Emerging wild virus of native grass bioenergy feedstock is well‐established in the Midwestern USA and associated with premature stand senescence

The North American native prairie grass Panicum virgatum (switchgrass) is a primary bioenergy feedstock candidate. Its widespread distribution and genetic diversity enable the possibility of developing this perennial grass for high production in a variety of conditions, including on marginal lands. A critical concern in feedstock development and deployment is the risk of novel pathogen emergence. Here, we investigate the landscape‐scale prevalence and epidemiology of a little‐studied North American virus that was first detected in switchgrass and other grasses in bioenergy trials in the US Midwest. Switchgrass mosaic virus (SwMV, Genus Marafivirus, Family Tymoviridae) is transmitted by leafhoppers and is phylogenetically a sister to Maize rayado fino virus, a significant pathogen of maize in parts of the Americas. Our goal was to determine whether SwMV is uniquely limited to specific bioenergy trials or well‐established and circulating more broadly. We used molecular diagnostics to quantify naturally occurring SwMV infection in leafhoppers and switchgrass in naturalistic stands throughout a large Midwestern landscape, and quantified leafhopper abundances and stand performance. Our analysis revealed that this apparently wild virus is well‐established and widespread. Infection was present at nearly all sites, across diverse landscape contexts, with its prevalence ranging as high as 33%–60%. Infection appeared to accumulate and persist in stands over time. It was associated with increases in premature stand senescence but not with reductions in stand height. Although wild viruses are believed to evolve benign relationships with their natural hosts, these data suggest that SwMV has the potential to impact yield components. Viruses are frequently overlooked in crop development efforts, but they represent the majority of emerging plant pathogens. For SwMV, it is imperative to quantify its impact on host performance, to identify the extent of any host resistance, and to assess any risks of virus spillover to agricultural plantings of other Poaceae species, including maize and sorghum.


| INTRODUCTION
Perennial grasses are important potential feedstock for sustainable production of cellulosic bioenergy (Lemus & Lal, 2005;Robertson et al., 2017;Somerville et al., 2010). In the United States and elsewhere, a primary feedstock candidate is switchgrass (Panicum virgatum L.)-a genetically diverse C4 prairie grass native to North America (McLaughlin & Adams Kszos, 2005;McLaughlin et al., 2002;Sanderson et al., 1996). Switchgrass is found in numerous natural habitats, from native prairies to riparian areas, across a wide geographic range and latitudinal gradient (Casler et al., 2004;Lowry et al., 2014). Switchgrass is a strong candidate for bioenergy production because it produces substantial biomass with modest crop inputs (i.e., nitrogen), can tolerate marginal lands, and provides multiple ecosystem services (Mitchell et al., 2008(Mitchell et al., , 2012Werling et al., 2014).
To develop switchgrass as a bioenergy crop, it is essential to evaluate the extent and nature of pathogen infections it may acquire in the field. Such infections could reduce switchgrass yield and quality, and potentially increase disease pressure on other Poaceae crops (Schrotenboer et al., 2011). To date, most attention has focused on identifying and controlling fungal rust infections, which are readily apparent in both lowland and upland switchgrass ecotypes (Hirsch et al., 2010;Kenaley et al., 2018;VanWallendael et al., 2020;Zale et al., 2008). However, initial data indicate that switchgrass is also susceptible to multiple pathogenic crop-infecting viruses known to damage cereals, sugarcane, and turf grasses, including members of the genera Polerovirus and Luteovirus (Families Solemoviridae and Tombusviridae) (Garrett et al., 2004;Schrotenboer et al., 2011), and the species Panicum mosaic virus (PMV, Family Tombusviridae) and its synergistic dependent satellite Panicum mosaic virus (SPMV) (Scholthof, 1999;Sill & Pickett, 1957;Stewart et al., 2015), and Sugarcane mosaic virus (SCMV, Family Potyviridae) (Agindotan et al., 2010). Some of these viruses are vectored by flying insects, including sap-feeding leafhoppers and aphids, that can spread infection locally and over long distances.
As a native prairie grass, switchgrass likely arose 2 million years ago in the Pleistocene and has had a long presence in North America, where it still can be considered a 'wild' noncrop species in contrast to domesticated grasses (Parrish et al., 2012). As indicated by recent high-throughput sequencing of crop and noncrop vegetation, noncrop plants harbor a rich diversity of plant viruses that are only beginning to be explored Min et al., 2012;Roossinck et al., 2010;Shates et al., 2019;Susi et al., 2017). Initial investigations of the switchgrass virome in Illinois (USA) identified two novel species: Switchgrass mosaic-associated virus (SgMaV, Genus Mastrevirus, Family Geminiviridae) (Agindotan et al., 2015), and Switchgrass mosaic virus (SwMV, Genus Marafivirus, Family Tymoviridae) (Agindotan et al., 2010(Agindotan et al., , 2012, which is the focus of this study. SwMV is transmitted by the grass-feeding leafhopper Graminella aureovittata (Agindotan et al., 2013b), a species associated with moist prairies in the central and eastern USA (DeLong, 1948).
Crop-associated viruses (henceforth 'crop viruses') that cause economic loss in crops have so far received most attention in plant virology Wren et al., 2006). Crop viruses can have significant negative effects not only on crops but also on noncrop vegetation (Malmstrom & Alexander, 2016). Crop-associated BYDV, for example, can stunt switchgrass root systems  and reduce the biomass production and integrated multiyear fitness of switchgrass plants accumulate and persist in stands over time. It was associated with increases in premature stand senescence but not with reductions in stand height. Although wild viruses are believed to evolve benign relationships with their natural hosts, these data suggest that SwMV has the potential to impact yield components.
Viruses are frequently overlooked in crop development efforts, but they represent the majority of emerging plant pathogens. For SwMV, it is imperative to quantify its impact on host performance, to identify the extent of any host resistance, and to assess any risks of virus spillover to agricultural plantings of other Poaceae species, including maize and sorghum.

