The U.S. Fish and Wildlife Service (USFWS) San Francisco Bay-Delta Fish and Wildlife Office (BDFWO) has completed or is in the process of developing recovery criteria, a species status assessment, and 5-year review for howellii. To inform these efforts, the overall goals of this chapter were to conduct a thorough survey of current and historical locations for the Antioch Dunes evening-primrose across the known range and explore the relationship of environmental variables to Antioch Dunes evening-primrose abundance and distribution. Specifically, we perform the following steps:

A comprehensive literature review of peer-reviewed papers, news articles, vegetation databases, USFWS grey literature and internal documents, and herbaria collections was completed to document the historical and current range of howellii and to determine field survey sites. For web search engines, combinations of keywords “Antioch Dunes evening primrose,” “Oenothera deltoides subsp. howellii,” “Antioch Dunes,” and “range” were used. Vegetation databases included CalFlora, USFWS Environmental Conservation Online System (ECOS), California Native Plant Society (CNPS) internal databases, and citizen-science platform iNaturalist. The USFWS provided a query of the California Department of Fish and Wildlife’s California Natural Diversity Database (CNDDB).

We also searched herbaria databases that contain location information, including the Consortium of California Herbaria One (CCH1), The University and Jepson Herbaria of the University of California at Berkeley, and the BIO Herbarium at the University of Guelph (OAC-BIO Herbarium). An in-person visit was made to the UC Davis Center for Plant Diversity Herbarium to view specimens. The UC Davis Library Map Collection database was queried to look for any relevant soil or land use maps that might give an indication of where suitable habitat may have existed historically. The USFWS BDFWO provided documents including reports from past howellii surveys, summary spreadsheets of management actions, and CNDDB exports.

All study sites determined from the literature review are in the Bay-Delta region of California, near the confluence of the Sacramento and San Joaquin Rivers (fig. A1). Sites were historically part of a 6,700-acre dune sheet (McNally, 2014). Although soil dating work has shown multiple deposits of sand over time, most sand is thought to have been deposited during the most recent glaciation period (12,000–26,000 years ago; Atwater, 1982; Powell, 1983). Today, only remnants of this system remain because of climate and land use changes (Powell, 1983).

We visited eight sites where howellii was known or thought to be present (fig. A1): Antioch Dunes National Wildlife Refuge (ADNWR, Stamm and Sardis Units considered separately), Brannan Island State Recreation Area, Browns Island, City of Oakley Parcel (Legless Lizard Preserve), Dutch Slough vineyard, Georgia-Pacific Corporation property in Antioch, and Kemwater/Olin Corporation property in Antioch. Additionally, population estimates from the East Bay Regional Parks Botanic Garden in Berkeley were obtained. Within the ADNWR Stamm unit, a recent constructed dune restoration site (Stamm Management Area 1) was considered separately from the off-dune habitat within that unit. This constructed dune restoration consisted of open sand features constructed from the placement of sand dredge material onto ADNWR Stamm unit property.

Field surveys were done during peak bloom in May and June 2019 to determine the presence or absence of howellii and to count adult (buds, flowers, or fruit present) and juvenile (no sexual organs present) individuals. All sites were visited before surveying to determine if plants were present and the intensity of sampling necessary to count each individual. At sites where howellii populations were detected during site visits, field surveys were done using a grid system loaded on a Leica Real-Time Kinematic Global Positioning System (RTK GPS; ±1 centimeter [cm] horizontal, ±2 cm vertical accuracy; Leica Geosystems Inc., Norcross, Georgia). At Brannan Island, Browns Island, and the City of Oakley Parcel, grid dimensions were 16 meters (m) by 16 m. At ADNWR, we used a pre-existing 20- by 20-m grid established during 2017 vegetation inventory surveys (Mathers and U.S. Fish and Wildlife Service, 2018). Surveyors walked transects through each grid counting the number of individuals, if present. A GPS point was taken at each individual plant to record location data and ensure that there was no double-counting. Where plants were clustered together (within 1 square meter [m²]), several individuals were recorded under one RTK point. On the constructed dune system, in the Stamm Management Area 1 of ADNWR RTK, surveys were not practical because of the high density of plants; RTK points were therefore only taken at individuals where leaves were collected for genetic analysis. All other individuals on the constructed dune were counted at the grid scale. To accomplish counts at the grid scale, flags were laid out at the corners of each grid using RTK, and all individuals were estimated in groups of 10s or, rarely, 100s.

