Crop wild relatives of the United States require urgent conservation action

The predicted distributions of the assessed US native CWR ranged from northern Alaska through southern Mexico and the Caribbean and out to Hawaii (Fig. 1). Taxonomic richness across all modeled taxa was concentrated in parts of the Northeast and Midwest, the Pacific Northwest and California, the Mountain West and Southwest, and the Gulf Coast region of the Southeast, with the predicted ranges of up to 91 taxa overlapping in the same ∼5 km² areas.

The assessed crop progenitors and closest wild relatives (1A taxa) displayed the highest predicted richness in the Northeast, Midwest, and Pacific Northwest, with up to 53 taxa overlapping in the same areas (SI Appendix, Fig. S1). Distant relatives (1B) were more evenly dispersed, with concentrations in the Mountain West, Southwest, Pacific Northwest and California, and Midwest and Northeast, with up to 32 taxa overlapping in the same areas. Those taxa with undetermined relation status (1C), which are likely in general to be distant relatives, had the highest predicted richness in the Southwest and Midwest, with up to 29 taxa overlapping. Richness patterns also varied by associated crop and crop type. For instance, the predicted ranges of wild relatives of cereals were concentrated in the western United States; fruits in the temperate regions of the Northeast and Pacific Northwest; vegetables in the Mountain West, Pacific Northwest, and Midwest; nuts in the eastern United States; sugar crops in the Southeast; and pulses, roots and tubers, and fiber crops in the Southwest desert borderlands (SI Appendix, Fig. S2). Interactive predicted distribution maps for each assessed taxon are provided in Dataset S1 (23).

Preliminary threat assessments, based on extent of occurrence (EOO) and area of occupancy (AOO) analyses (SI Appendix, Table S3) (24), identified 42 taxa (7.1%) as candidates for designation as critically endangered (CR) in their natural habitats, 297 (50%) as endangered (EN), 166 (28%) as vulnerable (VU), 66 (11.1%) as near threatened (NT), and the remaining 23 (3.9%) as of least concern (LC) (SI Appendix, Table S4). AOO was the primary determinant of these designations. Of the 1A taxa, 16 (6.3%) may be considered CR, 121 (47.8%) as EN, 71 (28.1%) as VU, 37 (14.6%) as NT, and 8 (3.2%) as LC.

With regard to ex situ conservation, more than 400 gene banks and botanical gardens worldwide safeguard one or more of the assessed US native CWR. The US Department of Agriculture (USDA) Agricultural Research Service, National Plant Germplasm System; national gene banks in India, Australia, Mexico, Morocco, Brazil, Bulgaria, Canada, Ecuador, Germany, the United Kingdom, Japan, and the Russian Federation; and international agricultural research institutes, including the International Potato Center and the International Center for Tropical Agriculture, hold the greatest numbers of accessions, with 35.9% of the total accessions maintained in the USDA’s system and 79.5% maintained in these repositories collectively (Dataset S2) (23).

Comparing the diversity conserved in these ex situ collections to the predicted native ranges of the plants, we found the great majority of taxa to be significantly underrepresented ex situ. Eighty-three taxa (14% of the total) were entirely absent from the available germplasm and botanical garden databases, and an additional 196 taxa (33%) had fewer than 10 accessions in conservation repositories, thus offering relatively limited genetic variation for research and education (SI Appendix, Table S4). A total of 454 taxa (76.4%) were assessed as urgent priority for further collecting to address gaps in ex situ conservation, with an additional 100 (16.8%) considered high priority, 33 (5.6%) considered medium priority, and only 7 (1.2%) considered low priority (Fig. 2). The mean final ex situ conservation score (FCSex) across taxa was only 13.9 (median, 5.7) on a conservation status scale of 0 (very poor) to 100 (comprehensive), with metrics ranging from 0 to 92.8. The ecological representation of taxa—based on an analysis of the proportion of potentially inhabited ecoregions (25) in which samples have been collected for conservation repositories—was considerably higher than geographic representation, with a mean ecological representativeness score (ERSex) of 24.9 (median, 7.7), compared with 9.7 (median, 0.5) for the geographic score (GRSex). Of the 1A taxa, 173 (68.4%) were identified as urgent priority for further collecting, 53 (21%) as high priority, 20 (7.9%) as medium priority, and only 7 (2.8%) as low priority.

