Phylogenetic farming: Can evolutionary history predict crop rotation via the soil microbiome?

Abstract Agriculture has long employed phylogenetic rules whereby farmers are encouraged to rotate taxonomically unrelated plants in shared soil. Although this forms a central tenet of sustainable agriculture, strangely, this on‐farm “rule of thumb” has never been rigorously tested in a scientific framework. To experimentally evaluate the relationship between phylogenetic distance and crop performance, we used a plant–soil feedback approach whereby 35 crops and weeds varying in their relatedness to tomato (Solanum lycopersicum) were tested in a two‐year field experiment. We used community profiling of the bacteria and fungi to determine the extent to which soil microbes contribute to phenotypic differences in crop growth. Overall, tomato yield was ca. 15% lower in soil previously cultivated with tomato; yet, past the species level there was no effect of phylogenetic distance on crop performance. Soil microbial communities, on the other hand, were compositionally more similar between close plant relatives. Random forest regression predicted log10 phylogenetic distance to tomato with moderate accuracy (R 2 = .52), primarily driven by bacteria in the genus Sphingobium. These data indicate that, beyond avoiding conspecifics, evolutionary history contributes little to understanding plant–soil feedbacks in agricultural fields; however, microbial legacies can be predicted by species identity and relatedness.


| INTRODUC TI ON
Sustainable farming methods often mimic patterns and processes that are characteristic of natural ecosystems with the assumption that unmanaged wild communities have undergone intense selection over evolutionary time, weeding out poor designs in favor of superior ones (Altieri, 1995;Denison, 2012). Thus, understanding how natural communities are structured and identifying the components that are disrupted by modern agricultural practices may offer novel insight on how to restructure cropping systems to enhance production (e.g., higher yield, greater water use efficiency, fewer inputs of pesticides, and/or fertilizer).
Perhaps the most dramatic difference between natural and agricultural systems lies in their varying levels of diversity. Even to the untrained eye, natural communities stand out in maintaining more plant species per unit area than crop fields. While richness and evenness are, historically, the two most popular means by which to quantify diversity, recent studies emphasize a more cryptic component: phylogenetic relatedness, defined as the amount of time since two species shared a common ancestor (Cavender-Bares, Kozak, Fine, & Kembel, 2009;Vamosi, Heard, Vamosi, & Webb, 2009). Relatedness offers a quantitative estimate for the degree of shared evolutionary history, either between two co-occurring individuals or averaged across an assemblage of species. This means that two communities can have identical richness, while differing drastically in relatedness.
These findings imply that in agricultural systems configuring polycultures based on random crop arrangement is "unnatural" and may in fact be less successful than employing a targeted approach to maximize the phylogenetic distance separating crops.
An analogous process, called phylogenetic underdispersion, is sometimes observed in nature when related species co-occur because they share traits allowing them to persist in unique or stressful environments (Forrestel, Donoghue, & Smith, 2014;Verdú & Pausas, 2007).
Most crop production guides, however, recommend avoiding consecutive plantings of related species (i.e., same genus or family) over time. Closely related species tend to be more ecologically similar, resulting in more intense competition for a limited pool of resources (Burns & Strauss, 2011;Losos, 2008). Seedling survival and growth, for example, are higher with increasing phylodiversity of neighboring vegetation (Castillo, Verdú, & Valiente-Banuet, 2010;Webb, Gilbert, & Donoghue, 2006), presumably due to increasingly divergent abiotic requirements and ecomorphological traits associated with acquiring those resources (e.g., rooting depth). Thus, plants on average stand to benefit from associating with distantly related species via niche partitioning (Cadotte, 2013). A second major driver is evading consumers. Closely related plants are more likely to share parasitic insects and pathogens (Gilbert, Briggs, & Magarey, 2015;Gilbert & Webb, 2007;Novotny et al., 2006;Yguel et al., 2011). As a result, increasing phylodiversity is a good rule of thumb for improving the statistical likelihood of cultivating nonhosts in a crop field without underlying knowledge of pest biology or diet breadth. In the context of rotations, host-specific, soil-borne microbial pathogens are considered the primary driver of negative feedbacks from close relatives.
Existing evidence for soil-mediated phylogenetic effects on plant performance and microbial communities are mixed and derive entirely from unmanaged systems. Phylogenetic influences on plant-soil feedbacks impacting plant growth are highly inconsistent across studies, likely depending on variables such as the amount of phylogenetic distance tested relative to the focal plant (Anacker, Klironomos, Maherali, Reinhart, & Strauss, 2014;Burns & Strauss, 2011;Dostál & Palečková, 2010;Fitzpatrick, Gehant, Kotanen, & Johnson, 2017;Kuťáková, Herben, & Münzbergová, 2018;Liu et al., 2012;Mehrabi, Bell, & Lewis, 2015;Mehrabi & Tuck, 2014;Münzbergová & Šurinová, 2015). Yet, phylogeny appears to play a relatively stronger role in structuring plant-associated soil microbes; close relatives tend to share more similar communities of rhizosphere bacteria and fungi, particularly for plant pathogenic taxa (Barberan et al., 2015;Gilbert & Webb, 2007;Peay, Baraloto, & Fine, 2013;Sarmiento et al., 2017;Schroeder et al., 2019). In combining these two approaches, one study found that increasing phylogenetic distance between neighbors improved focal plant growth in field-collected "live" soil, but after the soil was experimentally treated with fungicide the relationship dissipated (Liu et al., 2012). These data suggest that plant species-and/or genus-specific fungal pathogens mediate the negative consequences of growing in the same soil as close relatives.
Surprisingly, no field studies have quantified the effects of plant phylogenetic diversity on crop yield and soil microbiomes in agriculture, despite the fact that agronomists widely advocate rotations based on these factors. In a recent greenhouse study, we found that relatedness did not predict the soil legacy of 36 crop and weed species on short-term vegetative growth of potted tomato plants (Ingerslew & Kaplan, 2018). Here, we conducted a 2-year field experiment using the same agricultural plant community to assess whether species relatedness impacts soil microbial legacies and tomato yield. In keeping with the central tenets of the phylogenetic diversity hypothesis, we predicted that plants more closely related to tomato imprint similar soil microbiomes, resulting in correspondingly lower yield, compared to more distantly related taxa.