K E Y W O R D S
bioenergy, Graminella, leafhopper, marafivirus, Panicum virgatum, pathogen, plant virus ecology, prairie grass, Switchgrass mosaic virus, Tymoviridae . In contrast, almost nothing is known about the effects of noncrop 'wild' viruses such as SwMV on either crop or noncrop vegetation. It has been suggested that most noncrop virus infections have little negative impact on hosts and might even be beneficial (Fraile & García-Arenal, 2016;Roossinck & Bazán, 2017). Among the very few wild viruses of plants that have been studied, effects on hosts were found to be slightly negative to neutral  or contextually dependent (Gibbs, 1980). In the case of SwMV, there is a potential for damaging impact. Its nearest known relative, maize rayado fino virus (MRFV), is arguably the most important viral pathogen of maize in Latin America (Gámez, 1969;Rybicki, 2015), raising the question of whether SwMV likewise might be pathogenic in its hosts.
We discovered SwMV in Michigan switchgrass about the same time that Agindotan et al. (2010) reported infection in bioenergy trial plots in Illinois and Wisconsin. These parallel discoveries prompted us to investigate the distribution and impact of the novel virus to better understand whether it might pose a threat to perennial grass feedstock. We began by asking whether the elevated SwMV prevalence seen in the bioenergy trials might represent a unique situation, perhaps influenced by cultivation conditions, or whether SwMV infection was instead widespread, with these initial reports representing just the "tipof-the-iceberg" of its distribution. Because the first SwMV detection in Michigan was in a conservation planting, not a feedstock trial, we chose to investigate the distribution of infection in established naturalistic stands throughout our region. We reasoned that if infections were found throughout these little-managed stands, it would be good indication that the virus was well-established in our area and not unique to a few bioenergy trial plots. Since little is known about the virus' epidemiology, we further sought to identify possible impacts of infection on stands and to assess whether local site properties or the nature of the surrounding landscape might predict its distribution. To do this, we quantified the prevalence of SwMV, the abundance of potential leafhopper vectors, and relationships between SwMV prevalence and stand conditions at sites in different landscape contexts throughout a 37,000-km 2 area of Michigan, USA. We used a SwMV-specific molecular diagnostic to quantify SwMV prevalence in both switchgrass and the native Graminella leafhopper species that feed on it, including the known SwMV vector G. aureovittata (Agindotan et al., 2013b). Our study coincided with a severe summer drought, which we quantified at each location with a drought index.
We found that SwMV infection was widespread and present at all but one of our 15 sites. Moreover, infection prevalence was the best predictor of switchgrass senescence in the drought, suggesting that infection damaged stressed stands, perhaps by reducing their stress tolerance. An alternative explanation-that drought or poor stand growth increased infection prevalence-was not supported by statistical models. Landscape context did not predict prevalence patterns or abundance of known vectors, suggesting that virus and vector pressure is a synoptic phenomenon filtered by site properties. Taken together, these findings strongly indicate that SwMV infection is well-established in our region and merits attention as a pathogen of potential virulence. More broadly, these findings highlight the need to better understand how selection of new crops influences their relationships with endemic wild viruses and the risk of emerging infectious disease.  (Agindotan et al., 2012;Agindotan et al., 2013b). In plant hosts, marafivirus virions are most often found in phloem and xylem tissues (Nault & Ammar, 1989). In switchgrass, SwMV infection may produce straight fine white, creamy, or yellowish lines and dots in leaves, running parallel to the veins (Figure 1a and b) and similar to symptoms of MRFV in maize (Zambrano et al., 2013). However, some infections are asymptomatic (Agindotan et al., 2013b). Infection overwinters in switchgrass rhizomes and reemerges with new tillers in the spring (Ryskamp et al., in prep); thus, prevalence values represent infections accumulated over multiple years. Marafiviruses propagate within their insect vectors (insect hosts), as well as within the plant host, and vectors require a latent period of at least one week after virus acquisition before transmission (Nault & Ammar, 1989). Tests with the leafhoppers G. aureovittata, G. mohri, and Flexamia atlantica (all members of the Family Cicadellidae, Order Hempitera) found that only G. aureovittata ( Figure 1c) transmitted infection (Agindotan et al., 2013b). Marafiviruses are not known to be transmitted by seed (Nault & Ammar, 1989) or by mechanical means in the field, although vascular puncture transmission is possible in the lab (Weiland & Edwards, 2011). SwMV has been detected in several C4 Poaceae species in Illinois besides P. virgatum, including in the North American natives P. amarum (bitter panicum), Andropogon virginicus (broomsedge bluestem), and Sorghastrum nutans (Indian grass), and in several species of the non-native Miscanthus genus (Agindotan et al., 2013a). Beyond these findings, the biology of this emerging and apparently native virus remains largely undescribed.

