The Evolution of Plant Harvesting at The Dawn of Agriculture: Perspectives from Sickle Gloss Texture Analyses

Archaeobotanical and genetic analysis of modern plant materials are drawing a complex scenario for the origins of cereal agriculture in the Levant. This paper presents an improved method for the study of early farming harvesting systems based on the texture analysis of gloss observed on sickle blades. We identify different harvesting activities (unripe/semi-ripe/ripe cereal reaping and reed and grass cutting) and evaluate their evolution during the time when plant cultivation activities started and domesticated crops appeared in the Levant (12,800 to 7000 cal BC). The state of maturity of cereals when harvested shifted through time from unripe, to semi-ripe and nally to ripe. Most of these changes in harvesting techniques are explained by the modication of crops during the transition to agriculture. The shift of plant harvesting strategies was neither chronologically linear nor geographically homogeneous. Fully mature cereal harvesting starts to be dominant around 8500 cal BC in Southern Levant and one millennium later in Northern Levant, which ts with the appearance of domestic varieties in the archaeobotanical record. The evolution of plant harvesting better ts with the gradualist model of explanation of cereal agriculture than with the punctuated one. and barley (Hordeum spontaneum/vulgare). Domesticated-type scars were identied on the emmer, einkorn, and barley chaff at a frequency higher than expected in wild cereal species (21.1–41.2%). Based on these results it can be concluded that by 8,700–8,200 cal BC there are positive signs that barley, emmer and einkorn were being cultivated and partially domesticated at TQN 9 . In the contemporaneous levels of Tell Aswad, emmer and barley were exploited and the latter also shows high proportions of domestic type spikelets (30%) 33 34 . Gloss texture analysis nds higher proportions of ripe harvesting than would correspond to a context where fully domesticated cereals only reach 30–40% (see also below). Ripe cereal harvesting continues to dominate in the slightly later Early PPNB levels of Kharaysin (8,400-8,100 cal BC), in the Middle PPNB levels of the same site (8,000–7,700 BC) and one sickle from Nahal Hemar, while in the Late PPNB levels at Ba’ja ripe cereal cutting decreases and semi-ripe cutting stands out slightly. Thus, ripe harvesting starts to become dominant in Southern Levant from the Early PPNB, which ts well with the presence of domestic varieties of cereals from that period observed in archaeobotanical assemblages 33 . Unripe cereal cutting is marginal in the period in which ripe cereal harvesting dominates (mid 9th to mid-8th millennium BC), though it is still present in the EPPNB at Tell Qarassa North and Kharaysin.


Introduction
Wheat and barley were rst cultivated and domesticated in the Levant 1 2 . However when and where wild cereal cultivation and the exploitation of domestic varieties took place, which are crucial questions for understanding why and how hunter-gatherers became farmers, are a matter of vivid debate, opposing a progressive (protracted) model with an abrupt (core area) one 3 4 5 6 7 8 9 .
Archaeobotanical data suggest that cereal domestication was the result of a series of events occurring at different places over thousands of years, during which wild wheat persisted in cultivated elds, with the process occurring more or less rapidly in different areas of southwest Asia 10 . Genomic analysis is showing the enormous complexity of the evolutionary history of the cereal domestication process, indicating genetic ancestry from numerous geographic regions and diverse wild genepools 11 12 13 14 15 16 .
The analysis of sickle elements through use-wear analysis can offer a complementary source of information to archaeobotanical and genetic studies for disentangling this complex scenario 17 18 . Sickle blades were crucial in the process of cereal domestication, as reaping with lithic tools allowed the preferential selection of mutant seeds with a solid rachis that most probably led to domestication 19 . Sickle blades are present in the Natu an sites and become more common in Pre-Pottery Neolithic sites, thus from ca. 13,000 to 7000 cal BC, during the whole process of transition towards agriculture 20 . Plant (and cereal) cutting generates polish on the tool edge that, after some hours of use, can be observed macroscopically as a sheen that is called sickle gloss. The standard method of use-wear analysis relies on the microscopic observation and the visual comparison and matching of use-wear on experimental and archaeological tools 21 22 . Using this method, a tendency to observe atter and more abraded gloss in the later phases has been attested 17 18 . However, this qualitative use-wear analysis is insu cient to discriminate between tools used to cut different types of plants. During the last decade, a new quantitative method based on texture analysis of 3D use-wear polished surfaces, using confocal microscopy 23 24 , is offering unprecedented precision and greater objectivity in the identi cation of the plants that were cut with lithic tools 25 26 .
