Body size is one of the most important dimensions of biology. More specifically, body mass influences many structural, physiological, biomechanical and ecological features across different geographical and temporal scales (Hildebrand, 1974; Peters, 1983; Damuth & MacFadden, 1990; Vizcaíno et al., 2016). In this context, body mass is very useful to get insights into extinct species and ecosystems. For example, it was used, particularly in tetrapods, to infer biological characteristics in some species (e.g., brain size, home range, oral food processing, size musculature; e.g., Jerison, 1970; Hurlburt, 1996; Palmqvist & Vizcaíno, 2003; Vizcaíno et al., 2006; McGuire & Dudley, 2011; Cuff et al., 2017; Vidal-Cordasco et al., 2017) or study the composition and structure of the paleo-communities and -ecosystems (e.g., Dewar, 2003; Costeur & Legendre, 2008; Rodríguez-Gómez et al., 2016; Schroeder et al., 2021; Wyenberg-Henzler et al., 2022). In addition, inferring body mass in extinct species allows exploring many macroevolutionary questions on feature evolution and taxonomic diversification (e.g., Legendre, 1986; Laurin, 2004; Benson et al., 2014, 2018; Ksepka et al., 2020).
Since early in the 20th century (Gregory, 1905), different methods were developed to accurately estimate body size. They can be classified into two main categories: the extant-scaling and volumetric-density approaches (Campione & Evans, 2020). The former is rooted in the quantitative study of the relative growth and proportions of body parts. It was demonstrated that there is a simple and power law between the size of body elements (ES) and the whole body size (BS) of an organism: ES = aBSb, where a is a constant and b the allometric exponent (Huxley, 1924, 1932). Considering this scaling relation, body mass of fossils can be estimated from morphometric or morphogeometric variables of skeletal traits taken on bones and teeth (e.g., Myers, 2001; Sarko et al., 2010; Ercoli & Prevosti, 2011; Campione & Evans, 2012; Cassini et al., 2012). Firstly, scientists established proportions or ratio diagrams in extant species (e.g., Fisher, 1945; Amadon, 1947; Vucetich et al., 2015), and next, they used predictive equations (Gingerich, 1990; Engelman, 2022a). The latter should relate, as accurately as possible, the body mass of extant species. Then, the obtained equation is applied to fossils to estimate their mass through selected variables (e.g., Jerison, 1970; May, 1982; Anderson et al., 1985; Andersson, 2004; Christiansen & Fariña, 2004; Mendoza et al., 2006; Cassini et al., 2012; Smith, 2016; O’Brien et al., 2019; Engelman, 2022a; Filippini et al., 2022). For mammals, a clade that has been intensively analyzed in this context, there are many examples using as proxy the whole body (e.g., Jerison, 1970; MacFadden, 1986; Van Valkenburgh, 1990) or different body parts: teeth (e.g., Gingerich et al., 1982; Legendre, 1986; Van Valkenburgh, 1990; Rinderknecht & Blanco, 2008); skull (e.g., Van Valkenburgh, 1990; Mendoza et al., 2006; Rinderknecht & Blanco, 2008; Sarko et al., 2010; Cassini et al., 2012; Bertrand et al., 2016; Engelman, 2022a); and postcranial elements (e.g., McHenry, 1975; Anderson et al., 1985; Gingerich, 1990; Alberdi et al., 1995; Wroe et al., 1999; Andersson, 2004; Ercoli & Prevosti, 2011; Toledo et al., 2014).
