New evidence reveals dispersal of pearl millet from West Africa to South Asia by 2500 BCE

Abstract

Pearl millet (Pennisetum glaucum (L.) R. Br. syn. Cenchrus americanus (L.) Morrone) is the most widely cultivated millet in the world. It is a major source of food and fodder in Africa and South Asia, where it has been grown since prehistoric times. Genetic studies trace the domestication of this important crop to the western Sahel in Africa by 2800 BCE, although the earliest evidence of domestication traits have been dated to c. 2500 BCE in pottery temper. In South Asia, securely dated finds appear around 1900 BCE. Here, we report two absolute dates of pearl millet caryopses to 2578–2358 and 2576–2356 cal BCE from Bhando Qubo, an archaeological site in the lower Indus Valley (Sindh, Pakistan). These caryopses have proven to be contemporary to the oldest evidence of domesticated pearl millet in Africa and represent the earliest Pennisetum glaucum grains directly dated by ¹⁴C AMS worldwide. This discovery indicates that pearl millet expanded into South Asia at least five centuries earlier than previously thought and highlights the Indus Valley as a crucial corridor for prehistoric crop diffusion.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-20110-w.

Introduction

The importance of millets as nutrition-rich and climate-resilient crops is increasingly recognized by international development agencies¹. Their tolerance to limited rainfall and poor soil fertility and their short growing cycles² makes them reliable food resources in drylands. In the semi-arid regions of South Asia, pearl millet is currently a major source of food and fodder³. However, its arrival, role and significance in the agricultural practices of early societies in the subcontinent remain to be fully understood and clarified.

Pearl millet has recently been treated as Cenchrus americanus (L.) Morrone, following the merger of Pennisetum into Cenchrus^4,5. However, the name Pennisetum glaucum (L.) R. Br. is much more widely used in the botanical, agronomic, and archaeobotanical literature, and remains supported by long-standing nomenclatural discussions^6,7. To avoid confusion, we here use Pennisetum glaucum, while noting its current treatment as Cenchrus americanus in some taxonomic works.

The domestication of pearl millet has been traced to the western Sahel from the wild ancestor Pennisetum glaucum subsp. monodii (Maire) Sosef. and its earliest evidence as domesticate is reported from pottery tempers at the site of Er Negf in the Lower Tilemsi Valley (northeastern Mali). The temper was dated by ¹⁴C AMS to 2579–2369 cal BCE (3980 ± 31 BP; OxA-X-2287–27)^8,9. Additional investigations on the evolution of the flowering pathway of this crop support its domestication in West Africa around 2800 BCE¹⁰. Based on this chronology, genetic studies estimate the expansion of cultivated pearl millet eastwards by 2900 BCE¹¹. After 2200 BCE, pearl millet agriculture appears to expand quickly in more southern savannah regions¹².

Evidence suggests that the cultivation of wild pearl millet may have begun as early as 4500 BCE in the Araouane region of Mali, as indicated by the identification of wild Pennisetum chaff used as ceramic temper at the AZ22 site¹³. This hypothesis is reinforced by the presence of charred wild Pennisetum grains found at multiple sites across the central and western Sahara during the Early to Middle Holocene¹³. Further evidence supporting the exploitation and consumption of C₄ plants and possibly domesticated pearl millet by the mid-3^rd millennium BCE comes from isotopic analyses on the enamel bioapatite of human remains from Karkarichinkat Nord (KN05) in Mali¹⁴. While an earlier absolute date for a single Pennisetum grain morphologically indeterminate was reported at KN05, dated to 2621–2464 BCE (4011 ± 33; OxA16919)⁸, pottery impressions provide a terminus post quem for the cultivation of domesticated pearl millet at the site c. 2000 BCE^8,14.

The introduction of pearl millet into South Asia is suggested to have taken place via maritime connections within the Red Sea and the Indian Ocean trade networks through intermediary contacts in Arabia¹⁵. Pearl millet is documented in Gujarat (Northwest India) by at least 1700 BCE (Late Harappan horizon)^8,16,17 and possibly 2000 BCE^16,17. From Gujarat, the crop appears to have spread southward into peninsular India, where grains have been dated to 1600–1500 BCE¹⁸. Less secure archaeobotanical finds contextually associated with earlier chronologies are also reported from Hallur, in the Southern Neolithic Phase II of India (2200–1800 BCE)¹⁷ and in few Indus Valley Civilisation sites in Gujarat and Haryana (Northwest India): Kanmer (2600–1300 BCE)^19–21, Rangpur (1900–1300 BCE)²² and Khirsara (2100–1900 BCE)^23,24. However, most of these reports lack photographic documentation and explicit identification criteria, and none of the pearl millet grains have been directly dated using ¹⁴C AMS (Supplementary 1 Table 1).