Visual depictions of the predicted distribution and representation in conservation repositories of little sunflower (Helianthus pumilus Nutt.), a CWR of cultivated sunflower native to Colorado and Wyoming, are provided in Fig. 3 A and B as an example of taxon-level results. The species’ ex situ conservation occurrences are well distributed throughout its predicted range (GRSex of 83.2) and provide representation from all five of the ecoregions that it potentially inhabits (ERSex of 100), leaving collecting gaps mainly in the northernmost and other outlying areas of its predicted native range. Two additional taxon-level examples are offered in SI Appendix, Figs. S3 and S4, and individual results for all assessed taxa, including interactive maps, are provided in Dataset S1 (23).

The major geographic and ecological gaps in ex situ conservation of these wild taxa indicate the need for extensive further collecting throughout most of their predicted ranges. Spatial priorities for collecting thus largely mirror patterns of taxonomic richness, with uncollected populations of up to 89 taxa potentially found in the same ∼5 km² areas (SI Appendix, Figs. S5 and S6).

Based on their predicted distributions, we found that more than 32,000 protected land areas listed in the World Database of Protected Areas (WDPA) (26) and located in the United States potentially harbor the assessed taxa (SI Appendix, Table S5). Of these, protected areas in the Northeast and East (especially Delaware, Maryland, New Jersey, New York, and Pennsylvania, and Midwest (Illinois and Missouri) potentially provide protection to the greatest numbers of taxa, with a maximum of 89 predicted to occur in a 5-km² area within protected lands. Based directly on occurrence data rather than on modeled distributions, assessed taxa have been sampled from more than 3,800 protected areas in the United States. In this group, the most taxon-rich areas included the Patuxent Research Refuge as well as the Grand Canyon, Kings Canyon, Olympic, Mount Rainier, Indiana Dunes, Gulf Islands, Yellowstone, Rocky Mountain, and other national parks, shores, and wilderness areas.

Despite the large number of protected areas potentially harboring these CWR, the known occurrences and predicted ranges of most of the species were generally poorly represented in protected lands, with a mean final in situ conservation score (FCSin) across all taxa of 35.9 (median, 36.9) (SI Appendix, Table S4). Forty-three taxa (7.2%) were found to have no overlap with protected areas. In total, 66 taxa (11.1%) were designated as urgent priority, 487 (82%) as high priority, 34 (5.7%) as medium priority, and only 7 (1.2%) as low priority for further habitat protection (Fig. 2). Even more pronounced than in the ex situ analysis, ecological representativeness (ERSin) regarding habitat protection (mean, 89.8; median, 95.5) was much higher than geographic (GRSin) (mean, 11.1; median, 8.2), including 245 (41.3%) of the taxa potentially fully represented in protected areas in terms of the diversity of inhabited ecoregions. Of the 1A taxa, 31 (12.3%) were designated as urgent priority, 204 (80.6%) as high priority, 16 (6.3%) as medium priority, and only 2 (0.8%) as low priority for further habitat protection.

Representation of little sunflower in protected areas based on its predicted distribution, as well as the gaps in its potential habitat protection, are depicted in Fig 3C as an example of taxon-level results. The taxon was modeled as occurring in wilderness and other protected areas along its north-south gradient in the Rocky Mountains. These protected lands collectively occupy a relatively small portion of the species’ predicted range (GRSin of 5.3) but are fairly well distributed and thus represent all five of the ecoregions that it potentially inhabits (ERSin of 100). The most obvious in situ conservation gaps in its predicted range occur in its northern extents in Wyoming and in eastern lower elevation areas. SI Appendix, Figs. S3 and S4 provides additional taxon-level examples, and Dataset S1 (23) provides complete results, including interactive maps, for all taxa.

The most efficient establishment of additional in situ protection for the maximum number of US native CWR, based on predicted distributions falling outside current WDPA-designated protected areas, would focus on the Northeast and Midwest, the West Coast, and parts of the Mountain West and Southeast (SI Appendix, Figs. S7 and S8). Unprotected populations of up to 91 taxa for geographic gaps and up to 33 taxa for ecological gaps could potentially be found in the same ∼5-km² area.