| Study approach and locations
We evaluated naturally occurring virus dynamics in naturalistic stands that were planted in the past. This approach represents method 2 for studying plant virus effects in the field (experimental plants with naturally occurring virus infection) with some elements of method 1 (natural plant populations with naturally occurring virus infection), as not all information about planting material was known and the plantings had self-propagated and spread (Malmstrom & Alexander, 2016). The study examined established switchgrass-dominated communities with upland switchgrass ecotypes at 15 sites in 12 counties across Michigan's lower peninsula in the Great Lakes Region (USA) (Figure 2). Thirteen of these stands were established in the 1990s-early 2000s for conservation purposes (e.g., game bird habitat) in state game areas and on private property, and were left largely undisturbed or managed only lightly; ten of the latter were included in a related study of ecosystem service provisioning by switchgrass and prairie communities (Werling et al., 2014). In addition, we included two regularly harvested larger switchgrass plantings established in 2010 as "scale-up" sites for the US Department of Energy-supported Great Lakes Bioenergy Research Center (GLBRC). The GLBRC scale-up fields were seeded with upland P. virgatum cv. Cave-in-Rock, an octoploid natural-track cultivar from Illinois (Evans et al., 2015). Specific seeding records for the other fields were not available, but Cave-in-Rock was most commonly used in such plantings in Michigan during that time period. Field size ranged from 0.5-14 ha (median = 4 ha) with most fields being 2-6 ha in area; the two scale-up sites were the largest (13-14 ha). Nearest-neighbor distances between points ranged from 4.4 km to 63.0 km.
Sites were chosen to represent a range of landscape contexts with differing proportions of crop and noncrop cover types. To quantify landscape context, we evaluated the distribution of the 2012 US Department of Agriculture's Cropland Data Layer (CDL) land cover types (https:// nassg eodata.gmu.edu/CropS cape/) within circular buffers with radii of 0.5 km (79 ha area), 1.0 km (314 ha area), and 1.5 km (707 ha area) around each site; the 2012 CDL is a georeferenced raster with 30-m ground resolution. All GIS work was conducted in ArcGIS versions 10.6-10.8 (ESRI, Redlands, CA, USA). We aggregated the land cover types represented in our region into eight primary cover groups: (i) Agricultural cover, which included four dominant crops-maize, winter wheat, alfalfa, soybeansand lesser amounts of 19 other vegetable, fruit, and small grain crops; (ii) grass/meadow; (iii) developed; (iv) forest; (v) wetlands; (vi) miscellaneous perennials; (vii) barren; and (viii) open water (see Supplemental Materials for further description). Across all sites and at the three scales, agriculture represented 32-33% of the cover in this diverse landscape; grass/meadow, 17-31%; forest, 13-17%; developed areas, 12-14%; wetlands, 10-16%; and open water, 2-4% (Figure S-1A-C).

| Drought conditions
In 2012, the US Midwest experienced unusual dryness and drought (Rippey, 2015), and the effects in Michigan were spatially heterogeneous. To quantify how much drought each study site experienced, we calculated a drought index value based on spatially explicit estimates of the duration and weekly severity of drought conditions as published in the US Drought Monitor (https://droug htmon itor.unl. edu/). The Drought Monitor rates moisture conditions as no drought or dryness (no drought rating), abnormally dry (D0), moderate drought (D1), severe drought (D2), extreme drought (D3), and exceptional drought (D4). For each site's GPS location (i), we thus calculated a drought index (DI) as where j = the growing season week and d j = the Drought Monitor D value rating (0-4) for week j. Thus, weekly ratings of D0, D1, D2, D3, and D4 were valued as 1, 2, 3, 4, and 5, respectively. Weeks with no recorded drought or dryness were valued as 0. Switchgrass in this region typically sprout in early May, so we calculated DI for the 16 weeks from the week of May 1 through the week of August 14, when early August field measures were completed (i.e., n = 16).

| Switchgrass condition and sampling for SwMV detection
To quantify the relative differences in stand productivity among sites, we measured switchgrass height at 15 withinstand locations at each site in both sampling periods and calculated mean canopy values. To characterize stand condition, we quantified the degree of stand senescence by estimating the percentage of senesced switchgrass foliage (dry brown leaves) at 15 random points at each site.
For virus detection, foliar tissue was sampled from switchgrass in late August at the twelve accessible sites. At each site, we collected tissue from fifty plants sampled every 1.5 meters along two 70-m transects that were at least 20 m apart. At each transect point, we sampled the tiller closest to the point, without regard to size, condition, or symptoms. After collection, samples were transported on ice and then stored at −20°C until processed. At one site (Sw07), we were able to compare prevalence values with earlier 2010 collections from switchgrass (n = 41 plants) and big bluestem (A. gerardii) (n = 3 plants).

| Leafhopper collection & identification
Our prior field observations suggested that Graminella were most abundant in our area in late summer. To confirm that seasonal distribution, we sampled leafhoppers from June 2012-August 2012 at four sites (SW02, SW07, SW10, SW14) selected to represent diverse geographic regions. We sampled our larger network of sites twice in August, when Graminella numbers were greatest: In early August (August 2-8, 2012), we sampled all 15 sites, and in late August (August 23-29, 2012), we sampled 12 of the 15 sites, as SW09, SW17, and SW18 could not be accessed. All collections occurred during warm and sunny daylight hours (10:00 hours-04:00 hours); air temperature was recorded. There are multiple methods for capturing leafhoppers; we used sweep-netting because in our experience this method is the best approach in our system when fresh samples are required for virus analysis. For each collection, we captured leafhoppers from three separate transects of 50 sweeps each, spaced at 1 sweep per meter, for a total of 150 sweeps. Captured insects were killed by immersion for 10-15 minutes in a jar containing ethyl acetate, transferred into plastic bags, and stored in a cooler before long-term storage at −20°C.
We sorted leafhoppers from plant debris and other arthropods in the sweep samples with a sieve and microscope. In Illinois switchgrass stands, Agindotan et al. (2013b) found G. aureovittata, G. mohri, and F. atlantica and determined that only G. aureovittata transmitted SwMV. In our sweep collections, nearly all the leafhoppers were Graminella spp. We did not find any F. atlantica, and to the best of our knowledge, this species has not been recorded in Michigan. We also did not find any Dalbulus maidis or G. nigrifrons, which transmit MRFV (the cropinfecting relative of SwMV) to maize. G. aureovittata was readily identified by its characteristic shape and orange stripes ( Figure 1c). The remaining Graminella were a mixture of G. mohri and G. oquaka (Figure 1d), which look highly similar to each other (DeLong, 1948). To identify the species of these individuals, we dissected a subset and evaluated the male genitalia. Part of the abdomen of each sampled individual was removed, placed in a heated 10% potassium hydroxide solution for 30 min to expose the internal male parts, washed in distilled water, and then placed in glycerin for inspection under microscope following a modified version of the method of Oman (1949) (Trębicki et al., 2010). The aedeagus was then evaluated with the DeLong (1948) key. To preserve samples for RNA extraction, further sorting was nondestructive. G. aureovitatta was sorted easily based on morphological characteristics alone. Because G. mohri and G. oquaka could not be distinguished without destructive analysis and then only males could be properly identified, we grouped these two sister species together as G. oquaka/mohri. Sorted leafhoppers were then stored at 20°C until viral RNA could be extracted. For the early August sample, data for Graminella are complete but counts of total leafhoppers (all taxa collected) are missing from 3 sites (SW11, SW12, SWLA). In late August, the sample from SWLA was damaged partway through analysis so that from it only counts of G. aureovittata are available.

