The fossil seed of Ampelopsis was recovered from the Upper Pleistocene exposed near Zhenjiang town, Maoming city, Guangdong province, South China (Figure 1A; 21°52′47.5″N; 110°40′06.3″E). Accelerator Mass Spectrometry (AMS) of ¹⁴C dating showed that the geological age of this deposit was Late Pleistocene (29–27 ka BP, 33–30 ka cal. BP) (Huang et al., 2021a). The sedimentary succession comprised yellow sands with gravel and black, gray, and grayish-yellow mudstones with a total thickness of ~2.5 m, the base of which displays an angular unconformity with the underlying upper Eocene Huangniuling Formation (Figures 1B, C). The Ampelopsis mummified three-dimensional seed fossil was found in a gray mudstone. This sedimentary succession also yielded conifers belonging to Pinus Linn. and Keteleeria Carrière (Pinaceae) and angiosperms belonging to Canarium Linn. (Burseraceae), Elaeocarpus Linn. (Elaeocarpaceae), Liquidambar Linn. (Altingiaceae), and Fagaceae (Huang et al., 2021a,b; Bazhenova et al., 2022; Xiang et al., 2022).

The fossil described here was deposited in the Museum of Biology, Sun Yat-sen University, Guangzhou, China.

To clearly show the surface characteristics of the fossil seed, we immersed it in clean water and together with gentle brushing using an ultrasonic cleaner (Jiemeng JP−880) at a frequency of 42 kHz to remove adhered sediment particles. After cleaning, the seed was placed on absorbent paper and allowed to dry naturally and slowly in shade. Photographs of the seed were taken using a Nikon SMZ25 stereo microscope at the Museum of Biology, Sun Yat-sen University, Guangzhou, China. Length, width, length of ventral infold, and other morphological data were measured. A micro-computed tomography (Micro-CT) system was used to reveal the internal three-dimensional structure of the seed (Zeiss Xradia 520 Versa X-ray microscope (CT) housed at the University of Science and Technology of China in Hefei, Anhui province). Image data were processed by Dragonfly v.4.1 (https://www.theobjects.com/dragonfly) software.

Morphological terms were employed here to describe the seed of Ampelopsis [follow Tiffney and Barghoorn (1976) and Chen and Manchester (2011)] (Figure 2).

The identification of Vitaceae seeds is relatively reliable because of the characteristic pair of ventral infolds and a dorsal chalaza that is absent in other families (Chen and Manchester, 2011). The morphological structure of Vitaceae seeds is important in classification, and the variation of the shape and position of features such as the chalaza, the ventral infold, and the raphe has great value in the taxonomy of Vitaceae (Chen et al., 2007). Based on the external morphology, the seeds of extant Vitaceae can be identified to a specific genus and even a species (Tiffney and Barghoorn, 1976). A long or linear chalaza can be seen in seeds of Leea L., Cissus L., Cyphostemma Alston, Tetrastigma K. Schum., Rhoicissus Planch., and Cayratia Juss. (Chen and Manchester, 2011), however, our seed has a nearly elliptical chalaza. The whorled, spiny rugas are a distinctive feature present only in Acareosperma Gagnep. but absent in our fossil. In addition, Acareosperma has an extremely thickened endotesta in certain parts of the seeds (C41 > 0.15), but the maximum thickness (C41) of the endotesta of our fossil is only 0.08 (Chen and Manchester, 2011). Unlike our fossil, Ampelocissus Planch., Pterisanthes Blume, and Nothocissus Latiff are distinguished from those of all other genera of Vitaceae by having ventral infolds that are almost as long as the full length of the seed and chalaza that is located in the center of the dorsal side; the chalaza of Parthenocissus Planch is round and centrally positioned, a distinct rib, groove-like infold, and a raphe that tapers from the apex to the base can be seen in the ventral side; Yua C. L. Li has a pear-shaped seed that is flat on the dorsal and ventral sides, linear ventral infolds that extend two-thirds of the length of the seed from the base, ventral-infold circularity (C8) < 0.4 (up to 0.58 in our fossil); Clematicissus Planch. only has one long and linear ventral infold and a chalaza that is located centrally (Chen and Manchester, 2007, 2011; Chen et al., 2007).

