Fortelius and Solounias (2000) introduced the mesowear method to reconstruct the dietary behaviour of fossil herbivores by determining mechanical features of plant parts and their impact on tooth wear of selenodont and selenolophodont herbivores. These features can help, among other, to infer on vegetation structures and hence paleoenvironmental conditions.

Mesowear patterns are the result of a combination of attrition and abrasion, i.e. the relative contributions of tooth-on-tooth and food-on-tooth wear (Fortelius and Solounias, 2000). Attrition occurs when consumed plant parts, e.g., leaves, are soft, because they offer little resistance to teeth, while abrasion is caused by resistive parts of plants, e.g., phyhtoliths of dry grasses but also from external particles eventually consumed along with plant parts (Fortelius and Solounias, 2000 and Lucas et al., 2014).

The original mesowear method by Fortelius and Solounias (2000) comprises two variables: the height of the occlusal relief and the shape of the tooth cusps. Treated as separate variables by Fortelius and Solounias (2000), other authors began to combine the variables, because they are correlated. Sharp cusps are present in teeth with high occlusal relief, while high and low occlusal reliefs can feature rounded cusps. Blunt cusps occur in occlusal reliefs with low to no height (e.g., DeMiguel et al., 2010, Kubo and Yamada, 2014, Kubo et al., 2015, Mihlbachler and Solounias, 2006, Mihlbachler et al., 2011, Rivals et al., 2007 and Strani et al., 2018). In addition, the data obtained with the original method by Fortelius and Solounias (2000) has a major disadvantage. The same sample can be scored very differently by individual observers and is therefore not observer-independent. A standard that combines the two parameters, i.e. height of the occlusal relief and cusp shape, and defines stages and thus simplifies data analyses, solves these problems (Mihlbachler et al., 2011).

We transfer the classification system developed by Mihlbachler et al. (2011) to ruminants and design a specific mesowear ruler. The virtual ruminant ruler (supplementary material S1) is composed of the outlines of seven exemplary tooth cusps representing characteristic mesowear stages. Each mesowear stage is represented by a number simplifying statistical analysis, increasing the comparability with other data sets and decreasing inter-observational errors. The stages are based on combinations of the variables from the original method: high occlusal relief and sharp cusps (stage 0), high occlusal relief and rounded cusps (stage 2), low occlusal relief and rounded cups (stage 4), and low occlusal relief and blunt cusps (stage 6). Mesowear stages 1, 3, and 5 are intermediate stages improving resolution. The ruler illustrates attrition-dominated wear patterns in the lower stages and abrasion-dominated wear patterns in the higher stages. It is calibrated using the comparative extant database provided by Fortelius and Solounias (2000) to assign the specific mesowear stages to certain diet types (see supplementary material S2).

The ruminant ruler is used on photos. In total, 972 molars are photographed. They are fixated for contrast on dark underground using a small amount of silicon. It is important to take the picture at right angle showing the buccal site in the case of upper molars and the lingual site in case of lower molars (DeMiguel et al., 2008 and Kaiser and Solounias, 2003) to avoid parallax. The mesowear stage of each cusp is recorded by applying the mesowear ruler for ruminants. In order to record mesowear stages, the virtual ruler is superimposed on a photo of a molar and adjusted in size to the mesiodistal length of each cusp, using graphics software. All molars with at least one intact cusp are used regardless of tooth position, wear stage, and/or fossil site. Cusps that are completely worn down or damaged cusps are scored as ‘indeterminable’. When a cusp is initially scored between two mesowear stages the lower stage +0.5 is recorded.

Before reconstructing the diet of the barking deer, molars providing an interpretable mesowear signal have to be identified. Two aspects are under discussion, namely tooth position (e.g., Fortelius and Solounias, 2000, Franz-Odendaal and Kaiser, 2003, Kaiser and Fortelius, 2003, Kaiser and Solounias, 2003, Louys et al., 2011, Mihlbachler et al., 2011 and Rivals et al., 2007) and the stability of the signal throughout life (Rivals et al., 2007). We test the influence of both factors on the mesowear signal.

