Dietary reconstruction of pygmy mammoths from Santa Rosa Island of California

Abstract

Microwear analyses have proven to be reliable for elucidating dietary differences in taxa with similar gross tooth morphologies. We analyzed enamel microwear of a large sample of Channel Island pygmy mammoth (Mammuthus exilis) molars from Santa Rosa Island, California and compared our results to those of extant proboscideans, extant ungulates, and mainland fossil mammoths and mastodons from North America and Europe. Our results show a distinct narrowing in mammoth dietary niche space after mainland mammoths colonized Santa Rosa as M. exilis became more specialized on browsing on leaves and twigs than the Columbian mammoth and modern elephant pattern of switching more between browse and grass. Scratch numbers and scratch width scores support this interpretation as does the Pleistocene vegetation history of Santa Rosa Island whereby extensive conifer forests were available during the last glacial when M. exilis flourished. The ecological disturbances and alteration of this vegetation (i.e., diminishing conifer forests) as the climate warmed suggests that climatic factors may have been a contributing factor to the extinction of M. exilis on Santa Rosa Island in the Late Pleistocene.

Introduction

Endemic to the California Channel Islands, the pygmy mammoth (Mammuthus exilis, Maglio, 1970) was initially discovered on the Island of Santa Rosa and later on Santa Cruz and San Miguel in the Channel Island archipelago of California. M. exilis is a small mammoth considered to be a dwarfed form of its likely ancestor, the Columbian mammoth (M. columbi), which occupied the mainland of North America (Madden, 1977, Madden, 1981, Johnson, 1978). Today, the Channel Islands are comprised of eight islands (Fig. 1). In the Late Pleistocene, the four Northern Channel Islands formed a single super-island, dubbed Santarosae by Orr (1968) and lay closer to the mainland than today's Channel Islands. Even during periods of glaciation in the Pleistocene when sea levels were much lower than they are today, the islands were separated from the California coast by a relatively small water gap of around 6.5–8 km (Roth, 1996, Muhs et al., 2015). As sea levels rose due to the melting of continental ice, 76% of Santarosae disappeared (Johnson, 1972) leaving only the highest elevations exposed – now known as the islands of San Miguel, Santa Cruz, Santa Rosa, and Anacapa. Of these modern islands, all but Anacapa have produced mammoth remains (Agenbroad, 2001). The breakup of Santarosae is believed to have taken place about 11,000 cal. BP (Kennett et al., 2008).

Researchers have long been interested in the Channel Islands for several reasons. First, the islands are part of one of the richest marine ecosystems in the world and are home to over 150 endemic species such as the island fox (Urocyon litteralis) – a small fox with six subspecies each unique to the island it lives on. Hence, the Channel Islands are often referred to as the North American Galapagos. Second, because island species are generally regarded as more susceptible to human-induced extinctions than those on continents, and the Channel Islands were initially occupied by humans during a period of extensive extinction in North America (and elsewhere) around 13,000 cal. BP (Erlandson et al., 2011), data from such islands has proven useful for providing context for late Quaternary extinctions on continents such as the Americas and Australia (Steadman and Martin, 2003, Wroe et al., 2006). Third, islands are invaluable for studying evolution and diversification, including the effects of insularity on fauna (Palombo, 2008) such as the so-called “Island Rule”. The Island Rule was first stated by Foster (1964) after comparing numerous island species to their mainland varieties. He proposed that the body size of a species becomes smaller or larger depending on the resources available to it in its environment. The Island Rule posits that certain island species evolve larger size when predation pressure is relaxed (due to the absence of some mainland predators), while others evolve smaller size due to resource constraints regarding availability of food and land area (Whittaker, 1998, McNab, 2010).

The history of the discovery and excavation of pygmy mammoths from the islands is chronicled in Agenbroad (2001). Mammoth remains have been known from the Northern Channel Islands of Santa Rosa, San Miguel, and Santa Cruz since 1856 (Stearns, 1873). A spectacular find was made in 1994 by Park Service researchers led by Larry Agenbroad (a Santa Barbara Museum research associate at the time) of a nearly complete adult male M. exilis skeleton – a mature male of about 50 years of age (Agenbroad, 1998). After this discovery, a thorough pedestrian survey of the islands using GPS was begun, to document and pinpoint each discovery. More than 160 new localities were recorded with the majority of localities found on Santa Rosa Island (Agenbroad et al., 2007). Several mainland-size mammoth (Mammuthus columbi) elements in addition to remains of M. exilis were recovered as a result of this survey. The survey of Santa Rosa revealed a ratio of approximately 3:10 large mammoth:small mammoth remains (Agenbroad, 2012).