For a same extinct species, the body mass estimations from predictive equations vary across studies depending on several interrelated factors: (i) type of proxy (body part, measure or landmark configuration, uni- vs multi-variate data); (ii) reference sample (body mass data, taxonomic diversity, taxonomic/functional similarity, match of body mass range with target sample); and (iii) type of regression methods (quantity of predictors, methods of fitting; Smith, 2002; Ruff & Niskanen, 2018). Particularly, there has been much debate regarding which is the best estimator of size in mammals (e.g., Fariña et al., 1998; Elissamburu, 2012). Beyond biases linked to, for example, locomotor habits, the proximal limb bones and joints are weight-bearing structures and can be considered as better estimators of body mass than distal limb elements and teeth, possessing the lowest prediction errors (Damuth & MacFadden, 1990; Egi, 2001; Ercoli & Prevosti, 2011; Figueirido et al., 2011; Elissamburu, 2012; Goodwin & Bullock, 2012; Tarquini et al., 2018). For the cranium, the occipital condylar width, which is highly conserved across mammals and not biased by cranium size, appears to be an accurate predictor (Engelman, 2022a). With respect to the teeth, they usually present high prediction errors, which would be in relation to the non-direct link to body mass support, and additional bias factors such as variations in head-body proportions, physiological constraints, and food processing modes (e.g., Janis & Fortelius, 1988; Damuth & MacFadden, 1990; Janis, 1990; Van Valkenburgh, 1990; Lucas, 2004; Vizcaíno et al., 2006; Cassini et al., 2015; Croft et al., 2020). Among teeth, authors generally had preferred to use the first and second molars, especially the lower ones (e.g., Gingerich et al., 1982; Legendre, 1986, 1989; Egi et al., 2004; Costeur & Legendre, 2008). This locus selection was based on the fact that these teeth would be the best correlated with body size (Gingerich et al., 1982), and/or have the lowest prediction error (Egi et al., 2004) or the lowest coefficient variations (e.g., Gingerich, 1974; Gingerich & Ryan, 1979; Gingerich & Schoeninger, 1979; Polly, 1998; Egi et al., 2004; Natsume et al., 2008). In rodents, as the dimensions of first lower molar (m1) would be however inadequate for estimating body mass owing to high size variations in relation to the body mass of some groups (i.e., muroids, extinct Mylagaulidae; Morgan et al., 1995; Hopkins, 2008), the length and area of cheek tooth row were proposed as relatively accurate proxies (Hopkins, 2008; Millien & Bovy, 2010; Freudenthal & Martín-Suárez, 2013; Moncunill-Solé et al., 2014). According the results of Moncunill-Solé et al. (2014), the best estimator between the dimensions of the m1 and those of the cheek tooth row would depend of the taxonomic group. In any case, beyond these discussions, teeth, and in second instance skulls, are the only available proxies for estimating the mass of fossils in a vast majority of extinct mammals (Damuth & MacFadden, 1990; Engelman, 2022a).
Morphologies, body proportions, and patterns of scaling are highly variable in mammals (e.g., Hilderbrand, 1974; Gould, 1975; Fortelius, 1985; Bou et al., 1987; Pollock & Shadwick, 1994; Christiansen, 1999, 2002; Hilson, 2005). For the extant-scaling approach, a consistency of the relationship body mass-predictor variable(s) is assumed between the extant taxa used to generate the model and the extinct taxa for which a body mass estimate is desired (Campione & Evans, 2020). Hence, a taxonomic and/or functional similarity between the reference and target samples is usually recommended (e.g., Smith, 1985; Damuth & MacFadden, 1990; Lucas, 2004; Evans et al., 2007). Nevertheless, wide data bases with large taxonomical and ecological diversity can be considered in the case where the estimator is highly conserved (Campione & Evans, 2012; Engelman, 2022a). In addition, such reference sample showed a higher predictive power than more restricted ones using multiple regressions (stepwise method), where the quantity and quality of predictors would compensate the low or non-consistency (Mendoza et al., 2006; de Esteban-Trivigno et al., 2008). The abundance and non-independency of each taxon within the reference sample would also affect the regression analyses, especially in the context of diverse comparative datasets (Clutton-Brock & Harvey, 1977; Felsenstein, 1985; Smith, 2002). Several methods of differential weighting, including phylogenetic regressions, were proposed to correct these biases of abundance (McClelland, 1980; Mendoza et al., 2006) and non-independency (e.g., Cheverud et al., 1985; Felsenstein, 1985; Grafen, 1989; Diniz-Filho et al., 1998). Taxonomic and phylogenetic corrections in scaling studies can increase the accuracy and/or precision of body size predictions, in addition to their realism (Mendoza et al., 2006; de Esteban-Trivigno et al., 2008; Ercoli & Prevosti, 2011; Pyenson & Sponberg, 2011; Cassini et al., 2012; but see Engelman, 2022a and Nelson et al., 2023). The covered size range of the reference sample is another point to treat with caution (Reynolds, 2002). The mass predictions that require extrapolation beyond the range of modern taxa can be seriously biased if the relationship does not hold beyond the data range (Hahn, 1977; Campione, 2017). In addition, even if the assumed relationship is correct, the precision of the extrapolation is often extremely poor (Hahn, 1977). However, the impact of the aforementioned factors (i.e., consistency between the reference and target samples, taxonomic abundance and non-independency, and size range of the reference sample) on estimations would vary depending of the taxa. Results of studies in bovids and primates showed that considering these factors does not improve the predictive capacity (de Esteban-Trivigno & Köhler, 2011; Dagosto et al., 2018). The choice of predictor variable could affect more than the choice of reference sample and regression methods, and phylogenetic correction and extrapolation could differently impact the performance of estimator in this case (Dagosto et al., 2018).
