Technological and functional analysis of 80–60 ka bone wedges from Sibudu (KwaZulu-Natal, South Africa)

Fully shaped, morphologically standardized bone tools are generally considered reliable indicators of the emergence of modern behavior. We report the discovery of 23 double-beveled bone tools from ~ 80,000–60,000-year-old archaeological layers at Sibudu Cave in KwaZulu-Natal, South Africa. We analyzed the texture of use-wear on the archaeological bone tools, and on bone tool replicas experimentally used in debarking trees, processing rabbit pelts with and without an ochre compound, digging in sediment in and outside a cave, and on ethnographic artefacts. Debarking trees and digging in humus-rich soil produce use-wear patterns closely matching those observed on most Sibudu tools. This tool type is associated with three different Middle Stone Age cultural traditions at Sibudu that span 20,000 years, yet they are absent at contemporaneous sites. Our results support a scenario in which some southern African early modern human groups developed and locally maintained specific, highly standardized cultural traits while sharing others at a sub-continental scale. We demonstrate that technological and texture analyses are effective means by which to infer past behaviors and assess the significance of prehistoric cultural innovations.

Technological analysis. Apart from a single specimen manufactured from a very large mammal mandible ( Fig. 2C), the double-beveled tools from Sibudu were all made on pieces of limb bones. The taxon could not be determined for any of the bone tools. However, archaeozoological analysis, compact bone thickness evaluation, and the study of anthropogenic modifications indicate that the elongated robust shaft fragments (compact bone thickness averaging 8.44 mm, SD: 2.58 mm) derive mainly from medium, large, and very large mammals (Supplementary Table S2), and were shaped by scraping to straighten and flatten their lateral edges (Fig. 3A-B;  Table 1). Map insert made by LD using QGIS v. 2.14.3-Essen (Free Software Foundation, Inc., Boston -https:// downl oad. qgis. org/ downl oads/) using free vector and raster from Natural Earth (naturalearthdata.com).  Table S3). Traces of vigorous gouging are recorded on the flat and lateral sides of some of the objects (Figs. 2d,i,j,v, 3c-e). One end was shaped by grinding or scraping to produce a double-beveled rounded edge (Fig. 3f), ogival in lateral section (Fig. 4). On average, the tapering faces meet at an angle of 55.2° (SD: 9.9°; Supplementary Table S4). With the exception of three specimens displaying pristine traces of manufacture and little if any use-wear on their bevels or broad edge (Fig. 2d,p,u), utilization of the beveled end has smoothed manufacturing traces and completely erased them from most specimens, leaving a highly polished, striationfree surface, often extending over the entirety of both flat faces ( Fig. 3g-h). Heavily worn beveled edges on seven objects show micro-removal scars polished by wear (Fig. 3i), indicative of damage that occurred during use. Morphometric variation in complete double-beveled tools shows values ranging between 7 and 24 mm in width and between 5 and 11 mm in thickness, with a few broken tools originally displaying a thickness of up to 14 mm (Supplementary Table S2). This shows that Sibudu inhabitants selected robust limb bone shaft fragments to manufacture tools of different size, but with a highly uniform morphology (Fig. 4).  Table 1 for contextual information, www.nature.com/scientificreports/ Texture analysis and functional assessment. Evaluation of the acquired data from the six parameters used to analyze surface texture shows substantial similarities and differences between experimental, ethnographic, and worn and unworn archaeological tools (Fig. 5). Three parameters, Ymax, Sq, and AsFc, evaluate the texture as a whole, while Spc, Smr1, and Sal focus on aspects of the surface texture such as peaks ( Fig. 6; Supplementary Fig. S1). The former, which in simple terms account for entire surface complexity (Ymax), surface homogeneity around the mean height (Sq), and surface homogeneity at different scales (AsFc), reveal similar trends. First, worn and unworn archaeological specimens show clearly distinct ranges consistent with microscopic observations, indicating that lower values result from the smoothing of the traces of manufacture during tool use. Second, unused experimental tools present comparable roughness values for Ymax, Sq, and AsFc, indicating that the initial states of the working edges were the same regardless of the structure of the limb bone. Third, all experimentally used tools display roughness values higher than those recorded in their unused state. Fourth, tools used to debark trees and process a rabbit pelt without an abrasive product like ochre produce the lowest textural complexity. A higher degree of variability is observed among the wear patterns produced by processing a rabbit pelt tanned with an ochre powder and fat mixture, and on ethnographic bone tools used to debark trees. Tools used for digging in sediment display the highest values among all experimental and ethnographic tools. The values recorded on the tool used to dig in dry sediment on the Border Cave talus slope are consistently higher than those recorded on the tool used to dig in humus-rich soil away from the cave. The range of variation in texture recorded on archaeological tools is wider than that calculated for any of the experimental and ethnographic activities. No overlap is observed between the wear on archaeological tools and that on experimental tools used to debark trees and process rabbit skin without