| Detection of SwMV
We used molecular diagnostics to detect SwMV in a subset of the leafhopper and plant tissue samples collected. In total, we tested 180 switchgrass plants for infection and 192 leafhoppers (44 G. aureovittata and 148 G. mohri/oquaka). At each sampled site, we tested 16 switchgrass individuals (every third individual from each 50-plant collection, with field locations ≥4.5 m apart).
For leafhoppers, we tested all G. aureovittata collected, except for a few individuals reserved for species confirmation, because this species is known to transmit SwMV. In the early August collection, we also tested a subsample of 10 G. mohri/oquaka from each site. At sites where fewer than 10 individuals were collected, we tested all that were available. In late August, when G. mohri/oquaka were less apparent, we tested individuals from only one site (24 individuals tested of 29 collected).
2.6.1 | RNA extraction for virus detection From switchgrass, we extracted total RNA with the Spectrum Plant Total RNA extraction kit (Sigma-Aldrich), according to the manufacturer's protocol. Hundred milligram of frozen leaf tissue was homogenized for 2 minutes in the Mini-Beadbeater-16 (BioSpec Products, USA) in a 2-ml screw-cap tube containing liquid nitrogen and 1.0mm silica-zirconium beads. After homogenization, 500 µl of lysis solution containing 5 µl of 2-mercaptoethanol was added to each tube and vortexed for 30 s. The solution was incubated at 56°C for 5 min and then centrifuged for 12 min at 15,000 RCF to pellet cellular debris. Next, the supernatant was transferred to a filtration column and centrifuged for 1 min at 15,000 RCF. To capture RNA, the flow-thru lysate from the filtration column was mixed with 750 µl of binding solution and transferred to the binding column. Tubes were centrifuged for 1 min at 15,000 RCF. After washing, RNA was eluted from the binding column with nucleasefree water and 1 µl of RNaseOut ribonuclease inhibitor was added. RNA was stored at −80°C until further analysis.
From leafhoppers, we extracted total RNA using a modified Dellaporta method (L. Ingwell, pers. comm.) (Dellaporta et al., 1983). For each batch of 16 leafhoppers, 10 ml of Dellaporta extraction buffer was prepared in a nuclease-free glass container from 1.0 ml of 100 Mm Tris at Ph 8.0, 1.0 ml of 500 mM EDTA, 1 1.25 ml of 500 mM NaCl, and 6.75 ml of nuclease-free water. Immediately before use, 10 µl of 2-mercaptoethanol was added to the buffer. Each leafhopper was homogenized for 10 s using the Mini-Beadbeater-16 in a 2-ml screw-cap microcentrifuge tube containing 400 µl of Dellaporta extraction buffer and 1.0-mm silica-zirconium beads (BioSpec Products). To disassociate nucleoprotein complexes, each sample was next incubated with 52.8 µl of 10% SDS solution for 10 minutes at 65°C. After incubation, 128 µl of 5 M potassium acetate solution was added to facilitate protein and DNA removal, and the samples were centrifuged at 4°C for 10 min at 15,000 RCF, resulting in a pellet. Next, 480 µl of supernatant was transferred to a new tube and centrifuged for another 10 min at 15,000 RCF at 4°C. To precipitate RNA, each sample was incubated with 240 µl of cold 100% isopropanol at −20°C for 1 hr and then centrifuged at 4°C for 20 min at 15,000 RCF. The isopropanol was removed and discarded, leaving the pellet, which was washed with 70% ice-cold ethanol and centrifuged. After the ethanol was removed, pellets were air dried for 10 min. Finally, the RNA pellets were resuspended in 80 µl of nucleasefree water with 1 µl of RNaseOut ribonuclease inhibitor (Sigma-Aldrich) and stored at −80°C.

