Reference materials for this study includes eighteen species from thirteen genera and seven major plant families having regional and global dietary significance (Table 1). Not only have these taxa been documented in ethnographic literature, the majority have also been recovered from archaeological sites across western North America. Some of these materials were collected from preserved herbarium specimens while others were collected from live plants in the field. To capture variation within a species, starch granules were extracted from three geographically dispersed individuals (replicate samples) and described using an array of morphometric characteristics (e.g., size, shape, surface features, response to polarized light). Source and collection data for can be found in Supplementary Table S1. Plant nomenclature follows USDA GRIN http://www.ars-grin.gov/∼sbmljw/johnindex.html.

Different plant parts that most likely contain starch (seeds, fruits, nuts, endosperm, bulbs, tubers, caryopses, and taproots) were processed according to standard protocols (e.g., Torrence and Barton 2006) in the Natural History Museum of Utah (NHMU) Archaeobotany Lab. Material was ground using a sterile mortar and pestle and was sieved through a 125 µm mesh Endecott screen into a beaker using DH₂0. Sample material <125 µm was transferred to a sterile 50 ml test tube and each sample was centrifuged for 3 minutes at 3000 RPM. The supernatant was discarded, and the sample pellet was transferred to a sterile 15 ml test tube. Each test tube was re-suspended with a vortex mixer, adding 7 ml of lithium heteropolytungstate (LST; specific gravity 2.00–2.35), and then centrifuged for 15 min at 1000 RPM. The sample was extracted from the heavy liquid using a pipette, carefully removing the top 1–2 mm layer of organics containing starch and placed into new 15 ml test tubes. Each sample was rinsed three times until all residual heavy liquid was removed. Samples were then rinsed with acetone, mixed with a vortex, and centrifuged for 3 min at 3000 RPM. The acetone was decanted, and samples were covered and left to dry overnight. Once dried, the samples were mixed with 50% DH₂0 and 50% glycerol solution and then mounted on microscope slides.

Each slide was scanned using a transmitted brightfield microscope fitted with polarizing filters and Nomarski optics (Zeiss Axioscope 2, Zeiss International, Göttingen, Germany). A digital camera (Zeiss HRc) with imaging and measurement software (Zeiss Zen) were used to capture images of and measure starch granules. For each reference sample, randomly generated X-Y coordinates were used to measure and photograph approximately 100 granules from each individual plant (n=∼300 from each species). All starch granules present in these photographs were measured and examined for morphological characteristics.

Under ×400 magnification, the sizes, shapes, surface features and responses to polarized light of the starch granules were observed. Once identified and photographed, starch granules were described according to an established set of structural and surface characteristics (Reichert, 1913; ICSN, 2011) (Table 2). The size of each starch granule was measured as the maximum length through the hilum. Granules were also examined for the presence of morphological characteristics including 2-D shape, hilum position, central cavity, extinction cross, fissures, lamellae, pressure facets, and depressions along the margins (Reichert, 1913; ICSN, 2011). Starch granule size tends to be non-normally distributed and, therefore, relying on mean granule size is not appropriate for identification purposes (Louderback et al., 2017). Boxplots showing the distribution of starch granule size for each species was generated using the boxplot function in the “graphics” package for R v. 3.6.2 (Chambers et al., 1983; Becker et al., 1988; Murrell, 2005; R Core Team, 2019). To overlay individual granule size data, we used the jitter method within the stripchart function in the “graphics” package for R v.3.6.2.

Size distributions among all possible pairs of the study species were statistically compared using the non-parametric Kolmogorov-Smirnov (K-S) test. Applying a statistical analysis to starch granule identification promotes reproducibility and an overall increase in confidence (Louderback et al., 2017; Gao et al., 2021). It is becoming standard practice to measure the size, shape and morphological characteristics of reference starch granules from multiple populations so that variation and statistical significance can be assessed (Liu et al., 2014a; Brown and Louderback, 2020; Wilks et al., 2021).

