Phytolith analysis for differentiating between broomcorn millet (Panicum miliaceum) and its weed/feral type (Panicum ruderale)

Domestication of broomcorn millet (Panicum miliaceum) is one of the most significant events in prehistoric East Asia, providing sufficient food supply for the explosive growth of Neolithic populations and the transition into complex societies. However, to date, the process of broomcorn millet domestication is still largely unknown, partly due to the lack of clear diagnostic tools for distinguishing between millet and its related wild grasses in archaeological samples. Here, we examined the percentage of silicified epidermal long-cell undulated patterns in the glume and palea from inflorescence bracts in 21 modern varieties of broomcorn millet and 12 weed/feral-type Panicum ruderale collected across northern China. Our results show that the percentage of ηIII patterns in domesticated broomcorn millet (23.0% ± 5.9%; n = 63) is about 10% higher than in P. ruderale (10.8% ± 5.8%; n = 36), with quartiles of 17.2–28.3% and 5.1–15.5%, respectively. Owing to the increase in ηIII pattern percentage correlates significantly with a decrease in the grain length/width ratio, in the absence of exact wild ancestors of broomcorn millet, the characterization of phytolith differences between P. ruderale and P. miliaceum thus becomes an alternative approach to provide insight into origin of broomcorn millet.

features are often lost when seeds are oxidized into granules or ash, or broken into pieces in early Holocene archaeobotanical assemblages 8,9 . Since phytoliths are silica casts of plant cells created within and between cells of living plants tissues that can remain in sediments long after the living tissue has decayed, phytolith analysis makes the identification of decayed plant remains from archaeological samples possible and exact, i.e., rice, maize, foxtail millet and broomcorn millet can be identified according to shape, size, and other anatomical features [10][11][12][13] . In recent years, distinguishing wild and domesticated species in phytoliths occurring within archaeological residues has become a topic of great scientific interest 7 . However, owing to the highly similar phytolith shape within a given genus, it is not adequate to use morphological features alone to distinguish crop plants from their wild ancestors. The development of phytolith morphometry has enabled such distinctions to be made, and morphometry-based approaches are increasingly used to reveal crop domestication processes. These approaches include characterizing the fish-like scales of bulliform phytoliths of rice and the inflorescence phytoliths of millet and wheat [14][15][16][17][18][19][20] .
In the present study, we investigated 21 landraces of modern broomcorn millet and 12 specimens of P. ruderale from eight provinces cross north China in order to determine whether analyzing the phytoliths of inflorescence bracts can be used as an effective tool for discriminating broomcorn millet from its possible wild ancestor or weed/feral type, P. ruderale. Our findings provide a potential method to investigate the process of broomcorn millet domestication.