an ochre powder abrasive. Between one and three archaeological tools out of seven display variation compatible with the use-wear produced when experimentally digging humus-rich soil or processing rabbit skin with the ochre and fat compound, and debarking trees as recorded ethnographically. For each parameter, two to three archaeological wear patterns fall outside the range recorded for known activities. The parameters accounting for the sharpness of the peaks, Spc, and the proportion of peak material present above the core surface, Smr1 ( Fig. 6; Supplementary Fig. S1), reveal a clear difference between the wear present Table 1. Contextual data on the double-beveled bone tools from Sibudu. Excavation date, stratigraphic provenance, and cultural attribution of the bone tools reported in the present study. OSL ages (in ka) were reported in 49,55 . Four tools were described in previous publications but were not submitted to texture analysis. www.nature.com/scientificreports/ on archaeological specimens and that generated by scraping skin covered with an ochre mixture. They show a concurrence between some archaeological wear patterns and those produced experimentally when digging humus-rich soil, as well as those observed on ethnographic debarkers. Smr1 identifies one archaeological tool bearing a wear pattern that falls outside the experimental and ethnographic ranges. The last parameter, Sal, measures the horizontal distance in the direction in which the autocorrelation between slope and distance decreases the fastest. This parameter highlights significant overlap between archaeological, ethnographic, and experimental wear patterns. A Principal Component Analysis (PCA) based on the six textural parameters (Fig. 7) reveals that most of the archaeological use-wear patterns overlap with those measured on the ethnographic debarkers, on the experimental tools used to process skin with an ochre mixture, and those used to dig in humus-rich soil. The wear patterns produced experimentally when debarking trees, treating skin without ochre, or digging in dry soil fall outside When subjecting experimental and ethnographic textural data to a Flexible Discriminant Analysis (FDA), the model produced in training mode accurately attributes the function for which the tools were used 88.9% on average over 100 iterations; in validation mode, the accuracy is 79.8%. When the 25 best models out of 100 iterations are considered, the validation accuracy rises to 84.3%. This result implies that if the archaeological tools were used for one of the documented functions, the top 25 models would be able to correctly discriminate the function for which they were used 17 times out of 20 (Supplementary Table S5). Application of the two predictive approaches to archaeological use-wear produces similar results: they identify debarking of the kind represented on ethnographic tools as the most likely function for the Sibudu double-beveled tools (Supplementary Table S6). In four instances, tools may have been used to dig in humus-rich soil or for both activities. These predicted functions appear to have been similarly implemented in the three cultural horizons (PSB, SB, HP) in which the archaeological bone tools were found, despite their variation in size.

Discussion
The double-beveled bone tools discovered at Sibudu in layers dated between ~ 80-60 ka are among the earliest formal bone tools known. Textural and discriminant analyses indicate that most of the tools were probably used in debarking activities that produced use-wear comparable to our ethnographic sample, and possibly for digging in humus-rich soil, likely to extract roots, sedges or underground storage organs. Interestingly, the function of the Sibudu double-beveled tools is not linked to hunting or hide processing activities, with which the production of the first documented formal bone tools, e.g., projectile points, barbed points, smoothers, etc., have been traditionally associated, but rather domestic subsistence strategies devoted to the exploitation of vegetal resources. Extraction of underground plant resources, entailing contact of the tool with both the ground and vegetal matter,   68 . San hunter-gatherers use bark from various species for making fiber that is in turn used for construction, binding tools, or making snares. Some roots are excavated and scraped or cut to make traditional glues and adhesives 71 . Extracting and working roots in this way could produce earth-scraping traces and traces emulating bark removal. The differences between the roughness values recorded for the use-wear on experimental and ethnographic debarkers could either be due to differences in the length of utilization of the tools, or differences in the fibrous structure of the bark and sapwood on which the tools were used. While the discriminant analysis rules out a number of functions, e.g., skin processing with or without ochre, digging in dry sediment, and debarking some tree species, the PCA suggests that some Sibudu wedges were used in activities for which we do not yet have ethnographic or experimental correlates. This is consistent with the tiny size of one or two tools that appear too small for debarking or digging activities. Additional characterization of textural parameters on experimental and ethnographic bone tools must be undertaken to identify the functions for which some of these tools were used. Extension of this research strategy to other bone tools identified in the MSA and the European Middle Paleolithic, and for which a function has been proposed e.g., 21,22,30 , is necessary to strengthen previous interpretations or identify alternatives or complementary functions.