Additionally, representative photographs were taken of howellii at all sites where plants were located. These individual photographs were taken to document any morphological differences between howellii populations and as a record of occurrence. All count data were summarized as total populations within sites, and density of adult and juveniles (number of individuals in a grid/total grid area) to allow comparison between ADNWR and other sites. The ratio of juvenile to adult individuals was calculated for each grid by dividing the density of juveniles by the density of adults; ratios above 1 indicate higher density of juveniles than adults whereas ratios below 1 indicate higher density of adults than juveniles.

For each sampling grid, a visual estimate of the total percent cover of the plant community was recorded. This estimate included all plants and is a proxy for how invaded the sandy substrate was by invasive annual grasses. Notes about the dominant plant community were recorded as well, and any specific features that could influence howellii (in other words, road or ant mound in grid) were recorded when present. The relationship between howellii density and total plant cover was quantified by using linear regression in R v3.6 (R Core Team, 2016); density was log-transformed to meet assumptions of homoscedasticity.

Soil type and normalized difference vegetation index (NDVI) were assessed to determine relationships to current howellii populations. In addition, these layers were used to locate potential habitat across the historical Antioch sand sheet. Historical land cover (Stanford and others, 2011) and soil type (Soil Survey Geographic Database [SSURGO], accessed in 2019, https://websoilsurvey.nrcs.usda.gov/) describe the historical spatial cover of the sand sheet. Modern land cover was then used to estimate remaining land that might support populations of howellii. The 2018 4-band National Agriculture Imagery Program (NAIP) imagery data layer was used to create the NDVI layer. Mean NDVI was calculated within each sampling grid in ArcGIS (Environmental Systems Research Institute [ESRI], 2011, ArcGIS Desktop: Release 10. Redlands, California) and extracted to the howellii grid data. The relationship between howellii density and NDVI was quantified using segmented regression in R to determine the NDVI breaks where howellii occurred.

The literature review identified 16 sites to investigate as potential howellii populations. These sites historically contained howellii or were mentioned in internal USFWS documents, peer-reviewed literature, CNPS location records, or herbaria records (table A1). Of these sites, eight were visited to quantify total population size and demography of individuals (table A1). A summary of the literature review is provided in appendix A1.

We confirmed the presence of howellii at six of the eight sites we visited (figs. A2–5; table A2). In addition, Regional Parks Botanic Garden provided individual counts. A total of 54,216 plants were counted, 5,220 of which were adults (table A2). More than 98 percent of individual howellii were found within ADNWR, with the constructed dune restoration at Stamm Management Area 1 accounting for 96 percent of adult individuals and nearly 100 percent of juvenile individuals at the time of surveying. A shapefile of individual GPS points for ADNWR Sardis Unit, ADNWR Stamm Unit outside of the constructed dune in Management Area 1, Brannan Island, Browns Island, Georgia-Pacific Corporation, and the City of Oakley Parcel are available as appendix A2.

Populations across sites were not randomly distributed; decreasing total plant cover increased total howellii population numbers exponentially across all sites (fig. A6). Increased plant cover within a grid shifted populations toward adult plants with limited juveniles (fig. A7). Grids with more than 75-percent cover on average had fewer juveniles than adults and, therefore, could not replace themselves over time (fig. A7).

Given the strong relationship between total plant cover and number of howellii individuals, we modeled the acreage that would be necessary to reach adult howellii population sizes that match USFWS recovery criteria (draft at time of analysis, now finalized; U.S. Fish and Wildlife Service, 2019): 1,500 or 4,800 (table A3). 25-percent cover closely represents dune habitat, while 75-percent cover most closely represents an invaded annual grass habitat. A habitat at 90-percent cover is highly invaded. To support the same population of 4,800 individuals, a highly invaded habitat would require an order of magnitude more acreage (181–327 acres) than a dune habitat (13–22 acres; table A3).