As many as three species ostensibly of different sizes have been proposed by researchers over the years (Orr, 1956a, Orr, 1956b, Orr, 1968, Roth, 1982, Roth, 1996). However, Agenbroad (2009) showed convincingly, using less fragmentary material, that only two species are likely present – M. columbi and M. exilis (Fig. 2).

There has been much speculation regarding the origin of mammoths on the Northern Channel Islands, and the finding of Columbian mammoth remains is intriguing given that these remains may represent remnants of an ancestral population, unless they represent later migrants to the island. The oldest remains are found in the basal conglomerate of the Garanon Member of the Santa Rosa Island Formation. U/Th results formerly suggested an age of at least 200 ka (Orr, 1968), but these are not now considered reliable and new U/Th data indicate an age of ca. 80 ka (Muhs et al., 2015). Mammoth remains attributed to both M. columbi and M. exilis have been found throughout the entire Formation (the latest calibrated direct radiocarbon date on M. exilis being 10,700 ± 90 BP (B-14660), equivalent to c. 12,600 cal BP (Agenbroad, 2012).

Also intriguing is the speculation as to how mainland mammoths arrived on the islands. Initially, it was assumed that the ancestral form of M. exilis was either Mammuthus (formerly Archidiskodon) imperator or Mammuthus columbi. It has now been shown that M. imperator and M. columbi are conspecific (Slaughter et al., 1962, Miller, 1971, Miller, 1976, Agenbroad, 2003) and further research has suggested that M. columbi represents the likely ancestor of M. exilis (Johnson, 1978, Madden, 1981, Roth, 1982, Roth, 1996). This suggests that M. exilis evolved according to the Island Rule (Foster, 1964). That is, a large continental species (M. columbi) adapted to an island environment becoming in time a new smaller species – M. exilis (Fig. 2).

The question remains – how did Columbian mammoths reach the islands? Historically, it was assumed that ancestral mammoths could not have swum to the islands. Consequently, various land bridges linking the Northern Channel Islands to the mainland have long been hypothesized (e.g., Clements, 1955, Van Gelder, 1965, Valentine and Lipps, 1967, von Bloeker, 1967, Remington, 1971). This early idea, that insular mammoth remains proved the existence of a land bridge, persisted for many decades (see synopsis by Johnson, 1978). The idea was deeply entrenched but began to give way in light of accumulating geological and biological evidence (Savage, 1967, Johnson, 1978). For example, Johnson, 1972, Johnson, 1978 pointed out that a land bridge was not a sine qua non for explaining mammoths on the Northern Channel Islands, citing research that elephants are excellent distance swimmers, among the best of all land mammals and highly skilled at crossing water gaps. Hence, currently it is now understood that sea-level fall brought the islands only a few kilometers from the shore, enabling mainland mammoths an opportunity to gain access even if a land bridge was not present.

Today the local vegetation of the Northern Channel Islands is dominated by grasslands and shrublands with some patchy woodlands, and mild temperatures with little annual temperature fluctuation (Junak et al., 2007). To date, studies of the vegetation history of the island have included pollen analyses on Santa Rosa Island (Cole and Liu 1994; Kennett et al., 2008), San Miguel Island (West and Erlandson, 1994), continuous offshore pollen cores from the Santa Rosa Basin (Heusser, 1995, Heusser, 2000) and paleobotanical studies on Santa Cruz Island (Cheney and Mason, 1930). During the Late Pleistocene, when these islands existed as the single super-island of Santarosae, forest vegetation was present including cypress, pine, and Douglas fir (Cheney and Mason, 1930, Anderson et al., 2008). Climatic changes at the end of the Pleistocene have been invoked as the cause for the loss of coastal forest (Anderson et al., 2008), and major fires coincident with the last known occurrence of pygmy mammoths (around 13,000 cal BP) have been demonstrated (Kennett et al., 2008). Anderson et al. (2010) combined pollen analysis with an analysis of the fire disturbance regime of Santa Rosa Island and reported that a coastal conifer forest covered the highlands of Santa Rosa during the last glacial but that by ca. 11,800 cal. BP Pinus stands, coastal sage scrub and grassland replaced the forest as the climate warmed.

The purpose of this study is twofold: 1) to explore the paleodietary ecology of Mammuthus exilis from Santa Rosa Island, California using stereomicrowear analysis, and 2) to compare microwear patterns of M. exilis to those of extant elephants and to other mammoths and mastodons from North America and Europe, especially its ancestor M. columbi. To test the hypothesis that M. exilis was exploiting the forest vegetation during the Late Pleistocene, we employed light microscopy dental microwear on a large sample of pygmy mammoth teeth from Santa Rosa Island. If forest vegetation including woody browse was exploited on Santa Rosa Island during the Late Pleistocene, we predict that: 1. M. exilis would exhibit more scratches in the low scratch range (i.e., between 0 and 17) typical of extant mammals that feed on leaves; 2. Some individuals of M. exilis would display the unusually deep and wide scratches (i.e., hypercoarse scratches) found in extant proboscideans and other mammals known to ingest large quantities of twigs and bark.

