Article Figures & Data
Figures
- FIG 1
(A) Distribution of number of variants per LD group (at r2 ≥ 0.95), (B) distribution of genomic distance spanned by LD groups found on the chromosome or the megaplasmids (including only groups found only on one replicon), and (C) number of groups containing only PAVs, only SNPs, or both as well as the number of LD groups found within and across replicons. There were 22,057 SNPs and 10,674 PAVs that were not grouped with other variants and 9,501 LD groups with a median of three variants per group, and the largest group contained 6,970 variants. Half of all variants are in groups that contain ≤12 variants. Only variants used for association testing (minor allele frequency ≥ 5%, missingness ≤ 20%) were grouped.
- FIG 2
(A) Phenotypic distributions of the focal traits and (B) proportion of phenotypic variance explained (PVE) by relatedness among strains (i.e., the K-matrix) alone, as predicted by a linear mixed model, and by both relatedness and large-effect variants through a Bayesian sparse linear mixed model (BSLMM) implemented in GEMMA. PVE was calculated for all variants, only SNPs, and only PAVs. The gray lines indicate the lower 95% of the empirical null distributions from permuted data sets.
- FIG 3
Evaluation of the expected proportion of variance explained (PVE) for A17 biomass by the most strongly associated variants as determined by association testing and forward model selection. Panel A shows the cumulative PVE explained by the 10 most strongly associated variants (black line, more than 10 variants rarely explained more variation than expected by chance) as well as the cumulative PVE from each of 100 randomly permuted data sets that make up the empirical null distribution (gray lines). For A17 biomass, the actual data explain more variance than the permuted data; however, panel B shows that only the first 3 variants explain more of the residual PVE (i.e., after accounting for PVE of the previous variants) than expected by chance. In panel B, the vertical gray lines represent the lower 95% of the null distribution.
- FIG 4
The proportion of remaining phenotypic variance of the focal traits explained by adding each additional top variant, as in Fig. 3B.
- TABLE 1
Mean r2, a measure of nonindependence between segregating variants, is generally low between pairs of variants of different types or on different replicons, while the median size and spanned distance of LD groups is less on the megaplasmids than on the chromosome
Variant type or location Mean r2 between
variantsNo. ungrouped
variantsNo. of LD
groupsMedian no. of variants
per LD groupMedian LD group
spanned distanceaAll 0.06 32,821 9,501 3 N/A SNPs only 0.07 22,057 8,364 3 N/A PAVs only 0.02 10,764 632 2 N/A Between SNPs and PAVs 0.03 N/A 505 7 N/A Chromosome SNPs 0.24 789 900 7 173,406 pSymB SNPs 0.05 13,671 4,478 3 518 pSymA SNPs 0.12 7,597 2,912 3 1,063 ↵a Spanned distance calculated only for LD groups with SNPs that were all on the same replicon.
- TABLE 2
Candidate genes tagged by variants in LD groups that explained more variation than expected based on the empirical null distribution (see Fig. S7 for QQ-plots)
Trait Replicon Positiond Annotation (MaGea locus tag) 2-Aminoethanol pSymA 52580 fixI: nitrogen fixation protein FixI (SMEL_v1_mpb0065) Formic acid pSymA 38256 napA: nitrate reductase, periplasmic, large subunit (SMEL_v1_mpb0048) Gentamicin pSymA 796714 Transcriptional regulator, ROK family (SMEL_v1_mpb0963) pSymA 263510 nifH: nitrogenase Fe protein (SMEL_v1_mpb0322) pSymA 282760 Putative aldehyde dehydrogenase (SMEL_v1_mpb0345) Spectinomycin pSymA PAV Conserved protein of unknown function (SMEL_v1_mpb0259) Streptomycin Chrom. PAV Multisensor signal transduction histidine kinase (SMEL_v1_0575) Desiccation pSymB 1161576 Putative aldehyde or xanthine dehydrogenase (SMEL_v1_mpa1160) A17 biomass pSymA 269841 nifA: Nif-specific regulatory protein (SMEL_v1_mpb0330) 269869 270090 270096 270157 270283 270292 pSymA 271348 nifA: (SMEL_v1_mpb0330); unknown (SMEL_v1_mpb0331) pSymA 274195 Unannotated pSymA 276359 gabD: succinate-semialdehyde dehydrogenase I, NADP-dependent (SMEL_v1_mpb0338) 276443 276563 pSymB 1231268 queC: 7-cyano-7-deazaguanine synthase (SMEL_v1_mpa1230) pSymB 1376015 Diguanylate cyclase/phosphodiesterase (SMEL_v1_mpa1374) pSymB 669804 Sulfotransferase family (SMEL_v1_mpa0678) R108 biomass pSymA 305290 fixN: cytochrome c oxidase subunit 1 homolog (SMEL_v1_mpb0374) 305308 305353 AMTb pSymA 648346 Diguanylate cyclase/phosphodiesterase (SMEL_v1_mpb0802) 649133 APc pSymA PAV fixS: FixS2 nitrogen fixation protein (SMEL_v1_mpb0492) ↵a http://www.genoscope.cns.fr/agc/microscope/home/index.php.
↵b Annual mean temperature.
↵c Annual precipitation.
↵d Variants are sorted by genomic position, not ranking or LD group.
