfactory, ranging from 58% to a maximum of 69% (Table 3, column “LD”).

Discussion

The specimens in this study were processed using the same treatment method. If some deformation occurred as a result of this treatment, it is reasonable to expect that this would be the same for each species. The average values of linear measurements of 12 selected morphological characters of each Fasciola sp. were graphically compared with previously published values from the same parasites and the same hosts (see Fig. S1). This comparison showed that our samples were in the range of average values and suggested that the small number of specimens did not result in any sample bias.

To reduce the effects of growth on the phenotype, only fully mature stages were selected from each molecularly defined group. However, in addition to possible growth heterogeneity, other divergent influences could have affected phenotypic variations in our total sample, such as different geographic origins, microhabitats, host, and seasonality. We considered that, within each species or group, these environmental effects would probably have had much more of an impact on size than on shape. We used the geometric morphometrics approach to tentatively set apart these size effects from the morphometric data, extracting the remaining shape variables and using them exclusively to describe the groups distinguished by molecular data. The allometric residue that possibly remained in the shape variables was not removed because our comparisons concerned species rather than intraspecific groups. Species may differ by their allometric growth, which becomes part of their evolutionary divergence (Klingenberg, 2016).

Our tentative removal of the size effect from the linear measurements (either 20 LN or 12 LN) did not change the global classification scores as obtained by raw measurements, probably because it could not perform a consistent removal of size, as indicated by our allometric analysis. Size remained dominant in the LSR (∼61%) relative to the PR (∼41%) and OD (∼24%). Another inconvenience of LSR is that they do not allow a visual representation of shape changes between taxa (Rohlf & Marcus, 1993). Methods based on coordinates data allow such visual representation (see Fig. 2D).

Our samples represented limited geographic isolates and, because of the low sample sizes, could suffer from low representativity. Moreover, the number of variables could be greater than the number of individuals in our smallest sample (nine F. hepatica). We followed the “rule” for a multivariate analysis, especially discriminant analysis, which is to use a number of variables that does not exceed the number of individuals in the smallest sample. There are various methods available to deal with multidimensionality. This specific number of PC scores was selected so as to use a number of variables that does not exceed the number of individuals in the smallest sample. This procedure allowed multidimensionality to be reduced where necessary; this did not make our samples more representative, but at least the information they provided was not over-interpreted.

Since linear measurements can provide information about both size and shape, and because of the relatively small size of Fasciola intermediate forms (Fig. 2), it was expected that this sample could be identified using traditional techniques. Nevertheless, linear measurements did not just reflect global size variation since they could also allow an excellent score of recognition for the pair F. hepatica and F. gigantica, which are parasites of similar global size (Fig. 2, Table 4). The extraction of shape (LSR) did not produce different results, confirming the relatively poor efficiency of shape removal from our linear measurements. With or without shape extraction, traditional morphometrics appeared to be a powerful means of species determination. It does, however, require an excellent knowledge of the parasite’s internal anatomy.

The previously recommended number of linear measurements necessary to distinguish F. hepatica from F. gigantica and Fasciola intermediate forms was as high as 20 (Valero, Panova & Mas-Coma, 2005; Ashrafi et al., 2006). By considering the contribution of each variable to their first two PC, we were able to reduce this number from 20 to 12, with similar or better results (see Table 3). This improvement could be explained by the elimination of redundancy among measurements, and a possible lack of precision for some of them. Either way, our examination of repeatability produced satisfactory results for both shape (93%) and size (99%), although these results were lower than the results of the outline-based technique (98% and 99%, respectively).

The outline-based morphometrics analysis was based on manual contour digitization. Although this type of manual data collection is relatively fast, an automated capture technique might be preferable. Automated curve tracing is already included in many types of imaging software, and it may be more appropriate for generalizability (Yang et al., 2015), but its useful application is dependent on the complexity of the contours (Sheets et al., 2006; Firmat et al., 2010). We would recommend that a study be performed that is specifically aimed at comparing automatic and manual capture techniques. The fact remains, however, that whether manual or automatic techniques are used, the outline digitization process is a relatively easy task and does not require highly specialized skills from the user.

The landmark-based approach required a little more knowledge about the internal anatomy of the parasites and did not provide satisfactory results. The lower reclassification scores of the landmark data relative to the OD should not be regarded as a rule. In our case, very few reliable LM could be suggested, reducing the amplitude of shape capture, which could explain the unsatisfactory scores obtained. It is possible that better results might be obtained with, for instance, a different landmark configuration, or a different number of landmarks.

In our validated method of reclassification each individual was sequentially removed from the total sample and its identification performed using the model computed without it, a method also called a “leave-one-out” or “jack-knife” procedure (Manly, 1986). Thus, individual identification was repeated as many times as there were individuals in the total sample (see Data S1). Our “accuracy” for each method (Tables 3 and 7) was the percentage of individuals that were correctly identified at the end of each such procedure. For the ANN classification, each classification session was repeated 30 times, and the accuracies were presented as the average and standard deviation of the 30 sessions (Tables 3 and 7).

The validated classification applied to different kinds of data used different methods (Table 3). The one based on the Mahalanobis distance (Dist) generally produced higher scores with more correct identifications than the other classification techniques, which conforms with previous studies and highlights the frequent relative superiority of this technique (Dujardin, Dujardin & Kaba, 2017; MacLoad et al., 2007; Sonnenschein et al., 2015). However, reliable classification and assessment of group separation using this analysis is based on several assumptions, some of which frequently do not coincide with morphometric data (Mitteroecker & Bookstein, 2011). For this reason, we added additional classification techniques, such as the simple maximum likelihood (MxL) approach (Polly & Head, 2004; Dujardin, Dujardin & Kaba, 2017) and the ANN technique.

Since the MxL method assumes normally distributed data and independence among the characters used, our best results were obtained when applied to (a subset of) the PC of the log-transformed input variables, rather than when applied to the raw variables (detailed comparisons not shown). In agreement with previous comparisons (Dujardin, Dujardin & Kaba, 2017), the MxL technique provided a slightly lower accuracy than the Dist method.

In our study, the ANN produced slightly lower scores than the Dist or the MxL approaches. Our idea was to compare the taxonomic signal embedded in different types of data rather than to deeply explore the power of each classification method. So, for each classification method, we tried to subject each type of data to the same process. The neural network was thus configured in the same way, irrespective of which data were used as the input (see Data S2). Certainly, some improvements could be obtained by refining the configuration according to the kind of data being used. Moreover, a possible explanation for the relative weakness of the ANN method used here is the low sample sizes of the training data. The ANN method may require many data to achieve an accurate level of performance (Soda, Slice & Naylor, 2017). However, the general scores were satisfactory, and we showed that this technique could greatly benefit from using combined data as the input, for example using a few traditional measurements together with outline shape data (Table 7).

Conclusion

Both traditional and modern morphometric approaches can be helpful in the morphological identification of Fasciola sp. We showed that by reducing the number of linear measurements from the 20 commonly recommended ones to a selected subset of 12 measurements, the accuracy of identification could be improved.

In our study, the outline-based approach produced satisfactory classification scores, more so than the landmark-based approach. Moreover, we showed for the first time the apparent differences in shape between “intermediate forms” of Fasciola sp. and F. gigantica and F. hepatica. The relative simplicity of data collection (even manual collection) makes the outline-based approach an attractive alternative.

By comparing various data collection and classification techniques, our study suggests promising new strategies to follow in helping with the identification of species in the genus Fasciola based on their morphology.

Supplemental Information