The technical replicates were combined using an equal mass of DNA from each replicate prior to library prep

Potentially leachable soil nitrate levels were calculated for each field using nitrate concentrations from the top 15cm at the harvest sampling event, which occurred within the first three weeks of harvest. Though the plants continued to grow for the duration of the harvest, it is unlikely that nitrate from the top 15cm were used due to the soil’s low water content, and no precipitation or irrigation occurred for the duration of harvest. Bulk density in the top 15cm was assumed to be 1.2 g soil/cm3 as experimental bulk density was measured with 1m of soil and likely overestimated the bulk density at the surface of the soil.Soil sub-samples taken from 0-15cm and 30-60cm at midseason were set aside for DNA analysis. In addition to the experimental plots, samples were also taken from both depths at the nearest irrigated crop production areas and non-cultivated soils, such as hedgerows, field sides, etc. . Gloves were worn while taking these samples and the auger was cleaned thoroughly with a wire brush between each sample. Roots were also collected from one plant per plot and were dug out using a trowel from the top 15 cm of soil. These samples were stored on-site in an ice-filled cooler and transferred to a -80 degree C freezer immediately upon returning to the lab . Roots were later washed in PBS Buffer/Tween20 and ground using liquid N.Root DNA was extracted using a NucleoSpin Plant II kit . Soil DNA was extracted using a DNeasy PowerSoil Pro Kit . Two technical replicates were extracted for each sample for a total of 0.5g of soil and 0.2g of roots. All samples were sent to the University of Minnesota Genomics Center for sequencing using ITS2 primers.The ITS2 rRNA region was selected for amplification and fungal community analysis. This region has been successfully utilized in recent AMF community studies.

Though AMF-specific primers exist , we chose the more general ITS2 fungal primers for several key reasons. First, in the field, SSU primers detect more taxa in nonGlomeraceae families but give lower resolution in the Glomeraceae family. Because the four species in our inoculant are in the Glomeraceae family and this family is dominant in agricultural systems and clay soils, cannabis vertical farming we prioritized species resolution in Glomeraceae over other families. More broadly, the higher variability in the ITS2 region can lead to more unassigned taxa, but does not run as much of a risk that distinct taxa will be lumped together. Third, and of particular importance in our root samples, these primers are better able to select for fungal over plant material than other ITS primer options. Finally, ITS2 allowed us to also examine the broader fungal community in our samples, whereas SSU and LSU options are AMFspecific and cannot be used to characterize other fungi.Qiime2 was used for all bioinformatics. Reads without a primer were discarded, and primer/adapter sequences were trimmed off reads using cutadapt. Samples were denoised with DADA2, and taxonomy was assigned using the UNITE version 9 dynamic classifier for all eukaryotes. Taxa outside of the fungal kingdom were removed from all samples and SRS normalization was used to reduce each sample to 7190 reads. 7190 was chosen as a cutoff due to a natural break where no samples fell between 4000 and 7190 reads. Because depths below 4000 retained less than 90% of sample richness, 7190 was chosen, retaining over 95% of richness. The 22 samples out of 301 samples that fell below this cutoff were discarded. These samples included all 5 blanks, 3 samples from field 1A , 4 samples from field 1B , 2 samples from field 2 , 4 samples from field 3 , and 4 samples from field 4 .In addition to the variables of interest, each model had a random effect of field and block within field.

Yields were modeled using the total marketable fruit weight harvested from each plot at each harvest point, while BER was modeled using the proportion of fruits that were classified as non-marketable due to BER from each plot at each harvest point. Yield models and BER models treated weekly harvests as repeated measures, adding random effects of plot within block and harvest number. For hurdle models, random effects were treated as correlated between the conditional and hurdle portions of the model. Because PDW was measured at three time points, the initial PDW model treated the time points as a repeated measure and added a random effect of plot within block. However, given the nonlinear relationship between PDW and fruit quality described by farmers, further models used only PDW at the 6th harvest when fruit quality was at its peak and therefore did not include any repeated measures.The initial model for each outcome variable included plant spacing and PC1 for soil texture , along with PC1 for GWC and PCs 1 and 2 for nutrients at all four depths , as well as the interaction between texture and GWC. In this initial model, only one depth showed a statistically clear relationship with each outcome variable . To improve model interpretability, we then replaced the two PC’s from the depth of interest with the scaled transplant values of nitrate, ammonium and phosphate at that depth, also adding the ratio of nitrate to ammonium and an ammonium-squared term to allow for non-linearities in outcome response to nitrogen levels. Because all nutrient variables had variance inflation factors over 5 in this model , we dropped nutrient PC’s for each depth that was not of interest, leaving only the transplant nutrient values at the depth of interest in the model.

