Edaphic correlates of feedstock‐associated diazotroph communities

Miscanthus × giganteus and Panicum virgatum are potential promising bioenergy feedstock crops suitable for the temperate zone. The energy efficiency and sustainability of bioenergy production could be improved by reducing their fertilizer inputs – particularly energy intensive nitrogen fertilizers. Miscanthus is known to benefit from nitrogen fixation by associative diazotrophs. However, because the effects of edaphic‐, management‐, and plant‐related factors on feedstock‐associated diazotroph communities have not yet been characterized, it is not currently possible to optimize the nitrogen contribution to feedstock crops from associated diazotroph communities. To address this critical knowledge gap, we characterized the bacterial and diazotroph communities in the rhizosphere and endophytic compartments of both species at eight research sites across Illinois. We also quantified the nifH gene abundance in the rhizosphere soil as well as a range of soil chemistry parameters at these sites. Multivariate statistical analyses revealed that diazotroph and bacterial communities in the rhizosphere varied primarily among sites, with very small differences between host species. Conversely, diazotroph and bacterial communities in the endophytic compartments differed significantly between plant species, but did not vary substantially among sites. Finally, nifH gene abundance in the rhizospheres of both species varied substantially from site to site and was positively correlated with soil iron concentration as well as soil ammonium concentration, and negatively correlated with abundance of other soil nutrients including calcium, total nitrogen, and nitrates. These results indicate the potential edaphic drivers of associative diazotroph communities in feedstock rhizospheres and suggest that manipulating bioavailable iron content in the soil is a potential direction for investigating the optimization of these communities to improve their nitrogen contribution to crops.