Microwear has been employed for over three decades as a technique to visualize scars in dental enamel caused by food items such as plant phytoliths or by exogenous grit or dust coating the surface of vegetation (e.g., Rensberger, 1978, Walker et al., 1978, Solounias and Semprebon, 2002, Semprebon et al., 2004, Merceron et al., 2004, Merceron et al., 2005, Scott et al., 2005, Scott et al., 2006, Ungar et al., 2008, Ungar et al., 2010) including studies of mammoths and other proboscideans (Rivals et al., 2012, Rivals et al., 2015). Microwear turns over relatively rapidly, thus giving insight into the dietary behavior of the last days, or weeks, before an animal's death – the so-called “Last Supper Effect” (Grine, 1986). As such, it provides a window into the short-term dietary behavior of a taxon as opposed to the deep-time adaptation of gross tooth morphology, giving insight into what a taxon was actually eating despite what it might originally have been adapted to eat (Rivals and Semprebon, 2011). Importantly, this snapshot into the habitat and dietary behavior of a taxon is largely taxon-independent.

While microwear analyses have proven to be very reliable in diagnosing dietary behavior of mammals, two considerations are relevant to all microwear approaches: 1) potential error involved in the analyses, and 2) the causal factors producing microwear scars. Different microwear methodologies have different strengths and weaknesses, but all have value depending on questions being studied. The three methods currently used involve scanning electron microscopy (SEM), light microscopy for dental microwear (LDM), and dental microwear texture analysis (DMTA). Because SEM and LDM rely on observer measurements, while DMTA employs an automated approach, studies have been done to identify and/or quantify both inter- and intra-observer error involved in each (e.g., Grine et al., 2002, Galbany et al., 2005, Fraser et al., 2009, Mihlbachler et al., 2012 DeSantis et al., 2013, Williams and Geissler, 2014, Hoffman et al., 2015).

While these studies have provided valuable insight regarding elucidating potential error in microwear studies, it is important to not generalize results utilizing one methodology to another. Thus, microwear error studies aimed at testing LDM (other than Semprebon et al., 2004), while adding to the cannon of knowledge, have not duplicated the original LDM methodological regime (i.e., have employed different magnifications, used different counting areas, have not counted all features visible, and/or assessed microwear from photographs – sometimes using untrained observers).

The original methodology developed by Semprebon, 2002 and Solounias and Semprebon, 2002 (and utilized in this study) involves counting pits and scratches directly using a stereoscope at 35 times magnification following the same standard microscope calibration and systematic counting methods employed every day in diagnostic medicine and life science research to accurately count substances such as blood cells and platelets using light microscopy (where differential light refraction as used in this study is used to identify the objects being counted).

For example, the procedure used to construct the comparative extant ungulate and proboscidean databases (Semprebon, 2002, Rivals et al., 2012) and to analyze the fossil proboscideans used in this study relies on employing a calibrated ocular reticle subdivided into 9 subsquares which greatly enhances the ability to not lose track of features counted rather than attempting to count numerous features scattered within a larger, undivided field of view. Additionally, features are counted according to a strict protocol that ensures that features present are not skipped over or counted more than once (Estridge and Reynolds, 2012). Also, a relatively large counting area is employed and features are counted twice (to reduce error due to potential variable scar distribution). When this protocol is followed, both interobserver and intraobserver error has been demonstrated to be low (Semprebon et al., 2004).

In some studies, however, feature counts have been made from images using much higher magnifications than used in Semprebon et al. (2004) and counting error would naturally be expected to be higher due to the greater number of features visible at higher magnifications. For example, Mihlbachler et al. (2012) reported that interobserver error was higher than that reported by Semprebon et al., 2004, although evidence was found (2012) that error diminished significantly with increased experience of the observer and data collected by all observers was highly correlated (Mihlbachler et al., 2012). Also, while DMTA has been employed to extract 2D microwear data from photosimulations of scanned areas of DMTA, (e,g., DeSantis et al., 2013), stereomicroscopy was invented to provide depth of field and a three dimensional visualization of an object being examined and thus is widely used in microsurgery, dissection, forensics, medicine, industry, and quality control. In microwear analysis as employed here and in Solounias and Semprebon (2002) and Semprebon et al. (2004), stereomicroscopy allows relative depth and 3D shape of different features to be assessed.

