The five large stratigraphic layers (or phases) at Hortus cave record cycles of cold, alternating between more and less intense, combined with variation in levels of humidity and increasingly arid conditions. The major phases are interspersed between short interphases that exhibit a greater resemblance to temperate climates (Table 1). Paleosols evince a paleoclimate that was relatively cold corresponding to the ecological variation noted to have occurred in MIS 3 (Lumley, 1972 and Lumley, 1973).

Dental Microwear Texture Analysis (DMTA) has been previously utilized to reconstruct fossil hominin diets (Scott et al., 2005 and Ungar et al., 2012), including those of Neandertals (El Zaatari et al., 2011, El Zaatari et al., 2016, Estalrrich et al., 2017, Karriger et al., 2016 and Krueger et al., 2017). DMTA captures three-dimensional textures on worn enamel surfaces that correspond to the mechanical properties of the foods consumed (Schmidt et al., 2016, Scott et al., 2012, Scott et al., 2006, Scott et al., 2005 and Ungar et al., 2012). Antemortem dental microwear reveals the foods masticated in the last few days or weeks of an individual's life (Ungar, 2015). As such, short-term dietary changes and seasonality can be inferred from DMTA. The most profound differences in Neandertal DMTA signatures exist between individuals from paleoenvironments that reflect open habitats and those from sites that suggest greater vegetation coverage or wooded settings – the former characterized by high meat consumption and the latter emphasizing mixed meat and vegetable foods (El Zaatari et al., 2011, El Zaatari et al., 2016 and Estalrrich et al., 2017). A similar, habitat-driven difference in anterior tooth-use has also been documented among Neandertals using DMTA (Krueger et al., 2017). Krapina was shown to differ from Vindija in its elevated anisotropy, or patterning, suggesting substantial amounts of tough vegetation, such as plant fibers or underground storage organs, may have been consumed (Karriger et al., 2016). Additional research on Neandertal buccal microwear (Pérez-Pérez et al., 2003, Pérez-Pérez et al., 2017 and Pinilla et al., 2) and occlusal macrowear (Fiorenza, 2015, Fiorenza et al., 2011 and Pinilla et al., 2017) also suggests extensive variability in Neandertal diets with respect to chronology, paleoenvironment, climate and cultural factors.

Texture variables from DMTA capture different aspects of dietary diversity. For example, herders, who consume large amounts of meat, exhibit lower values for dental microwear textures than do agriculturalists (Schmidt et al., 2016). Meanwhile, plant foods, or food processing strategies, can introduce grit and other hard particles such as phytoliths and seed shells that tend to inflict damage to the enamel surface resulting in higher textural values (Calandra et al., 2012, DeSantis et al., 2013 and Karriger et al., 2016). Because different profilers, file parameters and generations of software compromise the complementarity between machines (Arman et al., 2016), all the textural data presented here are from a single profiler at the University of Indianapolis.

If the individuals preserved at Sub-Phase IVb and Phase V of Hortus consumed similar seasonally based resources, they should have a uniform dietary profile unless the increasingly cold and dry conditions influenced the availability of plant resources. Based solely on stratigraphy, it would be expected that individuals excavated from Sub-Phase Vb, Hortus III, Hortus V and Hortus VIII, should cluster in dental microwear textures. Hortus XI from Sub-Phase Va should be secondarily joined to the individuals in Sub-Phase Vb. It may be expected that Hortus IV, from Sub-Phase IVb which is older, should differ from the individuals from Sub-Phase V which was colder and more arid. A recent analysis of microwear textures from the El Sidrón Neandertals contextualized sample variation with respect to age, sex and maternal lineage, and found key differences in anisotropy (epLsar) by sex and some idiosyncratic variation in scale of maximum complexity (Smc) (Estalrrich et al., 2017). The extensive life-stage variation among the Hortus Neandertals provides an impetus for exploring microwear textures with respect to age as well as chronology.