Introduction
Diazotrophs are bacteria and archaea that convert atmospheric nitrogen to ammonia, thus making it bioavailable to other life forms. Communities of diazotrophs in soils and in plant tissues could play an integral role in enhancing the sustainability of biofuels production by reducing the nitrogen fertilizer needed for producing feedstock crops. Biofuels are expected to constitute a growing portion of the global energy infrastructure due to government mandates and incentives (Dimaranan & Laborde, 2012;Timilsina, 2014). Cellulosic biofuels, such as those from Miscanthus 9 giganteus (miscanthus) and Panicum virgatum (switchgrass), have the potential to have significantly lowerperhaps even negative -carbon intensity compared with conventional biofuels (Qin et al., 2006;Adler et al., 2007;Davis et al., 2009;Dimaranan & Laborde, 2012;Sanscartier et al., 2014;Timilsina, 2014). However, nitrogen fertilizers, which can improve feedstock yields (Arundale et al., 2014a), constitute significant energy expenditure in feedstock production. Reducing the need for nitrogen fertilizer can therefore substantially decrease the net carbon intensity of biofuels (Melillo et al., 2009;Davis et al., 2011). Improving the nitrogen contribution from associative diazotroph communities to feedstock crops is also desirable for enabling cellulosic feedstock production on low-nutrient soils (Chakraborty, 2014;Lowman et al., 2015) like nonprime farmland and marginal land, thereby reducing the competition between biofuels and food crops for prime farmland (Hardy et al., 2007;Saraf & Hastings, 2010).
Diazotroph communities are associated with the rhizosphere and endophytic compartments of miscanthus and switchgrass, similar to many other grasses (Bagwell et al., 1998;Erica et al., 2014;Gupta et al., 2014;Sarathambal et al., 2015), sugarcane (Thaweenut et al., 2010;Taul e et al., 2011), and cereals (Prakamhang et al., 2009;Venieraki et al., 2010;Rangjaroen et al., 2013;Rodr ıguez-Blanco et al., 2015). The composition of diazotroph communities in the rhizosphere and the endophytic compartment varies with edaphic and management variables, as well as between hosts. We have demonstrated that for genetically identical M. 9 giganteus cultivated in the United States across large geographic distances, and diazotroph community structure in the rhizosphere was affected by edaphic variables, while communities in the endophytic compartment remained largely similar across sites . Among diazotroph communities associated with native and uncultivated relatives of M. 9 giganteusnamely M. floridulus and M. sinensisrhizosphere communities have been shown to be affected largely by edaphic factors in the rhizosphere but endophytic communities to vary primarily between host plant species in the endophytic compartment (Li, 2015). We have also shown that 16% of the nitrogen in the first-year growth of M. 9 giganteus was derived from biological nitrogen fixation (BNF) (Keymer & Kent, 2014). Optimizing feedstock-associated diazotroph communities to increase BNF contribution to the 80% common in forage legumes (Carlsson & Huss-Danell, 2003) may be unrealistic. However, the 60% maximum BNF from associative diazotrophs observed in sugarcane (Taul e et al., 2011;Urquiaga et al., 2011), or 70% in elephant grass  suggest that there could be substantial room for improving the BNF contribution from feedstockassociated diazotroph communities. While BNF is conventionally expected from soil-dwelling diazotrophs, it has also been observed in the endophytic compartment of many plant species including wild rice (Elbeltagy et al., 2001), sugarcane (Boddey et al., 2003), grasses (Dalton et al., 2004), and switchgrass (Lowman et al., 2015). Together, these previous findings suggest that in order to enhance the nitrogen efficiency of cellulosic biofuels production, and it may be possible to optimize diazotroph communities in the feedstock rhizosphere by manipulating edaphic variables such as micronutrient concentration (Fe, Ca) organic matter content, and pH as well as conceivably those in the endophytic compartment through targeted plant breeding and engineering.
The essential first step toward optimizing feedstockassociated diazotroph communities is a thorough understanding of the effects of edaphic, biotic, and management variables on these communities. The structure and function of soil microbial communities are known to be affected strongly by edaphic variables including pH, soil organic matter, and soil type (Girvan et al., 2003;Lauber et al., 2009;Fierer et al., 2012;Kuramae et al., 2012;Schreiter et al., 2014). Nitrogen application (Wakelin et al., 2007;Shu et al., 2011;Ste z pie n et al., 2014), plant species (Mao et al., 2011(Mao et al., , 2013Li, 2015), and management practices like soil amendments, residue retention, and pasture grazing (Wakelin et al., 2007(Wakelin et al., , 2009 are also known to affect diazotroph abundance and diversity. Therefore, it is reasonable to expect that edaphic and management variables can affect rhizosphere feedstock-associated diazotroph communities. To identify potential edaphic drivers of variation in diazotroph communities associated with M. 9 giganteus and P. virgatum, we characterized the structure of the associated diazotroph and bacterial communities at eight research sites across Illinois with variation in soil chemical properties. We quantified the soil chemical parameters at each site and the abundance of the nifH gene in the rhizosphere. Additionally, variability in these communities was characterized across host species and habitats (i.e., rhizosphere vs. endophytic compartment). Our findings reveal significant variation in diazotroph community structure correlated with variation in soil chemical parameters. Furthermore, diazotroph abundance in the rhizosphere, as indicated by nifH gene abundance, was found to be strongly correlated with soil iron and ammonium concentrations. These findings indicate potential directions for further investigating the parameters related to optimal feedstock-associated diazotroph communities and could, over the long-term, inform site selection and optimization to enhance beneficial diazotroph communities in the biofuels feedstock rhizosphere.

Site description
The research sites for this study have been described in detail elsewhere (Heaton et al., 2009;Arundale et al., 2014a). In brief, field trials were established in 2002 (Dekalb, Urbana, Dixon Springs), 2004 (Havana, Orr, Brownstown, Fairfield), and 2005 (Dudley Smith Farm) with the locations chosen to represent the variety of soil chemical parameters and climate across Illinois. Each field trial consisted of randomized complete block design with four 100 m 2 plots each of M. 9 giganteus (var. Illinois) and P. virgatum (var. Cave-in-Rock) (except for Dudley Smith Farm, which consisted of four 0.2 ha plots of the two plant species), and four nitrogen application rates from 0 to 200 kg hectare À1 . Soil characteristics, agronomic management, and yield from these plots have been described previously (Heaton et al., 2008(Heaton et al., , 2009Arundale et al., 2014a). A small number of miscanthus samples were collected from subplots that received fertilizer application at 200 kg hectare À1 and switchgrass subplots received fertilizer applications ranging from zero to 13 kg hectare À1 . Key soil chemistry and site parameters are tabulated in Table S1.