Caviomorphs are one of the most varied groups of South American mammals (Eisenberg & Redford, 1999; Patton et al., 2015; Wilson et al., 2016). Among rodents, they display a remarkable eco-morphological variation, including an unparalleled size range that extends from ~ 50 g to 60 kg nowadays (Wilson et al., 2016). In addition to the hydrochoerine cavioid Hydrochoerus hydrochaeris, the heaviest living rodent, several caviomorph species have a large size (> 5 kg) in the context of rodents (e.g., Cavioidea: Cuniculus paca, Dolichotis patagonum; Chinchilloidea: Dinomys branickii, Lagostomus maximus; Erethizontoidea: Erethizon dorsatum; Octodontoidea: Capromys pilorides, Myocastor coypus; Vucetich et al., 2015; Álvarez et al., 2017: Supporting Information S3). In the past, the body size range of the group was higher with the records, in the four superfamilies, of small-(< 50 g) to large-sized species (> 60 kg), and even giant-sized species (> 100 kg) (Rinderknecht & Blanco, 2008; Antoine et al., 2012; Vucetich et al., 2015). The Hydrocheriinae, Erethizontidae, Eumegamyinae, Tetrastylinae, Neoepiblemidae, and “Heptaxodontinae” particularly showed large to giant representatives (Biknevicius et al., 1993; Sánchez-Villagra et al., 2003; Candela, 2004; Rinderknecht & Blanco, 2008; Vucetich & Deschamps, 2015; Vucetich et al., 2013, 2015; Pérez et al., 2017). The most remarkable examples are within chinchilloids, represented by the eumegamyine dinomyid Josephoartigasia, whose weight would have been at least 254 and up to 2584 kg according other estimations (Rinderknecht & Blanco, 2008; Millien, 2008; Vucetich et al., 2015; Engelman, 2022b), and the neoepiblemid Phoberomys, which would have weighted between 108 and 750 kg (Sánchez-Villagra et al., 2003; Hopkins, 2008; Millien & Bovy, 2010; Geiger et al., 2013; Vucetich et al., 2015; Engelman, 2022b).
In addition to different methodological approaches and sampling issues mentioned above, a recurrent problem when estimating of the size of giant caviomorphs is that the comparative sample of extant caviomorph species does not include such large sizes. Previous contributions included large extant rodents (e.g., capybaras, beavers, marmots; Hopkins, 2008; Millien, 2008; Millien & Bovy, 2010; Bertrand et al., 2016) improving the estimations for middle-large sized caviomorphs but without being able to solve the problem of falling into extrapolations (Reynolds, 2002). Thinking how to tackle the problem, Millien (2008) proposed to also consider herbivores of other order(s) such as artiodactyls or perissodactyls with living representatives with sizes similar to that reached by giant extinct caviomorphs. In order to infer the body sizes of extinct caviomorph rodents, including representatives of four superfamilies, we propose to construct predictive equations including almost all genera of extant caviomorphs (49/55; Mammal Diversity Database, 2022) and a diverse sample of small- to large-sized extant rodent and non-rodent herbivores. To accomplish this objective, we used dental and cranial measures in non-phylogenetic and phylogenetic frameworks. This research also contributes to respond to more specific questions, such as, (1) What are the best tooth loci and dental variables to estimate the body mass of caviomorphs?; (2) Is the length of the cranium a better proxy than teeth?; (3) Does the inclusion of non-rodent herbivores in the reference sample allow improving the estimates?; (4) Which regression models (multiple vs. simple; taking into account phylogeny or not) have better predictions?; (5) What are the dental and cranial allometric patterns?.