Materials and Methods
P. ruderale features dark grey or dark brown pericarps, stands 40 to 100 cm tall, has a large number of branches, and forms a loose panicle, whereas P. miliaceum (broomcorn millet) has white, yellow, brown or red pericarps (accounting for over 90% of the total), stands 100 to 150 cm tall, with close, loose, or lateral panicles and less branches 5,21 (Fig. 1). In this study, we selected 12 modern P. ruderale, all of which have dark pericarp coloring, and 21 landraces of modern P. miliaceum with light pericarps. For P. ruderale, nos. 1 to 6 were collected in 2012 and 2017 from field investigations conducted near roadsides and abandoned fields in Yangyuan County, Hebei Province, northern China, and nos. 7 to 12 were collected from Institute of Crop Sciences, Chinese Academy of Agricultural Sciences (ICSCAAS). Landraces of modern P. miliaceum were collected from ICSCAAS (nos. 13 to 20) and the Institute of Geology and Geophysics, Chinese Academy of Sciences (IGGCAS; nos. 21 to 33). Notably, nos. 21 and 28 were collected in the same location in Yangyuan County where P. ruderale specimens were collected. For detailed descriptions of the plants, see Table 1 and Fig. 2. For comparison of wild relatives of P. ruderale, we collected one sample of Panicum repens from IGGCAS.
For phytolith statistics, it is essential that all samples are subjected to identical physical treatment; thus, we applied the same method to all samples. We took 10 seeds from each plant specimen, and (i) cleaned the whole seeds with distilled water to remove adhering particles, (ii) placed all samples in 20 ml of saturated nitric acid and placed in a boiling water bath for 50 min to oxidize organic materials completely, (iii) added distilled water and centrifuged at 2500 rpm. for 5 min, (iv) decanted and rinsed with absolute ethanol and centrifuged at 2500 rpm. for 5 min (twice), and (v) transferred the resulting phytoliths to storage vials. The residual subsamples were mounted onto microscopic slides in Canada Balsam medium for observation.
The number of ηI, II, and III patterns were counted three times for each plant specimen. For the first count, we randomly selected 20 pieces of phytolith regardless of the size of the piece; for the second count, we selected 20 pieces of phytolith between 125 μm and 375 μm regardless of how many single undulated patterns occurred in one piece; for the final count, we selected 20 phytolith pieces sized not only between 125 μm and 375 μm, but also containing more than 30 single undulated patterns in each piece. For each count in each specimen, we calculated the percentage of ηIII from more than 1,000 single undulated patterns. Finally, we calculated 36 total values (12 specimens ×3 counts) representing the percentage of ηIII in P. ruderale, and 63 (21 specimens ×3 counts) values in domesticated millet, by counting over 160,000 single undulating patterns.
Phytoliths were observed by phase-contrast light microscopy at 400× magnification. The identification of ηI, ηII, and ηIII was aided by reference materials described in Lu et al. 13 . To unify the identification criteria, we further specified the undulating patterns of ηI, ηII, and ηIII based on reference materials in Lu et al. 13 (Fig. 3). As Fig. 3 shows, according to the numbers of sub-branches along the two sides of main branches, we classify three sub-types as follows: ηI exhibits only one main branch or the top part of the main branch is symmetrically expanded; the ηII subtype excludes the top-expanded branches, and if present, only one sub-branch occurs in one side of the main branch; ηIII have more than one sub-branch in one side of the main branch -usually two or three. Notably, the epidermal long cells within the joint part were avoided when counting the undulating patterns, because the undulating patterns around the joint part were too sinuous to allow correct identification.
Morphometric measurements were performed on an average of 60 grains for each sample. To minimize measurement error, we placed the grains on a soft salt layer in the culture dish, ensuring that the widest profile of the grains faced upward. Specifically, grain width (W) and length (L) were measured. The ratio of L/W was calculated. Descriptive statistics and Pearson tests on the quantitative data were performed. Figure 4 shows the identical features of undulating patterns of epidermal long cells present in the upper lemma and palea in both domesticated broomcorn millet and P. ruderale; they all exhibit undulating patterns of ηI, ηII and ηIII across the husk as defined by Lu et al. 13 . The undulations tend to increase in highly sinuous variation towards the central part of the lemma and palea. The different η-undulated regular patterns in general vary gradually across different parts from the base and top (η I), to the side (ηII), and to the center (η III) of the lemma and palea (Fig. 4).

Results
We statistically analyzed the percentage of ηIII in all counted undulated patterns (ηI, ηII, and ηIII). The percentage of ηIII in both domesticated millet and P. ruderale in all three counts exhibit normal distributions  (Fig. 5). The mean values of the three counts ranged from 9.5% to 12.2% in P. ruderale and 22.8% to 23.3% in domesticated broomcorn millet, indicating that the distributions of ηIII percentage data for P. ruderale (Fig. 5, W1-3) and domesticated broomcorn millet (Fig. 5, D1-3) were relatively concentrated within each group regardless of counting method and phytolith size. The mean values of all three counts were 10.8% ± 5.8% (SD; n = 36) and 23.0% ± 5.9% (SD) (n = 63), respectively. Notably, the highest and lowest quartiles among the three counts of P. ruderale and domesticated broomcorn millet were 5.1-15.5% and 17.2-28.3% without any overlap. The results show that ηIII percentages in domesticated broomcorn millet are markedly higher than those from P. ruderale, suggesting that the undulated patterns of epidermal long cells in the upper lemma and palea could be used to distinguish between the two species. The L/W ratio describes the shape, especially the roundness, of grains. The mean L/W of P. ruderale ranges from 1.44 to 1.78, whereas the ratio of domesticated millet ranges from 1.23 to 1.44 (Fig. 6). This shows that grains of domesticated broomcorn millet have rounder (i.e., fatter and wider) shape on average than P. ruderale.   Table 2), indicating that the ηIII percentage increases with a corresponding decrease in the L/W ratio (Fig. 7).
The percentage of ηIII in samples grown in regions of similar climate have further been compared. The percentage of ηIII in P. ruderale range from 7.0% to 14.0%, while in P. miliaceum, percentages range from 20.7% to 30.1%. This shows that regardless of climate region, the percentages of ηIII in P. ruderale remain markedly lower than those from domesticated landraces. Particularly, six P. ruderale and two P. miliaceum samples grown in the same location in Yangyuan County, Hebei Province, exhibited large differences in the mean percentage of ηIII -7.0% ± 2.1% and 21.2% ± 5.4%, respectively (Fig. 8) -despite the fact that they were grown under the same environmental conditions.