Our findings from the Lower Indus Plain archaeological site of Bhando Qubo (Sindh, Pakistan), a Late-Early (Transitional) to Mature Harappan mid-sized settlement discovered by Shaik and Veesar²⁵ and excavated in 2019 by the ModAgrO Project of the Spanish Archaeological Mission in Sindh (Fig. 1, Supplementary 2), revealed the presence of the oldest securely dated pearl millet caryopses in South Asia and, at present, worldwide. These findings suggest that pearl millet expanded into this region at least half a millennium earlier than previously acknowledged. This highlights the Indus Valley as a crucial corridor for prehistoric crop diffusion and marks the long-term importance of pearl millet in drylands.

Fig. 1.

Archaeological sites reporting pearl millet remains (fully domesticated or in the process of domestication, sensu Fuller et al.13) with radiocarbon dates earlier than 1500 BCE. The plotted values represent the medians of the modelled dates for the sites included in the Bayesian analysis (see Supplementary 1 Table 2). For the sites MT25, MK36, and Kassala we used the medians of the unmodelled dates, since these determinations were excluded from the modelling (see Methods). Map produced with QGIS 3.40 using data interpolated with the Inverse Distance Weighting (IDW) method. This map is offered under the Creative Commons Attribution 4.0 International License (CC-BY 4.0) and is derived from public domain data produced by the U.S. Geological Survey.

Results

Stratigraphic and chronological sequence

Bhando Qubo has a stratigraphic sequence of 7 m with 25 archaeological layers. These layers have been grouped into 12 stratigraphic units (SU) based on the identified sedimentary processes (see Supplementary 2 Fig. 1 and Table 1). Despite the absence of clear domestic architectonical structures, the presence of well-executed floors, fireplaces and grinding stones throughout most of the sequence suggests that the part of the settlement excavated was associated with domestic activities.

Based on pottery typologies, SUs 2–4 are chrono-culturally associated with the Mature Harappan phase (2600–2100 BCE), while SUs 5–12 correspond to the Transitional-late Early Harappan phase (2800–2600 BCE)²⁵ (A. Uesugi, personal communication, 2024).

Macroremains

A total of 20L of sediment were floated for each excavated layer and context (see Methods). Three pearl millet (Pennisetum glaucum (L.) R. Br.) caryopses, PM ARCH 1, PM ARCH 2 and PM ARCH 3, were identified in layer 4B-2 (SU3) (ubiquity 3,2%) based on a combination of anatomical and morphometrical characters (Fig. 2, Supplementary 3 and 4). Layer 4B is part of a sequence of floors, within which 4B-2 corresponds to an ash accumulation, possibly a fireplace (Supplementary 2 Table 1 and Figs. 2, 3). PM ARCH 2 was directly radiocarbon dated to 2578–2358 cal BCE (CIRAM-8334, 95.4% or 2 σ) while PM ARCH 3, gave a date of 2576–2356 cal BCE (CIRAM-8335, 95.4% or 2 σ) (Fig. 2, Supplementary 1 Table 2).

Fig. 2.

Comparison between charred pearl millet caryopses recovered from the site and modern reference specimens. Left: (a, b, c) Charred pearl millet (Pennisetum glaucum (L.) R. Br.) and (d, e) Pennisetum sp. grains from Bhando Qubo, layer 4B-2; Right: Modern pearl millet caryopses from Pakistan, UPF Laboratory for Environmental Archaeology seeds reference collection—ref. BGP 1377. Scale: 1 mm.

Fig. 3.

Earliest reported radiocarbon dates associated with fully domesticated pearl millet caryopses. Green dates correspond to the chronological distribution of West African sites, while red dates correspond to the chronological distribution of South Asian sites.

A further two caryopses were found in layer 4B-2 (PSP ARCH 4 and PSP ARCH 5), which were identified to genus level as Pennisetum sp. (Fig. 2, Supplementary 4), as well as four undetermined panicoid grains resembling Pennisetum. Other millets with a historical record for cultivation²⁶ were identified at Bhando Qubo and they belong to the genera Brachiaria, Echinochloa and Setaria. These grains are present in 52% of the layers/contexts, while 76 undetermined panicoid grasses were found in 29% of the contexts.