Our fossil displays a distinctive feature of the seeds of Ampelopsis that the seed coat is well-developed near the ventral-infold opening, as seen in the cross-section, but it is significantly thinner in the ventral-infold cavities and have an abrupt “keyhole” outline, which readily distinguishes it from Vitis L. and Ampelocissus (Manchester et al., 2013). Our fossil is consistent with Ampelopsis seeds, which have a pair of ventral infolds that are short and parallel to slightly divergent (equal-in-width or the raphe tapers between the ventral infolds), and a chalaza that is close to the notch and oval to pyriform in shape (Chen and Manchester, 2011; Manchester et al., 2013). Based on macroscopic morphology and structural characteristics, our fossil can be reliably assigned to Ampelopsis. We compared our fossil with some Asian species of Ampelopsis based on research of Chen et al. (2007), Chen and Manchester (2011), and our observation of specimens.

Both Ampelopsis grossedentata (Hand.-Mazz.) W. T. Wang (Figures 3Q–W) and A. chaffanjonii (H. Léveillé and Vaniot) Rehder have an oval chalaza, but the ventral infold in our specimen is nearly semicircular, which is different from the inverted oval of A. grossedentata. The skirt-shaped tip of the beak in our specimen is also different from the sharp beak of A. chaffanjonii. Our specimen is heart-shaped with an obvious apical notch, which is also different from the inverted oval seeds with a circular apex as shown in A. grossedentata and A. chaffanjonii.

Just like Ampelopsis gongshanensis C. L. Li and A. hypoglauca (Hance) C. L. Li, our specimen also has obvious rugas on the dorsal and ventral surfaces, but the raphe between the ventral infolds of our specimen is nearly equal-in-width along with its length, while it tapers from the apex to the base in A. gongshanensis. The chalaza of our specimen is close to the apical notch and different from that of A. hypoglauca in which the chalaza is in the middle of the dorsal surface. In addition, A. gongshanensis and A. hypoglauca differ from our specimen in having obovoid seeds with round and blunt apices and sharp basal beaks.

The chalaza of our specimen and that of Ampelopsis rubifolia (Wall.) Planch. is elliptical, with obvious rugas on the dorsal and ventral surfaces, but A. rubifolia has a prominent upper raphe, an obovoid seed, and an inconspicuous ventral infold. Both our specimen and A. tomentosa Planch. have short ventral infolds, but the seeds of A. tomentosa are inverted triangular oval in shape, and the dorsal raphe of the seed is prominent.

Both our specimen and Ampelopsis cantoniensis (Hook. et Arn.) Planch. (Figures 3L–P) have an oval chalaza, as well as obvious dorsal and ventral rugas. However, the seeds of A. cantoniensis are obovate, the apex is round, the ventral infolds are not obvious, and the tip of the beak is sharp, while our specimen does not have these characteristics.

Unlike our fossil seed, in Ampelopsis acutidentata W. T. Wang, the chalaza is located in the center, and the ventral surface has a prominent rib. Seeds of A. bodinieri (Levl. et Vant.) Rehd. have an acute beak. Seeds of A. acutidentata and A. bodinieri are obovate with a rounded apex and a smooth surface. A. aconitifolia Bge., A. megalophylla Diels et Gilg, and A. delavayana Planch have obovoid seeds with centrally positioned chalaza and ventral groove-like infolds. A. heterophylla (Thunb.) Sieb. et Zucc. seed is oblong with a nearly round top, and the ventral infold is oval and extends diagonally from the base to the top of the seed. A. humulifolia Bge. has a rounded seed apex, a chalaza located in the center, and the ventral infold is oval. The seed of A. glandulosa (Wall.) Momiy is narrowly elliptic. Seeds of A. vitifolia (Boiss.) Planch. have rounded tops, smooth surfaces, and distinctive ventral ribs.

The seed of Ampelopsis japonica (Thunb.) Makino is obovoid to heart-shaped, the surface of the seed is smooth or has obvious rugas, the chalaza is elliptical, the beak is prominent, has a nearly semicircular ventral infold located in the middle of the seed, and a nearly equal-in-width raphe is located between the ventral infolds. Our fossil shows the greatest similarity with this species in terms of all the above characteristics.

Chen (2009) divided Vitaceae fossil seeds into 14 types based on morphological measurement data. After measurement and comparison, our fossil seed can be classified as seed type st-Ampelopsis-rugose. Therefore, we compared our fossil with the seeds of Ampelopsis belonging to this type and the newly reported seeds of Ampelopsis which have obvious rugas on the surface (Table 1).

