Paleolithic human exploitation of plant foods duringthe last glacial maximum in North ChinaLi Liua,1, Sheahan Bestelb, Jinming Shic, Yanhua Songd, and Xingcan Chenb

aDepartment of East Asian Languages and Cultures, Stanford Archaeology Center, Stanford University, Stanford, CA 94305; bInstitute of Archaeology, ChineseAcademy of Social Sciences, Beijing 100710, China; cShanxi Museum, Taiyuan 030024, China; and dDepartment of Archaeology, Shanxi University, Taiyuan030006, China

Edited by Dolores R. Piperno, Smithsonian National Museum of Natural History and Smithsonian Tropical Research Institute, Fairfax, Washington DC,and approved February 11, 2013 (received for review October 18, 2012)

Three grinding stones from Shizitan Locality 14 (ca. 23,000–19,500calendar years before present) in the middle Yellow River regionwere subjected to usewear and residue analyses to investigatehuman adaptation during the last glacial maximum (LGM) period,when resources were generally scarce and plant foods may havebecome increasingly important in the human diet. The results showthat these tools were used to process various plants, including Triti-ceae and Paniceae grasses, Vigna beans, Dioscorea opposita yam,and Trichosanthes kirilowii snakegourd roots. Tubers were impor-tant food resources for Paleolithic hunter–gatherers, and Paniceaegrasses were exploited about 12,000 y before their domestication.The long tradition of intensive exploitation of certain types of florahelped Paleolithic people understand the properties of these plants,including their medicinal uses, and eventually led to the plants’ do-mestication. This study sheds light on the deep history of the broadspectrum subsistence strategy characteristic of late Pleistocene northChina before the origins of agriculture in this region.

ancient starch | late Paleolithic China | plant processing |usewear analysis | stone tool function

It has long been argued that foragers at the end of the Pleis-tocene broadened their resource base to encompass a wide

array of plant and animal foods and that this “broad spectrumrevolution” entailed the transition to farming (1). New studieshave shown that human exploitation of an extensive range ofplant foods can be traced back to much earlier dates in latePaleolithic times in southwest Asia, suggesting a long history ofintensive foraging for plants before domestication (2).Usewear traces and plant residues on grinding stones can

provide crucial information for understanding the plant-foragingstrategies of ancient people. By applying these methods, severalprojects have shown evidence of widespread plant use, includingwild cereals and tubers, by late Paleolithic populations in manyareas of the world, dating to a period between ca. 30,000 and 23,000y ago, including the Near East (3), Europe (4), and Australia (5).In China, the earliest grinding stones have been uncovered

from several Paleolithic site clusters distributed on the LoessPlateau region along the middle Yellow River valley. These in-clude Longwangchan in Shaanxi (6) and Shizitan and Xiachuanin Shanxi (7–9), dating to ca. 25,000–9000 calendar years beforepresent (cal. B.P.). A study of usewear traces and starch residueson grinding stones from Locality 9 (S9 hereafter) in the Shizitansite cluster (ca. 12,700–11,600 cal. B.P.) has demonstrated thatpeople used these tools to process various plant foods, includinggrasses, tubers, acorns, and legumes (10). Among the grass starchgranules uncovered at this site, some from Panicoideae may havebeen the wild ancestors of domesticated millets (Panicum mil-iaceum and Setaria italica ssp. italica). Therefore, it is importantto investigate the use of plants in an earlier period in this region,to trace possible continuity in a putative plant procurementstrategy that may have eventually led to domestication. In thisstudy, we demonstrate usewear patterns and residues on threegrinding stones (ca. 23,000–19,500 cal. B.P.) excavated fromLocality 14 (S14 hereafter) in the Shizitan site cluster (9), whichdisclosed the exploitation of Triticeae and Paniceae grasses,

Vigna beans, Dioscorea opposita yam, and Trichosanthes kirilowiisnakegourd roots.

Sites and Environmental ContextsS14 (110°32′40″ E, 36°02′11″ N; 655 ± 5 m in altitude) is amonga series of more than 50 late Paleolithic localities (ca. 25,000–9,000 cal. B.P.), referred to as the Shizitan site cluster, distrib-uted along the Qingshui River, a tributary of the Yellow River insouthern Shanxi (Fig. 1). All of these localities are characterizedprimarily by small flaked tools and a microlithic technology, withno pottery, dwelling structures, human burials, or storage facili-ties found, suggesting that the occupants were mobile hunter–gatherers (7–9, 11). The long time span of human occupationwith abundant material remains provides a wealth of data forstudy of the transition from the late Paleolithic to the earlyNeolithic in the middle Yellow River valley.Excavations at S14 in 2000, 2003, and 2005 revealed three

continuous cultural strata (II, III, and IV) within a depth of 50–100 cm in an area of 25 m2, dating to a time period of ca. 23,000–18,000 cal. B.P. (Table S1). Material remains uncovered include1,643 lithic artifacts and 2,776 animal bone fragments, in addi-tion to 17 ash and burnt surface areas (fireplaces). Severalgrinding tools (slabs) have been found near fireplaces in all strata(refs. 8 and 9; ref. 12, p. 20), and three of them (all made ofsandstone), analyzed here (Fig. 1), were uncovered from StrataIII and IV (ca. 23,000–19,500 cal. B.P.).The site is situated in a mountainous region with a temperate

climate and a diversity of flora resources today. Pollen data ob-tained from Shizitan suggest that vegetation coverage in the re-gion was dominated by grasses from 35,100–9,400 B.P. (dating bythermoluminescence analysis) (13). During the last glacial maxi-mum (LGM), the climate was dry and cold, featuring a steppeenvironment. During the period of ca. 18,500–13,200 cal. B.P., thisregion experienced the last deglaciation, characterized by a mildand arid or semiarid steppe environment with a small amount ofdeciduous and broadleaf species. This episode was then followed bya dry and cold period dating to 13,200–13,000 cal. B.P. before an-other era of improved climatic conditions (ref. 12, p. 167).

ResultsWe collected usewear and residue samples from three grindingstones (GS1, GS2, GS3), which were kept in the Shanxi Museumafter excavation. To compare used tools with unused natural stone,we also took samples from a sandstone rock (SS1) in a nonculturaldeposit near the site (Table S2).

Usewear Analysis. Polyvinyl siloxane (hereafter PVS or peel) wasapplied to the Shizitan grinding stones to provide portable and

Author contributions: L.L. and X.C. designed research; L.L., S.B., J.S., Y.S., and X.C. per-formed research; L.L. and S.B. analyzed data; and L.L. and S.B. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.1To whom correspondence should be addressed. E-mail: liliu@stanford.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1217864110/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1217864110 PNAS Early Edition | 1 of 6