High density howellii populations were found within a distinct range of NDVI values. In grids with NDVI less than 0.1, there was a relationship between NDVI and howellii adult plant density; with high NDVI (indicating high plant cover); however, there was no relationship to adult plant density and few plants were found (fig. A8). Additionally, NDVI spatial patterns mirrored those found with total plant community cover and high howellii plant densities (fig. A9). Low plant cover and sandy habitat, such as the constructed dune restoration site on ADNWR Stamm Unit, had low NDVI values indicating high habitat quality, whereas the more invaded sites had higher values, indicating poor habitat quality (figs. A9–10).

Unconsolidated entisols (mineral soils that have not differentiated into distinct horizons) underlie the survey region where interior dune communities were common pre-development (fig. A11). Total acreage of the historical sand sheet was 34.4 square kilometers (km²). Using 2018 imagery, the acreage of habitat within the historical sand sheet that has a modern NDVI within the range that supports howellii is 12.9 km² (fig. A12). With developed land removed, there is a maximum of 5.6 km² available that may support howellii, although this is an upper bound because it includes currently cultivated land (fig. A13). Normalized difference vegetation index values are likely to represent urban landforms and not quality dune habitat because most of the land within the NDVI range is developed (fig. A14).

Our results indicated that habitat creation, such as that undertaken at the Stamm Management Area 1 of ADNWR, has the potential to increase howellii populations. We found the highest population densities on the constructed dune habitat and the highest juvenile to adult ratios. The dune is not dispersal-limited because thousands of individuals naturally recruited into the site within a year of restoration. The proximity to other protected habitat may underpin the lack of dispersal barriers; constructed sites located far from any source populations may not show the same success because of lack of pollinators or other constraints (Pavlik and others, 1993). When dredge spoil or other dune construction projects occur beyond the natural dispersal range of howellii, populations could still be viable if transplanted. The Brannan Island howellii population, for example, occurs on old dredge spoil from the 1920s (U.S. Fish and Wildlife Service, 2002; California Department of Parks and Recreation, 2019). Dune construction can be effective by providing open space with mobile sand while reducing competition from other species, which are key drivers of juvenile germination or establishment (Pavlik and Manning, 1993; Greene, 1995).

We found strong relationships between total plant community cover and howellii population size and demography. Total plant cover therefore could be a quantifiable indicator of habitat quality for howellii and is relatively easy to measure in the field. Management of existing habitat to reduce total plant cover may increase howellii populations and shift demography toward juveniles. The difference in distribution and demography of juvenile howellii between areas with bare ground and those with high total plant cover supports previous findings that bare, sandy ground is important for juvenile success (Pavlik and Manning, 1993; Greene, 1995; Thomson, 2005). Populations in areas with low total plant cover had higher proportions of juveniles and therefore a better chance of growing or stable populations over time, compared to areas with high total plant cover made up mostly of adult plants. Because howellii is a short-lived perennial, recruitment is a key demographic transition for population growth and one that may be facilitated by low plant cover (Thomson, 2005).

The mechanisms underlying the impact of total plant cover on juvenile recruitment are unclear. It could be that light is necessary for successful germination or that invasive grasses and other vegetation compete strongly for water or nutrients to prevent successful establishment of seedlings (Pavlik and others, 1993; Greene, 1995). Disturbance of the soil also could be necessary for germination (Thomson, 2005) because dune stabilization by vegetation may reduce sand action that scarifies the seed coat to promote germination. Experimental seedbank germination studies are needed to test if howellii is present in invaded habitat but isn’t germinating because of dune stabilization (for example, competition or reduced light). If present in invaded habitat seedbanks, restoration activities that reduce total plant cover could be successful in jumpstarting howellii populations. Additionally, field studies that test the effect of total plant cover on seedling establishment and recruitment, and greenhouse studies that test the effects of light, nutrients, and water availability separately are necessary to fully disentangle the mechanisms by which plant cover influences howellii population demography. These studies are especially crucial in locations where adult and juvenile plants are present and land management activities that reduce plant cover are feasible. One possibility is that detection of juveniles is lower in invaded habitats because of difficulty in locating individuals; explicit tests of surveyor detection error are needed to constrain this possibility.

