Review

This is a critical review of the recently published article …

Wassermann B, Müller H and Berg G (2019) An Apple a Day: Which Bacteria Do We Eat With Organic and Conventional Apples? Front. Microbiol. 10:1629.

[TOC]

Beta Diversity

The authors used unweighted UniFrac distance. Calculating UniFrac distances requires a phylogenetic tree. This was neither provided nor was the method described. I used Muscle to create multiple sequence alignment.

Beta diversity, based on unweighted UniFraq distance matrix, was visualized by Principle Coordinates Analysis (PCoA) and statistical significance was calculated by Analysis of Similarity (ANOSIM). Source: Wassermann et al. 2019

PCoA

It is difficult to see any clusters and plotting ellipses for treatments at confidence level 95% did not help.

My main concern is the low percentage of variation that is explained by the first three axis. It would be much better if Weighted Unifrac would be used instead.

Explained Variance using PCoA

eigen.UF  <- ordinate(d.bac.rare, "PCoA", "Unifrac")
sum(eigen.UF$values$Relative_eig[c(1:3)])
eigen.WUF  <- ordinate(d.bac.rare, "PCoA", "WUnifrac")
sum(eigen.WUF$values$Relative_eig[c(1:3)])

Pairwise Comparison

I also failed to confirm all the results of the analysis of similarities. Not only tissue Seeds but also CalyxEnd were not significantly different between treatments. Unweigthed UniFrac is a qualitativ distance metric giving rare taxa equal weight compare to more abundant taxa. Interestingly, when I consider abundance (Weighted UniFrac) non of the pairwise comparisons is significant any more.

The data indicates a significant difference between location (inside/outside) of the tissue. An interesting finding, relevant for the interpretation of the results.

Pairwise Comparison using ANOSIM

ANOSIM Results (p-values without multiple-hypothesis correction)

Tissue              : R = 0.53, p = 0.001
Treatment           : R = 0.28, p = 0.001
Treatment no Seeds  : R = 0.40, p = 0.001
Treatment-Tissue    : R = 0.78, p = 0.001
Inside/outside      : R = 0.51, p = 0.001

Multilevel Pairwise Comparison using adonis (Permanova)

Table 1: Pairwise - Unweigthed UniFrac

Comparison Df SumsOfSqs F.Model R2 p.value p.adjusted sig
H-Stem vs C-Stem 1 0.5785735 3.4661078 0.36615976 0.030 0.04096552 .
H-StemEnd vs C-StemEnd 1 0.5630496 3.0985516 0.34055438 0.030 0.04096552 .
H-Peel vs C-Peel 1 0.5117392 2.2376209 0.27163436 0.032 0.04096552 .
H-FruitPulp vs C-FruitPulp 1 0.6023229 3.0952419 0.34031441 0.030 0.04096552 .
H-Seeds vs C-Seeds 1 0.3455386 1.3945509 0.18859170 0.061 0.06493548
H-CalyxEnd vs C-CalyxEnd 1 0.3542998 1.6847801 0.21923596 0.063 0.06600000

Table 2: Pairwise - Weigthed UniFrac

Comparison Df SumsOfSqs F.Model R2 p.value p.adjusted sig
H-Stem vs C-Stem 1 0.1866470 3.3325662 0.35709001 0.025 0.05250000 NS
H-StemEnd vs C-StemEnd 1 0.1871960 2.8463682 0.32175557 0.026 0.05250000 NS
H-Peel vs C-Peel 1 0.5025782 10.6873377 0.64044594 0.022 0.05250000 NS
H-FruitPulp vs C-FruitPulp 1 0.7930154 30.1887011 0.83420239 0.034 0.05250000 NS
H-Seeds vs C-Seeds 1 0.3537613 3.6847942 0.38047212 0.143 0.15472131 NS
H-CalyxEnd vs C-CalyxEnd 1 0.1389192 1.1419600 0.15989449 0.197 0.20315625 NS

Maybe a hierarchical clustering can help to better understand data structure.