It is expected that the six individuals from Hortus will have a greater resemblance to one another than to Neandertals from a diverse number of ecogeographic regions and temporal horizons, from MIS 7 to MIS 3. Given prior evidence of the role of ecogeography in explaining Neandertal diets (El Zaatari et al., 2011), there may be a difference between Neandertals from the circum-Mediterranean and those from farther north. To address this question, the Hortus assemblage is compared to Neandertals from the circum-Mediterranean sites of Kebara 2 and Tabūn Series III from northern Israel and as well as a wide range of continental sites including those in central Europe (Kůlna and Švédův stůl), the French Pyrenees (Montmaurin and Malarnaud), Southwest France (La Quina 5) and the Meuse River Basin of Belgium (Spy 1). It is expected that Neandertals, known for their heavy meat consumption (Richards and Trinkaus, 2009), should have lower textural values in colder, inland habitats with little forest resources than those in circum-Mediterranean ecozones (El Zaatari et al., 2011, Fiorenza et al., 2011 and Fiorenza et al., 2015). To reconstruct the diet of the Hortus assemblage, they are compared to the published mean values for complexity and anisotropy for twelve Holocene societies (Karriger et al., 2016).

The permanent and deciduous molars in this study included Hortus III, Hortus V and Hortus VIII from Sub-Phase Vb; Hortus XI from Sub-Phase Va; Hortus IV from Sub-Phase IVb; and Hortus VI which is unassociated with respect to phase (Lumley, 1973) (Table 2).

The comparative sample was chosen to provide climatic and temporal extremes of western Eurasia from MIS 7 to MIS 3, and includes the molars of Kebara 2, dated to approximately 62 kyr BP (MIS 4) and Tabūn Series III from MIS 5, both from northern Israel (Grün and Stringer, 2000 and Krueger et al., 2017); those of Kůlna and Švédův stůl (Ochoz 1) of the Moravian Mountains of the Czech Republic, both of which are likely to be from MIS 3 (Krueger et al., 2017; Table 3); and Spy 1 from the Meuse River Basin of Belgium, dated to near the terminus of MIS 3 at 36 kyr BP (Semal et al., 2009). The comparative sample also includes Neandertals somewhat closer to Hortus cave, including La Quina 5 from Level 3 of La Quina cave in Charente of Southwest France, discovered with middle Mousterian tools, and dated to MIS 3 at circa 43–47 kyr BP (Discamps and Royer, 2017), and the French Pyrenees sites of Malarnaud dated to MIS 5 (Petite-Marie et al., 1971) and Montmaurin, discovered in La Niche, level C3, which has been dated to MIS 7 (Crégut-Bonnoure et al., 2010) (Table 3). Although the comparative sample is limited (n = 8), the isolated fossil sites include much of the chronological and geographic range of Neandertals from western Eurasia.

Dental molds were created at the Centre européen de recherches préhistoriques de Tautavel and the Musée de l’Homme by FLW, using polyvinylsiloxane (Colètene-Whaledent). Roc de Marsal from the collections of the Musée national de préhistoire, Les Eyzies, was also molded, but postmortem taphonomic processes led to the exclusion of this fossil from the study. Dental molds of Spy 1 were loaned to FLW from the Institut royal des sciences naturelles de Belgique by Patrick Semal. Dental casts from the molds were created from centrifuged epoxy resin and hardener (Buehler) at Georgia State University and allowed to air-dry for 24 hours before extraction. Additional dental casts (Kebara 2, Tabūn Series III, Kůlna and Švédův stůl) were provided by Erik Trinkaus.

The dental casts were scanned at the University of Indianapolis using the “Indy” machine, a white-light confocal profiler (Sensofar Plμ) at 100 × magnification (Solarius Development Inc.). Scanning occurred on the enamel surface of the protocone of the maxillary molars and on the protoconid of the mandibular molars on phase II facets, which have been shown using DMTA to be more effective than phase I facets in distinguishing known dietary proclivities (Krueger et al., 2008). We focused on Facet 9 unless it was unsuitable for study. Then, we looked for Phase II facets adjacent to Facet 9. Only surfaces that were free of irregularities originating from fossilization, preparation or casting were considered. For each specimen, four scans were captured from a 138 × 102 μm viewing field. The total sampled area of 276 × 204 μm was condensed to a study area of 242 × 182 μm once the scans were stitched together using an automated process to form a single point cloud for each individual. To observe surface morphology, two- and three-dimensional surface representations were created (Fig. 1 and Fig. 2).

Once the scans were leveled within the program SolarMap Universal, the point clouds were analyzed using scale-sensitive fractal analysis (Schmidt et al., 2016 and Scott et al., 2006), which yielded four variables that describe the textural properties of the enamel surface. To consider roughness and relative areas at different scales of observation, an area-scale fractal complexity (Asfc) algorithm was employed which compares surface texture at 7200 μm² to 0.02 μm². The mastication of hard and brittle foods will generate higher complexity than the consumption of tough, fibrous foods (Schmidt et al., 2016, Scott et al., 2012, Scott et al., 2006 and Ungar et al., 2012). Scale of maximum complexity (Smc) measures aspects of enamel surface deformation that are not necessarily captured by complexity (Asfc), separating hard vs. hard-brittle food consumption (Scott et al., 2006).