Our specimen is similar to Ampelopsis crenulata (Reid and Chandler, 1933; Chandler, 1978) and A. turneri (Chandler, 1961) is known from the lower Eocene of the United Kingdom. They have a dorsal surface that is obviously raised on the side view, rugas arranged radially around the chalaza, and there are obvious grooves around the chalaza. However, the ventral infold of our specimen is wide and our specimen has a basal groove, while that of A. turneri is narrow without a basal groove. The apex of our specimen has an obvious notch, and the ventral infold accounts for approximately one-third of the total length, while that of A. crenulata is only slightly grooved and accounts for five-eighths of the total length. In addition, our fossil is heart-shaped, and the nearly equal-in-width raphe is located between the ventral infolds, while the shapes of A. crenulate and A. turneri are obovoid, and the raphe tapers from the apex to the base.

A fossil seed of Ampelopsis sp. reported from the middle Eocene of Germany (Collinson et al., 2012) is obovoid in dorsal and ventral views and lacks a prominent beak, while our specimen is heart-shaped with a prominent beak. Moreover, the seed of Ampelopsis sp. is 6 mm long, and it is longer than our specimen (4.38 mm). The ventral infold length of Ampelopsis sp. accounts for approximately one-half of the total seed length, while in our specimen, it only accounts for one-third. In addition, our fossil does not have 16–20 wide, sharp-crested ridges on the dorsal surface like those possessed by Ampelopsis sp.

Like Ampelopsis cf. malvaeformis (Schlotheim) Mai (Meller, 1998) known from the upper and middle Eocene of Austria, our specimen has nearly the same width of the raphe between the ventral infolds and has obvious dorsal rugas. However, the length (4.38 mm) and width (4.22 mm) of our specimen exceed those of A. cf. malvaeformis (2.5–3.2 mm in length and 1.8–2.6 mm in width). In addition, the seed of A. cf. malvaeformis is obovate to conical, with a prominent dorsal chalaza, while our specimen is heart-shaped, and the dorsal chalaza is obviously lower than the dorsal plane.

Both Ampelopsis ludwigii (A. Braun) Dorofeev found in the Lower Miocene of the Czech Republic (Teodoridis, 2003; Martinetto et al., 2014) and our specimen are heart-shaped, and both have a prominent beak at the base, with obvious radial rugas around the chalaza. However, in A. ludwigii, the raphe between the ventral infolds tapers from the apex to the base, which is different from our specimen. Moreover, the tip of the beak in our specimen is skirt-shaped, while that of A. ludwigii is sharp.

As with the Ampelopsis ludwigii reported from the Lower/Middle Miocene of Germany (Czaja, 2003), our specimen has a nearly semicircular ventral infold in the middle, a prominent beak at the base, and the raphe between the ventral infolds is nearly equal-in-width throughout its length. However, our specimen is heart-shaped, the ventral surface also has obvious rugas, and the chalaza is obviously lower than the dorsal plane, while A. ludwigii is obovoid, the ventral infolds are almost invisible, and the chalaza is convex.

There are 3–5 shallow and flat rugas on the dorsal surface of Ampelopsis cf. ludwigii, which was found in the Miocene of the Czech Republic. Unlike our specimen, this fossil has a narrow ventral infold, its basal groove is missing, and the raphe gradually narrows from the apex to the base (Bužek et al., 1976).

Miki (1956) reported Ampelopsis leeoides Planch. from the Pliocene of Japan. This fossil has 4–7 pairs of rugas arranged symmetrically on the dorsal surface, its raphe is nearly equal in width, and the nearly semicircular ventral infold with cracks along with the peripheral rugas is located in the middle of the seed. Our specimen is similar to it. However, the beak of A. leeoides is not obvious, the apex is nearly round, and the ventral infold occupies ~1/2 of the total length of the seed. Moreover, the overall shape of A. leeoides is from obovate to oval. Our fossil does not have these characteristics.

To sum up, the Ampelopsis fossil seed described here is different from all the above fossils and most extant species of Ampelopsis, but it shows the greatest similarity with extant Ampelopsis japonica, so our fossil seed can be assigned to this living species.

The Ampelopsis fruit is a berry with 1–4 seeds, the ovary is two-locular (but the septa do not fuse) with two anatropous ovules in each chamber, typically has the aborted ovule (Figures 3V, W) and possesses a funiculus extending into the chalazal end of the ovule (Wen, 2007; Ickert-Bond et al., 2014; Gerrath et al., 2015).