| MATERIAL S AND ME THODS
We conducted a 2-year field experiment at the Meigs-

| Soil conditioning
We cultivated 36 plant species in a randomized complete block design with 8 replicated blocks in a single field. We also included two plant-free fallow control plots per block, resulting in 304 total plots (= 36 species + 2 controls × 8 blocks). A block consisted of two adjacent 285 ft length rows (between-row spacing, 6 ft), with 19 plots equally split between the two rows. A plot was considered four consecutive plants of the same species in a row, with 3 ft between-plant spacing, and a 6 ft buffer separating plot treatments. There was no space between neighboring blocks, that is, each two-row block was immediately adjacent to the next. The field was tilled in late May 2017 before constructing raised beds covered in a double layer of black plastic mulch to reduce weed pressure with drip tape for irrigation. A preplanting fertilizer was added to the soil at the following rates: potash 0-0-60 (71 lbs/ac) and diammonium phosphate 18-46-0 (147 lbs/ac).
Seeds for each of the 36 species were germinated in the laboratory in the spring and fertilized weekly beginning 2 weeks after transplanting seedlings into pots in the greenhouse. See Ingerslew and Kaplan (2018) for details on germination procedures and seed sources. Because seedling size varied across species, we standardized germination times. On June 1, seedlings were transplanted into their randomly assigned field plots. Because pure species plot treatments were necessary for the experimental design, we applied the following herbicides between rows on July 7 and 31 to prevent natural weed infiltration: paraquat (Gramoxone SL 2.0) and S-metolachlor (Dual II Magnum). Other pesticides (i.e., insecticides, fungicides) were not applied in either year of the study. We hand weeded within and between rows as needed throughout the growing season to maintain treatments.
Between October 9 and 18, all plants were harvested and removed from the field. To do so, we uprooted plants, removing the main F I G U R E 1 Impact of identity of the conditioning plant species in year 1 (y-axis) on tomato yield (mean ± 95% CI) in year 2 (x-axis). Vertical dashed line represents the global mean across all plots (= 22.27); thus, 95% CIs that do not bracket this line over-or under-perform relative to the community average. Red bars represent plants that are closely related to tomato, that is, in the family Solanaceae, while the white bar denotes the plant-free fallow control. Hatched bars separate weed species from crops (unhatched) taproot and as much of the larger roots as possible. On November 27, the herbicides glyphosate, sulfentrazone, and metribuzin were applied to the whole field to kill any remaining plants.