With the TA subset (n = 747 cusps), we examined the link between wear stage and mesowear (Fig. 1). Mesowear stages generally increase by advancing the wear stage. Moreover, the variability of the mesowear signal increases, too.

t-Tests show a significant difference in the mean mesowear value when comparing cusps in wear stages 1–3 to the entire sample (wear stages 1–5) and/or the 'original sample' according to Fortelius and Solounias (2000; wear stages 1–4). Additionally, in a serial comparison of a particular wear stage with the next higher one, the samples always differ significantly, except for the comparison between wear stages 4 and 5.

The sample including all available cusps (n = 1533 cusps; TAe subset in Table 1) shows also a significant difference regarding the means of mesowear values, when comparing the cusps in wear stages 1–5 and 1–4. In addition, there is a significant difference between the means of mesowear values, when wear stages 4 and 5 are compared. The aforementioned increase in means of mesowear values from wear stages 3–4 is still clearly visible. We therefore decided to execute further analyses only on a reduced sample of wear stages 1–3.

With the TP subset (n = 497 cusps), we examined the link between tooth position and mesowear. The results are shown in Fig. 2. The box plots of upper molars ('M1 sup', 'M2 sup', 'M3 sup') only differ in the distribution of the dot densities and the length of the whiskers. All boxplots cover the same range, except the upper and lower third molars ('M3 sup', 'M3 inf'). The interquartile ranges differ in the lower first and third molars (M2 inf, M3 inf). The means of mesowear values range from 1.6 to 2.7, with lower third molars ('M3 inf') displaying the lowest mean value, while the lower first molars ('M1 inf') display the highest. Regarding ‘M2 sup’, the t-tests reveal significant differences to ‘M3 inf’.

This test was also executed with the subset including tooth cusps (n = 1042 cusps) from all sites (TPe). Fig. 2 shows that, in addition to TP, the mean mesowear value of ‘M3 sup’ differs significantly from that of ‘M2 sup’. Consequentially, we decided to continue with a reduced sample including upper and lower molars, but excluding third molars.

The remaining CA subset (n = 418 cusps) was divided by fossil sites (Fig. 3). The box plots for Lida Ajer and Sibrambang only differ in the distribution of the dot densities. The boxplot of Jambu differs from the other two by showing a shorter range and overall smaller variance. The means of mesowear values of Sibrambang and Jambu are below 2, while the one of Lida Ajer is higher than 2. The t-tests show no significant difference between Sibrambang and either of the other caves. There is, however, a significant difference between Lida Ajer and Jambu. The barking deer from all three sites can be classified into the category of mixed feeders. However, the diet of barking deer from Lida Ajer has a grazing component, while the diet of barking deer from Sibrambang and Jambu has a browsing component. When the CA subset is not divided by fossil sites, the overall mesowear signal of the barking deer is 2.02 (±1.39).

The TA subset in Fig. 1 shows that cusps in wear stages 1–3 should be used for the mesowear method, because the means of mesowear values of cusps in these wear stages fluctuate only marginally. This is also confirmed by t-tests. This trend is also visible in the TAe subset, however not as clearly as in the TA subset.

The fluctuations in the mean of the mesowear values between cusps in wear stages 1–3 can be explained using dot densities. With increasing wear stages, the range of mesowear stages increases. Cusps in wear stage 4–5 show an even stronger shift to higher mesowear stages.

The results from the serial comparison of a particular wear stage with the next higher one confirm our decision to exclude wear stage 4, because it is similar to wear stage 5 in terms of mesowear signal. The significant differences between the other wear stages are not surprising, as on the one hand they cover different time intervals and on the other hand it is visible in the distribution of dot densities in Fig. 1. It also shows that they are distinct from each other.

For the original mesowear method, Fortelius and Solounias (2000) suggested to exclude teeth that are either unworn or entirely worn out, corresponding to our wear stages 0 and 5. Unworn teeth cannot display a mesowear signal related to the diet. Teeth that are entirely worn out loose a specific signal as a result of excessive wear. Our results generally confirm the exclusion of higher wear stages, although our studies illustrate that a significant overprint of the mesowear signal begins earlier than expected, i.e. when the dentin is affected by tooth wear, corresponding to our wear stage 4 (Table 2).