All nutrient VIF values were below 5 in the resulting model. Reported models were run using unscaled nutrient values for ease of interpretation. Transplant nutrient levels were used rather than midseason/harvest both because they are the most relevant to farmer management and because their interpretation is more clear than later time points, when low levels can either indicate lower initial nutrient levels, or that plants have more thoroughly depleted those nutrients.Two fungal community descriptors were calculated for each soil depth and root fungal community: the Shannon index and the count of OTUs in the class Sordariomycetes, which was identified as an indicator of dry farm soils . Counts were scaled, and both community descriptors were added to the final model described in the “Variable selection” section to determine the impact of fungal community structure while controlling for water, nutrients, and texture. Because the metrics between roots and the two depths of soil fungal communities were highly correlated, three separate models were run: one with both fungal community metrics from 0-15cm, one with metrics from 30-60 cm, and one with root community metrics.After preliminary modeling with principal components , cannabis drying rack we determined that nutrients at 60-100cm had a statistically meaningful influence on yields and PDW, while nutrients at 30-60cm showed an influence on BER. We then regressed inoculation and nutrient levels from these depths of interest against each harvest outcome variable–yields, proportion BER and percent dry weight–while controlling for other soil and field characteristics , as well as random effects; see Table 6 and “model structure” above. We also added two fungal metrics to each model . Sordariomycetes counts at 30-60cm, a signature of dry farmed soils, showed a clear relationship with fruit quality, after controlling for all variables in Table 6 . Full results for each model can be found in the supplement. Where indicated, significant and positive coefficients in the hurdle portions of models signify that the outcome is more likely to be zero. Specifically, BER was less likely to occur in plots with higher ammonium levels , and Sordariomycetes counts were associated with plots where no marketable tomatoes were harvested on a given day .Of the AMF taxa that were identified to the species level in soils and roots, none was a species present in the inoculum. After removing samples that did not contain any AMF taxa, PERMANOVAs using Bray distances showed a statistically clear difference between community composition in inoculation vs. control roots but not bulk soils when stratifying by field and controlling for water, nutrients, and texture. No AMF taxa were significantly enriched in the inoculation or control condition. Taken together, these AMF community results suggest that the inoculum shifted the root fungal community at transplant and did not persist in bulk soils for the 9 weeks before DNA samples were taken.A PERMANOVA using Bray distances showed statistically clear differences in fungal community composition in irrigated, dry farm, and non-cultivated bulk soils as well as communities at 0-15cm and 30-60cm when stratifying by field and controlling for water, texture and their interaction, which also significantly differentiated between communities . Though dry farm, non-cultivated and irrigated soils each had more unique taxa than taxa shared with another location, dry farm and non-cultivated soils each had nearly twice as many unique taxa as taxa shared with a single other location, while irrigated soils had more taxa shared with dry farm soils than unique taxa . Abundance analysis showed that there were 466 taxa that significantly discriminated between the three soil locations. We then set the LDA threshold to 3.75 to highlight only the most stark differences, resulting in 13 discriminative taxa . All of the taxa identified as being enriched in dry farm soils were sub-taxa of Sordariomycetes, a fungal class that is highly variable in terms of morphology and function. We therefore identified Sordariomycetes as a dry farm indicator taxa, or a sort of dry farm “signature”.

We included the Sordariomycetes count in models as an indication of how much the soil had shifted towards a dry farm-influenced community . AMF taxa were notably absent as discriminative taxa and PERMANOVA did not show a difference in AMF community composition between the two depths, suggesting that AMF are not limited in their dispersal down to 60cm59.After identifying Sordariomycetes as an indicator taxa for dry farming, we further explored whether multiple years of dry farming enhance soils’ dry farm signature by comparing fields that had not received external water inputs for multiple years and those which had received regular external water inputs the summer prior to the study. The extent to which Sordariomycetes were enhanced was measured by the difference between counts in dry farm and irrigated soils in the study year . We found that fields that had not received regular external water inputs the previous year showed a significantly higher difference in Sordariomycetes counts between dry farm and irrigated soils , indicating that multiple years without irrigation enhance a soil’s dry farm signature.On-farm research across seven commercially managed dry farm fields allowed us to observe tomato, nutrient and soil fungal community dynamics in situ, opening a window into how dry farm systems function on working farms. Given the long-term specialized management that farmers have tailored to their dry farm practice and fields, this on-farm approach facilitated results that reflect this management paradigm across the region and are therefore broadly applicable to dry farm management choices and outcomes on the Central Coast of California.Marketable yields per plot surprisingly did not correlate with plant spacing, which runs counter to current common wisdom in extension publications. Because spacing ranged from 15-48 inches between plants , relatively consistent yields on a per-plant basis contributed to a wide range in yields on a per-area basis . As there are very few irrigated tomatoes in the Central Coast region due to its cool, moist climate, it is difficult to compare dry farm yields to what might be found in an irrigated system in the same region. However, in 2015 , the statewide average fresh market tomato harvest was 39 T/ha, a number that is surprisingly on par with the average dry farm yield in this study . Because there is a clear trade off between yield and fruit quality–the highest yielding fields also had the lowest fruit quality, and increasing ammonium concentrations improve fruit quality while lowering yields–it may be difficult to increase yields above the state average while still charging consumers a premium for dry farm quality.