Our methodology (see extended description in Methods and Supplementary information) is based on the quantitative analysis of gloss texture on experimental tools used for cutting ve plant categories. These included three types of cereals: 1) morphologically domestic (Triticum monoccocum, T. aestivum, T. diccocum and T. spelta), 2) morphologically wild but cultivated (T. boeoticum thaoudar) and 3) wild cereal species growing in natural stands (T.boeoticum, T. dicoccoides and Hordeum spontaneum), plus 4) reeds (Phragmites communis) and 5) another grass (Ampelodesmos mauritanica). The three categories of cereals were harvested at the point of maturity when it was possible to maximize the yield. This resulted in harvesting in an unripe state for spontaneously growing wild cereals, in a semi-ripe state for cultivated wild cereals and in a ripe state for domestic cereals. Gloss on experimental tools used in these ve plant-cutting activities were scanned using confocal microscopy and 3D surfaces were obtained from these readings. These surfaces were measured using 25178 ISO parameters of texture analysis. Discriminant function analysis was used to obtain predictive algorithms (based on Bayes' theorem) for the classi cation of archaeological tools in one of the ve experimental categories. To test the discriminant capacity of the method, one half of the 3D surfaces were classi ed against the other half, obtaining a good rate of correct grouping. This method was used to analyze 215 sickle elements from nineteen archaeological sites/periods in Northern and Southern Levant dating from ca. 12,800 to 7000 cal BC (Natu an to Late PPNB periods) (Fig. 1). These sites and time span cover the period and regions where cereal cultivation and domestication took place during the transition from hunter-gatherer to farming societies.

Methods (For a more detailed description see Supplementary information)
Plant cutting generates use-polish on the tool edge that -after some hours of use-can be observed macroscopically as a sheen that is called sickle gloss. Experiments of harvesting wild cereal growing in natural stands, cultivated wild cereals, domestic cereals, reeds and grass were carried out. Between six and ten areas with gloss (650x500 µm each) were scanned with a Sensofar Plu Neox confocal microscope. The 3D surfaces were sampled, extracting areas of 200x200 µm, which were processed and later measured with the Mountain 7 software, from Digital Surf. Several parameters of texture analysis de ned in ISO 25178 were selected on the basis of their discriminant capacity for the ve experimental categories. Quadratic discriminant function analysis was used to build a predictive model for group membership, based on the Bayes' theorem. The discriminant algorithms obtained from the texture analysis of experimental tools were used for the analysis of glossed tools recovered in archaeological sites.
The discriminant function analysis consistently groups the surfaces of use-wear polish resulting from cutting the three cereal types (domestic, wild cultivated and wild in natural stands), reeds and other grasses. Signi cant mean differences (Wilks' Lambda) were observed for all the predictor parameters and for discriminant functions, with 73% of the 3D surfaces correctly classi ed. Using the threshold of 45% of correct classi cation of samples for considering each experimental tool as correctly classi ed, all the experimental tools can be considered as correctly grouped. To test the real predictive capacity of the method, we blindly classi ed half of the samples against the other half. Eighteen experimental tools were correctly classi ed, while two can be considered as indeterminate, in which the threshold of 45% of the samples is not reached for any of the ve categories.
The archaeological tools were classi ed in one of the ve categories of plant cutting tools when more than 50% of the 3D surfaces were classi ed in one of the groups, otherwise, they were classi ed as indeterminate ( Table 1). The classi catory threshold was raised in the archaeological tools (50%) with respect to the experimental ones (45%) in order to reduce the rate of potential errors.
For ensembles of tools, we considered the proportion of 3D surfaces classi ed in each of the ve plant cutting groups (Figs. 2 and 3). The index of the degree of maturity of harvested cereals per level/site ( Fig. 4) was calculated considering exclusively the results attributed to cereal harvesting in the three stages of maturity. The proportion of unripe harvesting was multiplied by three, unripe harvesting by two and ripe harvesting by one and the addition was calculated. Thus, 300 would be the index of a site with exclusive unripe cutting and 100 in another one with exclusive ripe cutting.