Sample collection
Soil and root samples were collected in August 2009. Three rhizome samples with adhering rhizosphere soil were collected for switchgrass and miscanthus from each of the four unfertilized subplots at each site for a total of 96 rhizosphere samples from unfertilized switchgrass and miscanthus plots and an additional 15 samples from fertilized miscanthus plots. Six bulk soil cores for quantifying soil chemical parameters were collected to 6″ depth from each subplot. All soil cores and plant tissue samples were placed in sealed plastic bags on ice while in the field and transported back to the laboratory.

Sample processing
Representative subsamples from each sampled plant rhizome were aseptically transferred to sterile 1 l Nalgene bottles with sterile water. Adhering rhizosphere soil was eluted from the root surface by vigorously shaking the Nalgene bottles. The eluent soil slurry was collected and stored at À20°C until it could be lyophilized. Extraction of endophytic community DNA was carried out as described by Keymer & Kent (2014), a modification of the process described by Brulc et al. (2009). Rhizomes were washed vigorously with distilled water, placed in 1 l Nalgene bottles with sufficient 5.25% sodium hypochlorite solution and shaken for 15 min, and subsequently washed twice with sterilize distilled water. The cleaned tissues were homogenized in a blender with 40 ml phosphate-buffered saline + 0.15% with Tween 80. The homogenized slurry was poured into 50 ml tubes with glass beads and shaken on ice for 1 h. The extract was passed through a coarse sieve (No. 25 USA standard test sieve, pore size 710 lm; Newark Wire Cloth Company, Clifton, NJ, USA) to remove large pieces of plant tissue, and finer plant debris was removed by centrifugation for 5 min at 453 9 g. The resulting supernatant was centrifuged at 3900 9 g for 10 min in a Sorvall RC-5B refrigerated superspeed centrifuge with SS-34 rotor (DuPont) to recover endophytic bacterial cells in the reserved pellet.

DNA extraction
DNA from the endophytic microbial communities was extracted from the pellet resuspended in sterile water using the FastDNA kit from MP Biomedicals (Solon, OH). DNA was extracted from lyophilized rhizosphere soil samples using the FastDNA for Soil kit from MP Biomedicals (Solon, OH). The extracted DNA samples were purified using CTAB (cetyltrimethylammonium bromide) and 1 : 24 chloroform : isoamyl alcohol followed by ethanol precipitation to remove humic contaminants. Extracted and purified DNA was quantified using NanoDrop 1000 (Nanodrop Technologies, LLC, Wilmington, DE). The DNA was stored at À20°C until used for ARISA, T-RFLP, and nifH qPCR analyses as described below.

Nitrogenase gene abundance
As a marker of nitrogen fixation potentially available to the feedstock crops, quantitative polymerase chain reaction was used to measure abundance of the nifH gene in the rhizosphere. nifH was amplified using the primers PolF (5 0 -TGCGAYCCSAARGCBGACTC-3 0 ) and PolR (5 0 -ATSGCC ATCATYTCRCCGGA-3 0 ) (Poly et al., 2001). The 25 ll reaction volumes were prepared using 19 Sybr Green PCR MasterMix (Applied Biosystems, Foster City, CA, USA), 400 nM of each nifH primer, 20 ng of template DNA, and 500 lg ml À1 bovine serum albumin. The thermal cycling program consisted of an initial denaturation at 94°C for 5 min, 40 cycles of 94°C for 30 s, annealing temperature for 30 s, 72°C for 30 s. Annealing temperatures were stepped down during the initial cycles (1 cycle at 64°C, 2 cycles at 62°C, 3 cycles at 60°C, and 4 cycles at 58°C, and final 30 cycles at 56°C) to improve specificity. Negative controls and duplicate standard curves generated by serial dilution of purified PCR products were included in each plate. Copies of nifH were normalized on per nanogram of DNA determined by template DNA quantification using the Quant-iT TM dsDNA HS Assay Kit with a Qubit â fluorometer (Invitrogen/Molecular Probes, Eugene, OR, USA).