| RT-PCR amplification and
Sanger-sequencing of amplicons We used reverse-transcription (RT) to convert viral RNA from plant and insect samples to cDNA, which was then amplified with PCR. Total RNA concentrations were quantified with the Qubit Fluorometer 2.0 (Life Technologies). In reverse transcription, 1 µg of total RNA (to a maximum of 5 µl for more dilute samples) was added to a mixture containing 0.4 µl of 10 µM reverse primer (BO88-MRFV-10R: 5'-GCC CAC AGG TCT TAT GGC CGA CCT GCT ACC -3' (Agindotan et al., 2010)) and 4 µl of 10 mM dNTPs (Sigma Aldrich), previously mixed, and nucleasefree water was used to bring the total reaction volume to 12 µl. Mixtures were incubated for five minutes at 65°C and then in ice for five more minutes to promote annealing. Next, 7 µl of a master mix containing 4 µl of 5X firststrand buffer (Sigma), 2 µl of 0.1 M dithiothreitol, and 1 µl of RNaseOut ribonuclease inhibitor (Sigma) was added to each tube. Finally, each tube received 1 µl of SuperScript II enzyme (Sigma-Aldrich) for a final RT reaction volume of 20 µl. Samples were incubated at 42°C for 50 minutes to promote DNA polymerization, and then 15 minutes at 70°C to inactivate the enzyme.
We then performed PCR on the cDNA to amplify a 635bp region of the viral coat protein, following a modified version of the Agindotan et al. (2010) protocol. Briefly, 2 µl of diluted RT product (1/10 dilution in nuclease-free water) was added to a 0.2-ml PCR tube containing 18 µl of master mix: 2 µl of 10X PCR Buffer, 1.2 µl of 25 mM MgCl 2 , 1.6 µl of 10 mM dNTPs, 0.8 µl each of 10 µM reverse primer (used in RT) and forward primer (5'-GCTATTCCTGCTCCTCCTCGTGTGGTTGAAACC-3'), 0.2 µl of AmpliTaq Gold enzyme (Sigma-Aldrich), and 11.4 µl of nuclease-free water. Final reaction volume was 20 µl. RT product was diluted to limit inhibition of downstream PCR reaction. Amplification was performed using a Peltier Thermal Cycler (PTC-200, MJ Research) as follows: activation at 94°C for 10 minutes, followed by 40 cycles of denaturing (94°C, 30 s), annealing (60°C, 30 s) and extension (72°C, 45 s), with a final extension (72°C, 10 min). The PCR product was analyzed on a 1.25% ethidium bromide gel under UV light. DNA amplicons were purified with the QIAquick PCR Purification or the QIAquick Gel Purification kit (QIAGEN). Purified DNA was submitted with forward and reverse primers to the Genomics Technology Support Facility (Michigan State University, East Lansing, MI, USA) for Sanger sequencing.

| Statistical analysis and ecological predictors
Statistical analysis was conducted in JMP Pro version 15 (SAS, Cary, North Carolina USA), except as noted. We used generalized regression with native distributions and model selection methods with Akaike information criterion values (AIC c , corrected for small sample size) (Burnham & Anderson, 1998. The best distribution for each response variable was determined by comparing AIC c values and weights for fits with appropriate choices. Calculation of a global Moran's Index for each response variable in ArcMap 10.8 did not find evidence of spatial autocorrelation (Table S-1). N = 15 for all models except those in which SwMV prevalence in switchgrass was the dependent variable or appeared as a predictor in at least one candidate model, for which N = 12. For null models, we included those with intercept only, or with only intercept and latitude or longitude. We considered the best model to be that with the lowest AIC c value and present as competing models those for which ∆AIC c ≤ 2.
We first evaluated potential predictors of two aspects of switchgrass stand performance in early August: (i) mean stand height (Weibull distribution), a measure of stand growth related to productivity and (ii) mean stand senescence (mean percent dry leaves, log-normal distribution), a measure of stand condition. For both, we evaluated several models with intercepts and single explanatory variables describing local conditions (i.e., drought index, DI, or switchgrass infection prevalence). For stand senescence, we further considered stand height as a single predictive variable, and models with both switchgrass infection prevalence and drought index, and with both factors and their interaction.
We next evaluated models of local and landscape factors that might explain patterns of two key elements of the virus system: (i) Graminella abundance in early August (negative binomial distribution) and (ii) SwMV prevalence in switchgrass (exponential distribution with the single zero data value converted to 0.001). Abundances of the leafhoppers and virus prevalence both have the potential to be shaped by local stand conditions as well as by landscape-level supply and the extent of landscape provisioning of biocontrol. For Graminella abundance, we evaluated four relevant metrics of local site conditions (drought index, field size, height of switchgrass, temperature at time of collection). For SwMV prevalence in switchgrass, we considered three site properties (drought index, field size, stand height), three measures of vector abundance (abundances of G. aureovittata and of all Graminella in early August, and total August abundance of G. aureovittata), and early August measures of SwMV prevalence in G. aureovittata, G. oquaka/mohri, and in all Graminella. Finally, we evaluated the influence on both response variables (Graminella abundance and SwMV prevalence in switchgrass) of the proportions (within 0.5 km, 1.0 km, and 1.5 km buffers around each site) of three of the eight land cover groups previously described: (i) wetlands, (ii) grass/meadows, which might provide Graminella habitat and SwMV reservoirs, and (iii) agricultural cover, which likely would not. We also considered the contribution of two land cover groups whose proportions in 1.5-km buffers were associated with increased biocontrol in this region: (i) forests and (ii) an additional category of herbaceous perennial habitat (Werling et al., 2011b) that includes alfalfa, shrublands, clover/wildflower, and three cover types from the grass/meadows group (other hay, fallow/idle crop, and pasture/grass). Proportions were calculated as the proportion of all land cover in that buffer.