With regard to their combined ex situ and in situ conservation status, which represents an average of the results of the two conservation strategy assessments, individual determinations for taxa ranged from no protection at all (22 taxa; 3.7% of the total, including CWR of beans, blackberries and raspberries, blueberries, grapes, pecans, strawberries, and sunflowers, many of which are natural hybrid taxa) to a moderate level of conservation. For example, the final combined conservation score (FCSc-mean) was 69.3 on a scale of 0 to 100 for Oregon endemic wild strawberry Fragaria cascadensis K. E. Hummer, 62.9 for a Hawaiian blueberry (Vaccinium reticulatum Sm.), 61.8 for a Santa Cruz Island gooseberry (Ribes thacherianum [Jeps.] Munz) endemic to the Channel Islands, and 58.1 for a recently described wild sunflower (Helianthus winteri J. C. Stebbins) with a narrow range in central California (SI Appendix, Table S4). The FCSc-mean averaged across all taxa was 24.9 (median 23.1). Based on the average of their ex situ and in situ conservation status, 349 taxa (58.8%) were identified as urgent priority for further action, 220 (37%) as high priority, 25 (4.2%) as medium priority, and none as low priority (Fig. 2).

Of the 1A taxa, 135 (53.4%) were classified as urgent priority for further action based on combined ex situ and in situ conservation status, 101 (39.9%) as high priority, and 17 (6.7%) as medium priority, with an average FCSc-mean of 26.8 (median, 24.2) (SI Appendix, Fig. S9). Regarding associated crop types, US native cereal, fruit, nut, root and tuber, sugar, and vegetable CWR had the largest proportions of taxa determined to be urgent priority for further action (SI Appendix, Fig. S10). Regarding associated crops and wild food plants, those of avocado, chestnut, citrus, melon, pecan, potato bean, sugar maple, sugarcane, vanilla, and wildrice demonstrated the most urgent priorities on average across taxa, whereas relatives of beans, cherimoya, echinacea, sunflower, and zucchini were of somewhat lesser immediate concern (Fig. 4). Comparing the preliminary threat assessment results with the combined conservation gap analysis showed that the most threatened taxa (assessed as CR or EN) were also generally those with the most urgent priorities for conservation action (SI Appendix, Fig. S11).

Further conservation action for US native CWR is clearly needed, both to safeguard their diversity in ex situ repositories and to facilitate their continued evolution in their natural habitats. Among the taxa assessed to be of urgent conservation priority are wild genetic resources of cereal, fiber, fruit, nut, oil, pulse, root and tuber, spice, sugar, and vegetable crops that collectively generate more than $116 billion in annual US agricultural production value (27). In sunflower alone, whose CWR are exclusively native to North America, the direct annual economic benefits derived from use of the wild taxa have been estimated at $267 to 384 million (28). Here we discuss the critical steps needed to enhance conservation and facilitate use of these cultural-genetic-natural resources, including conducting further field exploration and validation, strengthening collaborative conservation, characterizing and facilitating access to the plants, and raising awareness about their existence, value, and plight.

Species distribution modeling and model-based conservation biogeography are increasingly critical to conservation planning (29), particularly for large-scale prioritization analyses such as this national study, given the increasing numbers of threatened species and decreasing numbers of field botanists (30). Occurrence, ecogeographic predictor, and conservation data deficiencies, as well as modeling method limitations, make field validation of modeling results an essential step before conservation action. (An extended discussion of data and modeling challenges is provided in SI Appendix, Materials and Methods.) Engagement of volunteer botanists, local botanical societies and gardens, students, and other citizen science stakeholders through collaborative initiatives with backstopping from species experts represents a promising approach to accomplishing the discovery, verification, monitoring, and collection of native CWR populations (31). Meanwhile, further investment by biodiversity, geospatial, and conservation information providers in making these data as complete, correct, and accessible as possible, including incorporating new data from emerging fieldwork, will only improve the potential of conservation biogeographic analyses.