The original extent of habitat that may have supported howellii or similar dune specialist species covered a broad range in the Antioch area. With the sand sheet area now under fast-paced development, the potential available habitat for dune specialists is much reduced and declining quickly. Our map of potential quality habitat gives a baseline search extent for further analysis that could potentially pinpoint habitat that can support howellii. Identifying NDVI values between −0.1 and 0.1 with low variation across time within this extent will allow further reduction of the focus area and direct attention on land parcels that may hold relict populations or could potentially support outplanting operations. The analyses presented here provide maximum potential habitat; future analyses will need to verify and constrain broad land cover classes to remove roads, levees, and other landforms not conducive to plant populations of concern.

Regardless of whether habitat creation or habitat management is used, the total acreage necessary to sustain populations of adult howellii is greatly reduced when that habitat has low total plant cover. To sustain population sizes matching delisting criteria (2 populations of 1,500 and 5 populations of 4,800 [15-year moving median]; U.S. Fish and Wildlife Service, 2019), quality of habitat is important. Modeling suggests 95 (95-percent confidence interval: 74–122) total acres are needed if the habitat is high quality with low (25 percent) total plant cover (figs. A2A, A15) and 1,369 (95-percent confidence interval: 1,017–1,842) total acres are required if the habitat is highly invaded with high (90 percent) total plant cover (figs. A2C, A15).

This project was designed as a companion to the field surveys documenting distribution and abundance of howellii described in chapter A. Our study goals were to assess the distribution of genetic diversity within and among occurrences of howellii and explore its evolutionary relationship to the putative new taxon with an adjacent distribution on the Antioch sand sheet by using genome-wide genetic markers.Specifically, we:

Leaf tissue was collected from all extant howellii occurrences (chapter A): Antioch Dunes National Wildlife Refuge (ADNWR Stamm dune in Management Area 1, Stamm excluding dune in Management Area 1, and Sardis Units considered separately), Brannan Island State Recreation Area, Browns Island, and City of Oakley Parcel (preserve for legless lizards). Samples also were collected from the Regional Parks Botanic Garden collection, and from cultivated plants obtained from Annie’s Nursery (www.anniesannuals.com; Richmond, California).

At Dutch Slough, a site that was previously thought to contain howellii, plants were morphologically distinct from howellii observed at ADNWR and other occurrences. After consulting the Jepson manual (Baldwin and others, 2012) and herbarium specimens, it was determined that these plants appeared to have traits that were a mix between howellii and cognata (fig. B1). For example, these individuals showed dentate leaf margins with green, not gray-green, strigose foliage (cognata traits) but contained free sepal tips greater than 1 millimeter (mm; howellii traits). Dutch Slough individuals also contained more variability in traits than observed at other occurrences; leaf traits were especially variable among individuals within a population. Therefore, collections were expanded to survey any occurrences of this potential hybrid or new lineage in the local area and to survey the closest geographical subspecies, cognata. In total, six occurrences were collected in addition to the original howellii occurrences. Three of these occurrences (Dutch Slough, Oakley Rose Avenue, and Big Break) contained the morphological hybrid or new taxon (clade X), and cognata was collected from three occurrences further south in the Central Valley (Corral Hollow, Monvero Dunes North and South: fig. B2; table B1). Monvero Dunes samples were collected by Ryan O’Dell, Bureau of Land Management, Hollister Field Office.

Leaf tissue for DNA analysis was collected from 20 individuals at each occurrence, where possible and 30 individuals from ADNWR Stamm unit occurrences. For each sample, three to five leaves were collected into coin envelopes in the field and transferred to a 45-degrees Celsius (°C) oven until tissue was fully desiccated. Once dry, envelopes were stored with color-indicating silica gel desiccant in plastic bags inside a dessicator cabinet (Bel-Art, SP Scienceware, Wayne, New Jersey) at room temperature.