Gloss texture analysis nds signi cant differences between the tools used for cutting the ve plant categories in terms of roughness, isotropy, wavelength, complexity, depth and density of furrows and slope of surface points ( Table 2). The capacity of gloss texture analysis to separate the ve plant-cutting activities is explained by the different mechanical characteristics of cereals, reeds and grass. Phragmites stems are 5-15 mm wide and rigid, cereal stems are 1.5-2.5 mm wide and more exible, while Ampelodesmos leaves are 3-9 mm wide, very supple but tenacious. The degree of humidity in the stems/leaves is higher in reeds as they grow in swampy areas, medium for the while evergreen Ampelodesmos grass and lower for cereals when they are cut in the late Spring or Summer. The capability to discriminate the three types of cereal-cutting tools is probably due to different degrees of humidity present in the stems when reaped. When the aim of cereal cultivation is maximizing the yield, cereals are harvested in a state of maturity that is as advanced as possible. Morphologically domesticated cereals are usually reaped when they are ripe and the grains are fully mature. In contrast, wild cereal stands are commonly harvested before the plant reaches maturation, that is green or semi-green, because once fully mature, the spikelets start to disarticulate, leading to grain-loss 48 17 . Interestingly, our method shows clear differences between the experimental tools used to harvest wild cereals from natural stands in Syria and Israel and those used to cut wild cereals cultivated in Jalès (France). Texture analysis places the use-wear created when harvesting cultivated wild cereals in an intermediate position, between wild spontaneous and domestic cereal cutting tools. To explain this discrimination, we suggest several possible explanations. 1) Synchronous planting and harvesting may lead to more homogeneous ripening even after very few generations of cultivated cereals 49 . In the experimental carried out in Jalès, wild cereals were sown very densely on rich soil, and the plots developed thick stands that matured uniformly 48 . 2) These wild cereal cultivation experiments demonstrated that the most productive yield is obtained when crops are in a semi-ripe stage, which lasts few days. During this short period the seeds are more developed than in an unripe or green stage. Harvesting is thus more productive, yet at the same time, grain-loss is avoided, as the plants are not mature enough for the ear to disarticulate during harvesting 48 18 . Besides, cultivated elds of wild cereals can be more thoroughly surveyed than natural stands, because they are located near villages, and thus they can be more easily harvested in a more advanced moment of maturity. 3) In experimentally grown natural stands, cereals ourished together with a mix of other plants that compete with them, including perennial grasses, while in cultivated elds the dominance of cereals is expected (by clearance of the eld or weeding). Thus, when harvesting spontaneously growing wild cereal stands, the diversity of plants harvested is greater than in cultivated elds, and include green perennial grasses, which could affect gloss texture, increase roughness, complexity, density of furrows and slope of surface points, while decreasing isotropy and wavelength (see Table 2).
Our method enables a high con dence in the classi cation of cereal harvesting tools in three states of maturity: unripe, the state in which wild cereals in natural stands were harvested; semi-ripe, the condition in which experimental wild cultivated cereals were cut at Jalès; and ripe, the state in which domestic cereals were reaped. However, for archaeological tools, it is not possible to automatically equate the degree of humidity of cereal when harvested with a speci c step in the process towards agriculture (i.e. unripe cutting=harvesting wild cereals in natural stands). Ripe stems of wild cereals could be harvested for feeding the livestock or for technical activities such as basketry or for roo ng houses. Similarly, domestic cereals can be cut in a semi-ripe state for making freekeh, a kind of roasted snack where the taste of the unripe grain is sought. Environmental humidity/dryness can in uence the degree of moisture of cereals when harvested. Another issue is the possibility that one sickle could have been used for cutting different types of plants. In this case, texture analyses would result in an indeterminate classi cation of the tool or in the classi cation of the tool in the group of dominant use. These factors have to be considered when interpreting the archaeological data. However, when information of gloss texture analysis is considered in conjunction with the archaeobotanical information, it is possible is to evaluate how plant harvesting systems changed with the development of plant cultivation and domestication. In this work, we interpret the evidence by assuming that people in the past were trying to achieve the highest yield, which was most probably the goal during the origins of agriculture as this is a prerequisite for explaining the process of genetic selection leading to domestication. The evolution towards riper cereal harvesting from 12,800 to 7,000 cal BC both in Northern and Southern Levant strongly suggests that these changes in harvesting strategies were related to crop management shifts that took place during the transition to agriculture.

Data Availability
The datasets generated during and/or analysed during the current study are available in the Digital CSIC repository, http://hdl.handle.net/10261/225522

Results
The results of the analysis are listed according to the classi cation of individual tools (Table 1) or the proportion of 3D surfaces classi ed in each of the ve plant working groups in the archaeological contexts (see Fig. 2 for Southern Levant and Fig. 3 for Northern Levant). An index of the degree of maturity of harvested cereals (Fig. 4) was calculated for the sites/levels in the Northern and Southern Levant (see Supplementary information). The tendency to harvest riper cereals through time can be observed in both regions. However, some differences appear, as from the Early PPNB harvesting in the South is consistently riper than in the North.  27 , while phytolith analysis showed the existence of remains of seed husks from mostly wild wheat and abundant wild barley, as well as phytoliths from stems of cereals/grasses, reeds and rushes 28 . The combination of archaeobotanical and use-wear lines of evidence suggests that wild uncultivated cereals were harvested at the site. The gloss texture analysis from Shubayqa 1 and 6 (Natu an) indicates a variety of plant-cutting activities, including the cutting of semi-ripe cereals, reeds and other grasses. The initial analysis of the Shubayqa 1 archaeobotanical assemblage suggests that cereal grasses were rarely exploited by its inhabitants, whereas Cyperaceae tubers dominated the assemblage 29 . This corresponds to the rarity of sickle int tools in the lithic assemblage, although there are many ground stone artefacts at the site 30 . Wild cereals and club-rush tubers were used to produce at bread-like products 31 . In these Natu an sites, gloss texture analysis shows a variety of plant-cutting activities (cereals, reeds and other grass), while the importance of unripe harvesting suggests the exploitation of wild cereals in natural stands in Hayonim Terrace.