Radiocarbon dates modelling

The sequential model produced an agreement index of 126.3, with all analyzed dates showing individual agreement values ranging from 72.1 to 105.2 (see Supplementary 1 Table 2). The modelled date from Karkarichinkat Nord (KN05) (Mali) indicates a chronological range of 2617–2461 BCE (95.4% or 2 σ), while the grains from Bhando Qubo (CIRAM-8334 and CIRAM 8335) show ranges of 2576–2356 BCE and 2575–2354 BCE (95.4% or 2 σ), respectively (see Fig. 3, Supplementary 1 Table 2). These results suggest that, although the Pennisetum sp. grain (which is lacking fully domesticated traits) recovered from Karkarichinkat Nord (KN05) represents the earliest direct evidence of the domestication process, this evidence precedes that of fully domesticated Pennisetum glaucum in South Asia by only a period of approximately 50 to 100 years, according to the chronological ranges of modelled dates (see Supplementary 1 Table 2). So far the earliest fully domesticated grain of Pennisetum glaucum with absolute radiocarbon dating in West Africa has been identified at Birimi²⁷. The modelled dating for Birimi provides a temporal range of 2340–1421 BCE (95.4% or 2σ) giving a median of the modelled date of 1819 BCE. The unmodelled date from Birimi shows a temporal range of 2402–1300 BCE (95.4% or 2σ) where the median value is 1800 BCE (see Supplementary 1 Table 2). The broad temporal range of both the modelled and unmodelled dates for Birimi makes it difficult to consider these as reliable chronological references for the first appearance of fully domesticated pearl millet in West Africa. Nevertheless, considering the medians of the modelled and unmodelled dates from Birimi, the spread of fully domesticated pearl millet may have occurred at some point around 2200–2000 BCE, suggesting that, over a period of approximately 300–500 years, this crop would have developed fully domesticated traits and spread throughout West Africa¹².

The presence of pearl millet at other sites in South Asia has been dated, contextually and not directly, later than 2500 BCE. For example, at Kanmer it has been recovered from a context directly dated on a barley grain (PLD-16352 / 2461–2290 BCE (95.4% or 2 σ modelled date)). Thus, the arrival of Pennisetum glaucum in the Indus Valley Civilisation settlements of the Lower Indus Plain (Sindh, Pakistan) would have occurred about a century earlier than in Gujarat and c. five to six centuries earlier than previously estimated.

Discussion

According to eFlora of Pakistan²⁸, there are five wild Pennisetum species present in the country: P. divisum (J.F. Gmel.) Verloove, Govaerts & Buttler, P. flaccidum (Griseb.) Roshev., P. hohenackeri Hochst. ex Steud., P. lanatum Klotzsch, and P. orientale Rich., as well as one cultivated species (P. glaucum). Both P. flaccidum and P. lanatum are reported as fodder grasses. The wild ancestor of pearl millet (P. glaucum subsp. monodii) is not present in Pakistan. The absence of this wild ancestor in the area suggests that this crop was introduced from Africa in its already domesticated form.

Based on an exhaustive literature review of all the published radiocarbon dates of pearl millet from Africa and Asia, the caryopses from Bhando Qubo are contemporary to the oldest secure evidence of domesticated pearl millet in Africa, the pottery temper from Er Negf (3980 ± 31 BP; OxA-X-2287–27). These caryopses also represent the earliest domesticated Pennisetum glaucum grains directly dated by ¹⁴C AMS (Fig. 3). The contemporaneity of the domesticated pearl millet remains from Bhando Qubo (Sindh, Pakistan) and Er Negf (Mali), together with the absence of evidence of domesticated pearl millet between West Africa and Asia pre 1500 BCE, seems to confirm not only an exceptionally rapid dispersal rate as already proposed by some authors^13,29,30 but also the need to redefine the domestication timeline of pearl millet in Africa and its early diffusion into South Asia. In this regard, the study of microbotanical remains has proven to be highly effective for the identification of economically important plants that tend to be less evident in the macrobotanical record (i.e. millets), or in contexts where preservation is hindered^31,32. Additional multi-proxy analyses (including starch and phytolith) are in progress and they will be able to enhance our understanding of early agricultural exchanges between Africa and South Asia.

Once domesticated, pearl millet was incorporated into pastoral systems, enhancing its potential for widespread dispersal³⁰. The crop appears to have spread eastwards across the Sahara, reaching Sudan by c. 1850 BCE³³ –but possibly a few centuries earlier− ³⁴ and India by c. 1700 BCE¹⁶. Pearl millet is believed to have reached India directly from Africa by sea, given the absence of archaeological finds in the Nile Valley and the Near East³⁵. The early presence of pearl millet in the Lower Indus Plain may suggest a key role of this geographical area in the diffusion of African domesticates. In this respect, Sauer’s hypothesis of “a lost corridor of mankind” between the coasts of India, Arabia and East Africa¹⁶ should be further explored taking into consideration also the Indus River and the Makran coast. Based on the new finding from Bhando Qubo, we argue that not only is an alternative diffusion pathway of African millets plausible, but also that the Indus plain might have been the first diffusion corridor for some African crops. These crops initially arrived on the coast of Sindh, where the Indus River, its delta, and the Rann of Kutch played a key role in their spread north into the Indus plains, east into Gujarat, and eventually to the rest of the Indian subcontinent.