Diazotroph community structure
The effects of site variables, plant species, and plant habitat on the community of nitrogen fixing bacteria were assessed with terminal-restriction fragment length polymorphism (T-RFLP) analysis of the nitrogenase gene. The gene was amplified from the extracted and purified microbial DNA using primers PolF (5 0 -FAM-TGCGAYCCSAARGCBGACTC-3 0 ) and PolR (5 0 -HEX-ATSGCCATCATYTCRCCGGA-3 0 ), which generate amplicons of approximately 360 base pairs (sequence position 115-476 relative to A. vinelandii nifH coding sequence (GenBank accession number M20568)) (Poly et al., 2001). The PolF/PolR primers were chosen for effective performance for fragment analysis as well as qPCR for the communities of interest in agricultural soils (Gaby & Buckley, 2012. PCRs were conducted as described in (Keymer & Kent, 2014). The resulting amplicons were purified with MinElute 96 UF PCR Purification Kit (Qiagen, Valencia, CA). Purified amplicons were digested for 16 h at 37°C using the restriction enzyme MboII and MnII following the procedures recommended by New England Biolabs (Beverly, MA) (Keymer & Kent, 2014). The lengths of digested gene fragments were quantified by denaturing gel capillary electrophoresis at the W. M. Keck Center for Comparative and Functional Genomics, University of Illinois, Urbana-Champaign, with the ABI 3730xl Genetic Analyzer (PE Biosystems, Foster City, CA, USA). ROX 1000 size standard (Applied Biosystems, Foster City, CA, USA) was used during capillary electrophoresis. Fragment size was determined for the assay as described below.

Microbial community structure
To understand the effects of site characteristics, plant species, and habitat on the bacterial community, we analyzed the community structure in each sample with automated ribosomal intergenic spacer analysis (ARISA) (Fisher & Triplett, 1999). The bacterial rRNA operon intergenic transcribed spacer (ITS) region was amplified from extracted and purified microbial DNA with PCR using 5 0 6-FAM labeled forward primer 1406f (5 0 -TGYACACACCGCCCGT-3 0universal, 16S rRNA gene), and reverse primer 23Sr (5 0 -GGGTTBCCCCATTCRG-3 0bacteria-specific, 23S rRNA gene). Denaturing capillary electrophoresis of the amplicons was performed at the W. M. Keck Center for Comparative and Functional Genomics, University of Illinois, Urbana-Champaign, IL. The ROX 1000 size standard (Applied Biosystems, Foster City, CA, USA) was used during capillary electrophoresis.

Fragment size analysis
Sizes of the ITS and T-RFLP amplicons were determined in GEN-EMARKER software (Version 2.4, SoftGenetics, LLC, PA). Minor run-to-run variations in observed vs. actual fragment length that results from capillary electrophoresis were resolved using the allele-calling features in GENEMARKER before analysis. To include the maximum number of peaks while excluding background fluorescence, a threshold of 200 fluorescence units was used. The area of each peak was normalized to account for run-to-run variations in signal detection by dividing the area of individual peaks by the total fluorescence (area) detected in each sample, expressing each peak as a proportion of the observed community (Yannarell & Triplett, 2005).

Soil chemical analysis
Soil chemistry parameters, including pH, total nitrogen, NO 3 À nitrogen, NH 4 + -nitrogen, total carbon percent, total organic matter, and Mehlich-3 phosphorus, potassium, calcium, and iron (Mehlich, 1984), were quantified from bulk soil cores that were obtained at the same time as the rhizosphere and plant tissue samples. Analysis was carried out by the Iowa State Soil and Plant Analysis Laboratory (Ames, IA). Soil moisture was quantified in-house with gravimetric soil moisture analysis, where samples were weighed before and after oven drying at 60°C for 24 h.

Statistical analyses
ARISA, T-RFLP, qPCR, and edaphic data of each of the samples were analyzed statistically using multivariate correlational and ordination methods in R (R Core Team, 2017) and with R packages vegan version 2.4-2 (Oksanen et al., 2017) and labdsv version 1.8-0 (Roberts, 2016). Multiple samples obtained from each 100 m 2 plot were treated as independent samples rather than technical replicates, to capture within-plot variation in the communities. Significant local variation in the communities is expected within these large experimental plots due to withinplot differences in edaphic conditions (Tripathi et al., 2014;Livermore & Jones, 2015;Nunan, 2017). The correlation of nifH abundance in the rhizosphere with environmental and soil chemical parameters was assessed with Pearson's product-moment correlation coefficient (Puth et al., 2014). The effect size and statistical significance of site, feedstock species, and habitat on the structures of the microbial and diazotrophic communities were determined with permutational multivariate analysis of variance (PERMANOVA) (Anderson, 2005). Community structure (ARISA and nifH T-RFLP) was analyzed with partial correspondence analysis (CA) (Legendre & Legendre, 2012) and visualized as ordinations to facilitate the assessment of community similarities. The correlation of soil chemistry parameters with community structure variation across sites was quantified with the function envfit and plotted on the ordinations, as well as tabulated (Table 2).