The bar charts in Fig. 1 show that most of the cusps are assigned to wear stages 2 and 3. The molars in these stages cover the entire range of mesowear stages between mesowear stages 0 and 6. The wear stages (Table 2) do not necessarily reflect equal periods of time. In order to check correspondences between tooth age and individual age, we assigned each of the molars in our sample to a particular age class by a classification system introduced by Anders et al. (2011). Because of the erupting sequence in a tooth row, molars with different tooth positions do not display the same degree of wear (Anders et al., 2011). Anders et al. (2011) introduced six individual dental age stages (IDAS: ‘prenatal’ (0), ‘infant’ (1), ‘juvenile’ (2), ‘adult’ (3), ‘late adult’ (4), ‘senile’ (5). Individuals in IDAS 0 are not usable for the mesowear method. We only have isolated teeth in our sample and no tooth rows, thus only estimates can be made to determine the dental age stage of individuals. We translated the IDAS system in series of wear stages (Table 3). The mesowear method can be used on individuals from IDAS 1–3 on all molars, while individuals of IDAS 4 show only a robust mesowear signal on the third molar. The other molars of individuals of IDAS 4 are in wear stage 4 or higher. Hence individuals from IDAS 5 are not usable for the mesowear method. However, because we excluded the third molar, in fact, only individuals in IDAS 1–3 show robust mesowear signals.

Fig. 4 shows the dental age distribution per cave (TP subset in Table 1 including all wear stages) according to Table 3. It illustrates that the wear stages do not represent equally long periods of time. Recent barking deer enter the weaning period with 61 days (AnAge, 2017 and Tacutu et al., 2018). This corresponds to the shift from deciduous to permanent dentition and is reflected in the transition from IDAS 1 to 2. Barking deer reach sexual maturity after a year, which is associated with the onset of an adult life period and IDAS 3. In the wild, barking deer possess a life expectancy of less than 17 years. Assuming that late adulthood and senescence would cover a maximum period of 5 years, individuals still spend approximately 11 years in IDAS 3. The cusps in IDAS 3 ‘adult’ represent the longest time interval, reflected by their high numbers (Fig. 4). Moreover, molars can be subjected to mesowear analysis from ‘juvenile’ to ‘adult’ life periods covering at least 2/3 of their life span. Because of its small size the sample from Jambu shows a different trend.

According to the original approach of Fortelius and Solounias (2000), a minimum of 20 teeth is required to obtain a robust signal and they only included the sharper cusp of the second upper molar. Fossil samples are normally much smaller than the sample in the Dubois Collectie, so it stands to reason to include as many tooth positions as possible. In the TP subset, we tested all other tooth positions against the ‘M2 sup’ subset. Fig. 2 shows that all tooth positions can be used, except the lower third molar.

We also tested the entire sample TPe (Fig. 2) and in this case ‘M3 sup’ provides a significantly different mean value when compared to the one of ‘M2 sup’. We therefore restricted reconstructions of paleodiet and inferences on the environment on a sample consisting of upper and lower first and second molars only.

Why ‘M3 sup’ and ‘M3 inf’ show significantly different mean mesowear values than ‘M2 sup’ remains to be further investigated. A variety of reasons seems to be conceivable. Firstly, significant differences may be attributed to the marginal position of the third molars in the tooth row, thus supporting a specific involvement in the mastication process that differs from the one of rather central positions. Secondly, the variations may be related to the eruption sequence. M3 erupts later than other molars and is thus not as long in wear.

Several authors tested additional tooth positions. Kaiser and Solounias (2003) included the fourth premolar as well as all molar positions in the upper tooth row for their extended mesowear method. Kaiser and Fortelius (2003) tested additionally the same tooth positions in the lower tooth row. In the latter case, a calibration factor was required, because the mesowear signal obtained from the lower tooth row shifted considerably towards a grazing signal. However, both studies were performed on equids. Premolars are strongly molarized in equids. In contrast to that, molars and premolars are more heterogeneous in Muntjaks. Therefore, we excluded premolars from our sample. Differences between upper and lower tooth row were not as pronounced in our sample as observed by Kaiser and Fortelius.

