Corroborating evidence refutes batch effect as explanation for fetal bacteria

• Published in: Microbiome Vol. 9; no. 1; p. 10
• Main Authors: Rackaityte, E, Halkias, J, Fukui, E M, Mendoza, V F, Hayzelden, C, Crawford, E D, Fujimura, K E, Burt, T D, Lynch, S V
• Format: Journal Article
• Language: English
• Published: England BioMed Central Ltd 12.01.2021; BioMed Central; BMC
• Subjects: Analysis; Bacteria; Bacteria - genetics; Datasets; Feces; Fetuses; Humans; Hybridization; Intestines; Letter to the Editor; Lymphatic system; Lymphocytes; Meconium; Metadata; Microbiomes; Micrococcaceae; Neonatology; Principal components analysis; RNA; rRNA 16S; Scanning electron microscopy; Sectioning
• ISSN: 2049-2618; 2049-2618
• DOI: 10.1186/s40168-020-00948-0

Inclusion of samples with lower sequence reads in a study of very low bacterial burden can artificially inflate false negatives due to inadequate community coverage – this is particularly pertinent since “Batch 2” samples had significantly lower sequence reads per sample than those of “Batch 1” (excluding mock controls; median read depth “Batch 2” = 19,601, “Batch 1” = 39,194; P < 0.0001; specimen read depth available in Supplemental Table 2 of our original manuscript), plausibly explaining the higher rate of Micrococcaceae-negative samples in “Batch 2”. Because our qPCR- and FISH-based analyses of fetal specimens had indicated a sparse, low-burden bacterial presence, the 16S rRNA analysis performed and reported in our study used a multiply rarefied dataset (to ensure that the 16S rRNA profiles were representative) and included only those samples with ≥1000 16S rRNA sequence reads to permit confident detection of bacterial signals. Normalization of read-depth is recommended for analysis of zero-inflated microbiome data and enables clustering of samples according to biological metadata [5], yet this was not performed by De Goffau and colleagues. [...]the batch effect described by De Goffau which the authors claim explain the fetal Micrococcus 16S rRNA signal in Figure 1 and Figure 2, is predicated upon Principal Component Analysis (PCA), an ordination method based on Euclidian distance which assumes linear relationships and a normal data distribution. [...]16S rRNA reads for Micrococcaceae OTU10, found to be enriched in meconium specimens compared to a multitude of technical (n = 48) and biological (n = 35) control samples processed in parallel, were not exclusively detected in “Batch 1”, but existed in both “Batches” (Extended Data Figure 3h of our original manuscript). [...]this approach requires thin-sectioning which dilutes a rare signal. [...]we turned to scanning electron microscopy, which permits the ability to scan the surface of thick sections and obtain high magnification resolution of structures.