Genomic DNA was extracted using the E-Z 96® Plant DNA Kit (Omega Bio-tek Inc., Norcross, Georgia) with Antifoam Y-30 (Sigma-Aldrich, St. Louis, Missouri) added to the homogenizer plate and a 60-minute room temperature incubation before elution in 100 microliters (μL) of water. Reduced representation genomic libraries were prepared using a modified double-digest restriction-site associated DNA sequencing (ddRAD) scheme (Peterson and others, 2012) in which 700 nanograms (ng) of DNA were digested with EcoRI and MseI restriction enzymes. In-line barcodes were ligated onto the EcoRI cut site and fragments were size-selected on a Pippin Prep (Sage Science, Inc., Beverly, Massachusetts) at a 400±30 base pair (bp) range. Each library of 12 multiplexed samples was amplified by polymerase chain reactions (PCR) for 15 cycles with the addition of a unique library index sequence. Libraries were quantified, pooled, and sequenced at the University of California, Berkeley, Vincent J. Coates Genomics Sequencing Lab (Berkeley, California) on two Illumina NovaSeq SP 100bp single-end lanes (Illumina, Inc., San Diego, California).

Raw sequence trimming, demultiplexing, quality filtering, and genotyping was performed using Stacks v2.41 on the USGS Yeti High Performance Computing platform. Clustering, assembly, and filtering parameters were optimized using a subset of individuals following the r80 method by Paris and others (2017). This subset was comprised of 30 percent of all samples evenly distributed across collection locations with read coverages that fell within 2 standard deviations of overall mean coverage per locus. This subset also was used to create a locus catalog (cstacks) for the full Stacks genotyping pipeline. The following parameters were used: maximum number of mismatches between stacks within individuals, M=3; maximum number of mismatches between stacks between individuals, n=3; minimum percentage of individuals across populations required to process a locus, R=0.65 (Catchen and others, 2011). One single nucleotide polymorphism (SNP) was randomly chosen from each locus, exported in a variant call format (VCF) file, and further filtered for coverage depth by removing the bottom 1 percent, top 10 percent, and requiring 6 times the minimum coverage. Whitelists of loci and samples that passed the VCF filter were then rerun in the populations program to calculate genetic diversity statistics, including number of private alleles, expected heterozygosity (H_e), inbreeding (F_IS), and genetic differentiation among sampling locations (F_ST).

Two genetic clustering analyses were used to explore the pattern of genetic structure among samples. A principal component analysis (PCA) was implemented in adegenet v2.1.3 (Jombart, 2008) by using R v3.6.3 (R Core Team, 2016) to visualize the genotypes in multidimensional space. FastSTRUCTURE v1.0, a Bayesian clustering method (Raj and others, 2014), was run for values of K from 1 to 10 to identify differences in allele frequencies across populations. A range of values for the optimal number of clusters (K) was determined using the chooseK.py script in the fastSTRUCTURE package. We report pairwise F_ST and H_e for the identified genetic clusters and tested for differences in H_e by using the adegenet function Hs.test.

Field observations at clade X occurrences suggested that these plants had traits that were intermediate between howellii and cognata (fig. B1). We tested whether these plants represented an active hybrid zone between howellii and cognata, or a distinct taxon, using the likelihood optimization function snapclust in adegenet, with the option to explicitly model F1 (equal contributions of each parental population) and backcross (equal contributions of an F1 and a parental population) hybrids. To further investigate the lineage of clade X, we used a maximum likelihood based phylogenetic approach implemented in Raxml v8.2.12 (Stamatakis, 2014) and executed in the CIPRES Science Gateway v3.3 (www.phylo.org). Briefly, a concatenated gene matrix was compiled for individuals with heterozygous sites coded using standard ambiguity codes. Analyses used a single model of evolution (General Time Reversible with a gamma distributed rate variation among sites; GTR+g), were not partitioned, and were set to 100 rapid bootstrap inferences, before a thorough ML search. Resulting trees were visualized in FigTree v1.4 (https://github.com/rambaut/figtree/releases/tag/v1.4.4) and highly supported bootstrap values (>80) were mapped onto the most likely tree. Finally, to identify genetic loci that discriminate among clusters, a discriminant analysis of principal components (DAPC) was used to maximize the differences between genetic clusters. To determine the optimal number of principal components (PCs) to retain, we used the interactive adegenet web server to perform a cross-validation procedure with the following settings: dataset split 90-percent training set, 10-percent validation set, 30 replicates, and selected the number of PCs with the lowest root mean squared error and highest mean successful assignments. We then used an average clustering threshold to identify SNPs that load onto each discriminant axis (DA).