ANTH

ROPO

LOGY

durable records for microscopic analysis. Previous research, us-ing PVS on usewear patterns from grinding stones in China (10,14, 15) and other parts of the world (16–19) established valuablereference data for the study of ancient tools. In recent years, wehave also conducted a series of experimental studies on grindingstones (20) to produce references for processing seeds, tubers,nuts, wooden objects, minerals, shells, and stone implements. Basedon these studies, the analytical variables examined in the currentproject include: stage of polish development (low, medium, high),polish reticulation pattern, polish topography, striations (furrow,sleek, fine), pitting and pecking, and surface microtopography(Table S3). The results of the analysis of the three grinding stones(GS1–GS3) are as follows.GS1.GS1 is a fragment of a slab (weight, 0.74 kg), and both sidesshow microscopic traces of use, mainly small areas of polish. Onthe top side, the microscopic surface is relatively flat, and higherplateau areas show more polish than lower valley areas. In someareas, very fine striations run multidirectionally on the high levelof the grain surface; in other areas, crystal grains are clearly flat-tened with polish (Fig. 2, 1 and 2). Based on our experimental study,similar polish and very fine striations appear on grinding slabsafter processing dry tubers (Dioscorea opposita yam, Trichosantheskirilowii snakegourd root, and Pueraria lobata kudzu-vine root) for1 h or longer (Fig. S1, 5 and 6).GS2. GS2 is a small slab (weight, 0.25 kg), and the top surface isslightly concave. Three PVS peels taken from the top surfaceshow similar usewear traces: the microscopic surface is relativelyflat, and the high plateau areas show more polish than the lowervalley areas. On slightly lower areas, striations are sometimesvisible, mostly very fine, faint, short, and running in differentorientations (Fig. 2, 3). These patterns are consistent with thosefrom processing plants, including dehusked wheat and millet, inour reference data (Fig. S1, 2 and 3). Grinding movements weremultidirectional, probably including linear bidirectional (back-and-forth) and circular motions. Pitting is visible in some areas,probably caused by processing hard-shelled seeds or nuts. In raresituations, some wide angular striations (furrows) are present(Fig. 2, 3), resembling the traces of stone-on-stone abrading donein our experimental study (Fig. S1, 1). These furrows, however,may have been caused by unintentional contact with hand stones.GS3. GS3 (weight, 5.64 kg) is the largest one in the assemblage. Itwas found face-up on top of GS2, situated above a fireplace. Two

PVS peels were taken from the used surface. The microscopicsurface is rather flat with a few polished areas, and striations arerare. On peel one (P1), some crystal grains show medium-levelpolish without striations, and on P2, small areas of shallow stria-tions with U-shaped cross-sections (sleeks) are present (Fig. 2, 4and 5). Compared with the references in our experimental studies,the polish is similar to that from tuber grinding, and the striationsresemble traces from large seed processing, such as with beans(Fig. S1, 4).

Summary. Three slabs examined all show usewear traces, whichclearly differentiate them from unused stone surfaces (Fig. 2, 6).These slabs were apparently used to process various plant foods,such as seeds, tubers, and nuts. To understand exactly what plantswere involved here, we have to rely on residue analysis.

Residue Analysis. Ten residue samples were extracted from thethree tools either by applying a small amount of distilled water tothe tool and then extracting the sample with a pipette or by re-moving sediment adhering to the tool surface (21, 22).Phytolith remains on the S14 grinding implements were minimal

and mostly not diagnostic (Supporting Information and Table S4).Grass husk phytoliths were not recovered, suggesting that thesegrinding stones were probably not used to dehusk cereals. Charcoaland/or burnt phytoliths were present on all three tools, perhapsreflecting the location of the implements near fireplaces or foodprocessing areas.

Yellow River

Qingshui River

JixianDongchengxiang

Qianxialing

ShiziheXijiaotou

Dahuling

Jingmen

36

Shizitan site

modern villagemodern town

2 km

110° 30' 110° 40'

S1 S14

S9S29

Baishansixiang

China

A

BGS1 GS2 GS3

10 cm 10 cm10 cm

PVS sampling location starch sampling location

P1

P2

P1

P2

P3

1

2

1

2

P1

1

Fig. 1. Site location and artifacts analyzed. (A) Major localities of the Shi-zitan site cluster in Jixian, Shanxi. (B) Grinding stones analyzed and samplinglocations on the tools.

1 2

3 4

5 6

Fig. 2. Usewear patterns from S14 grinding stones. (Magnification: 200×)(1 and 2) GS1, fine striations running diagonally and a polished crystal sur-face, resembling tuber processing. (3) GS2, small pitting (right), short furrows(lower left), and very fine and faint striations running vertically (upper left),likely related to pounding small, hard objects, abrading stone objects, andprocessing plants, respectively. (4) GS3-P1, a medium level polished area,similar to tuber processing. (5) GS3-P2, sleeks running vertically, similar todry bean or large seed processing. (6) Uneven surface of natural crystals onan unused sandstone, for comparison with used tools.

2 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1217864110 Liu et al.

Six of the 10 residue samples yielded a total of 136 starchgranules. In general, samples taken from used surfaces tend toproduce more starch residues than those from unused surfaces.Two sediment samples (GS1-3 and GS3-4) and one water sample(GS2-3), all taken from unused surfaces, contained no starch.GS1 and GS2 revealed only 10 starch granules (7.4% of thetotal), likely because the tools were washed after excavation,a situation that apparently affected the survival rate of starchresidues. In contrast, 92.6% of total starch granules were foundin the samples from GS3, which was unwashed before sampling.Sixty-five granules (48% of the total) show characteristics ofdamage, many resembling starch granules after grinding, as seenin previous studies (23) and our reference data. The samplestaken from a natural rock contain a small number of starchgranules with no sign of damage. These observations suggest thatthe profiles of starch residues recovered from used surfaces ofancient grinding stones are most likely associated with toolfunction rather than with the enclosing soil matrix, an inferenceconsistent with several previous studies (e.g., refs. 10 and 24–26).Among over 900 specimens in our modern reference collec-

tions, we specifically analyzed those starch-rich and economicallyimportant samples relevant to the research area, including 156samples belonging to 83 species in 45 genera of 18 families. Ofthe 136 starch granules uncovered from S14, 121 (89% of thetotal) are identifiable compared with our references. The ancientstarch granules are classified into six types on the basis of theirmorphology and size. The unidentifiable starch granules (15;11%) lack diagnostic features comparable to available references(Table S5). Because the starch remains from grinding stones arelikely to have derived from ground plants, we compared them

with starch extracted from ground seeds/tubers in our modernreference samples (Supporting Information).

Type I Starch. Type I starch granules (n = 45; 33% of the totalstarch) are round or oval in shape and relatively large in size(14.56–39.21 μm). The surface is rather flat, the hilum is centric,lamellae are visible on large granules, and the extinction cross isshaped as either “×” or “+.” Most of the type I granules appeardamaged (33; 74%), showing broken edges, deep fissures, pro-nounced lamellae, and/or a dark central area on the extinctioncross (Fig. 3, 1–4). These starch granules, in morphology andsize, resemble many taxa in the Triticeae tribe of the grass familyindigenous to north China and still found in Shanxi today (27).This includes many genera of Agropyron, Elymus, Roegneria, andLeymus in our references, as well as those reported in otherstudies (ref. 28, figure 4). The characteristics of damaged gran-ules are also consistent with those found in ground Leymus andAgropyron from our references (Fig. 4, 1–4).It is possible that type I starches belong to more than one genus

in the Triticeae tribe, although the large granules are particularlysimilar to Agropyron cristatum and Agropyron desertorum in form(Fig. 4, 2 and 3) and size (Fig. 5, 1).

Type II Starch. Type II starch granules (n = 20; 15% of the totalstarch) are irregularly oval or nearly kidney-shaped and relativelylarge in size (17.31–40.76 μm). Fissures and lamellae are visiblein most cases, and the extinction cross often exhibits many arms.Seven granules (35%) are damaged, characterized by brokenedges and/or a large dark area at the center (Fig. 3, 5–8), con-sistent with bean starch granules damaged by grinding (23). Type

1 2 4

5 6 7 8

9 10 11 12 13 14

15 16 17 18

19 20 21 22 23

24 25 26 27 28

TypeI

TypeII

TypeIII

TypeIV

TypesV&VI

3

Fig. 3. Starch grains uncovered from Shizitan S14,compared with tuber starch grains from Shizitan S9,Shigu, and Egou (under DIC and polarized filters). (1–4) Type I starch (Triticeae), showing damages (3 and4). (5–8) Type II starch (Vigna sp.), showing damages(7 and 8). (9–14) Type III starch (Paniceae), showingdamages. (15–18) Type IV starch (D. opposita).(19–21) Type V starch (T. kirilowii). (22 and 23) TypeVI starch (T. kirilowii). (24 and 25) Type VI starchfrom Shizitan S9. (26–28) Types V and VI starch fromPeiligang culture sites at Shigu (26) and Egou (27 and28). Panels 26–28 are reproduced with permissionfrom ref. 14. (Scales bars: 9–14 and 19–28, 10 μm;others, 20 μm.)