Diazotroph communities in the rhizosphere
The variation in the structure of diazotroph communities in the feedstock rhizosphere was influenced by site, and differed by a very small, though statistically significant, extent between the host plant species (Table 1). The interaction of site and host species also had a small, but significant effect on the variation in rhizosphere diazotroph communities. Due to the small size of the effect of the host species, on rhizosphere community structure, we describe below (Tables 1 and 2; Fig. 1) the communities in the miscanthus and switchgrass rhizosphere together. Variation in diazotroph community structure was also correlated with edaphic factors (Table 2; Fig. 1a). The correlation with site latitude reflects the effect of the combination of edaphic factors on the site-to-site variation in rhizosphere diazotroph community structure. Among specific soil chemical properties, the ones best correlated with the variation in diazotroph community structure were phosphorus, iron, and potassium. Notably, total nitrogen content was not correlated with variation in diazotroph community structure, while active nitrogen species exhibited small amount of correlation. Finally, nifH abundance was significantly correlated with variation in community structure. The strong statistical correlation of phosphorus concentration with diazotroph community structure may be due to the strong gradient in phosphorus concentrationfrom 48.75 to 5 ppm, across our study sitesa greater percent difference than any other soil chemical parameter. The graphical representation of the edaphic factors ( Fig. 1) is useful to depict their relative trends with respect to each other across the geographic area studied (e.g., nifH abundance and Fe concentration are positively correlated with each other, but negatively correlated with total N and most other soil chemical parameters), the tabulation is provided for the quantitative understanding of strength and significance of these correlations with the structure of the communities.

Bacterial communities in the rhizosphere
The variation in the structure of bacterial communities in the feedstock rhizosphere was influenced by site, and differed by a very small, though statistically significant, extent between the host plant species (Table 1). The interaction of site and host species had a small effect on the variation between rhizosphere diazotroph communities. Among edaphic and environmental factors, site latitude was strongly correlated with the variation in bacterial community structure (Table 2; Fig. 1b), reflecting the effect of the combination of edaphic factors on the site-to-site variation of bacterial communities. Among specific soil chemical properties, the ones most correlated with the variation in bacterial community structure were pH, phosphorus, and calcium. In contrast with diazotroph communities, total nitrogen content as well as nitrogen species was correlated with variation in bacterial community structure. nifH abundance was also significantly correlated with variation in bacterial community structure.

Communities in the endophytic compartment
Diazotroph communities in the endophytic compartment (Table 1; Fig. 2) varied between the host species as well as among sites. Endophytic bacterial communities were dissimilar between the two host species and were affected to a smaller extent by site. As expected, habitat (rhizosphere vs. endophytic compartment) was a significant factor in the variation in community structure for diazotrophs as well as bacterial communities.

Nitrogenase gene abundance in the rhizosphere
Abundance of nifH gene in the feedstock rhizosphere revealed interesting patterns of correlation with edaphic factors (Table 3; Figures S1 and S2). Notably, nifH abundance was greater in sites with greater iron content and was also correlated with greater concentration of ammonia in the soil. Conversely, sites with greater nitrate concentration exhibited lower nifH abundance, though the correlation was less than statistically significant for communities in the miscanthus rhizosphere. Similarly, nifH abundance was generally lower in the miscanthus plots that received nitrogen fertilizer treatment. Greater concentration of conventional markers of soil fertilitysuch as soil organic matter, total nitrogen, potassium, and phosphorus, as well as calciumwas also correlated with lower nifH abundance. The marked northsouth variation in soil chemical properties is reflected in the correlation of nifH abundance with site latitude.
Comparing across host species, nifH gene abundance was not significantly different in the rhizosphere of the two hosts (P = 0.24).