We found minimal population structure among howellii occurrences and found that clade X and cognata occurrences form distinct genetic clusters. The first principal component (PC1) of the PCA described 14.6 percent of the total genetic variation and PC2 described 6.22 percent. There was a clear pattern of three distinct clusters where the difference between cognata and howellii was attributed to PC1 and the difference between howellii and clade X was on PC2 (fig. B3). PC3 described 5.54 percent of the variation and distinguished the Corral Hollow and Monvero Dunes cognata occurrences (appendix B1). These three main clusters separated by principal component (PC) axes 1 and 2 were further supported by a fastSTRUCTURE analysis that found the optimal number of genetic clusters in the dataset was three. This analysis showed that clade X occurrences share some similarity with howellii but remain a distinct genetic group (fig. B4).

Similarly, estimates of genetic differentiation (F_ST) were low among sampled howellii occurrences (range 0.02–0.096), especially the ADNWR locations. Notably, the Regional Park Botanic Garden (RBG) samples appeared least differentiated from the ADNWR locations. The Annie’s Nursery (ANH) sample was the most differentiated from the howellii occurrences but this could be because of its unknown cultivation history. Pairwise differentiation estimates among all occurrences were highest between Corral Hollow and non-cognata occurrences (table B2). We also found that cognata was more differentiated from howellii and clade X, when samples were grouped by genetic cluster, (pairwise F_ST for cognata-howellii=0.085, cognata-clade X=0.087, howellii-clade X=0.043; p<0.05), rather than by occurrence.

Genetic diversity (expected heterozygosity [H_e]) for the sampled howellii occurrences ranged from 0.094±0.003 standard error (SE) to 0.113±0.003 SE (table B1). The three ADNWR occurrences had similar H_e and were significantly higher than the remaining three occurrences; Browns Island, Brannan Island, and Oakley (p<0.001). Of the additional howellii collections, the RBG samples had a diversity estimate similar to the non-ADNWR occurrences (0.093) but the samples from ANH were much lower at 0.058. Genetic diversity estimates for clade X and cognata occurrences were all above 0.1 with the exception of Corral Hollow (0.07±0.003 SE). Inbreeding estimates (F_IS) were essentially zero for all occurrences in the dataset (table B1).

When samples were considered by genetic cluster, rather than by occurrence (excluding samples from RBG and ANH because of unknown cultivation history), H_e for cognata was significantly greater than clade X (p<0.033) and howellii (p<0.037) and clade X was significantly greater than howellii (p<0.024). The number of private alleles found for each type was 351 (howellii), 328 (clade X), and 795 (cognata). These combined results indicated that cognata is more genetically diverse than howellii and clade X, and clade X is more genetically diverse than howellii.

The snapclust hybrid model converged with no evidence of F1 or backcrossed hybrids at any location. Group membership probabilities for the three hybrid classes (F1, howellii-backcross, cognata-backcross) were statistically zero across the dataset, and all clade X samples were assigned to the howellii group (appendix B2). The maximum likelihood phylogenetic analysis strongly supported four clades, with howellii and clade X most closely related to each other and two clades within cognata that separated the two geographic locations sampled (fig. B5). There were no other well-supported clades within any of these lineages that corresponded with sampling sites within clades (appendix B3). These results, in addition to the population structure results, indicated that clade X is likely a separate distinct group from howelli and cognata with no indication of modern gene flow among these clades. However, we did find high genetic similarity and evidence of recent gene exchange or origin among occurrences within howellii and within clade X. Although the intermediate morphological features of clade X could indicate that this lineage was of hybrid origin, the genetic structure suggests it has since been isolated from each parental lineage for some time. Alternatively, a common ancestor of howellii and clade X could have first diverged from cognata and then howellii and clade X further diverged from each other.

Given these well-defined genetic clusters and the need for genetic identification in future taxonomic investigations, we used a discriminant analysis of principal components (DAPC; fig. B6) to identify the alleles with the highest contribution to the differences between each group. We found 22 SNPs that defined discriminant axis (DA) 1, and 5 SNPs on DA 2. DA 1 differentiated cognata from remaining samples, and DA 2 differentiated clade X from howellii. The list of SNPs and the original 95bp locus sequences are included in appendix B4.