Liu et al. PNAS Early Edition | 3 of 6

ANTH

ROPO

LOGY

II granules resemble beans in the Phaseoleae tribe, particularlyVigna species, in shape and size (Figs. 4, 5–8 and 5, 2), such asVigna angularis, Vigna unguiculata, and Vigna radiata. It is diffi-cult to identify type II starch to the level of species.

Type III Starch. Type III starch granules (n = 18; 13% of the totalstarch) are characterized by small sizes (8.95–18.81 μm) andpolygonal shapes. The hilum is centric, with star or Y-shapedfissures often radiating toward the periphery, whereas extinctioncrosses are mostly “+”-shaped with straight arms. Most starchgranules (13; 72%) in type III are damaged, showing a large

central depression with or without a small circular protrusion inthe middle, pronounced fissures, widened extinction cross, and/or visible lamellae on part of the granule (Fig. 3, 9–14). Type IIIstarch granules are similar in shape to several genera within thePaniceae tribe. Some genera in our reference samples, such asSetaria, Panicum, Echinochloa, Pennisetum, and Digitaria, allexhibit both faceted and spherical granule shapes, the presenceof fissures, a centric hilum, and straight cross shape, similar totype III starch granules. However, the damaged type III granulesparticularly resemble, in morphology, those from green foxtailgrass (S. italica ssp. viridis), which is the wild ancestor of do-mesticated foxtail millet (S. italica ssp. italica), and barnyardgrass (Echinochloa crusgalli) Fig. 4, 9–14).In general, wild Paniceae starch granules (approximately <14

μm) are smaller in size range than those from domesticated millets(ref. 10, figure 4; refs. 28–30). However, type III starch granulesfrom S14 show sizes greater than 14 μm in half of the cases ex-amined (9 of 18). These large granules only overlap well with thesize ranges of domesticated foxtail millet and a ground green foxtailsample from Zhengzhou (Henan) in our references (Fig. 5, 3).Based on our experimental study, starch granules from several

wild and domesticated Paniceae species become larger in sizeafter grinding, with the maximum length of granules increasing27–56% among several reference samples of Echinochloa, Pan-icum, and Setaria spp. It is known that milling can cause struc-tural change in starch granules (31), and in several cases, ancientPaniceae starch granules from grinding stones appear larger thanthose from the modern corresponding species (ref. 28, p. 254).Given that grasses taken from 23,000-y-old tools at S14 are un-likely to be domesticated species, it is possible that some type IIIstarch granules belong to wild Paniceae grasses, including Setariaand other genera, and that their large sizes may have beencaused by grinding (Supporting Information).

Type IV Starch. Type IV starch granules (n = 24; 18% of the total)are characterized by a large granule size range (19.33–59.05 μm),irregular triangular or oval shapes, an extremely eccentric hilum,the presence of lamellae in most cases, and an extinction crosswith bent arms (Fig. 3, 15–18). One granule appears to bedamaged, showing broken edges, rough surface, and a dark areain the center of the extinction cross. These starch granules mostresemble the D. opposita yam, which has many cultivated andwild variations in north China (Fig. 4, 15 and 16). Among five D.opposita samples we collected from Henan, the domesticatedones show smaller granule size ranges (8.83–33.16 μm, 14.01–53.16 μm, 11.16–48.22 μm) than those of two wild ones (16.27–63.08 μm, 22.9–70.62 μm). Type IV starch is more comparablewith wild yam in morphology and size (Fig. 5, 4).

Types V and VI Starch. Type V starch granules (n = 4; 3% of thetotal) are characterized by a bell shape, medium size (14.75–22.96

1 2 3 4

5 6 7 8

9 11 12

13 14 15 16

17 18 19 20

10

Fig. 4. Modern starch references (under DIC and polarizedfilters). (1) Leymussecalinus (ground). (2)A. cristatum (ground). (3 and 4)A. desertorum (ground).(5 and 6) V. radiata. (7 and 8) V. unguiculata (ground). (9) S. italica ssp. viridis,showing undamaged granules. (10) P. miliaceum (wild), showing undamagedgranules. (11 and 12) S. italica ssp. viridis (ground), showing pronounced fis-sures. (13) S. italica ssp. viridis (ground), showing central depression with aprotrusion. (14) E. crusgalli (ground), showing central depressions on mostgranules. (15 and 16) D. opposita (wild). (17–20) T. kirilowii. (Scale bars: 9–14,10 μm; others: 20 μm.)

2: Type II compared with Vigna beans

3: Type III compared with Paniceae grasses and mellit

1: Type I compared with Triticeae grasses

4: Type IV compared with Dioscorea opposita yam

5. Types V & VI compared with Trichosanthes kirilowii snakegourd root

Size

in µ

m

5

10

15

20

25

30

35

40

SZT1

4Ty

p eV

(N=

4)

SZT 1

4Ty

p eVI

(N=

1 1)

Shiz

itan

9T y

pesV

- VI (

N=

4)

Egou

-Sh i

guT y

pes V

- VI(

N=

10)

T.ki

ril o

wii

Hen

an ( N

=148

)

10

20

30

40

50

60

70

80

SZT1

4Ty

pe IV

( N=

20)

D.o

p po s

i ta

(dom

.) Lu

o yan

g(N

=10

7)

D. o

p pos

i ta

Yans

hi(N

=13

4)

10

20

30

40

50

60

70

SZT1

4Ty

pe II

(N=

20)

V.a n

gula

ris

(dom

.)(N

=16

9)

V .un

g ui c

u lat

e (d

om.)

(N=

159)

V.ra

diat

a(N

=142

)

0

10

20

30

40

0SZ

T14

Type

I(N

=32

)

Agro

pyro

ncr

ista

tum

(N=1

77)

Leym

usse

cali n

us(N

=154

)

10

20

SZT1

4Ty

pe II

I (N

=18

)

Echi

noch

loa

cru s

gall i

JP(N

= 123

)

Pani

cum

mil i

aceu

mw

ildB J

(N=

153)

Seta

r ia

f abr

i (N

= 142

)

S. it

alic

avi

ridi

sI. M

ongo

lia(N

=14

4)

S. i.

viri

disB

J(N

=17 9

)

S.i.

viri

disZ

heng

zhou

(N=1

30)

S. i.

i tal

ica

(dom

.) Sh

aanx

i(N

=147

)

Fig. 5. Ancient starch types I–VI compared with modernreferences after grinding (all reference samples are wildforms unless indicated as domesticated).

4 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1217864110 Liu et al.

μm), an eccentric hilum, absence of lamellae, and an extinctioncross with bent arms (Fig. 3, 19–21). Type VI starch granules (n =11; 8% of the total) are regularly round or regularly oval in shape,and medium in size (11.47–29.72 μm). The hilum is eccentric inmost cases, lamellae are visible on large granules, and the ex-tinction cross has bent arms (Fig. 3, 22 and 23). One granuleappears damaged, as indicated by its broken edges.Types V and VI starch are most comparable in morphology and

size with the root of snakegourd (T. kirilowii in the Cucurbitaceaefamily) in our reference data. Starch granules from modernT. kirilowii are varied in shape, and include spherical, oval, bell-shaped, and polygonal morphologies. The hilum is eccentric inmost cases, fissures in various forms are sometimes present, la-mellae are visible on some large granules, and the extinction crossoften has bent arms (Fig. 4, 17–20). When sizes are compared,types V and VI also fall into the range of T. kirilowii (Fig. 5, 5).Types V and VI particularly resemble the large granules inT. kirilowii, which are often regularly oval and bell-shaped.