Franz-Odendaal and Kaiser (2003) tested the extended mesowear method on ruminants and concluded that the method is not applicable to ruminants, because their heterodontia is more distinct than in equids. The authors suggested to restrict mesowear studies of ruminant teeth to upper second and third molars. The dot densities in Fig. 2 show that the sample tested here consists of a higher number of sharp cusps in the upper molars, i.e. mesowear stage 0, than in the lower molars. This is in accordance to the results of Franz-Odendaal and Kaiser (2003). They found that mixed feeders show a tendency to maximise sharpness in upper teeth, whereas this is not the case in more specialised diets. This is also confirmed in our sample, at least by ‘M3 sup’.

Louys et al. (2011) examined African antelopes and found a significant difference between ‘M3 sup’ and ‘M2 sup’. They concluded that ‘M3 sup’ is the most suitable tooth position for mixed feeders. Browsers were reclassified most correctly using ‘M3 inf’, frugivores using ‘M1 inf’ and grazers using ‘M2 inf’. According to the ruminant ruler, the barking deer can be classified into the category of mixed feeders. The authors used a reclassification approach while we scored directly. The application of different methods may offer an explanation for the different conclusions. Besides, there might be also taxon-specific differences between bovids and cervids.

We moreover found insignificant differences between upper and lower tooth rows. This is in accordance to the results of Louys et al. (2011), who found minor differences in mesowear scores between upper and lower molars across the tooth positions and trophic categories in general.

In this study, the barking deer from Lida Ajer, Sibrambang and Jambu are classified as mixed feeders. Up to present, there are no other paleodiet reconstructions for barking deer of this timeframe. Janssen et al. (2016) studied younger barking deer from Wajak (MIS 3 to Holocene interglacial) using δ ¹³C values. They suggest a mixed C3–C4 diet or even a pure C3 open canopy diet. Hence, the reconstructed diet in Janssen et al. (2016) and our study are in accordance with the recent diet of barking deer, which is well known. It consists of tender leaves of trees and shrubs, herbs, forbs, tender shoots, flowers, buds, bark, twigs, mushroom, grass, and fruits (Farida et al., 2003 and Hoogerwerf, 1970 and references therein; Ilyas and Khan, 2003 (Kitchener et al., 1990, Nowak, 1999 and Oka, 1998). Barrette (1977) reported that fallen fruits represent the staple food component of barking deer. This is, however, challenged by other authors (e.g., Ilyas and Khan, 2003). The observations of Barrette (1977) may be correct for a particular population. Yet, as Ilyas and Khan (2003) report, the diet depends on the seasonal availability of fruits in the habitat. This may also explain why other authors like Odden and Wegge (2007) describe the barking deer as a selective browser.

There is no mesowear signal on the ruminant ruler for frugivore animals. Fortelius and Solounias (2000) suggested “tip crushing wear” as result of biting on pits, which leads to damaged, but eventually rounded or even blunt cusps. In a preliminary study (unpublished data), we tested the ruminant ruler on recent wild Philantomba monticola from the Senckenberg Collection in Frankfurt. Its frugivorous diet appeared as a mixed feeder signal on the ruminant ruler. The results obtained with the ruminant ruler are in good accordance with the diet of the recent species, as inferred from field observations.

As Rivals et al. (2011) note, a mixed feeding signal is rather an abrasion index resulting from seasonal shifts in vegetation than a discrete diet type. During rainy seasons attrition dominates, whereas in dry seasons abrasion prevails. Such combination would appear as a mixed feeding mesowear signal. This effect should be more prominent in browsers than in grazers. Molars of grazers are more adapted to abrasive wear than teeth of browsers. Thus the differences between seasons should be less noticeable. While the mesowear signal of a browser is shifted towards the mixed feeder spectrum by consuming dry grass and rather hard food in the dry season, the mesowear signal of a grazer may not change significantly, even though the individual is consuming browse during the rainy season. In the future, this hypothesis should be further investigated with other methods that provide a higher resolution, like microwear, isotopes, or mesowear III (Davis and Pineda-Munoz, 2016, Sánchez-Hernández et al., 2016 and Solounias et al., 2014).