Anisotropy (epLsar) or the “exact proportion of Length-scale anisotropy of relief” serves as a proxy for the degree to which foods are dragged along the occlusal surface which creates a concentrated patterning of enamel surface texture (Scott et al., 2012 and Scott et al., 2006). Surfaces characterized by numerous parallel scratches are expected to yield high values for anisotropy.

Textural fill volume (Tfv) compares the volume of the scanned surface using square cuboids with different lengths (10 and 2 μm) to estimate the degree to which damage to the enamel surface is the result of dental microwear as compared to facet curvature (Scott et al., 2006). High textural fill volume characterizes hard-object consumers (Scott et al., 2012) and may reflect the consumption of mechanically challenging particles of different sizes (El Zaatari and Hublin, 2014).

Texture variables are plotted against one another using a convex hull around individuals from Hortus and the values for the eight Neandertals from isolated sites. A convex hull is generated using the outermost values to define a closed set encompassing the smallest possible area demarcated by the entire sample. Two bivariate plots are shown, complexity (Asfc) compared to anisotropy (epLsar) and scale of maximum complexity (Smc) versus textural fill volume (Tfv). The Hortus assemblage, grouped with a convex hull, is also compared to a schematic representation of complexity and anisotropy using hunter-gatherers, agriculturalists and pastoralists, including Early Bronze Age England, Late Bronze Age England, Iron Age England, Iron Age Nepal (Mebrak), Bronze and Iron Age Greece, Mongol Xiongnu (Xiongnu Period Mongolia) and Bronze Age/Iron Age Mongolia from Schmidt et al. (2016); Natufian and Neolithic Israel from Chiu et al. (2012); Indiana Middle Woodland, Indiana Middle Woodland East and Indiana Middle to Late Archaic from Frazer (2011); and Kentucky Archaic from Karriger et al. (2016).

Individuals from Hortus fall into two groups for anisotropy (epLsar) and textural fill volume (Tfv). Those with elevated anisotropy (epLsar) include Hortus IV with the highest value, followed by Hortus VIII, Hortus VI and Hortus V (Fig. 3). Those with elevated anisotropy are all near adults or young adults, estimated to be between 17 and 34 years by de Lumley (1973) (Table 2). The second group includes those individuals who are not in the young adult category such as Hortus III aged to 6.5–7.9 years (Ramirez Rozzi, 2005) with the lowest value, and Hortus XI aged to 50+ years (Lumley, 1973) with the second lowest anisotropy value (Fig. 3; Table 4).

These patterns are replicated for textural fill volume (Tfv) where young adults Hortus IV, Hortus VI and Hortus VIII are polarized from Hortus III and Hortus XI which exhibit the lowest, nearly identical, values (Table 4; Fig. 4). Hortus V falls between these extremes but is closer to the young/old grouping with relatively low values for textural fill volume (Tfv). Young adults Hortus IV and VIII have the highest textural fill volume (Tfv) and anisotropy (epLsar).

The Hortus assemblage shows a uniformity of complexity (Asfc) and scale of maximum complexity (Smc) when compared to isolated Neandertal sites (Fig. 3 and Fig. 4). Hortus XI and Hortus VI exhibit elevated complexity (Asfc) in the range of Montmaurin and Tabūn Series III, whereas Spy 1 followed by Kůlna 1, are outliers (Fig. 3). In contrast, Hortus V exhibits the lowest complexity (Asfc) of the Hortus assemblage which is similar to those of Malarnaud and Svédův stůl, followed by Hortus VIII and Hortus III (Fig. 3). When compared to the isolated sites, Hortus adults exhibit among the highest values for anisotropy (epLasr) as well as the lowest, exhibited by Hortus III (Fig. 3). For scale of maximum complexity (Smc), the Hortus assemblage contrasts with Kebara 2, which is not shown in Fig. 4 given its extreme value (7.3), followed by Kůlna 1 and Malarnaud (Table 4).