The shapes of Ampelopsis seeds can be significantly affected by the number of seeds per berry (Chandler, 1961; Tiffney and Barghoorn, 1976) (Figure 4). If a berry contains several seeds, they will squeeze each other during the growth and development process, forming multiple planes with edges and corners on the ventral surface (Figures 3N, O, 4A, B). If a berry contains only one seed in a locule, the seeds usually have a flat surface (Figures 3S–U, 4B–D). However, our fossil has a flat ventral surface without edges and corners, and the edge is rounded (Figures 3E, F, H, I). We, therefore, infer that there may have one to three seeds in the berry from which our fossil seed is derived, which are in one (Figure 4D) or two locules (Figures 4B, C), respectively, indicating that at least one ovule in one locule of the fruit is aborted. Aborted ovules commonly occur in the fruits of extant Ampelopsis. According to the structural characteristics of our fossil, we speculate that this phenomenon of Ampelopsis fruit development existed in the Late Pleistocene.

The vascular strand that connects the seed with the berry culminates in a relatively dark knot on the upper part of the seeds, marking the chalaza (Gerrath et al., 2015). There is a cavity structure at the chalaza in extant Ampelopsis seeds, and the chalaza can be clearly distinguished from the adjacent seed coat tissue. This special structure is also found in our fossil seed. The obvious cavity structure at the chalaza can be seen in the seed section imaged by Micro-CT (Figures 3G, J), and this part of the tissue is relatively loose, clearly distinguishing it from the surrounding seed coat tissue (Figures 3C, D, G, J). This indicates that the special chalazal structure of Ampelopsis seeds existed in the Late Pleistocene, and there has been little change in this structure since Late Pleistocene. To a certain extent, it reflects the stability of plant reproductive organs' structure.

Ampelopsis is mainly distributed in Eurasia, North America, and Central America (Chen, 2009; Nie et al., 2012), with most occurring in East Asia and China [17 species (13 endemic species)] (Chen et al., 2007; Fang et al., 2009) (Figure 5). Ampelopsis has abundant fossil records (Figure 5), which are mainly from the Paleocene to Pleistocene in Europe. The earliest fossil record of this genus is Ampelopsis cf. monasteriensis Kirchheimer, a seed reported from the Paleocene of Germany (Mai, 1987). In addition, fossils ascribed to this genus have been reported from the Eocene of Britain and Germany, the Oligocene of Poland, the Miocene of Austria, Germany, Poland, the Czech Republic, and Bulgaria, as well as the Pliocene to the Pleistocene of Italy (e.g., Reid and Chandler, 1933; Teodoridis, 2003; Martinetto et al., 2014; Kowalski et al., 2020). A. rooseae Manchester was discovered in the middle and upper Eocene of Oregon (Manchester, 1994; Manchester and McIntosh, 2007), and a seed fossil of the genus was also reported from the Miocene of Washington state (Tcherepova and Pigg, 2005). In Asia, Ampelopsis fossils are only reported from the Eocene to the Miocene of Siberia, Russia (e.g., Dorofeev, 1963; Nikitin, 2006), the Upper Miocene of China (Huang et al., 2017), and the Pliocene of Japan (Miki, 1956).

To date, apart from the fossil seeds from the Upper Miocene of Zhaotong Basin, Yunnan province (Huang et al., 2017), there have been no reports of Ampelopsis fossils in China. The Ampelopsis fossil seed found in Maoming Basin is the first fossil record of the genus from South China, indicating that Ampelopsis had reached South China by the Late Pleistocene. Previous fossil records of Ampelopsis have mostly been distributed in the mid-latitudes (Figure 5), but our specimen was discovered in the low-latitude area of South China. Four earth's climate states are identified during the past 66 Ma: Warmhouse (66–56 and 47–34 Ma), Hothouse (56–47 Ma), Coolhouse (34–3.3 Ma), and Icehouse (3.3–0 Ma) (Westerhold et al., 2020). During the Pleistocene, especially during the last glaciation, the global temperature experienced a rapid decline (Liu et al., 2001), and the temperature decline also occurred in the Maoming (Huang et al., 2023), the climate in Maoming might become suitable for the Ampelopsis. Thus, we infer that global cooling during the last glaciation led to the southward spread of the genus into the low-latitude area of South China.