Summary. The recovery rates of starch from the three tools areuneven, partly relating to whether or not the artifact was washedin the postexcavation process. GS3 (unwashed) yielded the moststarch granules (126) with all six types present. Taken as a whole,the S14 starch assemblage is dominated by Triticeae grasses(33%), followed by yam (18%), beans (15%), Paniceae grasses(13%), and snakegourd root (11%). Although the percentage ofstarch granules from certain plants cannot be used to determinethe proportion of plant use in the foraging strategy, they help usto understand, in a broad view, the components of starchy foodin the hunter–gatherers’ diet.Among the damaged starch granules, a majority comes from

seeds (grasses and beans), and granules with apparently enlargedsizes occur mostly in Paniceae tribe grasses. These observationsare consistent with our experimental grinding of plants, in whichPaniceae starches demonstrate the highest percentages of dam-aged granules and greatest increase in granule sizes amongvarious seeds and tubers after grinding (Tables S6 and S7).

DiscussionThe presence of a high percentage of grass starch (46% of thetotal) on S14 grinding tools, as the direct evidence for human useof these grasses for food, is consistent with the grass-dominatedecosystem of the region suggested by pollen analysis (13). Pan-iceae starch on S14 grinding stones constitutes the earliest directevidence for human consumption of these types of grasses.Starch granules identifiable to Panicoideae subfamily grassesalso have been found on grinding stones from Shizitan S9 (10),suggesting a tradition of foraging these small seeds long beforetheir domestication during Neolithic times in the region.Triticeae starch residues on grinding stones have also been

found in the late Paleolithic site at Shizitan S9 (10). Despitea long history of exploitation of Triticeae grasses in northernChina, these plants have never gone through a domesticationprocess in this part of the world.Legume starch granules, including Vigna sp., have been found

on grinding stones in many sites along the Yellow River region,including the Late Paleolithic site at Shizitan S9 (10), andNeolithic sites at Egou (14) and Shangzhai (32). Beans appear tohave been one of the earliest plant foods used by hunter–gathersin north China, but their taxonomy cannot be determined to thelevel of species based on starch data only. The earliest knownmacrobotanic remains of Vigna beans in China have been iden-tified as Adzuki (V. angularis), dating to the late Neolithic inShandong (33). Beans rich in starch from genera other thanVigna have not been found in the archaeological contexts inprehistoric north China. However, the use of Vigna for foodbefore 2,500 BC in China still awaits confirmation by macro-botanic discovery in the future.The most interesting funding of this study is the presence of

starch from two taxa of tuber (yam and snakegourd root), witha significant percentage in the starch assemblage (together 29%).

Tubers rarely survive identifiably in macrobotanic remains;therefore, starch analysis is the most effective method by whichto recover them from their archaeological contexts. Starchgranules identifiable to Dioscorea yam have been uncoveredfrom chipped stone tools dated to as early as 28,000 cal. B.P.(34), and from grinding stones (10, 14), and pottery vessels (35)found across north China, dating from the late Paleolithic,through the Neolithic, to the Bronze Age. This plant was ap-parently widely used very early as an important source of food.The presence of starch from snakegourd root provides the

earliest evidence to date that this tuber was used for food inChina. Starch granules from this tuber have previously beenfound from grinding stones in a late Paleolithic site at Shizitan S9(10), as well as in early Neolithic sites at Shigu and Egou inHenan (ref. 14, figures 7D and 8 F and H), but they were un-identifiable at the time of discovery, because of a lack of com-parative references (Fig. 3, 24–28). Like yam, snakegourd rootseems to have been widely foraged by both hunter–gatherers andearly farmers in the middle Yellow River region.Yam and snakegourd today are distributed widely in north

China, and both are used as traditional medicine (ref. 36, pp. 218and 244–245; ref. 37, pp. 103–105). Whereas yam has beencommonly cultivated and cooked as food, snakegourd root hasnot been regularly consumed in north China except when used asfamine food. The traditional method of processing and cookingsnakegourd root can be found in Jiuhuang Bencao (Herbal forRelief of Famines), written by Zhu Su (1361–1425). Snakegourdroots were skinned, cut into slices, and soaked in water for 4 to5 d, with the water changed each day. The roots were thenground with tools and sieved with textile to produce very fineflour. Alternatively, the roots were dried and ground before beingleached more than 20 times to make very fine flour. The flourthen could be used to make cakes or noodles (ref. 38, pp. 64–65).It is important to note that snakegourd root needs to be groundto flour for consumption, a scenario in line with the starch foundon grinding stones in ancient sites. We can now trace the use ofsnakegourd root back to 23,000 y ago, and such a long tradition ofuse may have helped people to recognize its medicinal propertieslong before the historical period. We are currently unable todistinguish cultivated yam and snakegourd from wild ones, basedon starch. This is a topic worthy of further research.When both usewear patterns and starch residues found on

these tools are compared, they provide supporting evidence forone another in most cases. It is particularly interesting to notethat starch granules from yam and snakegourd root occur in highfrequency on the used surface of GS3. Usewear traces from thissurface (GS3-P1) show a unique polished pattern, resemblingtuber grinding in our experimental study.Usewear traces on GS2 include pitting, probably related to

processing seeds with hard shells; however, we did not uncoverany starch from nuts, such as acorn. Given the absence ofbroadleaf trees in the pollen profile during the LGM in the re-gion (13), it is possible that nonstarchy shelled seeds other thanacorn were processed on this tool.

ConclusionsAround 23,000–18,000 y ago during the LGM, the QingshuiRiver valley appears to have been an area with a wide range offaunal and floral resources, which attracted small hunting-gathering groups. In addition to hunting, people collected andprocessed many types of plants, including grass seeds of Triticeaeand Paniceae, Vigna beans, D. opposita yam, and T. kirilowiisnakegourd roots, among others.All of these plants from S14 seem to have been continuously

exploited by hunter–gatherers for the next few millennia to thebeginning of the Holocene in the Shizitan area (S9), as well as byearly farmers in Neolithic times over a much broader region. Themajor difference between the starch assemblages of S14 and otherlater sites is the presence of Quercus starch in the latter, likelyresulting from the onset of the Holocene, when broadleaf treesbegan to be distributed in the region. Acorn needs to be processed

Liu et al. PNAS Early Edition | 5 of 6

ANTH

ROPO

LOGY

by grinding and leaching before human consumption (39). Suchfood-processing techniques appear to have developed much ear-lier, as shown by this study’s demonstration that the consumptionof certain tubers such as snakegourd root, which requires grindingand leaching, can be traced back to 23,000 y ago.Northern China is considered the area where foxtail and

broomcorn millets were first domesticated (40, 41). Recentstudies on macro- and microbotanical remains from Donghulin(Beijing) have suggested that millet cultivation/domestication maybe traced back to ca. 11,000–9,450 cal. B.P. at this site (30, 42). Thepresence of starch likely from Setaria and other Paniceae seeds atS14 suggests that these wild grasses were probably used for at least12,000 y before their cultivation/domestication recognizably af-fected seed morphology.A broad-spectrum subsistence strategy was already practiced

by people at Shizitan during the LGM. The intensive exploitationof Paniceae grasses and tubers for more than 10 millennia beforethe Neolithic would have helped people to develop necessaryknowledge about the properties of those plants, which eventuallyled to millet’s domestication and medicinal uses of tubers.