There are noticeable differences between the box plots of Jambu on the one hand and Sibrambang and Lida Ajer on the other one (Fig. 3). Although the barking deer from all three caves can be generally classified into the category of mixed feeders, the diets of barking deer at Sibrambang and Jambu are shifted towards the browsing spectrum, while barking deer from Lid Ajer have a grazing component. The differences between Lida Ajer and Jambu are significant, while Sibrambang is not significantly different from any of the caves.

The significant difference between Lida Ajer and Jambu might be explained by the different sample sizes and resulting variances. Although the size of the sample from Sibrambang is larger than that of the one from Lida Ajer, the difference between Sibrambang and Jambu is statistically not significant. In Sibrambang, the number of cusps in mesowear stages 0–2 is distinctly higher than in Lida Ajer. In this respect, it is more similar to that in Jambu. Overall, the differences between the three samples can be attributed to variable sample sizes and variances.

The sample from Jambu is smaller than those from either Sibrambang or Lida Ajer. Bacon et al. (2015) noted that a selection during excavation for most complete and best preserved teeth should be considered. Bacon and colleagues speculate that the larger portion of fossil teeth from Jambu has been left in the field. Even though they could not be subjected to mesowear studies, comparing the proportions of highly abraded and therefore poorly preserved molars among the samples from the three caves can make a valuable contribution to this discussion. Moreover, other taxa collected from the three caves should show similar effects if the collection protocol at Jambu differs from the one applied at other sites. Despite Lida Ajer being relocated, it is not possible to assign the fossils from the Dubois Collectie to distinct stratigraphic layers. There are several fossil-bearing layers at Lida Ajer (Westaway et al., 2017). The geology and formation of both of the other caves are still unknown. Differences between the samples may result from taphonomic causes as well. One approach would be to compare the percentages of porcupine gnawing marks on the teeth per cave. The impact of porcupines as accumulation agents may not be equal among the three caves. Because we have little (Lida Ajer) to no (Sibrambang and Jambu) information about the geology, except being located in karstic limestone and especially the layers the teeth were found in,the time span covered by the samples remains hidden as well. In case of Lida Ajer, Louys et al. (2017) concluded that the teeth must have been redeposited inside the cave in mass-movement events initiated by rainfall and earthquakes, which further complicate dating.

Taking all this into account, it should be noted that the number of teeth used in the Jambu sample is large enough to provide a robust signal. It is therefore likely that differences in the mesowear signals between the caves result from changes in the environment. Because they were not the aim of this study, we do not discuss ecological parameters like competition by other herbivorous animal. We will focus on the effects of the YTE and by the environmental changes accompanying the transition from MIS 5 to 4.

There are not many studies on the effects of volcanic ash on tooth wear. Spradley et al. (2016) studied how volcanic dust affects tooth wear in howling monkeys (Alouatta pallita) and concluded that volcanic ash has an important impact on the tooth wear of anterior teeth but less on molars. Another observation is that heavy wear on the premolars is due to fluorosis as a result of a surge in fluoride caused by volcanic ash. This would affect the premolars in a different way, because molars fully mineralize earlier in life history (Spradley et al., 2016). The sample tested here is restricted to molars. Nevertheless, further studies may look into the wear of premolars in these samples.

Spradley et al. (2016) studied the immediate impact of a volcanic eruption on tooth wear, but it is unrealistic that the barking deer of Lida Ajer or Sibrambang consumed leaves covered with volcanic ash in the first place. In the case of the YTE, the thickness of the initial ash layer is estimated at 10–15 cm, which is deadly for some types of vegetation (Timmreck et al., 2012), resulting in the absence of certain types of plants for a relatively short period of time. Even if the plants were covered by a thick layer of ash, the ash may be directly removed by wind and/or precipitation (Timmreck et al., 2012). However, volcanic ash may have had another important impact worth to be considered. Timmreck et al. (2012) note that the nutrients from the ash fall can induce quick vegetation recovery.

