High resolution identification of microbes and prediction of community dynamics in activated sludge plants

Microorganisms are ubiquitous and fundamental to life on earth and the ecological
processes in the ecosphere. The most important natural resource to mankind is
arguably water and our consumption of it directly and negatively impacts natural
water systems globally if not properly treated in engineered biological wastewater
treatment plants (WWTPs). Complex microbial communities (MCs) are mainly
responsible for removing and recovering organics and nutrients from the wastewater
through entirely biological processes and our understanding of them is still far from
complete. Since the early 2000’s advances in high throughput DNA sequencing
(HTS) technologies have enabled the exploration of MCs at unprecedented
resolution and have been fundamental to the work presented in this PhD thesis.
Precise identification of individual microorganisms is the first step when
investigating MCs and is important to be able to link identity with function. Current
ribosomal small subunit (rSSU) taxonomic reference databases are far from
complete and incapable of classifying the majority of microorganisms in activated
sludge (AS) at species level, which is critical for properly understanding the AS
MC.
The overall aims of this PhD project were twofold. Firstly, to develop methods and
tools that facilitate the exploration of complex MCs in WWTPs specifically at a
much higher resolution than previously possible, and to populate the tree of life.
Secondly, to develop a method that allows accurate prediction of MC dynamics at
species level in the near-future.
The first was achieved by utilizing HTS and molecular techniques to obtain millions
of high quality full-length 16S rRNA gene reference sequences directly from
activated sludge (AS) samples from a global-scale survey of 740 WWTPs. A novel
method named “AutoTax” was developed, which enables generation of de novo
taxonomy for previously unclassifiable bacteria based on sequence clustering at
different identity thresholds. AutoTax was used to generate a comprehensive
ecosystem-specific taxonomic reference database, MiDAS4 (Microbial Database of
AS), which, for the first time, has enabled species-level classification of a large
portion of important bacteria commonly present in the WWTP ecosystem.
A featureful R package “ampvis2” was developed to enable more convenient
analysis of MC data at higher taxonomic resolution compared to other software
tools, with additional features for analyzing the MCs of WWTPs in particular, i.e.
core community analyses. Bacteria believed to be comprising the “core” functional
community in MCs of AS systems were estimated at both a regional scale in Danish
WWTPs as well as at global scale. The AS ecosystem was found to be highly
diverse, however most of the diversity constituted bacteria with overall low
abundance, and a much smaller number of bacteria were abundant across many
WWTPs, which were believed to be the core community.
Lastly, a deep learning model based approach was developed to, for the first time,
allow accurate predictions of detailed MC dynamics of AS in the near future (2-3
months). DL models were trained and evaluated on detailed longitudinal community
datasets obtained from one of the longest running sampling campaigns of 19 Danish
WWTPs sampled over a period of 3-6 years. The most abundant bacterial species in
each dataset were modeled and the majority of predictions were accurate. The
abundance dynamics of a few very important bacteria for wastewater treatment were
accurately predicted as a demonstration, which provides valuable knowledge for
WWTP operators to prevent severe problems such as foaming or bulking, and in
general improve our understanding of the complex MC dynamics in AS WWTPs.