Materials and MethodsPVS samples were taken from different locations on the tool to documentboth used and unused surfaces. The PVS peels were examined under acompound (reflected light) Olympus microscope at magnifications of 50×,100×, 200×, and 500×. Images were collected with a Zeiss Axiocam ICc3 digitalcamera and Zeiss Axiovision software Version 4.7.

Residuesampleswereprocessed forstarchandphytolithextractionusingtheheavy liquid sodium polytungstate (in a specific gravity of 2.35). Extractionsobtained from residue samples were mounted in 50% (vol/vol) glycerol and50% (vol/vol) distilled water on glass slides and scanned under a Zeiss AxioScope A1 fitted with polarizing filters and differential interference contrast(DIC) optics. Imageswere taken using a Zeiss AxiocamMRc5 digital camera andZeiss Axiovision software Version 4.7.

Most images of usewear and starch from modern references used in thisresearch were also produced with the same microscopy to make the ancientand modern samples comparable.

ACKNOWLEDGMENTS. We thank Ofer Bar-Yosef and two reviewers for theirconstructive comments. This work was supported by the Stanford Archae-ology Center, the Australian Research Council, and the Shanxi ProvincialCultural Relics Conservation and Scientific Research Project.

1. Flannery KV (1969) Origins and ecological effects of early domestication in Iran andthe Near East. The Domestication and Exploitation of Plants and Animals, eds Ucko PJ,Dimbleby GW (Duckworth, London), pp 73–100.

2. Weiss E, Wetterstrom W, Nadel D, Bar-Yosef O (2004) The broad spectrum revisited:Evidence from plant remains. Proc Natl Acad Sci USA 101(26):9551–9555.

3. Piperno DR, Weiss E, Holst I, Nadel D (2004) Processing of wild cereal grains in theUpper Palaeolithic revealed by starch grain analysis. Nature 430(7000):670–673.

4. Revedin A, et al. (2010) Thirty thousand-year-old evidence of plant food processing.Proc Natl Acad Sci USA 107(44):18815–18819.

5. Fullagar R, Field J (1997) Pleistocene seed grinding implements from the PleistoceneAustralian arid zone. Antiquity 71(272):300–307.

6. Zhang J-F, et al. (2011) The Paleolithic site of Longwangchan in the middle YellowRiver, China: Chronology, paleoenviornment and implications. J Archaeol Sci 38(7):1537–1550.

7. Linfen Cultural Bureau of Shanxi Province (1989) Shanxi Jixian Shizitan Zhongshiqiwenhua yizhi [The Shizitan mesolithic culture site in Jixian, Shanxi]. Kaogu Xuebao[Acta Archaeologica Sinica] 3:305–323. Chinese.

8. Shizitan Archaeology Team (2002) Shanxi Jixian Shizitan jiushiqi shidai yizhi S14 di-dian [Paleolithic Locality S14 at Shizitan in Jixian, Shanxi]. Kaogu [Archaeology] 4:15–28. Chinese.

9. Shizitan Archaeology Team (2013) Shizitan yizhi dishisi didian 2002-2005 nian fajuejianbao [Preliminary report of the 2002-2005 excavations at Locality 14 of the Pa-leolithic Shizitan site in Jixian, Shanxi]. Kaogu [Archaeology] 2:3–13. Chinese.

10. Liu L, et al. (2011) Plant exploitation of the last foragers at Shizitan in the MiddleYellow River Valley China: Evidence from grinding stones. J Archaeol Sci 38(12):3524–3532.

11. Shi J (2011) Shanxi zhongyao kaogu faxian–Jixian Shizitan jiushiqi shidai yizhi [Im-portant archaeological discoveries in Shanxi–Paleolithic sites at Shizitan in Jixian].Zhongguo Wenwubao [Chinese Cultural Relics News] January 7:6–7. Chinese.

12. Song Y (2011) Shanxi Jixian Shizitan Yizhi Shiyingyan Shizhipin Yanjiu [Study on thequartzite artifacts in Shizitan Site, Jixian, Shanxi]. PhD dissertation (Chinese Academyof Sciences, Beijing). Chinese.

13. Xia Z, Chen G, Zheng G, Chen F, Han J (2002) Climate background of the evolutionfrom Paleolithic to Neolithic cultural transition during the last deglaciation in themiddle reaches of the Yellow River. Chin Sci Bull 47(1):71–75.

14. Liu L, et al. (2010) What did grinding stones grind? New light on Early Neolithicsubsistence economy in the Middle Yellow River Valley, China. Antiquity 84(325):816–833.

15. Liu L, et al. (2010) A functional analysis of grinding stones from Donghulin, northChina. J Archaeol Sci 37(10):2630–2639.

16. Dubreuil L (2004) Long-term trends in Natufian subsistence: A use-wear analysis ofground stone tools. J Archaeol Sci 31(11):1613–1629.

17. Fullagar R (2006) Residues and usewear. Archaeology in Practice: A Student Guide toArchaeological Analyses, eds Balme J, Paterson A (Blackwell, Malden, MA),pp 207–234.

18. Fullagar R (1991) The role of silica in polish formation. J Archaeol Sci 18(1):1–25.19. Field J, Fullagar R (1998) Grinding and pounding stones from Cuddie Springs and

Jinmium. A Close Look: Recent Australian Studies of Stone Tools, ArchaeologicalMethods Series 6, ed Fullagar R (Univ of Sydney, Sydney).

20. Fullagar R, et al. Stone tool-use experiments to determine the function of grindingstones and denticulate sickles. Bulletin of the Indo-Pacific Prehistory Association, inpress.

21. Fullagar R (2006) Starch on artifacts. Ancient Starch Research, eds Torrence R,Barton H (Left Coast Press, Walnut Creek, CA), pp 177–203.

22. Loy T, Fullagar R (2006) Residue extraction. Ancient Starch Research, eds Torrence R,Barton H (Left Coast Press, Walnut Creek, CA), pp 197–198.

23. Babot MdP (2003) Starch grain damage as an indicator of food processing. Phytolithand Starch Research in the Australian-Pacific-Asian Regions: The State of the Art,Terra Australis, eds Hart DM, Wallis LA (Australian National Univ, Canberra, Aus-tralia), Vol 19, pp 69–81.

24. Haslam M (2004) The decomposition of starch grains in soils: Implications for ar-chaeological residue analyses. J Archaeol Sci 31(12):1715–1734.

25. Barton H, Matthews PJ (2006) Taphonomy. Ancient Starch Research, eds Torrence R,Barton H (Left Coast Press, Walnut Creek, CA), pp 75–94.

26. Piperno DR, Ranere AJ, Holst I, Iriarte J, Dickau R (2009) Starch grain and phytolithevidence for early ninth millennium B.P. maize from the Central Balsas River Valley,Mexico. Proc Natl Acad Sci USA 106(13):5019–5024.

27. Liu T, Yue J, eds (2004) Shanxi Zhiwuzhi [Flora of Shanxi] (Chinese Science andTechnology Press, Beijing), Vol 5. Chinese.

28. Yang X, et al. (2012) From the modern to the archaeological: Starch grains frommillets and their wild relatives in China. J Archaeol Sci 39(2):247–254.

29. Ge W, Liu L, Jin Z (2010) Jizhong hebenke zhiwu dianfenli xingtai bijiao jiqi kaoguxueyiyi [Morphological analyses on starch granules of several grass species and theirsignificance in archaeology]. Quat Sci 30(2):377–384. Chinese.

30. Yang X, et al. (2012) Early millet use in northern China. Proc Natl Acad Sci USA109(10):3726–3730. Chinese.

31. Babot MdP (2006) Damage on starch from processing Andean food plants. AncientStarch Research, eds Torrence R, Barton H (Left Coast Press, Walnut Creek, CA),pp 66–70.

