Ezinfo

Since GC-MS analysis could only work on relatively volatile and thermo-stable compounds (MW < 580 Da) effectively, untargeted analysis of sludge samples using UPLC-MS approach was conducted to profile and identify more compounds with higher MW (< 2 kDa).
First, the filtered samples were lyophilized, from which the fractions of polar metabolites and non-polar lipids were extracted following the steps in Tipthara et al. During data processing, the software of EZinfo (version 3.0.3.0, Umetrics) was used for multivariate analysis of compounds in raw WAS, 172 ºC thermal pretreated sludge, digested sludge samples with and without 172 ºC pretreatment respectively. The most significant compounds were selected by orthogonal projection to latent structures discriminant analysis (OPLS-DA) and then exported to the software of Progenesis QI for further identification and analysis.

The metabolites fraction and lipids fraction were extracted from the lyophilized substances following the procedures in Tipthara et al. (2017) , and analyzed by ACQUITY UPLC (Waters, USA) coupled with a Xevo G2-XS QToF mass spectrometer (Waters, U.K.) with an electrospray ionization (ESI) source. Detailed description and analysis of Ultra Performance Liquid Chromatography Mass Spectrometry (UPLC-MS) operation process can be found in Tipthara et al. (2017) .
EZinfo (version 3.0.3.0, Umetrics) was used for multivariate analysis of compounds in raw WAS, ALK-ULS pretreated sludge and digested sludge samples respectively. The most significant components were determined by orthogonal projection to latent structures discriminant analysis (OPLS-DA) and then exported from S-plot back to Progenesis QI for further analysis and identification. Polar and lipid biomolecules were identified using (1) MetaScope by searching compounds against structural databases (SDFs) for E. coli (http://ecmdb.ca/), LIPID MAPS (http://www. lipidmaps.org/), HMDB (http://www.hmdb.ca/), and ChEBI (https://www.ebi.ac.uk/chebi/) and ( 2)
Chemspider by searching compounds against the KEGG database (http://www.genome.jp/kegg/).

Data on growth performance and targeted analysis were analysed by one-way ANOVA followed by the Student Newman-Keuls test (P < 0.05). After data preprocessing of untargeted metabolomics, multivariate analysis was performed to find discriminative features among groups by means of the EZ-Info software (Umetrics, Sweden). First, Principal Component Analysis (PCA) was used to ensure the absence of outliers and the correct classification of QCs after normalization. Then, all the four experimental groups were joined in a single file and Partial Least Squares-Discriminant Analysis (PLS-DA) was conducted to maximize the separation of dietary groups. The contribution of m/z features to the PLS-DA model was assessed by means of variable importance in projection (VIP) measurements. A VIP score > 1 was considered an adequate threshold to determine discriminant variables in the PLS-DA model (Wold et al., 2001; Li et al., 2012; Kieffer et al., 2016) . Additionally, orthogonal PLS-DA (Wiklund et al., 2008) with a high threshold (P [corr] > 0.7) was carried out to highlight the most discriminant compounds. To end, differences in normalized peak areas of m/z features were analysed by One-way ANOVA followed by Benjamini-Hochberg multiple testing correction analysis (Benjamini and Hochberg, 1995) .