While fully automated methodologies such as DMTA have been developed and have proven useful to reduce subjectivity and interobserver error (Ungar et al., 2003, Galbany et al., 2005) it is important to note, that DMTA reduces interobserver error only when the same exact location on the same tooth surface is scanned twice, as results obtained on the same tooth and same facet may be quite different when different regions of that tooth and facet are scanned (see Fig. 8 in Scott et al., 2006). This difference is due to the natural heterogeneity of microwear on tooth surfaces, a propensity which affects all microwear methodologies but is more problematic at higher versus lower magnifications and when only small areas of a tooth are analyzed. If relatively low magnification and a large counting area is assessed such as in the Solounias and Semprebon (2002) methodology, both inter- and intra-observer error may be minimized (see Williams and Geissler, 2014).

Regardless, the effects of potential error should be evaluated within the context of whether accurate dietary categorization of living mammals with known diets can be obtained by single observers. A variety of different variations on the original LDM methodology have produced great fidelity in terms of partitioning living mammals (e.g., carnivores, cetaceans, proboscideans, primates, marsupials, ungulates, and rodents) into their known trophic groups (Semprebon et al., 2004, Merceron et al., 2004, Merceron et al., 2005, Rivals et al., 2007, Rivals et al., 2012, Townsend and Croft, 2008, Fraser et al., 2009, Williams and Patterson, 2010, Fraser and Theodor, 2011, Fraser and Theodor, 2013, Williams and Holmes, 2011, Bastl et al., 2012, Fahlke et al., 2013, Christensen, 2014, Williams and Geissler, 2014).

The causal factors involved in producing microwear scars are only beginning to be understood. Microwear patterns must be influenced by the differing physical properties of enamel and ingested abrasives such as biogenic silica (i.e., phytoliths) from plants, or abiotic silica from dust or grit. While it is also possible that differing tooth morphologies and masticatory biomechanics may impact the nature of microwear scars, a similar pattern of microwear scars have been observed across various orders of mammals consuming similar dietary items which seems to argue that such effects are relatively minor (e.g., Semprebon et al., 2004) although further study is needed. A number of studies have tested the effects of abrasives in producing microwear scars and have expanded our thinking away from considering phytoliths alone as contributing to microwear scar topography (Covert and Kay, 1981, Kay and Covert, 1983, Maas, 1991, Maas, 1994, Gügel et al., 2001, Mainland, 2003, Sanson et al., 2007, Lucas et al., 2013, Schulz et al., 2013, Müller et al., 2014, Hoffman et al., 2015). Studies examining the relative hardness of enamel versus phytoliths have, however, produced contradictory results (Baker et al., 1959, Sanson et al., 2007, Rabenold and Pearson, 2014, Xia et al., 2015). The most promising studies regarding teasing apart causal factors involve controlled feeding experiments. These studies have elucidated patterns in such widely diverse taxa as American opossums (Covert and Kay, 1981, Kay and Covert, 1983), rabbits (Schulz et al., 2013), and ungulates (Hoffman et al., 2015). Such studies have documented an increase in scratch numbers in rabbits when their diet contained grass with more biogenic abrasives (Schulz et al., 2013) and in opossums that were fed fine-grained pumice (Covert and Kay, 1981, Kay and Covert, 1983). Hoffman et al. (2015) provided the controlled feeding experiment with domesticated sheep, and tested the effects of abiotic abrasives of different sizes. This study reported a significant increase in pit features that was correlated with an increase in grain size. This observed grit effect supports the interpretation of increased grit consumption by extant ungulates from semi-arid and arid environments due to increased pitting observed in these forms (Solounias and Semprebon, 2002).

LDM (after Semprebon, 2002, Solounias and Semprebon, 2002) was used in this study as it allows for the attainment of relatively large sample sizes (numbers of specimens), and a very large comparative stereomicrowear database exists which is comprised of extant artiodactyls, perissodactyls, and proboscideans of known diets compiled by a single observer (GMS). Relatively few studies have concerned themselves with proboscidean microwear patterns (examples are Palombo and Curiel, 2003, Palombo, 2005, Green et al., 2005, Todd et al., 2007 and Calandra et al., 2008, Calandra et al., 2010). Rivals et al., 2012, Rivals et al., 2015 studied dietary diversity patterns in Pleistocene representatives of Mammut, Anancus, Palaeoloxodon and Mammuthus from a variety of localities in Europe and North America. In this study, we test the effect of insular dwarfism on dietary niche occupation by comparing Pleistocene mammoths from the North American mainland with island mammoths from the Channel Islands off the coast of California.