Three‐dimensional X‐ray‐computed tomography of 3300‐ to 6000‐year‐old <i>Citrullus</i> seeds from Libya and Egypt compared to extant seeds throws doubts on species assignments

The watermelon (Citrullus lanatus subsp. vulgaris) is among the world's most important fruit crops. We here use C-14 dating and morphometric analysis to test whether ancient seeds can be identified species level, which would help document food expansion, innovation, diversity in Northeastern Africa. dated a Libyan seed 6182–6001 calibrated years BP, making it oldest Citrullus known. Morphometric could not reliably assign particular species, but several showed breakage patterns characteristic of modern cracked by human teeth. Our study contributes understanding early history humans, who may have mostly snacked on seeds, cautions against morphology alone identify archaeological samples. With production 161 million tons 2019, crops (FAO, accessed 4 May, 2021). Based geographic occurrence all its wild relatives (Chomicki & Renner, 2015), originated African continent. Molecular phylogenies show plants from Sudan as earliest branches within C. lanatus, pointing northeast Africa likely region domestication (Guo et al., 2019; Renner 2021; Wu 2019). alone, however, do reveal full when how humans began certain plant or animal. Instead, are starting points for further exploration through archaeological, linguistic, population-genetic research (Paris, 2015; Prendergast Archaeological watermelons has focused thick, hard coats preserve well found at settlements where people either ate pulp themselves used bitter oil extracted colocynthis (Akbar, 2020; Cox van der Veen, 2008; 1995, 1996, 2011; Wasylikowa 2004). Besides colocynthis, genus comprises five other least two also humans. One mucosospermus (or mucosospermus), crop West soft-coated so-called egusi soups stews (Achigan-Dako Oyolu, 1977; Prothro 2012). amarus, thick rind boiled with sugar make jam (Laghetti Hammer, 2007) source animal fodder (McGregor, 2016). three genus—C. ecirrhosus, naudinianus, rehmii—are (authors taxonomic names Table 1). 3 2 fragments, d, f, g assigned come site Uan Muhuggiag Tadrart Acacus range, occupied pastoralists during mid-Holocene (van 1995). layer were was radiocarbon 7784–7545 8162–7520 calBP (calibrated before present), stratigraphic evidence suggested that had been disturbed contained younger intrusive material (Wasylikowa 2004; paper mentions single more both later). Other 1500–1480 BC, came area Aswan Dam Reservoir (Lake Nubia; Van Zeist, 1983) appear lost (S. Blomsma, Groningen Institute Archaeology, pers. comm. SSR 30 March 2017). Amara (desert 2-R-65) Northern Sudan, 3455–3373 (P. Ryan, Royal Botanic Gardens, Kew, 12 Nov. 2019) H. Carter P.E. Newberry 11 baskets Tutankhamun's tomb Thebes Egypt, mixed Ziziphus clearly intended snack (http://www.griffith.ox.ac.uk/gri/carter/369-p1626.html). about 3300 BP (Ramsey 2010). Some basket number 527a kept Kew (accession 4.1965), others Dokki Agricultural Museum Cairo 4198; reproduced Warid, When found, “they still remarkably good condition flat 8 mm (5/16 in) long. fresh, they lightly roasted eaten cracking brittle outer cover front teeth” (Hepper, 2009, p. 56). identification these sites naturally based drawings, photographs, comparison seeds. These efforts, limited (i) incomplete sampling extant some which, such rehmii, only discovered recognized 1990s later; (ii) difficulty describing comparing three-dimensional (3D) shapes; (iii) preservation artifacts damage caused resulted broken (Cox 2008). In contrast linear measurements real objects photos, 3D morphometrics unveil morphological variation otherwise remains hidden, analyses proven particularly useful samples because complete shapes reconstructed even fragmented (e.g., Bonhomme 2017; Caracuta Analyses via micro-X-ray-computed tomography (micro-CT) therefore allow reliable distinction colocynths, watermelon, semidomesticated (above) thereby shed light uses watermelons. this study, we apply technique tomb, representing except soft-seeded mucosospermus. specific question wanted answer if micro-CT distinguish reported North (Kahlheber, Nixon 1995; Wasylikowa, 1993; Distinguishing between implications Neolithic period their highly divergent properties (above). distinguishing species' led one scholar idea sweet India brought Islamic period, view pre-Islamic might represent (Watson, 1983). Fossil collected provided K. Palaeobotanical W. Szafer Botany, Polish Academy Sciences, Kraków. presumed Samples 14 25, Sample 6 listed (1993). Seeds Thebes, lent 1932 Leonard Boodle, remain deposit Kew's Economic Botany Collection (de Vartavan, 1999). Modern purchased commercially obtained germplasm collections colleagues working Oman Sudan. Herbarium vouchers numbers scanning shown 1. taxonomy follows al. (2017). 6, sent Laboratory Ion Beam Physics, ETH Zurich (lab code ETH-102414), age calculated using measured 14C content after correction standards, blank values, fractionation (δ13C values semi-simultaneously graphite). There too few dating, needed CT scanning. conventional (before 1950 AD CE) calendar OxCal Version 4.3.2 (Bronk Ramsey, 2017) IntCal13 atmospheric curve (Reimer 2013). Image fossil performed micro-CT. mounted untreated batches inside pipette tubes filled acryl foam (Staedler 2013), an approach reduces facilitate downstream 2018). Micro-CT scans XCT-200 system (Zeiss Microscopy) Department Biodiversity Research University Vienna. X-ray detection scintillator crystals convert X-rays into visible light. then (Nikon optical lens) detected charge-coupled device (CCD) camera. 25 35 kV L9421-02 microfocus Hamamatsu electronics (Dataset S1 shows conditions). XMReconstructor (XRadia Inc.) perform reconstruction raw scan data. Resulting DICOM datasets imported Amira Software 5.4.5 (FEI Visualization Sciences Group, Burlington MA, USA). Datasets converted files (*.am) visualized Volren module Project View. cropped contain maximum four Crop Editor. To specimens XYZ planes comparable standard biological longitudinal (XZ, YZ) cross (XY) sections, transformed biologically relevant orientations (along symmetry) largest Transform Editor (Mode: extended, Preserve: voxel size, Padding Value: 0). Excess space resulting transformations removed, along standardization sizes (4.3234 × 4.3234 μm) coordinates (0, 0, 0) Segmentation create “outer” (exotesta) “inner” (endotesta) label fields each dataset. For segmentation, Magic Wand tool → Remove Islands, Fill Holes, sufficient present. lower regions and/or datasets, Blow Tool Paintbrush used. Labels smoothed Smooth (Size: dimensions). Surfaces generated SurfaceGen All surface models Figure 1 deposited MorphoSource under project id 0000C1170 (https://www.morphosource.org/projects/0000C1170?id=0000C1170&locale=en&page=2#media). Landmarks (N = 32) chosen accurately shape symmetry. They placed specimen surfaces Object (one set). OrthoSlice find midsection slice XY YZ planes, Surface Cut cut midsections place landmarks. incomplete/partial specimens, missing landmarks recorded manual NA entries post-processing. Landmark sets dataset saved *.landmarkAscii nonproprietary text (File S1). combined file containing formatted like TPS (*.tps) row landmark (ex: LM 3) identifier ID=C_colo_foss). TPS-formatted R 4.0.2 (R Core Team, 2014) carried out packages geomorph 4.0.0 (Adams Collyer, 2018) Morpho 2.8 (Schlager, 2020). pairs across symmetry, estimated fixLMmirror cases present mirrored provide (see Section limitations assuming symmetry botanical specimens). remaining landmarks, thin-plate-spline interpolation fixLMtps, generalized procrustes alignments gpagen retain eliminate statistical differences scale, translation, rotation (Bookstein, 1992). Principal component (PCA) aligned gm.prcomp quantify Figures created plotting functions (geomorph.prcomp.plot, plotRefToTarget, plotOutliers; accompanying available github.com/aubricot/Citrullus_3D_GMM), point cluster labels manually added Adobe Photoshop CS6. A (identified included our instead dating) 5313 uncalibrated calBP. inclusion (Figure 1d,f,i,g,h) substantially increased PCA morphospace S2 File S1), excluded final 2). excluded, principal components 39.64% 17.22% cumulative variance Two overlapping clusters apparent, comprising (including Muhuggiag, e 1) naudinianus ecirrhosus those tomb), amarus (j overlap almost completely. rehmii (p 2) distinct simple visual inspection (Figures S2). unusual (j) outlier S2) fragmented, placement plot due estimation error (compare 1a,b,c,e,j Notably, condition, initially properly morphology. Several consumption masticatory item figure 7 2008 1d,f,g). provides yet identified. Libya layers previously (Wasylikowa, results originally 2004), cannot species. Nevertheless, settlement sometimes traces teeth (our 1d,f,g) suggests actively cultivated mid-Holocene. suggesting snack, match Thebes. exception soft- large-seeded taxon lacking (C. per se unambiguous assignments. instance, measure 6.3 (5.6–7.1) 3.8 (3.1–4.4) size (10 measurements; authors out, fall range so diagnostic. More recent exhibit considerable (Van 2011: figures 3.18 3.19: length Late Roman seeds: 8–13.3 mm; width 5–7.3 mm). hampered small sample (Table 1), var. cordophanus, being cultivation. tried grow self-fertilize them, thus achieve set, without success. herbarium obvious issues drying pressing mature fruits. able include than any previous (none taxa colocynthis; Ziyada Elhussien, images first “cyber specimens” open-source repository future study. debated regard cause DNA sub-fossilized tissues. Certainly, within-species large enough, relying photographs initial option following However, if, case, there little differentiate alternative allows placements capture sample. Also, physical need handled once, cyber infinitely reused damaging archival material. mammal bones minimal imaging low dosage brief exposure (Immel 2016), although know no tests yet. Geometric paleontological often face dilemma retaining partial (broken) possibly introducing bias excluding losing valuable information. fragments mirroring axes Such estimates rarely perfectly symmetrical. Even so, geometric leveraged flowers (Savriama, 2018; Savriama Klingenberg, 2011) fossils (Dumoncel Although studied, approximation overall Using less computationally intensive entire procedure studies date. computing power coming years, new automated approaches compare meshes exciting opportunity nuanced introduced selection (Boyer 2015, Gao Gonzalez 2016; Renaud Methods will continue improve widespread imaging, collection digitization advances machine learning technology Schlager Senck time being, assignment image possible, fully domestication. Solutions require integrating collections-based phylogenomics records, DNA, genomics traits Paris, thank Abel Bosman, Ondřej Mikula, Antonio Profico morphomet2 Google group troubleshooting problems package geomorph; Irka Hajdas Zurich, dating; Susanne Pamperl, Research, Vienna, support scanning; Douglas Boyer, Mackenzie Nieto Aguilar, Julie Winchester, Jocelyn Triplett depositing MorphoSource; anonymous reviewers helpful comments, greatly improved manuscript. Financial DFG 603/27-1 Elfriede Franz Jakob Foundation Botanical Garden Munich. KAW analyses, GC conceived YMS contributed expertise analysis, KW archaeobotanical expertise, MN JS access scanner, expertise; worked version MS. Dataset S1. parameters. ETH-102414). This analysis. S2. Outliers Procrustes fit specimens. Ancient i, h, removed 2. p represents expected differ Refer see filenames indicated estimating parts. Filenames precise geometry “_exclu”. Please note: publisher responsible functionality supporting information supplied authors. Any queries (other content) should directed corresponding author article.