32. Yang X, et al. (2009) Starch grain analysis reveals function of grinding stone tools atShangzhai site, Beijing. Science in China Series D: Earth Sciences 52(8):1164–1171.

33. Crawford GW, et al. (2005) Late Neolithic plant remains from northern China: Pre-liminary results from Liangchengzhen, Shandong. Curr Anthropol 46(2):309–317.

34. Guan Y (2011) Shuidonggou Jiushiqi Shidai Wanqi Yizhi Gurenlei Zhiwu ZiyuanLiyong de Ge’an Yanjiu [The Case Study of Plant Resource Utilization at the LatePaleolithic Site of Shuidonggou, China]. PhD dissertation (Chinese Academy of Sci-ences, Beijing). Chinese.

35. Ge W (2010) Dianfenli Fenxi zai Kaoguxue zhong de Yingyong [The Application ofStarch Analysis in Archaeology]. PhD dissertation (Univ of Science and Technology ofChina, Hefei). Chinese.

36. Lu A, Chen S, eds (1986) Zhongguo Zhiwuzhi [Flora of China] (Science Press, Beijing),Vol 73.1. Chinese.

37. Pei J, Ding Z, eds (1985) Zhongguo Zhiwuzhi [Flora of China] (Science Press, Beijing),Vol 16.1. Chinese.

38. Zhu S (1406) Jiuhuang Bencao [Herbal for Relief of Famines], vol. 4, in Siku Quanshu di730 ce [Complete Library of the Four Treasuries, vol. 730]; reprinted (1983)(ShangwuPress, Taipei). Chinese.

39. Mason SLR (1996) Acornutopia? Determining the role of acorns in past human sub-sistence. Food in Antiquity, eds Wilkins J, Harvey D, Dobson M (Univ of Exeter Press,Exeter, UK), pp 12–24.

40. Crawford GW (2006) East Asian plant domestication. Archaeology of Asia, ed Stark M(Blackwell, Malden, MA), pp 77–95.

41. Zhao Z (2011) New archaeobotanic data for the study of the origins of agriculture inChina. Curr Anthropol 52(S4):S295–S306.

42. School of Archaeology and Museology of Peking University and Center for Archae-ological Study of Peking University, eds, Beijing Donghulin [Donghulin Site in Beijing](Cultural Relics Press, Beijing), in press. Chinese.

6 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1217864110 Liu et al.

Supporting InformationLiu et al. 10.1073/pnas.1217864110S14 AMS DatesThree AMS dates were obtained from charred bone samplesexcavated from the S14 site (Table S1).

Preliminary Record of Starch and Usewear SamplesPreliminary observations and basic information of the objectsstudied were documented when starch and usewear samples werecollected (Table S2).

Experimental Study of Usewear PatternsA series of experimental study using grinding stones (sandstone)to process various materials was conducted to produce useweartraces as references (Table S3), which have been used in thecurrent research (Fig. S1).

S14 Phytolith RemainsSamples from three grinding slabs and one natural sandstonewere scanned using 40× objective for phytoliths and other plantcellular remains. The results are shown in Table S4.Phytolith remains on the Shizitan grinding implements were

minimal. Charcoal and/or burnt phytoliths were present on allthree grinding implements, perhaps reflecting the location of theimplements near to a cooking and food preparation area orhearth. Aside from this, phytolith and other plant remains werepoorly preserved on grinding stone GS2, which did not exhibiteither eudicot or grass phytoliths. Multicellular eudicot phytolithswere present onGS1 andGS3, with tracheid phytoliths also presenton GS3. Eudicots represent ∼75% of the 300,000 floweringplants (ref. 1, p. 16). In this case, the eudicot phytoliths on GS1and GS3 were not diagnostic.Grass phytoliths in the form of single bulliform or long-celled

phytoliths from stems or leaves of grasses were present onGS1 and GS3. A multicellular stem or leaf grass phytolith waspresent on GS3, with a protruding plant hair, but was notdiagnostic. Grass bulliform and straight- or slightly sinuous-sided long cells are indicative of leaves or stems of grasses only.Grass husk phytoliths were not recovered. This suggests thatgrinding slabs 1–3 were not used to dehusk cereals. The ab-sence of phytoliths is consistent with the processing of plantsor items that do not produce large amounts of silicified plantcells. This includes many storage organs that contain starch,such as tubers.With regard to other plant remains, bordered pits were present

on GS3 only. Bordered pits are common in conifers (ref. 2, pp.260–261) and, as such, may indicate possible wood-working,although further research is needed to confirm this.

S14 Starch AnalysisStarch granules (136) uncovered from S14 samples were classifiedinto six types (Table S5) and compared with modern referencesfor identification.A total of nine starch granules were uncovered from the natural

rock (SS1)butnot listed inTableS5.These starchgranules, noneofwhich shows damage, can be identified as type I (8 granules) andtype IV (1 granule).

Experimental Study of Starch Size Change and DamageRatioMany studies have documented that starch granules change in sizebecause of various factors (3), such as environmental conditions(4), plant development and maturation (5, 6), milling/grindingprocesses, and heading temperatures (7–9). Milling/grinding,

which is particularly relevant to the current research, damagesthe structure of starch granules and causes morphological al-terations, such as breaking or flattening of the granules (10–12).Studies of starch granules recovered from grinding stones inarchaeological contexts also reported size changes in Paniceaeplants (ref. 13, p. 254; ref. 14, pp. 3528–3530). Because starchgranule size may be used as one of the criteria for distinguishingdomesticated millet from wild (15), we need better informationregarding the extent to which grinding may cause size changes inmillet starch. In addition, there is variation of starch damageratio among different starch types from S14, probably caused bygrinding and/or aging (16); thus, we also need to improve ourbasic understanding to the grinding-induced starch damagethrough experimental research.A preliminary experimental study of starch-size change after

grinding was conducted in 2011 and results show a significantincrease of starch granule size in a domesticated Panicum mil-iaceum sample (ref. 14, figure 4f ). We recently carried outa systematic laboratory experimental study of 14 plant samples,focusing mainly on wild species available in our referencecollections. This experiment involved two Triticeae and sevenPaniceae grasses, three Vigna beans, and one Dioscorea and oneTrichosanthes tuber (Table S6). This study was intended toachieve two objectives: first, to understand granule size changesconsequent to grinding the plants, as relevant to the starch res-idues from S14; and second, to document ratio of damagedgranules attributable to grinding those plant samples. To exam-ine the size changes in the same sample, we used a wet grindingmethod to process a very small amount of plant for each sample,and the same method was used consistently for all samples. Theprotocol of this experiment is as follows:

i) A small piece of tuber or one to three seeds were placed ina clean plastic bag, and a small amount of distilled water wasadded, soaked for several hours.

ii) Pregrinding starch extraction: tuber/seeds were gently poked/crushed with a pipette tip to release starch, 20-mL solutionwas extracted from the bag, mounted with 10% glycerol and90% distilled water (vol/vol) on a glass slide, and sealed withnail polish for microscopic examination and recording.

iii) Postgrinding starch extraction: a wooden roller was used topress the bag containing the remaining tuber/seeds until thesubstance was flattened and starch released; 20-mL solutionwas extracted from the bag, mounted with 10% glycerol and90% distilled water (vol/vol) on a glass slide, and sealed withnail polish for microscopic examination and recording.

iv) Granule size comparison: the maximum lengths (through thehilum) of 100–180 starch granules were measured for eachsample, the measurements of the pre- and postgrinding sam-ples from the same specimen were entered into the JMPsoftware for comparison.

v) Damaged starch ratio: the percentage of damaged starchwas recorded for each sample by counting granules show-ing characteristics of damage (e.g., broken surface, pro-nounced lamellae, deep and wide fissures, a large centraldepression, an enlarged extinction cross) against all visiblegranules (120–200) within randomly selected fields underthe microscope.