| Tomato response
Because we aimed to measure tomato responses to soil legacy effects from year 1 species treatments, we grew tomatoes throughout the entire field with seedlings transplanted in the exact location where the previous year's plants grew. Some species were persistent in reestablishing from belowground rhizomes (e.g., thistle, horsenettle, some grasses); these plants were repeatedly pulled by hand as needed to avoid competing with tomato. On June 1, we transplanted 1,216 tomato seedlings (var RG 611) into the field (i.e., 304 plots × 4 plants per plot). Two weeks later, they were fertilized through the drip irrigation with a soluble fertilizer (30 gallons of 10-34-0 NPK).
During transplant, we collected soil from each plot for microbial and nutrient analyses to quantify the soil legacy from year 1 treatments. Bulk soil was collected rather than rhizosphere soil to isolate the species temporal legacy without the confounding influence of tomato conditioning, while also measuring the initial soil properties experienced by the roots of a new tomato seedling. To do so, we sampled from the top 3-inch profile of the soil layer at each of the four locations in a plot where plants previously grew; then, we combined these samples, creating a single ca. 350 g soil sample per plot.
Sterile nitrile gloves were used to avoid microbial contamination between plots. In the field, we temporarily stored samples in sealed plastic bags in a cooler, before placing them in a −20°C freezer in the laboratory until analysis. After manually homogenizing samples, a 2 g subsample was isolated for microbial analysis (see below sections plots, rather than using total plot yield, which assumes plant density is identical. One of the 36 species treatments-spinach, Spinacia oleracea-did not persist in year 1 and thus was removed from the analysis. The impact of phylogenetic distance on plant-soil feedback for tomato yield was tested using regression (Proc Reg in SAS v. 9.4).
As a response variable, we used species means for the plant-soil feedback effect size, calculated as ln(species treatment/fallow control).
This was followed up with categorical tests (Proc GLM) comparing tomato yield: (i) fallow (1)

| Amplicon library preparation, sequencing, and bioinformatics
A 250 mg soil subsample was analyzed by Argonne National Laboratory for community profiling of bacteria and fungi. Raw sequence data are accessible in the Qiita repository (ID 12546; Gonzalez et al., 2018).

| 16S rRNA sequencing for bacterial community
Briefly, PCR amplicon libraries targeting the 16S rRNA encoding gene present in metagenomic DNA were produced using a barcoded primer set adapted for the Illumina HiSeq2000 and MiSeq (Caporaso et al., 2012). DNA sequence data were then generated using Illumina paired-end sequencing at the Environmental Sample Preparation and Sequencing Facility (ESPSF) at Argonne National Laboratory.
Specifically, the V4 region of the 16S rRNA gene (515F-806R) was PCR amplified with region-specific primers that include sequencer adapter sequences used in the Illumina flow cell (Caporaso et al., 2011(Caporaso et al., , 2012. The forward amplification primer also contains a twelve base barcode sequence that supports pooling of up to 2,167 different samples in each lane (Caporaso et al., 2011(Caporaso et al., , 2012

| ITS sequencing for fungal community
Genomic DNA was amplified using an internal transcribed spacer (ITS) barcoded primer set, adapted for the Illumina HiSeq2000 and MiSeq. These primers were designed by Kabir Peay's lab at Stanford University (Smith & Peay, 2014). The reverse amplification primer also contained a 12 base barcode sequence that supports pooling of up to 2,167 different samples in each lane (Caporaso et al., 2011(Caporaso et al., , 2012

| Phylogenetic relatedness does not affect tomato yield
Of the 35 conditioning species cultivated in year one, tomato was the only species whose soil legacy impacted tomato yield in year two; that is, 95% CI does not bracket the community mean in Figure 1. Importantly, none of the soil nutrients showed strong evidence for explaining species-level differences in crop performance, namely the lower yields in tomato plots. When comparing the nutritional profiles of tomato versus nontomato soils, none were significant at the Bonferroni-corrected p = .003 level. The only mineral trending in this direction (p = .030) was potassium, which was lower in tomato (322.25 ppm ± 27.67 SE) than nontomato soils (394.28 ppm ± 6.21 SE).

F I G U R E 4
Random forest classification correctly identifies soil planting history 27.6% of the time via fivefold cross-validation. Confusion matrix shows the predicted planting history of each sample, as the proportion of times that samples in each class were predicted to belong to each possible class

| Plant species imprint unique phylogenetic signatures on the soil microbiome
Both bacterial and fungal alpha diversity (i.e., within-sample biodiversity) were impacted by block, which was highly significant for all response variables, and secondarily by plant species, which affected some but not all responses ( Table 2) Planting history also exhibited a substantial impact on both bacterial and fungal beta diversity (i.e., between-sample dissimilarity).
Plant species had a significant effect on all beta diversity metrics F I G U R E 5 Random forest models identify microbial features predictive of planting history. The top 15 most predictive bacterial and fungal features are shown (minimum importance score 0.010), and the heatmap displays their normalized average relative abundances within each plant species. Samples and features are hierarchically clustered by UPGMA of pairwise Bray-Curtis dissimilarities for both bacterial and fungal communities (PERMANOVA p < .001; Table 3) and accounted for between 15% and 23% of the variation in beta diversity. Block also significantly impacted beta diversity (p < .001), accounting for between 6% and 25% of the variation, which was less explanatory power than plant species for all metrics except for fungal Bray-Curtis dissimilarity.
Next, we identified features that differentiate plant species to determine how planting history alters the relative abundance of specific microorganisms in the soil. To achieve this, we trained ran-