| Abundance of Graminella leafhoppers
The June-August time series of collections at four sites confirmed that Graminella abundance was greatest in late summer (Figure 3). In June, leafhoppers were captured at all four sites, but no Graminella were found, and in July, Graminella counts were low. Graminella abundance peaked at three of the four sites (SW02, SW07, SW10) in early August. At the remaining site (SW14), abundance was greatest in late August.
The extensive August collections across the full network of 15 switchgrass sites yielded more than 1218 leafhoppers in total, of which 914 individuals were Graminella (Figure 4a, Table S-2). Graminella were found at all sites, and this genus was the dominant taxon at most, comprising 40-100% of the leafhoppers in all but two of the collections (Figure 4b). Among the Graminella, the known SwMV vector G. aureovittata was much less abundant than its congeners G. oquaka and G. mohri (Figure 4a and c), representing ~4.8% of total Graminella captured across both dates. In the total August collection, we found no G. aureovittata at all at three sites (SW09, SW13, SW14) where other Graminella (N = 36-151 individuals) were collected.
Graminella were most abundant in early August, in which we caught 637 individuals (Figure 4a, Table S-1). Graminella were found at all sites except in one northeastern location (SW11, Huron County). At the other 14 sites, collection counts ranged from 2-162 individuals per 150 sweeps (median = 37). Of the Graminella caught across all sites, 96.9% (617/637) were G. oquaka/mohri and just 3.1% (20/637) G. aureovittata (Figure 4a and b, Table S-1). On a per-site basis, the numbers of G. aureovittata never exceeded those of G. oquaka/mohri and were generally much smaller (Figure 4a and b). The percentage of Graminella that were G. aureovittata thus ranged from a high of 50% at SW12 where only 2 Graminella were caught (one of which was a G. aureovittata) to 0% (4 sites), with a median value of 2.3%.
In late August, leafhoppers were less abundant overall and the total number we caught, as well as the number of Graminella, fell at most sites (median per-site decline −28% and −31% respectively) (Figure 4a and b). At the 11 collection sites, Graminella counts ranged from 3-96 per 150 sweeps (median = 19). G. aureovittata abundance remained low but did not decline, and we caught 24 individuals across 12 sites (median = 1).

| Switchgrass mosaic virus (SwMV) prevalence in switchgrass and Graminella
Reverse-transcription (RT)-PCR tests of 180 switchgrass samples and 413 individual leafhoppers revealed that SwMV was widely distributed across our study region. The virus was detected at 14 of the 15 switchgrass sites we sampled, either in switchgrass foliage, in leafhoppers, or in both (Figure 5a). Infection was found in switchgrass leaves at 11 of the 12 sites at which the species was sampled, with prevalence ranging 6.7%-60%. At one site with notable infection (Sw07), we were able to compare prevalence values from 2010 and 2012 and found little change (63.4% to 60.0%) (Figure 5b). Two of the three 2010 samples from big bluestem-a species not previously known to host SwMV-were infected as well (Figure 5b).
Graminella leafhoppers were caught in sweeps at 14 of the 15 sites but patterns of virus detection in them were bifurcated. At sites where virus prevalence in P. virgatum foliage was less than ~20%, we detected little to no virus infection in the leafhoppers, except at one site (SW14) where virus was found in all Graminella tested ( Figure  5a). In contrast, when foliar prevalence exceeded 20%, the majority of Graminella tested positive (75-100%). At two of the three sites where plant data were missing (SW17, SW18), a large proportion of Graminella were positive for virus, suggesting that prevalence in the P. virgatum was likely also notable.

| Best predictors of stand properties
The extent of premature stand senescence, reflecting stand condition in early August, was best predicted by SwMV prevalence, not by drought index, latitude or longitude, stand height (a measure of growth related to productivity), or multifactor models (Table 1, Figure 6). Stand height was not associated with SwMV prevalence or drought index in either early August (Table 2, Table S-3) or late August (data not shown).

| Influence of local site factors on vector and virus prevalence
Graminella abundance was best predicted by stand height in early August (Table 2, Table S -4). SwMV prevalence in switchgrass was negatively associated with field size and positively associated with SwMV prevalence in G. aureovittata, G. oquaka/mohri, or Graminella overall (competing models,

| Influence of land cover context
Land cover analysis showed the diversity of landscape contexts for the sites in this study. At the 1.5km-scale, agriculture was the largest category of land use for 7 of the 15 sites ( Figure S-4). Wetlands were dominant at three others, grasslands/meadows at two, forest at one, and the remaining two had notable developed land use nearby ( Figure S-4). However, neither Graminella abundance nor SwMV prevalence was predicted by proportions of any of the five land cover groups, representing possible habitat/ reservoirs or sources of biocontrol, which we had evaluated (Table 1 and Tables S-4 and S-5).

| DISCUSSION
Viruses cause the majority of emerging infectious diseases in plants (Anderson et al., 2004), and these diseases are likely to only increase in importance despite efforts to control them (Nicaise, 2014). At present, the leading driver of viral pathogen emergence is anthropogenic introduction of viruses to new hosts or regions, sometimes called 'pathogen pollution' (Anderson et al., 2004). Other current drivers include introduction of or increases in vector

G. aureovittata (n)
populations, altered agricultural practices, and virus evolution (Rojas & Gilbertson, 2008). Deeper in time, however, virus emergence was likely driven by human domestication of plants and the rise of agriculture. Gibbs et al. (2008), for example, found evidence that agriculture drove the emergence of potyviruses and their prevalence in crops. Recent geometagenomics analysis supports this idea, finding associations of several virus groups with agricultural land use . Because switchgrass is still close to its roots as a wild prairie grass, having experienced only a few cycles of selection for forage, conservation, and bioenergy (Parrish et al., 2012), its development as a bioenergy feedstock presents unique opportunities to watch domestication in action but also raises risk of driving new viral disease emergence. Our finding that the wild marafivirus, Switchgrass mosaic virus (SwMV), is well-established in Mid-Western USA agroecological landscapes raises crucial questions about its potential impact on bioenergy feedstock, its epidemiological drivers, and risk of spillover to other crops such as maize.