The radiocarbon date for Er Negf (3980 ± 31 BP; OxA-X-2287–27) in the Lower Tilemsi Valley (northeastern Mali) has been considered too old due to potential contamination from older carbon adhering to the temper used for dating⁸, while the date from Karkarichinkat Nord (KN05) (4011 ± 33 BP / 2621–2464 BCE) was obtained from a Pennisetum sp. grain which does not exhibit clear domestication traits⁸.

Other radiocarbon dates in West Africa on fully domesticated pearl millet caryopses or their securely associated archaeological contexts older than 1500 BCE have been obtained from the sites of Windé Koroji complex³⁶, Birimi²⁷, Ounjougou³⁷, and Kachama³⁸. In South Asia, the earliest dates in archaeological contexts where pearl millet has been identified come from Kanmer and Khirsara^{19–21,39,40}. However, these dates were not carried out directly on the pearl millet but on wheat/barley grains from the same layers.

The pearl millet caryopses from Bhando Qubo represent the earliest direct evidence of fully domesticated pearl millet (P. glaucum) with absolute dating in South Asia. These new dates suggest an out-of-Africa spread of this domesticated crop from somewhere in West Africa prior to 2500 BCE, as further supported by genetic studies^10,11.

Methods

Macrobotanical remains

Macrobotanical remains were collected using a bucket flotation system with a manual water pump. The recovery procedure was adapted from UPF Laboratory of Environmental Archaeology’s protocol (Supplementary 5).

A total of 20L of sediment were floated for each excavated layer and context. Light fraction was separated at the UPF Laboratory for Environmental Archaeology using a sieve stack of 2, 1 and 0.5 mm meshes. Sorting was carried out using a Leica EZ4W microscope. Heavy fractions were processed with the naked eye at the Archaeology and Anthropology Museum of Shah Abdul Latif University (Sindh, Pakistan).

Identification was carried out using published botanical literature^17,41,42 and modern reference collections at the UPF Laboratory of Environmental Archaeology and the Archaeobotany Lab, Seoul National University (Supplementary 3 and 4).

Radiocarbon calibration and Bayesian modelling

Twelve radiocarbon dates have been modelled. The selected dates correspond to archaeological contexts in which pearl millet grains have been identified as fully domesticated. An exception was made for Karkarichinkat North, which corresponds to a radiocarbon date on a pearl millet grain that was possibly domesticated, and may represent one of the earliest pieces of evidence for the domestication of this crop⁸. All the dates used in this study fall within a temporal range of ~ 2500–1500 BCE, in order to chronologically adjust the origin and expansion of the crop from the presumed place of origin to its spread into South Asia.

Radiocarbon dates obtained from pottery temper have been discarded (MT25, MK36, Er Negf), as they may yield, in some cases, artificially older results due to the possible inclusion of exogenous carbon in the sample (originating from the clay or other organic materials), producing dates that are not contemporary with human occupation^8,13. Evidence from Kassala and Shaqadud Cave has also been excluded, as it does not correspond to fully domesticated pearl millet, but rather to Pennisetum cf. glaucum. All radiocarbon dates presented in the main text of this paper and the Bayesian age model were processed using OxCal v.4.4 software⁴³ and calibrated with the INTCAL20 curve⁴⁴.

Radiocarbon dates have been processed using a sequence model in which all dates are assigned to a single phase^45,46. This assumes that the dates do not directly correspond to the domestication process itself, but rather reflect the distribution of direct dates on fully domesticated caryopses.

The charcoal sample (Windé Koroji—GX-19990) has been considered an outlier. Outliers are presumed to be older than the formation of the archaeological deposit from which they were recovered, thus preceding the layer they date (terminus ante quem)⁴⁶. To address these assumptions, the radiocarbon determinations on charcoal were treated as being older than the deposit, following the Charcoal t-type Outlier model.

The radiocarbon dates obtained from grains have been considered outliers. A 5% (0.05) prior likelihood of being outliers was assigned to these samples within the General t-type Outlier Model^45,46. Therefore, it has been assumed that the grains may not directly correspond to the specific event being dated, as the formation of multiple archaeological surfaces involves continuous remodeling and successive anthropogenic actions, creating cumulative palimpsests⁴⁷. In this context, the radiocarbon dates obtained may not necessarily reflect the activity directly associated with the grain, but rather the presence of the grain in the archaeological record. Thus, the focus is on dating the grain itself, irrespective of the human activity that may have occurred in the same context.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

C.J.A., M.MA., C.L. and O.P., conceived and designed the study. M.MA. led the fieldwork and sampling strategy, C.J.A. and C.L performed flotation, and all authors except M.MO. participated in the excavation. C.J.A. carried out the macrobotanical sampling and identification. O.P. performed the radiocarbon modelling. M.MO. designed the maps. C.J.A and O.P. drafted the manuscript, and all other authors contributed to the final version. M.MA. obtained funding.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Competing interests

The authors declare no competing interests.