Discussion
Informed site selection and soil amendments appear to be promising tools for optimizing rhizosphere diazotrophic communities and thereby improving the sustainability of cellulosic biofuels production. The systematic analysis of the differences in the correlations of edaphic factors with the variations in overall bacterial communities vs. diazotroph communities is a novel and notable finding of this work. For future research, it will be important to take into account the distinct responses of diazotroph communities and total bacterial communities to edaphic factors. The quantification of absolute and relative differences in the correlations of edaphic factors with community structure variation was possible due to the strong north-south gradient in soil chemical parameters in Illinois which stems from its distinctive paleogeography. The greater age and weathering of the southern and western sites in this study -Dixon Springs, Brownstown, Fairfield, and Orrcorresponds to their greater iron content and lower calcium, carbon, nitrogen, phosphorous, and potassium content (Martinson et al., 1987;Curry & Baker, 2000). Long-term weathering of soils generally increases their extractable iron content (Arduino et al., 1986), while calcium carbonate content of soils nifH T-RFLP and ARISA, respectively. The ordination controls for site-to-site variation to highlight the effect of habitat and host species (M. 9 giganteus -MxG, Panicum virgatum -PV). The difference in community structure between rhizosphere communities of the two host species is small, though statistically significant. (Table 1 lists effect sizes.) (n = 207 rhizosphere samples, n = 202 for endophytic compartment samples). decreases over time. In line with the massive body of literature evidence (Berge et al., 1991;Girvan et al., 2003;Fierer & Jackson, 2006;Wakelin et al., 2010;Fierer et al., 2012;Kuramae et al., 2012;Schreiter et al., 2014), the variation in rhizosphere bacterial community structure was strongly correlated with the measured soil chemical parameters, with pH being the factor best correlated with variation in bacterial community structure. The overriding effect of site variables on the feedstock rhizosphere microbiomerelative to the effect of plant speciesis in agreement with previous findings by our group (Li, 2015;Li et al., 2015) and others (Girvan et al., 2003;Kuramae et al., 2012;Schreiter et al., 2014). However, the quantified correlation of individual soil chemical parameters with diazotroph community properties presented here offers novel directions for future research into optimization of feedstock production. It is particularly notable that soil iron concentration was relatively more strongly correlated than most other edaphic factors with the site-to-site variation in the structure of diazotroph communities, than to the site-tosite variation among total bacterial communities. As the cofactor in nitrogenase, iron is critical for its functioning (Shah & Brill, 1977;Chan et al., 1993;Hinnemann & Nørskov, 2004) and consequently can be a limiting factor for diazotroph communities in iron-limited conditions. Lack of iron is known to affect nifH expression and nitrogenase activity in several diazotrophic microorganisms including K. pneumoniae (Miller & Orme-Johnson, 1992), C. anabaena (Fillat et al., 1995), and H. seropedicae (Rosconi et al., 2006). The effect of the variation in soil iron concentration on diazotroph communities was also evident in its correlation with nifH abundance. Iron bioavailability appears to be an important driving factor behind the variation in nifH abundance, given the strong positive correlation of nifH abundance with soil iron and negative correlation with soil calcium and pH. Furthermore, the abundance of nifH was also positively correlated with soil ammonium concentration, but negatively correlated with fertilizer input, total N, and nitrate concentration, suggesting a causal link between greater abundance of nifH and greater nitrogen fixation at these sites. While total iron content in most soils is generally not a limiting factor, bioavailability of soil iron is facilitated by longer weathering of parent material as well as pH below about 5.5which in turn is often associated with lower calcium carbonate content that can result from weathering (Rai, 1988;Loeppert et al., 1994). In addition, the low nitrogen content and low soil fertility likely also confer an additional competitive disadvantage for nondiazotrophs at these sites. The four sites with the highest iron content also exhibit the greatest rhizosphere nifH abundancesuggesting that geological history could be employed as a powerful heuristic for identifying feedstock production sites with greater potential for hosting beneficial diazotroph communities. An interesting outlier site to consider is Havana. The soil at the Havana research site is sandy with very low organic matter content. Even though Havana soils have low nitrogen and calcium content, rhizosphere communities from Havana research site also exhibit one of the lowest nifH abundance and ammonium content. This