Successions after volcanic events are well studied in two cases: the eruptions of Krakatoa in 1883 (e.g., Bush and Whittaker, 1991) and Mount St. Helens in 1980 (e.g., Dale et al., 2005). However, both case studies are not applicable to the effects of the YTE. Mount St. Helens is located in a different climate zone and on the mainland. Krakatoa is in the same climate zone as Sumatra, but its small size and geographic isolation allowed only for airborne species as pioneers (Bush and Whittaker, 1991). In very contrast, Sumatra is the sixth largest island on the Earth (van der Kaars et al., 2012). Succession started in Sumatra from parts of the island, which were less affected by the immediate effects of the YTE. van der Kaars et al. (2012) recognized an immediate and extensive impact on the vegetation of northern Sumatra. Sumatran pine forests (Pinus mekusii) were decimated in the highlands; however, the effect on other forest types appeared to be either rather limited or the effects were compensated by succession and regeneration within a couple of hundred years. Succession and regeneration happen on a time scale that cannot be resolved by the fossil sample in this study (Davis & Pineda-Munoz, 2016). Because of this, the differences in the box plots (Fig. 3) cannot be accounted to the YTE.

The samples of Lida Ajer and Sibrambang show a shift to a mixed feeder diet with a less pronounced browsing component. This could result from episodic shifts towards drier conditions. In fact, a climatic change is likely to have occurred during the relevant period. Large amounts of fossil orangutan and fossil wild boar led Bacon et al. (2015) to reconstruct the environment of the three caves as an aseasonal rainforest. The diet of orangutans consists of about 60% of fruits and leaves. A large population density therefore indicates a regular and abundant fruit source. High atypical densities of modern populations of wild pigs in Asia have been observed in aseasonal lowland dipterocarp rainforests with abundant food resources (Bacon et al., 2015). The authors found no significant differences between the fauna of MIS 5 (Lida Ajer, Sibrambang and Jambu) and MIS 4 (Duoi U’Oi). However, Bacon et al. (2015) treated all three caves as a single fossil site, calling it “Sibrambang”, for the purpose of their study, in which they compared different fossil sites in mainland and insular Southeastern Asia. It would be interesting to compare the numbers of orangutan and wild pig specimens from each of the caves to check for variations in abundances.

Van der Kaars et al. (2010) studied the palynological record from deep-sea core BAR94-42 and concluded that the vegetation of Southwest Sumatra consisted of rainforest with open herbaceous swamps along river courses and surrounding lakes during MIS 5. The climate was humid, with short dry seasons. Although vegetation does not fundamentally change in southwestern Sumatra from MIS 5 to 4, van der Kaars et al. (2010) note extended dry periods, increased fire activity, and a weaker monsoon in MIS 4. Kaolinite is a proxy for the availability of water, which was used by van der Kaars et al. (2010) among others. Combining this data with the presumed dates for the caves (Fig. 5) and a potential behavioral response of the barking deer, different options for chronological links between the caves can be discussed.

Fig. 5 shows a match between the wetter conditions and a rather specialized mesowear signal for the barking deer in the time period before 70 ka. The mesowear signals of barking deer of Sibrambang and Lida Ajer show a mesowear signal, which points to a more generalized mixed feeder diet. Combining this information with the kaolinite data and the dating of the caves enables us to narrow down the chronology suggested by Bacon et al. (2015) and Westaway et al. (2017).

Lida Ajer has presumably an age between 63 and 73 ka. The same applies to Sibrambang. As already pointed out, the resolution of the mesowear method is not high enough to illustrate the effects of the YTE; hence certain timelines cannot be ruled out. Sibrambang may be older or younger than Lida Ajer. Also, the dates for Sibrambang and Jambu remain to be confirmed and should be considered with caution. Based on biostratigraphic correlations with the fossil record in Java, Lida Ajer was initially considered to be older than 81 ka years (Bacon et al., 2015 and references therein). Nevertheless, the faunal response of barking deer show that the fauna of Jambu is associated with a period of higher kaolinite availability, hence more humid conditions and less extensive dry periods prior to 71 ka. The faunas of Lida Ajer and Sibrambang should have been active in either one or both of the episodes of lower kaolinite availability between 71 and 63 ka (Fig. 5). Moreover, Sibrambang may predate Jambu. Then again, there is no significant difference between Lida Ajer and Sibrambang in mesowear signals of the barking deer. The climatic conditions should therefore have been similar.