All Sub-Phase Vb molars, including Hortus III, Hortus V and Hortus VIII, are characterized by low complexity (Asfc) while the others from Hortus cave exhibit higher values (Fig. 5). The lowest complexity (Asfc) values for Holocene humans are for pastoralists (Schmidt et al., 2016), followed by Middle/Late Archaic groups from North America as well as Bronze/Iron Age Eurasians (Karriger et al., 2016). Hortus adults largely exceed the published mean anisotropy (epLsar) values for agricultural societies, which range from 0.0034 for Neolithic Israel to 0.0041 for Early Bronze Age England, albeit with considerable variability (Karriger et al., 2016). In contrast, the anisotropy (epLsar) for Hortus III and Hortus XI resembles those of Holocene hunter-gatherers more than agricultural and pastoral groups (Fig. 5).

It is noteworthy that Hortus III and XI both have the lowest anisotropy (epLsar) and textural fill volume (Tfv). Only Hortus adults exhibit high values for these two textures. The polarization of the juvenile (Hortus III) and older adult (Hortus XI) compared to young adults suggest that differences in food processing regimes or dietary behavior was based on age. Hortus III is from Sub-Phase Vb and should have been grouped with adults Hortus V and Hortus VIII if only phase explained the results. However, the young adults are grouped regardless of phase, suggesting long-term cultural specializations and/or dietary adaptations. Furthermore, Hortus IV from the earlier relatively warmer and wetter period of Sub-Phase IVb, does not differ substantially in anisotropy (epLsar) or textural fill volume (Tfv) from other young adults that derive from Sub-Phase Vb, but does differ from older adult Hortus XI from Sub-Phase Va. It might be expected that Hortus IV (Sub-Phase IVb) and Hortus XI (Sub-Phase Va) should be the most similar to one another given their proximity in time (Table 1). However, the opposite is true and these two have disparate textural values, both grouping instead with life stage.

However, these sub-phases are distinct with respect to complexity (Asfc). Food hardness was lower in Sub-Phase Vb, a period when extreme cold and aridity prevailed suggesting a greater reliance on softer foods, such as meat (Table 1; Fig. 5). Hortus XI from Sub-Phase Va has the highest degree of complexity of the assemblage and resembles Indiana Middle Woodland societies reconstructed as consuming substantial amounts of nuts and seeds (Frazer, 2011 and Karriger et al., 2016).

A principal aim of this study was to examine whether the individuals from Hortus cave cluster together or whether they are dispersed. The dental microwear textures of scale of maximum complexity (Smc) do cluster the assemblage from Hortus, whereas complexity, anisotropy and textural fill volume disperse the Hortus assemblage into two relatively distinct groupings. None of the individuals from Hortus are outliers, including Hortus VI, which is unassociated with respect to phase, suggesting that indeed, the assemblage forms a group and where differences are present; they are largely based on age, and to a lesser extent, chronology.

Hortus cave can be reconstructed as exhibiting a Mediterranean climate. The circum-Mediterranean tends to be relatively arid with seasonal rainfall. In many respects, the Hortus assemblage is similar to Kebara 2 and Tabūn Series III indicating an ecogeographic grouping of individuals from the cirum-Mediterranean. Inland climates tended to be colder during MIS 3 and may have relied more heavily on meat (Fiorenza et al., 2011 and Fiorenza et al., 2015). Whereas hard food consumption would be expected to increase complexity, lower values would indicate the ingestion of fewer hard foods, such as phytoliths, hard seed shells, marine shells, grit, clay and sand (Calandra et al., 2012, DeSantis et al., 2013 and Schmidt et al., 2016). Higher levels of anisotropy (epLsar) would indicate a greater reliance on tough foods, such as plant fibers (Karriger et al., 2016). The individuals from Hortus did not have a soft diet with elevated values for textural fill volume (Tfv) and, to a lesser degree, complexity (Asfc). Like Neandertals from Krapina and Vindija, adults and near adults from Hortus likely processed or consumed tough fibrous foods or fibers (Karriger et al., 2016).

While data on sex is not available for the Hortus sample, differences in non-masticatory anterior dental wear features (Estalrrich and Rosas, 2015) and molar microwear textures (Estalrrich et al., 2017) have been reported among some Neandertals. Extensive non-masticatory wear is well documented in the anterior dentition from Hortus (Estalrrich and Rosas, 2015 and Lumley, 1973). At El Sidrón, elevated anisotropy (epLsar) among females may correspond to sex-based divisions of non-masticatory or dietary behaviors (Estalrrich et al., 2017) and a similar scenario could explain some of the variation in anisotropy at Hortus.