These results imply that, although many of these tools were apparently used for the same purpose, a doublebeveled end was also sought for functions that we are unable to identify at present. It also implies, in light of the stratigraphic distribution of the studied artefacts, that this normative bone tool tradition was transmitted and maintained for at least 20 millennia and associated with three different MSA lithic technological traditions, namely the PSB, SB, and HP. Aside from two morphologically similar tools of uncertain age found at Broken Hill, Zambia, and interpreted as 'gouges' 36 , none of the numerous MSA sites from South Africa with PSB, SB and HP layers, including those in which well-preserved faunal assemblages and bone tools were found, has yielded     80 . Although complex, the stratigraphy at Sibudu is clear and well-preserved. Geoarchaeological analysis suggests exceptional stratigraphic integrity with minimal vertical mixing between the anthropogenically-formed layers 69,81 . Such integrity can be appreciated from the preservation of laminated, articulated phytoliths and centimeter-thick layers of undisturbed, carbonized bedding, sometimes extending laterally for meters 69 . Disturbance caused by the recent digging of pits, animal burrowing and rockfall was easily recognized and clearly delimited during the excavations. Sibudu features excellent organic preservation, with bone, charcoal, carbonized seeds and other plant remains found throughout the sequence 80,82,83 . The faunal assemblage from PSB layers is dominated by suids. The oldest PSB deposits also include a diversity of small game 84 . SB and HP layers show a high frequency of blue duiker (Philantomba monticola) together with suids and small-medium sized bovids 84,85 . The PHP faunal assemblage is interpreted as reflecting a more open environment with a decrease in small prey and a predominance of large and very large bovids 85 . Raw material such as dolerite and hornfels dominate the lithic assemblages, although quartz and quartzite were also used for manufacturing stone tools, particularly during the HP 80,86 . The lithic assemblages were subjected to detailed analyses that identified diachronic changes in stone tool technology [87][88][89] and the use in the HP and SB of bone pressure flakers to shape stone tools 46,86,88 . A recent chemical characterization of residue present on grindstones found in PSB and SB layers suggests they were used to grind ochre 90 . From 1998 to 2011, more than 9,200 ochre pieces were recovered from the MSA layers 91,92 . Sample selection. Archaeological sample. The archaeological material analyzed in the present study is curated at the Evolutionary Studies Institute at the University of the Witwatersrand in Johannesburg, South Africa, and comes from the excavations conducted by one of us (LW) at Sibudu (Amafa excavation permit 007/09). Faunal material from these excavations has been systematically analyzed in search of pieces bearing traces of modification. This led to the identification of a variety of bone tools from different archaeological layers described elsewhere 44,46 , as well as our sample comprising 23 complete and broken bone tools featuring one double-beveled end and modified adjacent edges, four of which were described in previous publications 44,46 . Our sample was recovered between 2004 and 2011 from PSB, SB and HP layers (Table 1). On seven tools, a selected portion of the worn area featuring an excellent state of preservation was molded with Coltène President light body dental  Supplementary Fig. S2). The manufacture of these bone tools entailed rounding the prominent areas of the distal epiphysis to ensure adequate grasp while working, and sawing the diaphysis in half at an oblique angle to create a bevel. Hardening of the bone was achieved in some cases by slow-heating the tools in warm ash. An iron blade was inserted in some tools close to the epiphysis 93 . After incising the bark longitudinally with the iron blade or a cutting tool, and then around the trunk at the two ends of the longitudinal incision, the debarker was inserted between the bark and the sapwood-often of oak trees-to detach the bark using pushing and wedging movements. This activity developed a characteristic polish on the beveled area of bone in contact with the bark and sapwood. The debarkers included in our sample were selected for the excellent state of preservation of their active area and the invasive use-wear which has obliterated traces of manufacture. On four tools the worn area was molded with Coltène President light body dental elastomer (Coltène, Switzerland). These tools were The limb bone fragments were selected to match the thickness of most archaeological bone tools. The pieces were dry when shaped. The bevel was polished using a self-adhesive flock polishing cloth (ESCIL, Chassieu, France) covered with a fine diamond solution. The surfaces thus produced were examined with an optical microscope in reflected light, and polishing was repeated until micro-striations were barely detectable at 40× magnification, i.e., striations measuring less than 1 µm in width. Previous research has demonstrated that differences in the original surface texture of bones influence the development of wear 61 . In addition, measuring the development of a usewear pattern on experimentally used bone tools shaped with prehistoric techniques is challenging because it is difficult to acquire measurements in exactly the same location before and after use. To overcome these issues, our experimental design aimed to produce homogeneous, comparable surfaces on which modifications that developed during experimental use could be precisely located and measured ( Supplementary Fig. S4). Textural analysis of the unused active areas confirms that they were characterized by almost identical roughness values ( Fig. 6; Supplementary Data S2). This