Heritability of microbiome community characteristics
To estimate how much maize genetics shapes its overall phyllosphere community, we estimated the heritability of alpha and beta diversity metrics using the Best Linear Unbiased Predictors from the data(Some data was log-transformed due to giving a better fit; see Methods.)
Heritability in this sense does not refer to vertical transmission but rather to how much of total phenotypic variation is due to plant genetics.
In essence, we are treating characteristics of the phyllosphere microbiome as a plant phenotype and estimating how much control the plant itself exerts over the trait
When testing the overall microbial community instead of individual microbes, this is the same as the community heritability (H 2 c ) advocated by van Opstal and Bordenstei)
although our methods estimate narrow-sense heritability (h 2 , the proportion due to additive genetic effects) instead of broad-sense heritability (H 2 , the proportion due to total genetics
We tested all alpha diversity metrics available in QIIME, along with the first 5 principal coordinates from Bray-Curtis, Unweighted UniFrac, and Weighted UniFrac beta diversity analyses
Each metric was individually fit as the response variable in a mixed linear model in TASSEL
Narrow-sense heritability (h 2 ) was estimated by including a kinship matrix generated from public genotype data on these maize lines and estimating the proportion of total variance explained by the kinship matrix
Alpha diversity metrics were extremely poorly heritable, with values ranging from 0 (most traits) to 0.05 (log-transformed Strong’s dominance index) ( Supplemental Table 4 ).
Beta diversity was significantly more heritable, with maximum heritability of 0.598 for Weighted Unifrac PC1, 0.521 for Unweighted UniFrac PC2, and 0.19 for Bray-Curtis PC1
To determine the significance of these heritability scores
we performed a permutation analysis where the phenotype data was randomly shuffled relative to the kinship matrix and the heritability recalculated. We initially screened all metrics against 100 random permutations; metrics with an empirical p-value ≤ 0.02 were then rerun against 10,000 random permutations to more precisely estimate the null distribution. Metrics with a final empirical p-value ≤ 0.001 were considered significant. (That is, at most 1 random permutation in 1000 had a heritability as high as or higher than the actual data.) None of the alpha diversity metrics passed the initial filter, while only two beta diversity metrics passed the second: Weighted Unifrac PC1 (h 2 =0.598, p = 0.0003), and Unweighted Unifrac PC2(h 2 =0.521, p = 0.0003).
Heritability of bacterial OTUs and clades
Using the same methods as for alpha and beta diversity, we performed heritability analysis on 185 individual OTUs and 196 higher taxonomic units(species, genus, etc.)
We tested several methods to deal with the non-normality of many OTU counts, including log transformation and several variations on negative binomial regression
In the end, we chose log-plus-1 transformation of normalized OTU counts because it was the most robust across OTUs. (The various negative binomial variants had a high tendency to fail due to the generalized linear model not converging
Most taxa are not significantly heritable, with only 2 OTUs and 3 higher-level taxonomic clades showing significant heritability at an empirical p-value ≤ 0.001
All five of these are in the class Alphaproteobacteria; four of the five are in the order Rhizobiales, and three are in the family Methylobacteria. An additional 376 OTUs or higher-level clades did not show significant heritability,
indicating that the majority of taxa living on the maize leaf surface are minimally influenced by the genotype of their host
Of the 16 core OTUs, only one showed significant heritability (h 2 = 0.356, p = 0.0008): OTU 575956, an unnamed Methylobacterium. The heritability of other core OTUs ranged from 0 to 0.411, and only two others come close to statistical significance: an unnamed Comamonadaceae at h 2 =0.411, p= 0.0013, and Microbacterium chocolatum OTU 1045797 at h 2 = 0.333 and p=0.0027
Heritability of inferred metagenome content
It is increasingly recognized that the functional capacity of a microbiome may be both more important and more consistent than its taxonomic makeup
We used PICRUST (Langille et al. 2013) to infer the metagenome content of the maize leaf microbiome using annotations from both the Kyoto Encyclopedia of Genes and Genomes (KEGG Orthology, KO) (Kanehisa and Goto 2000) and Clusters of Orthologous Genes (COG)
We then calculated the heritability of the inferred metabolic annotations as above, including binning individual metabolic annotations into groups according to their respective hierarchies
Few phyllosphere traits show robust genome-wide associations
Our inclusion of flowering time as a positive control confirms that our analysis pipeline is capable of identifying robust, biologically relevant hits. However, the small number of hits for phyllosphere traits
indicates that these traits are more difficult to map than flowering time
Even the most heritable phyllosphere traits have h 2 values around 0.7, compared to 0.92 for flowering time, so that the phyllosphere traits have intrinsically lower power for mapping
They may also have a more distributed genetic architecture, or at least fewer large-effect alleles, so that each individual locus explains less variance and is harder to identify. This may explain why nitrotoluene degradation, one of the most robustly associated metagenome traits, has no GWAS hits.
Of the hits we do identify, many are close to or within known genes Unfortunately, the majority of these genes have no known function, making interpretation of their association difficult. In addition to individual hits, however, we also identified two strong hit clusters on chromosomes 7 and 10 that are associated with multiple traits. The first is located at position 7:165,427,159 (maize genome AGPv3 coordinates) and includes 21 traits with a correlation-corrected score of 2.26, roughly equivalent to just over two completely independent traits mapping to the same location. The other is at 10:92,038,994 and includes 6 traits with a corrected score of 3.2.
表型性状是否是复杂性状（数量性状 连续性状变化））（单基因或者多基因） 表型数据 先看是否服从正态分布 近似正态 如果没有