The results of the experiment show that starch granules fromthe Paniceae samples (including genera of Echinochloa, Panicum,and Setaria) exhibit the greatest increase in size after grinding;the maximum lengths increased by 27–56% and average lengths

Liu et al. www.pnas.org/cgi/content/short/1217864110 1 of 6

by 3–38%. In contrast, the seven samples from Triticeae grasses,Vigna sp. beans, and the Dioscorea opposita and Trichosantheskirilowii tubers show 1–29% increase in maximum length and0–21% increase in average length (Table S6).The ground Paniceae and Triticeae grass samples also show

higher percentages in damaged granules (30–63% and 10–21%,respectively) than those from beans and tubers (4–15%) (TableS7). These ratios seem to be similar to those from S14 ancientsamples, in which type I (Triticeae) and type III (Paniceae) starches

show 74% and 72% of damages, respectively, whereas other typesonly exhibit 0–35% of damages.These results suggest thatunder thesimilar conditionsofgrinding,

the ground Paniceae starches are more likely to be damaged andshow greater change in size compared with other plants examined.Therefore, some of the starches from S14 type III samples, many ofwhich are large in size and appear to be damaged due to grinding,may have derived fromwild Paniceae grasses, including Setaria spp.,which have smaller starch sizes when undamaged.

1. Piperno DR (2006) Phytoliths: A Comprehensive Guide for Archaeologists andPaeoecologists (Altamira Press, Lanham, MD).

2. Evert RF (2006) Esau’s Plant Anatomy: Meristems, Cells, and Tissues of the Plant Body:Their Structure, Function, and Development (John Wiley and Sons, Hoboken, NJ).

3. Lindeboom N, Chang PR, Tyler RT (2004) Analytical, biochemical and physicochemicalaspects of starch granule size, with emphasis on small granule starches: A review.Starch/Stärke 56(3–4):89–99.

4. Haase NU, Plate J (1996) Properties of potato starch in relation to varieties andenvironmental factors. Starch/Stärke 48(5):167–171.

5. Karlsson R, Olered R, Eliasson A-C (1983) Changes in starch granule size distributionand starch gelatinization properties during development of maturation of wheat,barley and rye. Starch/Stärke 35(10):335–340.

6. Moorthy SN, Ramanujam T (1986) Variation in properties of starch in cassava varietiesin relation to age of the crop. Starch/Stärke 38(2):58–61.

7. Choi H, Kim W, Shin M (2004) Properties of Korean amaranth starch compared towaxy millet and waxy sorghum starches. Starch/Stärke 56(10):469–477.

8. Muralikrishna G, Malleshi NG, Desikachar HSR, Tharanathan RN (1986) Effect ofpopping on the properties of some millet starches. Starch/Stärke 38(2):48–51.

9. Choi S-G, Kerr WL (2004) Swelling characteristics of native and chemically modified wheatstarches as a function of heating temperature and time. Starch/Stärke 56(5):181–189.

10. Rakszegi M, et al. (2010) Effect of milling on the starch properties of winter wheatgenotypes. Starch/Stärke 62(2):115–122.

11. Babot MdP (2006) Damage on starch from processing Andean food plants. Ancient StarchResearch, eds Torrence R, Barton H (Left Coast Press, Walnut Creek, CA), pp 66–70.

12. Babot MdP (2003) Starch grain damage as an indicator of food processing. Phytolithand Starch Research in the Australian-Pacific-Asian Regions: The State of the Art,Terra Australis , eds Hart DM, Wallis LA (Australian National Univ, Canberra, Aus-tralia), Vol 19, pp 69–81.

13. Yang X, et al. (2012) From the modern to the archaeological: Starch grains frommillets and their wild relatives in China. J Archaeol Sci 39(2):247–254.

14. Liu L, et al. (2011) Plant exploitation of the last foragers at Shizitan in theMiddle YellowRiver Valley China: Evidence from grinding stones. J Archaeol Sci 38(12):3524–3532.

15. Yang X, et al. (2012) Early millet use in northern China. Proc Natl Acad Sci USA109(10):3726–3730.

16. Collins MJ, Copeland L (2011) Ancient starch: Cooked or just old? Proc Natl Acad SciUSA 108(22):E145.

1

3 4

5 6

2

Fig. S1. Usewear patterns on standstone slabs after grinding various materials in our experimental study. (Scale bar: 50 μm.) (1) Abrading stone tool, 8 h,showing reticulated long, parallel furrows and deep gouges, sharp edges on crystal grains. (2 and 3) Grinding dehusked wheat and millet, respectively, 2.5 h,showing medium to low level polish with few visible striations, rounded edges. (4) Grinding dry mungbean, 30 min, medium level polish with long, fine, andslightly crooked striations, rounded edges. (5) grinding dry yam, 1 h, showing medium level polish, rounded edges. (6) Grinding dry root of snakegourd, 1 h,showing medium level polish with very fine parallel striations, rounded edges.

Liu et al. www.pnas.org/cgi/content/short/1217864110 2 of 6

Table S1. AMS dates from S14

Sample no. Material Stratum 14C date Cal. B.P. (68.2%) Cal. B.P. (95.4%)

BA101589 Charred bone II 15,030 ± 150 18,358–18,051 18,611–17,901BA01158 Charred bone III 17,210 ± 290 20,700–20,000 21,150–19,550BA101591 Charred bone IV 19,050 ± 80 22,699–22,434 23,021–22,353

Cal. B.P., calendar years before present

Table S2. Comparative record of Shizitan S14 grinding stones and a natural rock

Artifact ObservationSize, cm; andweight, kg Lithology Sampling

GS1 no. 2468stratum III

Washed; top surface is slightly concave,showing pecking under low magnification;the bottom side is flat with white residuesediments; the tool edge shows unifacialflaking

Length: 13 Fine-grained purplesandstone

One water sample from top; one watersample from bottom; one sedimentsample from side (0.03 g); two PVSsamples from top and bottom

Width: 9.2Thickness: 3Weight: 0.74

GS2 no. 3592-2stratum IV

Washed; nearly rectangular in shape; topsurface is used, concave, and rough;bottom side is weathered severely; it isbroken into three pieces and gluedtogether after excavation

Length: 13.3 Fine-grained graysandstone

Two water samples from top; one watersample from bottom; one sedimentsample from back and side (0.01 g);three PVS samples from top

Width: 8.4Thickness: 1.5Weight: 0.25

GS3 no. 3592-1stratum IV

Unwashed; irregularly rectangular in shape;top surface is concave and used; bottomside is weathered

Length: 40 Fine-grained graysandstone

Two water samples from top; one watersample from bottom; two PVS samplesfrom top

Width: 28Thickness: 4Weight: 5.64

SS1 near the site Unwashed natural rock; unused Fine-grained graysandstone

Two water samples; one PVS samplefrom each side.

PVS, polyvinyl siloxane.