| SwMV infection is widespread
As of this writing, SwMV infection has been discovered in switchgrass in bioenergy plantings in four Midwestern US States: Illinois (Agindotan et al., 2010;Agindotan et al., 2013a), Michigan (two sites in this study), Missouri (Malmstrom, Lowry, et al., unpublished data), and Wisconsin (Agindotan et al., 2010) (Figure S-5). Our study is the first to examine the distribution and prevalence of SwMV in more naturalistic conservation plantings across a diverse agroecological landscape. We found SwMV infection to be ubiquitous in these systems with its prevalence reaching 30-60% in a quarter of the stands (Figure 5a) and persisting across years (Figure 5b). These findings demonstrate that this recently identified virus is not uniquely limited to a few bioenergy plantings but rather demonstrates characteristics of an established and endemic wild virus. This conclusion is reinforced by longitudinal studies in progress that document significant virus presence in stands over time (Malmstrom et al., unpublished data). Developing the understanding of SwMV ecology and epidemiology are thus important in assessing risk of significant disease emergence and impact.
In bioenergy trial plots in Illinois, Agindotan et al. (2013a) reported SwMV infection in ten different switchgrass cultivars (lowland and upland ecotypes), as well as in three other native grasses (A. virginicus, P. amarum var. amarum, S. nutans), and several introduced species (Miscanthus spp. and Saccharum ravennae), indicating that this virus is a multihost generalist, not a switchgrass specialist. Our findings expand knowledge of its host range to include A. gerardii (big bluestem), meaning that at least three of the four dominant species of North American tallgrass prairie (A. gerardii, P. virgatum, S. nutans) support SwMV infection. The fourth native dominant-Schizachyrium scoparium-has not yet been evaluated but may also prove to host SwMV because it belongs to the same Saccharinae subtribe (tribe Andropogoneae, subfamily Panicoideae) as three other hosts (Miscanthus, Sorghastrum, and Saccharum). It is possible that SwMV has been endemic in the tallgrass prairie for an extended period, but the extent of its influence requires further investigation. We speculate that at present the virus may be more common in the moist Eastern side of the tallgrass prairie region as infection has not yet been reported from virus surveys of switchgrass in drier regions, including Kansas (Malmstrom and Alexander, unpublished data) and Oklahoma (Muthukumar et al., 2009) (Figure S-5).

F I G U R E 6
Prevalence of SwMV infection is best predictor of stand senescence (see Table 1

| Potential impact on switchgrass
The impact of wild plant viruses on their natural hosts remains poorly understood. It is increasingly suggested that wild viruses serve as commensalists or mutualists, either little perturbing or benefitting their hosts (Fraile & García-Arenal, 2016;Roossinck & Bazán, 2017), particularly when infections are asymptomatic or latent (Takahashi et al., 2019). In switchgrass, however, SwMV infection frequently is symptomatic (Figure 1). Given its ubiquity, prevalence, and phylogenetic relatedness to the maize pathogen maize rayado fino virus (MRFV), SwMV merits consideration as a potential pathogen of note. To investigate effects of natural SwMV infection in the field, we examined relationships between SwMV prevalence and switchgrass height, as a proxy for stand growth, and between prevalence and the extent of stand-level senescence in early August, as a metric of growing-season condition and stress. In our area, switchgrass can remain green until hard frosts in October, so senescence in early August is ca. two months premature. We found no relationship between SwMV prevalence and stand height, suggesting that infection did not detectably limit initial canopy development (Table 2, Table S-3). Notably, SwMV prevalence was the best predictor of stand-level senescence (Table 1), with greater prevalence associated with greater senescence. These findings suggest two possibilities about the nature of SwMV influence on switchgrass. One possibility, in keeping with the hypothesis that wild viruses confer benefits on hosts, is that SwMV serves as a mutualist that permits infected plants to better tolerate drought-a frequently posited benefit of infection (Westwood et al., 2013;Xu et al., 2008). In this conceptual model, stands with greater senescence might have greater SwMV prevalence because infected plants were favored under drought stress and persisted while uninfected individuals succumbed. However, this scenario is not the most congruent with our data. For example, it would be most consistent with a statistical model that included prevalence, drought index, and their interaction, but the prevalence-only model (with intercept) was the AIC c -best fit (Table 1). Moreover, in this multistemmed perennial species, infected individuals do not vanish abruptly but rather senesce gradually, and we sampled senescing individuals. Therefore, even if senescence happened faster in uninfected plants (i.e., if infection increased stress tolerance), our measure of prevalence would probably not have been much influenced. Related hypotheses that drought or poor stand growth increased infection prevalence were not supported by statistical models.
A more straightforward explanation of the data is that SwMV is a pathogen that does not impede stand height gain but instead provokes premature senescence, perhaps exacerbated by drought. While this suggestion is at odds with the idea that wild viruses typically are not pathogenic in their natural hosts, it is supported by the frequent expression of symptoms in infected switchgrass. The extent to which premature senescence might translate to reduced bioenergy production depends on the interplay among its effects on biomass yield, nutrient resorption, and conversion efficiencies (Ong et al., 2018). While any optimization of harvest timing can be complex and specific to the conversion process (Ong et al., 2018), premature senescence shortens the growing period and is likely to reduce yield potential in all cases. Moreover, if premature senescence extends the time between senescence and harvest date, it provides additional opportunities for dry biomass to be lost to wind or drop to the ground beyond the reach of harvesting equipment (Adler et al., 2006;Anderson et al., 2013). Overall, we conclude that SwMV appears to be an endemic virus with capacity to achieve notable prevalence and potential to reduce yield quantity or quality in switchgrass. It merits careful attention in feedstock development and raises fundamental questions about factors that might allow maintenance of virulence in wild viruses.