| D ISCUSS I ON
Given the purported relationship between planting history and soil microbiome composition, we hypothesized that microbial community structure is linked to the evolutionary history of the host plant. Such a relationship would justify phylogenetically informed crop rotation practices, under the assumption that microbes associated with plant species represent host-specific pathogenic or performance-reducing taxa. Overall, we found mixed support for this hypothesis. Although phylogeny predicted some of the variation underlying plant legacy on soil microbial communities, phylogenetic relationships were entirely uninformative past the species level for predicting differences in tomato performance.
Thus, we conclude that plant phylogeny is moderately important in structuring the microbiome of agricultural soils, but has no value in forecasting changes to yield in multi-species cropping systems. Notably, these general conclusions mirror those from a recent study conducted on short-term vegetative growth of potted, greenhouse tomatoes using the same experimental plant community (Ingerslew & Kaplan, 2018).
Several putative factors could lead to a "phylogenetic breakdown" across levels whereby the same evolutionary pattern is not passed along from microbes to plants (i.e., if phylogeny structures microbes and microbes mediate plant health, then why is phylogeny unrelated to tomato performance?). First, microbes may not be the primary mechanism responsible for changes to crop yield. While it is widely assumed that soil microbes influence plant health and performance, plant-soil feedbacks are also driven by variation in growth-limiting micronutrients (van der Putten, Bradford, Brinkman, Voorde, & Veen, 2016). We consider this explanation less likely, compared with natural ecosystems, since fields were fertilized, which should dilute nutritional differences across plant species treatments. Yet, phosphorus levels were 18% lower in tomato soils compared with all other plots, which could serve as a factor contributing to lower yields in self versus non-self treatments. Phosphorus deficiency is known to reduce tomato growth and reproduction (Biddinger, Liu, Joly, & Raghothama, 1998;Menary & Staden, 1976 be an additional factor that reduced the apparent differentiation of fungal profiles between closely related crop species in this study.
Regardless of the specific microbial groups driving phylogenetic patterns, a few notable outcomes emerged from the overall analysis.
For one, the random forest regression revealed that predictive models were far more accurate at identifying distant plant relatives based on the soil microbiome than close relatives (i.e., compare solid versus dashed lines in Figure 8). The model consistently predicted that other solanaceous plants were more distant relatives than their true evolutionary history indicates. In addition, the regression illustrates that the model is more consistent in assigning an evolutionary classification to close and distant relatives; those plants intermediate in their relatedness to tomato were highly variable. In other words, the spread among the datapoints along the y-axis is more pronounced in the middle-0.75 to 1.75-distances compared with close and distant relatives where the predicted values tend to cluster around a central point. This outcome is further supported by random forest classification where certain plant species imprint highly distinct microbial signatures (see dark purple squares along the diagonal in Figure 4), whereas others leave virtually no discernible legacy. To our knowledge, the reasons underlying why some plants generate long-term, species-specific legacies on the soil microbiome and others do not are unknown.
As a whole, our data indicate that phylogenetic relatedness should not be used as a proxy for plant complementarity in multi-species crop rotations. The only rule consistent with current dogma is that consecutive plantings of the same species in shared soil should be avoided whenever possible. This means that crops should be rotated with a different species, but the identity of that rotation partner over time is not necessarily contingent on whether they are congeners or come from opposite ends of the plant kingdom. We suspect that certain crop pairings are beneficial and synergize based on somewhat idiosyncratic aspects of how those species respond to the others' legacy. An important caveat to these conclusions is that our experimental design, due to the large number of species treatments, was established using a plant-soil feedback framework based on how plants respond from one year to the next. True rotation studies, however, implement long-term rotations that simulate how farms produce crops in reality. If several nontomato Solanum species were rotated over a 5-or 10-year period, this could lead to the development of soil-borne diseases that we did not observe in a simple, two-year feedback study. Similarly, low-input (e.g., organic) systems with different abiotic and biotic pressures could change the relationship between phylogeny and yield. Our experiment was also conducted with a single crop and soil type; more studies are needed using a wider diversity of crop species and locations before broader conclusions can be drawn. Nevertheless, these data clearly illustrate the limitations to applied phylogenetics in agriculture and suggest that future cropping system studies test rotations that vary relatedness as part of their experimental design.
Microbiome bioinformatics work was supported in part by NSF grant # 1565100 to JGC. We thank Steve Smith at Red Gold for providing tomato seedlings and Larry Bledsoe for assistance in harvesting.

CO N FLI C T O F I NTE R E S T
None declared.

DATA AVA I L A B I L I T Y S TAT E M E N T
Raw sequence data are accessible in the Qiita repository (ID 12546; Gonzalez et al., 2018).