suggests that sites with organic carbon below a certain threshold may not be able to support significant populations of diazotroph, or even bacteria in general. The variation in nifH abundance in the rhizosphere implies that the relative contribution of diazotrophs to nitrogen requirements, and thus potentially on cellulosic feedstock production efficiency may also vary from site to site. Our previous study (Keymer & Kent, 2014), which demonstrated a 16% contribution from BNF to the nitrogen uptake during the first year of M. 9 giganteus growth, was conducted at Urbanathe site with the highest average soil N content and lowest average nifH abundance. It is plausible that the contribution of BNF to feedstock growth would be greater in sites with higher nifH abundance.
Notably, the yields of M. 9 giganteus and P. virgatum (data not shown, summarized in (Arundale, 2012)) across sites were found to not be significantly correlated with nifH abundance or with soil nutrient values including those of total nitrogen, ammonium, and nitrates at these sites. We hypothesize that greater nifH abundance compensates, in large part, for the site-to-site variation in nitrogen and mitigates the effect of low soil fertility on feedstock yield. Nitrogen fertilization has been found to have conflicting effects on miscanthus productivity. Across the research sites covered in this study, with stand age greater than 5 years, nitrogen fertilization at 202 kg ha À1 was found to be correlated with 25% increase in M. 9 giganteus yield relative to unfertilized plots (Arundale et al., 2014b). However, a longterm trial of nitrogen inputs between zero and 120 kg ha À1 was found not to affect miscanthus yields for 14 successive harvests (Christian et al., 2008). Efficiency analysis has revealed that nitrogen use efficiency and energy use efficiency are simultaneously highest for miscanthus when no fertilizer was applied (Lewandowski & Schmidt, 2006). The high energy and nitrogen efficiency of miscanthus appear to be due to high levels of BNF associated with miscanthus as revealed by modeling (Davis et al., 2009) as well as experimental (Keymer & Kent, 2014) studies. On the other hand, in previous studies involving switchgrass, nutrient balance modeling (Davis et al., 2009) did not suggest a significant contribution of BNF to nitrogen balance, and optimum biomass yields were achieved at the 120 kg ha À1 nitrogen fertilization (Vogel et al., 2002). It is plausible that the differences between the contribution of BNF to miscanthus and switchgrass can be attributable in part to interspecies differences in associating productively with rhizosphere and endophytic diazotrophs.
In the endophytic compartments of M. 9 giganteus and P. virgatum, the community structure of bacteria and diazotrophs is distinct between the two species, with a substantially smaller site-to-site variationsimilar to previous reports by us and others (Rangjaroen et al., 2013;Li, 2015;Li et al., 2015;Rodr ıguez-Blanco et al., 2015), suggesting that interspecies physiological differences play a role in shaping these distinct endophytic communities.
In the context of prior findings, the results presented here indicate that there may be an advantage to prioritizing feedstock production at sites that are low-fertility for conventional agriculture, but have a high potential for supporting beneficial associative diazotroph communities. Reducing the need for fertilizers would simultaneously reduce potential nitrate leaching (Behnke et al., 2012) as well as improve the carbon intensity of biofuels production (Lewandowski & Schmidt, 2006). Improving the productivity of biofuels feedstock production on low-fertility land would also minimize competition with production of food crops (Hardy et al., 2007;Saraf & Hastings, 2010). In addition to site selection, the findings here also suggest a potential for shaping the diazotroph communities through specific soil amendments that are correlated with potential desirable outcomes such as greater nifH abundance. Sites that are otherwise suitable for feedstock production but have low bioavailable iron could be amended to increase iron content and reduce soil pH in order to promote an increase in the population of diazotrophic microbes, potentially increasing the productivity of bioenergy feedstock crops.

Conclusions
This study demonstrates that soil chemical properties, particularly iron and calcium concentration, are significantly correlated with nitrogenase gene abundance as well as rhizosphere microbial community structure across a large geographic scale. These findings imply that informed site selection as well as soil amendments that improve associative diazotroph communities could improve the sustainability of biofuels feedstock production. The potential enhancements in cellulosic biofuel feedstock production from such efforts may improve the economic feasibility and productivity of feedstock farming on nonprime land, reduce potential competition for prime farmland with food and commodity crops and thus improve the energy infrastructure without adversely affecting food prices.