Liu et al. www.pnas.org/cgi/content/short/1217864110 3 of 6

Table S3. Experimental study of usewear patterns on grinding stones (sandstone) after working on different materials

Image inFig. S1

Material, action(sample no.) Polish Striation Pitting

Edges ofgrain Microtopography

1 Stone abrading 8 h(Exp10 slab)

Medium-high-level polish,highly reticulated

Parallel furrows anddeep gouges

Absent Sharp Leveled plateau

Shell abrading, 20 min(T005 slab)

Medium-level polish onhigh points, reticulated

Long parallel furrows,gouged

Absent Sharp Leveled plateau

Fresh bone abrading10 min (T005 slab)

Medium-high-level polishon high points

Parallel furrows Absent Sharp Leveled plateau

Hematite pounding20 min (T008 pestle)

Absent Deep gouges Absent Rounded Convex plateau

Bamboo abrading 2 h(T005 slab)

Medium-level polish onhigh points, reticulated

Long, parallel, wide,and deep sleeks

Absent Slightlyrounded

Leveled plateau

Soft wood abrading 2 h(T005 slab)

High-level polish onhighest points,

Parallel, wide andshallow sleeks

Absent Veryrounded

Convex plateau

Dry acorn poundingand grinding 1 h(T901 handstone)

Medium-level polish onhigh points

Some fine striationson both high andlower areas

Present, varioussizes on highareas

Rounded Uneven surfaceon plateau

Wet acorn with shells,pounding and grinding20 min (XS14-G slab)

Medium-level polish onhigh point

Few visible striations Present, varioussizes on highareas

Rounded Convex andleveled plateau

Oats with husks grinding20 min (XS10 slab)

Medium-level polish onhigh areas, reticulated

Clear and parallel sleeks Absent Rounded Leveled plateau

2 Dehusked wheat grinding2.5 h (EGS1W slab)

Medium-level polish onhigh points

Some fine striations,but rare

Absent Rounded Leveled plateau

3 Dehusked millet abrading2.5 h (T901 handstone)

Low-level polish on highpoints

Some fine striations,but rare

Absent Rounded Leveled plateau

4 Dry mungbean, grinding30 min (T018-04A slab)

Medium-level polish onboth high and low areas

Fine and long striationsrunning in variousdirections on highand lower areas

Absent Rounded Leveled plateau

Wet mungbean, grinding30 min (T010-06A slab)

Medium-level polish onhigh areas

Some fine and shortstriations on high areas

Absent Rounded Leveled plateau

5 Dry yam, grinding 1 h(EGS1-SY slab)

Medium-level polish onsome high areas

Absent Absent Rounded Slightly leveledplateau

6 Dry snakegourd root,grinding 1 h(EGS1-T slab)

Medium-level polish,reticulated

Very fine, long, andparallel striationson a few areas

Absent Rounded Leveled plateau

Dry kudzu-vine root,grinding 1 h(EGS1-GG slab)

Medium- and high-levelpolish, reticulated

Very fine, long, andparallel striationson a few areas

Absent Rounded Leveled plateau

Table S4. Phytolith remains on samples from S14

Type of residues GS1 GS2 GS3 SS1 (natural stone) Total

Charcoal Present Present Present Present PresentBrown/burnt phytoliths Present Present Present Present PresentEudicot phytoliths: multicellular 3 2 1 6Eudicot: platy 2 2cf. eudicot phytolith 2 2Tracheid: multicellular 1 1Grass phytoliths: bulliform 1 2 3Grass phytoliths: long cell 1 2 3Grass phytoliths: multicellular 1 3 4Bordered pits Present Present

Liu et al. www.pnas.org/cgi/content/short/1217864110 4 of 6

Table S5. S14 starch types and counts.

Type I Type II Type III Type IV Type V Type VI Unidentified Total Damaged

GS1-1 (front) 1 1GS1-2 (back) 3 2 5 1GS1-3 (side sediment) 0GS2-1 (front) 0GS2-2 (front) 3 1 4 1GS2-3 (back) 0GS3-1 (front) 33 18 13 10 3 9 13 99 60GS3-2 (front) 1 1 1 13 1 2 19 1GS3-3 (back) 5 2 1 8 2GS3-4 (side and back sediment) 0Total 45 20 18 24 4 11 14 136 65Percentage 33 15 13 18 3 8 10 48Damaged 33 7 13 1 0 1 10 65Damaged (%) 74 35 72 4 0 9 71 48

Table S6. Starch size change

N Min Median MaxIncrease inmax (%) Mean

Increase inmean (%) SD Provenance

Triticeae grassesAgropyron cristatum Grain 167 4.92 20.73 41.08 20.41 8.01 Inner Mongolia

Ground 177 5.04 20.12 45.05 10 20.95 3 8.07Leymus secalinus Grain 155 3.00 11.36 19.37 10.76 4.21 Inner Mongolia

Ground 154 2.12 11.33 22.02 14 11.48 7 4.73Paniceae grasses/milletEchinochloa crusgalli Grain 145 2.90 6.96 9.91 6.87 1.26 Japan

Ground 123 5.03 7.96 12.62 27 8.13 18 1.64Panicum miliaceum Grain 129 4.62 7.6 10.15 7.54 1.04 Beijing

Ground 153 4.46 7.67 13.65 34 7.79 3 1.66Setaria italica ssp. italica (d.) Grain 112 4.04 9.46 16.85 9.59 2.32 Shaanxi

Ground 157 3.84 11.13 26.30 56 11.52 20 2.77S. italica ssp. viridis Grain 128 3.35 5.72 9.84 5.89 1.20 Beijing

Ground 179 4.04 7.06 14.15 44 7.39 25 1.91S. italica ssp. viridis Grain 155 4.10 8.08 14.39 8.21 1.89 Zhengzhou Henan

Ground 130 4.86 9.24 20.08 40 10.01 11 2.74S. italica ssp. viridis Grain 137 3.45 7.37 10.41 7.26 1.40 Inner Mongolia

Ground 144 5.05 8.66 14.81 42 8.79 21 1.89Setaria faberi Grain 139 2.06 6.06 10.83 6.07 1.70 Beijing

Ground 142 4.74 7.9 15.58 44 8.37 38 2.06Vigna beansVigna radiata Grain 110 5.25 15.64 25.87 15.50 4.97 Yanshi, Henan

Ground 142 6.03 17.46 33.30 29 18.26 18 5.10Vigna unguiculata (d.) Grain 161 5.29 19.77 38.39 20.00 6.83 Yichuan, Shaanxi

Ground 162 8.55 19.61 39.65 2 20.01 0 5.72Vigna angularis (d.) Grain 132 9.53 26.07 71.8 29.89 14.59 China

Ground 169 12.04 34.87 72.44 1 36.12 21 11.88TubersT. kirilowii Whole 136 4.18 15.03 31.17 15.26 5.03 Yanshi, Henan

Ground 148 5.13 15.78 36.99 19 16.10 6 5.90D. opposita Whole 134 15.84 42.60 70.62 42.95 12.14 Yanshi, Henan

Ground 134 15.02 45.68 83.84 19 43.82 2 12.39

Note: all samples listed are wild forms unless indicated as domesticated (d.). Max, maximum; min, minimum.

Liu et al. www.pnas.org/cgi/content/short/1217864110 5 of 6

Table S7. Ratios of damaged granules in ground plant samples

Total granulescounted

Damagedgranules

Damaged(%)

Triticeae grasses

A. cristatum 143 14 10L. secalinus 140 29 21

Paniceae millets/grassesE. crusgalli 176 84 48P. miliaceum 204 80 39S. italica ssp. italica(dom)

128 38 30

S. italica ssp. viridis 170 97 57S. italica ssp. viridis 178 113 63S. italica ssp. viridis 148 87 59S. faberi 154 74 48

Vigna beansV. radiata 142 13 9V. unguiculata (dom.) 146 10 7V. angularis (dom.) 208 9 4

TubersT. kirilowii 137 21 15D. opposita 136 14 10

Note: all samples listed are wild forms unless indicated as domesticated(dom). Scientific names for foxtail millet (S. italica ssp. italica) and greenfoxtail grass (S. italica ssp. viridis) are derived from Germplasm ResourcesInformation Network (GRIN) Taxonomy for Plants (www.ars-grin.gov/cgi-bin/npgs/html/index.pl), which are different from those (S. italica and Setaria viridis)listed in Flora of China (www.efloras.org).

Liu et al. www.pnas.org/cgi/content/short/1217864110 6 of 6