Given the diversity of US native CWR prioritized for action, ambitious collaborative conservation efforts are needed among gene banks, botanical gardens, community conservation initiatives, and organizations focused on habitat conservation (32). Botanical gardens, employing an extensive network of conservation botanists and managing >80,000 acres in North America (33), offer unique opportunities to complement public gene banks in mobilizing field collecting activities and protecting native CWR that do not store well in freezers, in vitro, or in liquid nitrogen, as well as for long-lived, large plants whose propagules may be more easily distributed from adult individuals. Hobby gardeners and other citizen conservationists could be further engaged to curate CWR, especially those with edible fruit, ornamental value, or other attractive traits. Meanwhile, a primary emphasis on in situ conservation is required for taxa that are difficult to maintain outside of their specific natural habitats, such as the federally listed endangered Texas wildrice (Zizania texana Hitchc.).

While extensive field verification, population monitoring, and management planning are needed before a comprehensive national assessment of the in situ status of these taxa is complete, our analyses indicate that expansion of habitat protection in the country, especially within richness hotspots, is needed to safeguard the evolutionary potential of various native CWR over the long term. While the widening of current protected area boundaries or the establishment of new protected spaces may be necessary to accomplish these aims, the challenges to their implementation owing to cost and competing land uses indicate that enhancement of protection on existing open spaces—whether officially protected areas or other effective area-based conservation lands (34)—may be the most feasible approach. Assisted migration of populations into suitable habitats within conservation areas may also be considered. Greater awareness of native CWR by land managers is needed, as the plants are sometimes viewed as weeds or nuisances and may be mistaken for invasive species (18).

Given that the primary justification for conservation of these plants is their usefulness to people—both as genetic resources for research and as direct contributors to human diets and culture—it stands to reason that greater awareness of and access to native CWR for use, and more extensive characterizations of their traits, should bolster support for their protection. In turn, conservation of these plants in their natural habitats will also help safeguard ecosystems, other species, and holobionts, providing additional known as well as currently unrecognized benefits to society.

Thus, wider awareness of and access to CWR must be integrated into their conservation. For research and education, access via ex situ conservation repositories is related to the degree to which samples are available and have been characterized for their known or potential traits of value. Both aspects can be strengthened through the enhancement of existing online information and ordering systems, including better integration of these platforms so that the overall diversity of taxa across ex situ repositories can be more easily explored. The limited characterization and evaluation of these plants for projected needs represents a major bottleneck in current use of the taxa (1). Access for study of these plants in their natural habitats is also needed (8).

As many native CWR are culturally significant, providing food, spice, and other values to wild harvesters and their communities and markets, ensuring continued access to these resources from their traditionally harvested places is necessary. Consideration and participation in conservation planning of communities who use these plants are essential, given that they are important influencers of the viability of CWR populations and that their exclusion in the name of conservation has been shown to damage this diversity (35, 36).

Achieving CWR's full potential as conservation champions will require greater public awareness of these plants. While all involved organizations will need to enhance their public outreach around native CWR, botanical gardens, which receive more than 120 million visitors a year in the United States (33), could play a particularly pivotal role in introducing these species to people, communicating their value and plight, and better connecting the concepts of food security, agricultural livelihoods, and services provided by nature for the public (37).

We developed a current national inventory of CWR of the United States by verifying taxonomic names (38) listed in a published baseline (11), updating gene pool assignments (39), and ensuring inclusion of all target taxa occurring in the country and its territories (SI Appendix, Table S1). We categorized taxa based on relative crossability with and phylogenetic relation to associated crops, as well as on occurrence status, with category 1A comprising native close relatives of globally important agricultural crops (including the taxa listed as primary or secondary relatives or used as root/graft stock), as well as important wild food plants (11, 39). Category 1B included distant (tertiary) native relatives of these crops, while 1C included any other taxa in the same genera but with undetermined relationships. In total, 594 taxa are reported in the main results, including 253 in category 1A, 188 in 1B, and 153 in 1C (SI Appendix, Tables S2 and S4).

Occurrence data were compiled from biodiversity and conservation repository databases (40–46) and recent literature (17, 18, 47). Identifiable duplicates and nonwild records were removed, and taxonomic names were standardized (38). We classified each record as an existing ex situ accession (labeled “G,” because most records were from gene banks) or as a reference observation (labeled “H,” because most records were from herbaria) (10). Occurrences were clipped to their native states/provinces (38) within the United States, Canada, and Mexico, as well as to US territories and Caribbean islands. We compiled and processed a total of 834,673 occurrence records for the 594 target taxa, including 32,786 G and 801,887 H (SI Appendix, Table S4). Of these, 276,312 (8092 G and 268,220 H) had coordinates located in the target native areas of the taxa and were used for distribution modeling and spatial conservation analyses. The final occurrence dataset is available in Dataset S2 (23).