Discussion
Plant domestication is an evolutionary process that involves a series of profound morphologic and genetic changes [22][23][24][25] . Changes in grain size, usually manifesting as a reduction in L/W ratio and greater grain width, are a well-recognized aspect of domestication syndrome in cereals and other seed crops 23,26,27 . Our results show domesticated millet grains have a lower L/W ratio (fatter grains) than that of P. ruderale, while the percentage of   ηIII from domesticated millet is higher than from P. ruderale. The high correlation between the L/W ratio and the mean value of ηIII percentage suggests that the change in L/W is one of the main factors accounting for the changes in ηIII percentage in epidermal long cells. We speculate that this is because the domestication process enlarges the millet grain width, thus increasing the area of the central part of the husk and inducing the epidermal long cells within the lemma and palea to expand accordingly, resulting in a higher proportion of ηIII. Phytoliths in the lemma and palea are replicas of epidermal long cells, and their size and undulated patterns change according to the size and shape of seeds. Compared to the top and base of the husk, the central part of the husk is more expanded and rounded. Accordingly, the undulated patterns have the most sinuous type and the largest size, such as ηIII and ΩIII types, as shown in common millet, foxtail millet and green foxtail 13,15 . As Figs 3 and 4 illustrate, seeds of P. ruderale are not only smaller and thinner than those of domesticated millets, indicating the central part of the husk of P. ruderale is less expanded, but also significantly correlate with the proportion of ηIII type in the lemma and palea, thus causing the percentage of ηIII in P. ruderale to be lower than that in the domesticated type.
P. ruderale has been considered either the wild ancestor or a wild/feral type derived from back-mutation from the domesticated broomcorn millet [28][29][30] . Generally, feral derivatives of crop varieties may show a similar phenotype to that of the crop ancestor, e.g., thinner and smaller seeds, and shattering 30,31 . Thus, our quantitative data    suggest that the direct ancestor of P. miliaceum may have a lower proportion of ηIII type in the lemma and palea. Owing to the significant correlation between phytolith pattern and seed shape mentioned above, in the absence of exact wild ancestors of broomcorn millet, the characterization of phytolith differences between P. ruderale and P. miliaceum thus becomes an alternative approach to provide insight into origin of broomcorn millet that has long been ignored and unexplored.
Analysis of various heritability studies indicates that grain size is affected by both genetic and environmental factors, but that the contribution of heritable genetic traits is greater than the variation attributable to environmental differences 32 . In this paper, we compared the percentage of ηIII from the samples in different originated provinces from semi-humid to arid regions of north China. As Fig. 8 shows, irrespective of climatic region, the percentages of ηIII of P. ruderale are always lower than those from domesticated millet. In particular, the percentages of ηIII from P. ruderale and P. miliaceum samples grown in the same location of Yangyuan County are significantly different and can be distinguished within error range. These quantitative data confirm that environmental conditions may have less effect than genetic variation. Combined with Fig. 4, our data suggest that the percentages of ηIII are not associated with environmental conditions, but significantly correlate with the L/W ratio of grains. This indicates that the percentage of ηIII is a useful parameter facilitating the differentiation of P. ruderale and P. miliaceum.
Former phylogenetic study indicated that the allotetraploid origin of P. miliaceum, with the maternal ancestor being P. capillare (or a close relative) and the other genome being shared with P. repens 33 . Thus, in order to develop a framework to prevent the misidentification of P. ruderale, P. miliaceum, and P. repens, we further examined the silicon structure patterns in the lemma and palea from the inflorescence bracts in modern P. repens. The results show that the η type of epidermal long cells also occur in P. repens, which is identical with P. ruderale and P. miliaceum, whereas phytoliths exhibit the 'wave-type' morphology in the junction of the epidermal long cellthis is different from the 'finger-type' occurring in the other two species 13 (Fig. 9). This morphological comparison among the three species indicates that the phenotype of epidermal long cells in lemma and palea in similar in P. ruderale and P. miliaceum, and suggests that the combination of η type of epidermal long cells plus 'finger-type' morphology the junctions of epidermal long cell may be peculiar to P. ruderale and P. miliaceum. However, owing to the absence of P. capillare, which is a native plant to most of North America 34 , we only compared the epidermal long cells from P. repens, and additional investigation of Panicum species are needed to confirm the observations.
In conclusion, regardless of climate region, the percentage of ηIII patterns in P. miliaceum is markedly higher than the one from P. ruderale. Because of the significant negative correlation between ηIII pattern percentage and the ratio of grain length/width, in the absence of exact direct wild ancestors of broomcorn millet, the characterization of phytolith differences between P. ruderale and P. miliaceum thus becomes an alternative approach to provide insight into origin of broomcorn millet.

Data Availability
The datasets generated during the current study are available from the corresponding author on reasonable request.