- TABLE 3
For most traits, phenotypic variance explained by genome-wide relatedness (“PVE LMM”) was greater than the phenotypic variance explained by just the top variants
Trait PVE top variantsa PVE LMMb 2-Aminoethanol 0.05 0.00 Gentamicin resistance 0.14 0.49 Spectinomycin resistance 0.43 0.50 Streptomycin resistance 0.34 0.58 Annual mean temperature 0.09 0.12 Annual precipitation 0.10 0.23 Formic acid 0.08 0.30 Desiccation tolerance 0.19 0.31 A17 biomass 0.33 0.74 R108 biomass 0.19 0.53 R108 nodule number 0.06 0.19
Supplemental Material
FIG S1
(A) Distribution of number of variants in each LD group at r2 = 0.80. There were 27,509 LD groups total, and half of the variants were found in LD groups with 37 or fewer variants. (B) Number of LD groups containing PAVs or SNPs, and number of LD groups containing SNPs found on each replicon, also at r2 = 0.80. (C) Distributions of number of variants per LD group for LD groups at r2 = 0.95 containing only SNPs, only PAVs, or only SNPs on one of the replicons. Download FIG S1, PDF file, 1.3 MB.
Copyright © 2018 Epstein et al.
This content is distributed under the terms of the Creative Commons Attribution 4.0 International license.
FIG S2
Distributions for all 20 phenotypic traits. The five focal traits are in red, and the 11 traits with PVE > 0 are in bold. Download FIG S2, PDF file, 0.3 MB.
Copyright © 2018 Epstein et al.
This content is distributed under the terms of the Creative Commons Attribution 4.0 International license.
FIG S3
PVE explained by top variants from GWA after model selection. As in Fig. 4, the points represent the proportion of remaining variance explained (“PRVE”) by adding each additional variant, and, for clarity, only the top 10 of 25 variants are shown. The gray lines indicate the lower 95% of PRVE estimates from analyses of phenotypes randomly permuted 100 times. The five focal traits have red titles, and the ten traits for which at least the first variant explained more phenotypic variance than expected by chance have bolded titles. Download FIG S3, PDF file, 0.9 MB.
Copyright © 2018 Epstein et al.
This content is distributed under the terms of the Creative Commons Attribution 4.0 International license.
FIG S4
PVE by LMM (the K-matrix alone) and by the BSLMM. The gray lines indicate the lower 95% of random permutations. Note that random permutations were only conducted for BSLMM on the five focal traits because of computational limitations. The five focal traits are in red, and the nine traits with PVE by either the K-matrix or BSLMM >95% of the random permutations are in bold. Download FIG S4, PDF file, 0.6 MB.
Copyright © 2018 Epstein et al.
This content is distributed under the terms of the Creative Commons Attribution 4.0 International license.
FIG S5
Size and composition (brown SNPs, blue PAVs) of top 10 LD groups from the testing of the real phenotype data (column 1). (Column 2) The distribution of the mean proportion of SNPs in the top 10 LD groups from 100 GWA runs with randomly permuted phenotypes (LD groups containing both SNPs and PAVs were counted fractionally). The vertical line indicates the mean proportion of SNPs in the real data, and the percentage of random runs with a value greater than the real data is annotated to the right of the line. The genome-wide mean proportion of SNPs is 0.73. (Columns 3 and 4) The distribution of mean LD group size and mean MAF, respectively, for the permutations compared to the real data. Download FIG S5, PDF file, 1.3 MB.
Copyright © 2018 Epstein et al.
This content is distributed under the terms of the Creative Commons Attribution 4.0 International license.
TEXT S1
Supplemental methods describing phenotype measurement methods in detail. Download Text S1, PDF file, 0.1 MB.
Copyright © 2018 Epstein et al.
This content is distributed under the terms of the Creative Commons Attribution 4.0 International license.
FIG S6
Positions of strains and loadings of the 19 core bioclim variables on the first four PCA axes (variance explained in parentheses). Precipitation-related variables are shown in blue, and temperature-related variables are shown in red. Annual mean temperature (AMT) and annual precipitation (AP) were chosen as the representative variables for further analysis. Variable abbreviations: AMT, annual mean temperature (temp.); MDR, mean diurnal range; IT, isothermality; TS, temp. seasonality; TWaM, max. temp. warmest month; TCM, min. temp. coldest month; TAR, temp. annual range; TWeQ, mean temp. wettest quarter; TDQ, mean temp. driest quarter; TWaQ, mean temp. warmest quarter; TCQ, mean temp. coldest quarter; AP, annual precipitation (precip.); PWeM, precip. wettest month; PDM, precip. driest month; PS, precip. seasonality; PWeQ, precip. wettest quarter; PDQ, precip. driest quarter; PWaQ, precip. warmest quarter; PCQ, precip. coldest quarter. Download FIG S6, PDF file, 0.3 MB.
Copyright © 2018 Epstein et al.
This content is distributed under the terms of the Creative Commons Attribution 4.0 International license.
TABLE S1
Descriptions of the 20 phenotypes selected for association analyses. The phenotypes are listed by category. For binary traits, instead of a mean and standard deviation, the number of resistant strains (strains that grew) and the number of susceptible strains (strains that did not grow) are reported (resistant/susceptible in the Mean column). Download Table S1, PDF file, 0.1 MB.
Copyright © 2018 Epstein et al.
This content is distributed under the terms of the Creative Commons Attribution 4.0 International license.
FIG S7
QQ plots for the P values from likelihood ratio tests in linear mixed model GWA. The five focal traits have red titles, and the eleven traits with PVE greater than expected by chance have bolded titles. The P value distributions for four of these 11 traits showed no evidence for inflation, five traits (2-aminoethanol utilization, annual mean temperature, R108 nodule number, and plant biomass on both plant genotypes) showed some mild inflation, and resistance to spectinomycin and streptomycin both showed substantial inflation; that is, the distribution of P values was strongly skewed toward lower (more significant) values. Download FIG S7, PDF file, 0.9 MB.
Copyright © 2018 Epstein et al.
This content is distributed under the terms of the Creative Commons Attribution 4.0 International license.