The data from Shubayqa 6 and Kharaysin 1 (PPNA) indicate cutting of semi-ripe cereals. This could potentially be interpreted as evidence for harvesting cultivated wild cereals, which ts the archaeobotanical information indicating wild cereal cultivation during this period in southern Levant 32 . Nevertheless, few tools from these sites were analyzed and the archaeobotanical study of both sites has not been completed yet so this conclusion has to be considered as preliminary. Like Shubayqa 1, sickle blades are rare in the Shubayqa 6 lithic assemblage, although there are many ground stone artefacts at the site. Further analysis of additional sickles from these and other Natu an and PPNA sites in the southern Levant are necessary to interpret the evidence obtained.

Evolution of plant harvesting techniques in the Northern Levant
In the Northern Levant, the sickle blades from the Late Natu an at Abu Hureyra and Mureybet show use-wear patterns that correspond to experimental sickles used to cut both semi-ripe and unripe cereals. The presence of sickles showing evidence for cutting unripe cereals is not surprising considering that if wild cereals were harvested, cutting them green would be most advantageous to obtain the maximum yield. However, it is interesting to highlight the presence of sickle blades that, based on texture parameters, coincide with those that were used to harvest cultivated wild cereals in our experiments. Whether Natu an groups were cultivating wild cereal stands has been a matter of discussion since the 1950s 35 36 37 . To date, there is no clear empirical archaeobotanical data that supports this possibility 38 39 . In fact, the proportion of cereals among other plant resources in the archaeobotanical record of Abu Hureyra 1 and Mureybet I is small 36 40 . Taken together, our results could suggest that Natu an groups at these sites were exploiting both wild cereal stands and, perhaps, managed elds, where cereals were growing in denser stands and could therefore be optimized for harvesting in a semi-ripe stage. Further archaeobotanical and gloss texture analysis is needed to verify this hypothesis.
From the end of the 10th to the end of the 9th millennium cal BC, ripe, semi-ripe and unripe harvesting are present at Mureybet III (PPNA), Jerf el Ahmar (Late PPNA), Mureybet IVA (Early PPNB) and Dja'de el Mughara (Early PPNB). In all these contexts, archaeobotanical evidence suggests the large-scale cultivation of morphologically wild cereals 41 42 43 44 34 . The prevalence of semi-ripe harvesting in these sites/levels (dominant in Mureybet III and IV and in Dja'de) ts well with the cultivation of wild cereals 44 . The identi cation of unripe cereal cutting, especially in the PPNA levels of Mureybet III and Jerf el Ahmar, suggests that spontaneous wild cereals were also exploited. Sickle gloss texture analysis also indicates that ripe harvesting was common in these contexts even though domestic cereals were not present. Unripe cereal cutting tends to diminish through time and it is marginal or inexistent in the PPNB contexts/sites (Mureybet IVA, Dja'de, Abu Hureyra and Halula). The appearance of relevant proportions of ripe harvesting before domestic varieties were dominant is also observed in the Early PPNB levels of Tell Qarassa North, in the southern Levant. This could suggest that before the appearance of domesticated cereals, prior to the xation of the tough nature of rachis, cultivated wild cereals were evolving traits that progressively allowed for a more mature harvesting of the crops. This scenario would have been possible because persistent planting and harvesting together favored plants to grow in synchronization 45 and reduced the proportion of immature grains harvested 46 . Wild cereals that could be harvested in a drier state because they reached ripeness more homogeneously would match gloss texture of our experimental tools used for harvesting domestic cereals.
In the Middle PPNB occupations of Abu Hureyra the harvesting of semi-ripe cereals is still more common than of fully mature ones.
In this site/level domestic varieties such as free-threshing Triticum sp appear, accompanied by grains of H.spontaneum/distichum and T.dicoccoides/dicoccum 37 . In the Middle and Late PPNB levels of Tell Halula (from 7600 to 7000 cal BC) sickle elements mainly fall into the group of ripe harvesting, though the cutting of semi-ripe cereals is also relevant. Two-rowed barley (Hordeum distichum), naked wheat (Triticum aestivum/durum) and emmer (Triticum dicoccum) were the main cereals exploited at the site. The presence of T. aestivum/durum demonstrates the presence of domesticated cereals, though wild cereals were also exploited 47 . This is the period when sickle gloss texture shows, for the rst time in the area, the dominance of ripe harvesting over unripe and semi-ripe ones.  Figure 1 Map with the archaeological sites from which sickle blades have been analyzed. Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.