| Ecology and epidemiology of SwMV
A broad literature documents the influence of surrounding landscapes on pest/natural enemy dynamics and implications for pest management (Bianchi et al., 2006;Gurr et al., 2017;Haan et al., 2020;Karp et al., 2018;Landis et al., 2000;Meehan et al., 2011;Werling et al., 2011aWerling et al., , 2014. In this study, however, we found local site factors to be better predictors of Graminella abundance or SwMV prevalence than any landscape elements reflecting habitat or reservoir opportunities, or biocontrol sources. Graminella abundance was predicted only by the local factor of switchgrass stand height. None of the other local factors or any of the landscape factors, each with potential to influence vector abundances, was found to be influential. Studies of leafhopper responses to local and landscape factors in other systems reveal a range of relationships that differ among species, with some likewise demonstrating no clear associations with landscape factors (e.g., Vaidya et al., 2017). In our system, Graminella abundance may increase with stand height because of the increased structural complexity of the vegetation or changes in microclimate; Graminella nymphs, for example, seem to prefer the shade within deeper canopies (E. Cole, personal observation). Alternatively, Graminella abundance might increase with stand height to the degree that height reflects stand productivity and carrying capacity.
SwMV prevalence in switchgrass was best predicted by the proportions of Graminella leafhoppers testing positive for the virus (several positive relationships), and by field size (a negative relationship), but not by Graminella abundance. The negative relationship with field size is counterintuitive, and we suspect that field size is serving indirectly as a measure of time since stand establishment because the largest stands in our study were the most recently established. Since infection can overwinter in rhizomes (Ryskamp et al., in prep.) and persist for several years as evident at one of our sites, prevalence might be expected to accumulate over time. However, we could not more precisely test the influence of stand age because the specific establishment years of the older stands were unknown.
Competing models of switchgrass infection prevalence that contained leafhopper factors included prevalence in Gr. aureovittata (a known vector), in Gr. oquaka/mohri (one designated non-vector, one untested), and in all Graminella (Table 2). This result mirrors the identification of vector infectivity proportions as critical parameters in disease risk assessment in agriculture (e.g., Frost et al., 2013). The congruity of virus prevalence in leafhoppers and plants underscores the biological linkage between the two populations and suggests that sampling either one can provide useful information about infection prevalence within a stand. Insects have proven to be valuable integrators of virus signals within plant communities in both vector-and predator-enabled metagenomics (Ng et al., 2011;Rosario et al., 2013Rosario et al., , 2015. It is possible that laboratories experienced with insect identification might find testing leafhoppers for virus to be simpler than working with plant samples, which requires overcoming issues with tissue toughness and biochemistry (Lacroix et al., 2016). Interestingly, both our study and that of Agindotan et al. (2013b) found that in switchgrass stands with notable infection (>20%), the virus was detected in a greater proportion of leafhoppers than of plants, whereas in stands with lesser levels of infection (≤20%) the opposite was true: we generally, but not always, detected the virus less frequently in leafhoppers than in plants (low prevalence stands were not evaluated in Agindotan et al.). For disease monitoring, these results imply that detection of notable SwMV prevalence in Graminella indicates high likelihood of notable infection within the stand itself.
For epidemiological analysis, a key question is the degree to which virus signal detected in different Graminella species reflects their capability to transmit the virus (either effectively or in a limited manner), or merely reflects ingestion of virus particles. One species we sampled, G. aureovittata, was previously found to transmit SwMV while G. mohri was not (Agindotan et al., 2013b). The transmission efficiency of G. oquaka remains untested and merits attention. In our study, G. aureovittata comprised only 3.1% of the Graminella population, leading us to wonder whether the more abundant G. oquaka/mohri group might contribute to infection spread. We could not determine the relative proportions of G. oquaka and G. mohri in our study because the two species look highly similar and the destructive identification measures needed to distinguish them were incompatible with virus testing. We therefore recommend expanded testing of the transmission efficiencies of both species and suggest that particular attention should be paid to nymphs, which are the most efficient vectors of other Marafiviruses (Nault & Ammar, 1989). Alternatively, if G. aureovittata proves to be the primary vector, its low numbers in 2012 may have been a short-term anomality caused by the drought, as DeLong (1948) reported that this species has greater affinity for damp environments than G. oquaka or mohri.
The lack of detectable effects of landscape cover (0.5-1.5 km distances) on Graminella abundance and SwMV prevalence is intriguing. We considered the influence of two land covers that might harbor Graminella and/or Poaceae hosts of SwMV (wetlands and grass/meadow) and one (agriculture) that likely would support neither. We also evaluated the influence of two potential sources of biocontrol: forest cover and perennial herbaceous cover (Werling et al., 2011b). In contrast to the local predictors, none of these landscape metrics showed any significant relationship with Graminella abundance or SwMV prevalence and did not contribute to the AIC c -best fit models. However, there was a marginal (p = 0.068) negative effect of forest area within a 1.5-km buffer on SwMV prevalence (but not Graminella abundance) worth future investigation. More generally, the lack of significant landscape signal at the scales we considered, along with the widespread finding of SwMV infection, suggests that SwMV and Graminella may be broadly dispersed across this landscape with site conditions serving as modulators that amplify or diminish their presence. As winged insects, leafhoppers can be widely distributed and "rain" across many vegetation types within a landscape (e.g., Keene et al., 2020). Moreover, the natural pest suppression supply generated by forests and herbaceous perennial landscapes (Werling et al., 2011a) may be ineffective at controlling leafhopper populations.

| Implications for disease emergence
Our data indicate that SwMV deserves attention as a potential driver of yield or quality loss in switchgrass and a possible emergent pathogen in feedstock development. Selection of native plant material for production (domestication) may inadvertently increase virus susceptibility (Schrotenboer et al., 2011), although not always (Nygren et al., 2015). The phylogenetic relatedness of SwMV to the maize pathogen maize rayado fino virus (MRFV) indicates the need to consider the risk of a host jump by SwMV to maize or related crops. The factors currently limiting spread of SwMV infection to maize are not known but may reflect vector distributions. MRFV is not only transmitted primarily by Dalbus maidis but also by the widespread Graminella nigrifrons, an abundant herbivore of Poaceae in the Eastern US (DeLong, 1948). SwMV is transmitted by at least one Graminella species and potentially others. Given the capacity for RNA virus evolution, it is important to consider the possibility that widespread planting of switchgrass might create opportunities for SwMV to develop capacity for transmission by other leafhoppers, including Gr. nigrifrons, and thus potentially to infect maize. Identification of resistance to SwMV and selection for it during feedstock development might reduce risk of these scenarios.