Starch granule measurements from individual species were pooled for overall size and morphological analyses (Supplementary Data S1). Smaller granules tend to exhibit few diagnostic characteristics, therefore, relying on the top 20% granule lengths allowed for greater differentiation between taxa because morphological characteristics occur more frequently in larger granules (Liu et al., 2014a; Louderback et al., 2017). Granule size distributions of all plant taxa are plotted in Figure 1. All comparisons were significantly different (p < 0.05), except between Cymopterus bulbosus and Lomatium triternatum (p = 0.08), Lomatium donnellii and Quercus agrifolia (p = 0.70), Lomatium donnellii and Quercus douglasii (p = 0.35), Quercus douglasii and Quercus agrifolia (p = 0.28), and Calochortus nuttallii and Erythronium grandiflorum (p = 0.38).

Morphological characteristics, including 2D shape, hilum position, central cavity, extinction cross, fissures, lamellae, pressure facets, and depressions were documented for each granule for all 18 taxa (Supplementary Data S1). The frequency of those characteristics occurring on granules was calculated and expressed as a number between 0.0 and 1.0. (Figures 2A,B). Definitions for these characteristics were compiled primarily from ICSN (2011) and Reichert (1913), but some definitions were refined for this particular study based on previous work (Piperno et al., 2004, 2009; Holst et al., 2007; Perry and Quigg 2011; Musaubach et al., 2013; Yang and Perry 2013; Louderback et al., 2017; Brown and Louderback 2020; Joyce et al., 2021). All characteristics described above occur more frequently in the top 20% size range of starch granules and, therefore, we report frequencies for those granules (Figures 2A,B).

A dichotomous key for identifying starch granules of the seven plant families analyzed in this study was generated (Box 1). Because it is based only on a few species from each family, the key will undoubtedly be revised as more families, genera, and species are systematically analyzed using the characteristics presented herein. Furthermore, new techniques will lead to new diagnostic characteristics resulting in further revision and greater application of new diagnostic keys. BOX 1

Different characteristics dominate the array of plant families that we have examined (Figure 3). For example, Amaranthaceae produces miniscule granules (<5 µm) with centric hila and a circular/oval shape. Because they are so small, it is difficult to visually discern additional morphological characteristics. Furthermore, Amaranthaceae starch are often formed in dense clusters of amyloplasts (sheets) bounded by cell walls, so it is rare to see isolated granules (even in reference material, Figure 3) (Reichert, 1913; Louderback, 2014; Capparelli et al., 2015; López et al., 2015). Apiaceae also produces starch granules with centric hila, circular/oval shape, and are frequently observed with transverse or stellate fissures, visible lamellae, and pressure facets (Herzog, 2014; Joyce et al., 2021). Another family that produces circular/oval granules with centric hila is Pinaceae. What distinguishes these granules from those in other families is the obvious presence of a round central cavity (Tinsley et al., 2021). Fagaceae, on the other hand, have starch granules with slightly eccentric hila and are characterized by triangular-trapezoidal shape with confused crosses, mesial longitudinal clefts, and visible lamellae (Messner 2011; Liu et al., 2014b; Brown and Louderback 2020). Liliaceae and Solanaceae produce starch granules with clearly eccentric hila, longitudinal fissures, and visible lamellae, but Liliaceae granules tend to be more circular/oval while Solanaceae are almost always elongated (Messner 2011; Louderback et al., 2017; Ahituv and Henry 2022).

Of the plant families examined by this study, Poaceae exhibits the most variation in starch granule size, shape, surface structures, and responses to polarized light. For example, Achnatherum hymenoides and Sporobolus airoides produce very small granules (<10 µm) with centric hila, but their shape is often angular, thus distinguishing them from Amaranthaceae. Elymus elymoides (and most Triticeae grasses—barley, wheat, rye) produce distinctive starch granules that can be quite large (up to ∼30 µm), are circular/oval with oblong central cavities, indistinct crosses, and very visible lamellae. Additional characteristics (e.g., pits) have also been observed on these granules (Perry and Quigg, 2011; Yang and Perry 2013; Brown and Louderback, 2020). Finally, Zea mays is another grass genus that produces unique granules. These have a significantly different size range from Elymus, Achnatherum, and Sporobolus (Figure 1) and they also have an irregular shape with transeverse/stellate fissures that are not observed in other Poaceae taxa (Musaubach et al., 2013; Wilks et al., 2021).