demonstrates that neither the experimental shaping nor the original bone structure influenced roughness parameters prior to the experiments. Three experimental beveled tools were used to remove sections of bark from three living, endemic, South African trees, namely Harpephyllum caffrum, Lannea antiscorbutica, and Mystroxylon aethiopicum. H. caffrum seeds were found at Sibudu in late MSA layers and M. aethiopicum seeds and charcoal throughout the sequence 67,82 . L. antiscorbutica is traditionally used as a medicinal plant 66,[117][118][119][120][121] . Two beveled tools were used as digging tools; one in the ground on the Border Cave (KwaZulu-Natal) talus slope and the other outside the cave. The talus slope floor consists of fine, sorted, abrasive sedimentary particles 122 , while the ground outside is a humus-rich topsoil with sparse gravel supporting grass growth. Two more beveled tools were used to remove fat and connective tissue from two rabbit pelts. After skinning the animals, the pelts were dried for 48 h. One of them was then scraped and cleaned using one of the beveled tools. The other was covered with a mixture of ochre powder and wet-rendered lard and left to dry for an additional 24 h before being scraped with the second beveled tool. All of the experimental tool tips were molded with Coltène President light body dental elastomer (Coltène, Switzerland) before use, and after 10 min and 20 min of use. www.nature.com/scientificreports/ Texture analyses. When necessary, the elastomer molds of the beveled surfaces were cut with a scalpel to expose the worn areas. The resulting samples were analyzed with a Sensofar S Neox confocal microscope equipped with a long-distance 50× lens (numerical aperture = 0.80), allowing a lateral resolution of 0.26 µm and a vertical accuracy of 3 nm. Three-dimensional surface acquisitions were made on both beveled surfaces on an area measuring 0.98 × 0.74 mm located close to the tool's working edge, i.e., between 2 and 3 mm. Areas displaying anatomical features such as capillaries, and post-depositional damage on the archaeological specimens were avoided as much as possible. Scan quality was reviewed after each acquisition. Scans with 95% or more of the surface measured were retained for further analysis. Less accurate acquisitions were remeasured by improving the light and exposure parameters until 95% of the targeted surface area was acquired.

Methods
Post-acquisition surface treatment followed a procedure similar to that described by Martisius et al. 61 , and was performed with Sensomap Mountains 7.4 software. Using built-in operators, it entails levelling the surface with the least square method, mirroring the y-and z-axes to obtain the original surface from molds, removing isolated outliers and those around edges, filling in non-measured points by interpolating from neighbors, and finally removing form using a polynomial of third order. Concavities due to the presence of capillaries and post-depositional damage were manually excluded from the analyzed area. A Gaussian filter of 80 µm was applied as a cut-off to remove waviness from the microtopography. The resulting acquisition was subdivided in four identical quadrants. The following roughness parameters (ISO 25178) were calculated: root mean square height [Sq], autocorrelation length [Sal], arithmetic mean peak curvature [Spc], and upper material ratio [Smr1].
Fractal analysis provides an alternative to the measure of roughness. It allows one to document and quantify irregular shapes at multiple scales. In archaeology, this approach was successfully applied to document worn surfaces of bone and stone tools as well as teeth 60,[123][124][125][126][127][128][129][130][131][132][133][134] . The Sensomap Mountains 7.4 software directly analyses the surface and quantifies the fractal parameters [AsFc] and [Ymax], which respectively correspond to the area-scale fractal complexity and the developed interfacial area ratio, i.e., a parameter equivalent to [Sdr] that can be measured on surfaces with missing values. The fractal analysis does not require previous application of a Gaussian filter to remove waviness.
Statistical analysis. Variation in each textural parameter was documented and compared for experimental, ethnographic, and archaeological specimens. Trends through time in the evolution of this variation were also evaluated for experimental bone tools.
Textural data (Supplementary Data S1) were processed with the stats, mda, plyr and permute packages in R-CRAN 135 . Principal Component Analysis was made with the stats package. The mda package allows one to perform different types of discriminant analysis. Flexible Discriminant Analysis (FDA) was performed. This extension of linear discriminant analysis uses non-linear combinations of predictors, e.g. splines, and is useful to model multivariate non-normal or non-linear relationships between variables within each group 136 . Every textural parameter gathered from the ethnographic objects and the experimental specimens used for 20 min was included in the analysis. To account for the small sample size of some groups, we performed the FDA in three steps. First, each group was divided into two randomly selected sub-samples, one half for training the model and the other half to validate it. Second, confusion matrices were produced and allowed us to calculate the accuracy of the model for both the training and validating samples. Finally, the model was used to predict in which group the archaeological data were most likely to belong. With plyr and permute, the three-step FDA was replicated 100 times. The predictions from the 25 models that yielded the highest accuracy values both in training and validation modes were then compared to assess the most likely function of the archaeological specimens.

Data availability
All data are available in the main text or the supplementary materials.