We produced species distribution models with the MaxEnt algorithm (48) using 26 bioclimatic and topographic predictors (SI Appendix, Table S6) (49, 50) at a spatial resolution of 2.5 arc-min, using a subset of variables (51) (SI Appendix, Table S7) as well as the number and location of pseudoabsences specific to each taxon. Median models across MaxEnt replicates were evaluated using the area under the receiver operating characteristic curve (AUC), the SD of the AUC across replicates, and the proportion of the potential distribution model with an SD of the replicates >0.15 (17) (SI Appendix, Table S4). Interactive models and evaluation metrics for each taxon are available in Dataset S1 (23).

For the preliminary threat assessment, we calculated the EOO and AOO of each taxon (SI Appendix, Table S3), adapted from the International Union for Conservation of Nature (IUCN) Red List criteria (AOO cell size 4 km²) (24) and run through the R package “Redlistr” (52). Taxa were classified as CR when EOO <100 km² or AOO <10 km², as EN when 100 km² < EOO < 5,000 km² or 10 km² < AOO < 500 km², as VU when 5,000 km² < EOO < 20,000 km² or 500 km² < AOO < 2,000 km², as NT when 20,000 km² > EOO < 45,000 km² or 2,000 km² < AOO < 4,500 km², and as LC when EOO ≥45,000 km² and AOO ≥4,500 km².

We assessed the degree of representation of each taxon in ex situ and in situ conservation systems, with four scores calculated for each conservation strategy (SI Appendix, Table S3). All scores were bound between 0 and 100, with 0 representing an extremely poor state of conservation and 100 representing comprehensive protection (10, 17). The sampling representativeness score ex situ (SRSex) calculates the ratio of germplasm accessions (G) available in ex situ repositories to reference (H) records for each taxon, making use of all compiled records irrespective of whether they include coordinates. The GRSex uses 50-km-radius buffers created around each G collection coordinate point to estimate geographic areas already well collected within the distribution models of each taxon and then calculates the proportion of the distribution model covered by these buffers. The ERSex calculates the proportion of terrestrial ecoregions (25) represented within the G buffered areas out of the total number of ecoregions occupied by the distribution model. The FCSex was derived by calculating the average of the three ex situ conservation metrics.

In situ conservation was analyzed based on extent of taxon range representation within protected areas listed in the WDPA (26). The sampling representativeness score in situ (SRSin) calculates the proportion of all occurrences of a taxon within its native range that fall within a protected area. The GRSin compares the area (in km²) of the distribution model located within protected areas versus the total area of the model. The ERSin calculates the proportion of ecoregions encompassed within the range of the taxon located inside protected areas to the ecoregions encompassed within the total area of the distribution model. The FCSin was derived by calculating the average of the three in situ conservation metrics.

The FCSc-mean was calculated for each taxon by averaging its final FCSex and FCSin scores. Taxa were then categorized with regard to the two conservation strategies as well as in combination, with UP for further conservation action assigned when FCS <25, HP assigned when 25 ≤ FCS < 50, MP when 50 ≤ FCS < 75, and LP when FCS ≥75. An extended description of methods and materials, including references and links to the ecogeographic and spatial input data and code; the US inventory; occurrence data; and further results, including interactive taxon-level models and conservation metrics, are provided in the SI Appendix.

We thank the botanists, taxonomists, plant collectors, geospatial scientists, and genetic resource professionals who compiled and made available through open-access repositories the occurrence and ecogeographic information used in this analysis. We also thank D. Amariles, S. Diulgheroff, A. Meyer, M. Obreza, S. Sotelo, and M. R. White for data retrieval support; J. Sullivan for map template inputs; J. Bamberg, D. G. Debouck, L. J. Grauke, C. Heinitz, and L. Marek for evaluation of results; and H. Dempewolf, A. Jarvis, and A. Novy for inputs to the text. C.K.K. was supported by USDA National Institute of Food and Agriculture (Grant 2019-67012-29733/Project Accession 1019405). The USDA is an equal opportunity employer and provider.