ML IDENTIFIES INFANT ROOMS 2018

This is a follow up study from a pilot we published in 2014. This study improved upon the pilot in that we followed more infants and added a layer of quantification to our survey via droplet digital PCR (ddPCR). During the pilot, I was painfully introduced to the world of low biomass microbial surveys. Often times PCR reactions would fail, and it was difficult to discern the cause. This was at a time in the microbiome field when very few labs were publishing their negative controls and I had a suspicion that samples derived from hospital surfaces may simply be mirroring the extraction kit-ome! It turns out, my suspicion was confirmed, that a substantial proportion of samples collected in the NICU are likely indistinguishable from negative controls. From the paper, “In total, approximately 3% of generated OTUs and 33% of samples present too weak of a signal to confidently distinguish them from negative control signatures.”

Below is a behind-the-scenes figure that never made it to the publication. Left, I’m plotting 16 rRNA gene amplicon generated OTUs in NICU room samples versus negative controls. Right, I’m plotting the same data but am applying a “ratio OTU method” that uses the ddPCR quantification data to transform the relative abundance OTU matrix. You can see in both cases, common kit contaminants should be removed (i.e. Shewanella and Halomonas). The E. coli OTU is likely from a positive control we were using. In hindsight, I wish we would have used a non-human associated microbe as the positive control.

Aside from the technical finding that low biomass samples should have thoughtful, paired negative controls (🙄 !!), the major biological finding is shown to the right. In short, despite NICU babies being immobile (i.e. they spend most of their time in an incubator), each NICU room harbored a unique microbial fingerprint. We speculated that this is from room-associated microbes being trafficked to infants, growing in infants, and then being disbursed in the room by healthcare providers. While this is a difficult claim to make using 16S rRNA gene data, it was validated later in a parallel study that used metagenomically assembled genomes.