Samples were collected from a full-scale methane bioreactor using sludge generated by the wastewater pre-treatment process of a beer brewing factory. Once collected, the samples were incubated for 24 h at 32°C. The samples were then incubated at 70°C for 1 h in order to selectively reduce the abundance of potential hydrogen-consuming microorganisms, primarily methanogenic Archaea. This community (mostly consisting of bacteria) was applied as enriched microbial inoculum (EMI) used in a 5% volume for photo-fermentation experiments.

For the photoheterotrophic degradation experiments axenic microalgae Chlorella vulgaris MACC360 was added to the dark fermentation effluents. Freshly grown algae (OD₇₅₀: 0.7 equal to 2.77 * 10⁸ algae cell mL⁻¹) were added to the dark fermentation effluent in a 10% volume (2 mL in 20 mL). The green algae strain was obtained from the Mosonmagyaróvár Algal Culture Collection (MACC; Institute of Plant Biology, University of West-Hungary, Hungary) and grown on TAP (TRIS-Acetate-Phosphate medium) medium as described in our previous work (Lakatos et al., 2017). The algae liquid cultures and plates were continuously incubated under 50 μmol m⁻² s⁻¹ light intensity at 25°C.

The microalgae-driven photoheterotrophic degradation of the dark fermentation effluent (EFF) was conducted under a number of different experimental conditions (Table 1). The dark fermentation effluent (EFF) was the result of a microbial dark fermentation of a brewery's raw initial wastewater (RIW) (Boboescu et al., 2014). The batch-mode experiments were conducted in 40 mL serum vials with 20 mL of dark fermentation effluent (either 3x diluted or non-diluted effluents), inoculated with 2 mL freshly grown Ch. vulgaris green algae (5.54 * 10⁸ algae cell). For certain samples 1 mL (5 %) of enriched microbial inoculum (EMI) was added. Photoheterotrophic fermentation was performed at 24°C in batch mode for a period of 72 h under continuous illumination with 50 μmol m⁻² s⁻¹ light intensity, the vials were shaken at 120 rpm. Samples were taken at 72 h for the determination of substrate degradation rates and microbial community composition. All experiments were performed in triplicate.

Biomass in each individual culture was estimated by daily measurement of optical density (OD₆₀₀) with a Jenway 6320D Spectrophotometer. The algae cell numbers were counted on Tris-Phosphate solid media using serial dilutions. All experiments were repeated three times.

Hydrogen was directly measured by gas chromatography using an Agilent Technologies 7890A GC system equipped with a thermal conductivity detector and argon as a carrier gas. The temperatures of the injector, detector and column were kept at 30, 200, and 230°C, respectively. An HP MolSieve column was used (Agilent). Since a concentration gradient of H₂ gas can form in the vial headspace, gas samples (0.5 mL) were taken out after mixing of the headspace gas by sparging several times with a gas-tight syringe. Both daily (by sparging the headspace with nitrogen in every 24 h after the measurement) and total accumulated (without headspace sparging) hydrogen production was measured.

Wastewater degradation efficiency was monitored as a result of BOD analysis (Biological Oxygen Demand) as well as total N and total P measurements. BOD measurements were performed according to the instructions of Hach-Lange Cuvette Test LCK555 at the start of the photo-fermentation experiment and at 72 h. Total nitrogen and total phosphorous were measured according to the instructions of Hach-Lange Cuvette Tests LCK138 and LCK349, respectively, at the start of the photo-fermentation experiment and at 72 h.

DNA from complex samples was extracted and purified according to described methods with some modifications (Sharma et al., 2007). Samples (0.5 g) were extracted with 1.3 mL extraction buffer (100 mM Tris-Cl pH 8.0, 100 mM EDTA pH 8.0, 1.5 M NaCl, 100 mM sodium phosphate pH 8.0, 1% CTAB). After thorough mixing, 7 μL of proteinase K (20.2 mg mL⁻¹) was added. After incubation for 45 min, 160 μL of 20% SDS was added and mixed by inversion several times with further incubation at 60°C for 1 h with intermittent shaking every 15 min. Samples were centrifuged at 13,000 rpm for 5 min and the supernatant was transferred into new Eppendorf tubes. The remaining soil pellets were treated three times with 400 μL extraction buffer and 60 μL SDS (20%) and kept at 60°C for 15 min with intermittent shaking every 5 min. Supernatants collected from all four extractions were mixed with an equal quantity of chloroform and isoamyl alcohol (25:24:1). The aqueous layer was separated and precipitated with 0.7 volume isopropanol. After centrifugation at 13,000 rpm for 15 min, the brown pellet was washed with 70% ethanol, dried at room temperature and dissolved in TE (10 mM Tris-Cl, 1 mM EDTA, pH 8.0).

Total DNA from selected samples was prepared for high-throughput next generation sequencing analysis performed on the Ion Torrent PGM platform (Life Technologies). To estimate coverage in sequenced metagenomes we used Nonpareil (Rodriguez-R and Konstantinidis, 2014).

Taxonomic profiling and assessment of metabolic potential were conducted using the public MG-RAST software package, which is a modified version of RAST (Rapid Annotations based on Subsystem Technology) (Meyer et al., 2008). The sequence data were compared to M5NR using a maximum e-value of 1 × 10⁻⁵, a minimum identity of 95%, and a minimum alignment length of 15, measured in amino acids for protein and base pairs for RNA databases.

All results for taxonomy and functional assignment were downloaded from MG-RAST and visualized in R package (Version 3.5.1) using Vegan (Oksanen et al., 2016) and ggplot2 (Wickham, 2016). This also allowed us to filter assignments with read counts lower than five reads per taxonomy unit and incorrectly assigned reads.

All reads were then trimmed to increase quality and remove contaminants using bbduk (http://jgi.doe.gov/data-and-tools/bb-tools/). Reads matching algal genomes were identified and separated. Multiple draft algal genomes were downloaded from NCBI and were used as reference to identify algal specific reads using bbduk. This allowed for assembly of the prokaryotic microbiome and eukaryotic algal reads individually.

The microbiome and algal reads were assembled using Megahit (Li et al., 2015). This is one of the few programs that can assemble single end metagenome reads. We then used MetaWrap to bin the bacterial contigs into species specific bins (Uritskiy et al., 2018). MetaWrap accepts bacterial assembled contigs and extracts individual draft genomes by using multiple binning software (metaBAT2, MaxBin2, and CONCOCT) and identifying those bins with the highest evidence across all software. The extracted bins were uploaded to the RAST server to annotate draft bacterial bins.

The three bacterial bins were also subsequently annotated using PROKKA (Seemann, 2014). Annotated protein sequences were then uploaded to KEGG database and BlastKoala (Kanehisa et al., 2016) was used to identify specific genes involved in biosynthesis pathways such as cobalamin, biotin, and thiamine biosynthesis.

All raw fastq files were uploaded to NCBI with SRA accession number PRJNA521112.

An important question was whether selected microalgae strains were able to proliferate and generate considerable amount of biohydrogen when grown on dark fermentation effluents. Microalgae proliferation and survival in the effluent was followed by an algae selective cell counting assay (Figure 1). The results showed that microalgae were able to propagate in the dark fermentation effluent either the effluent was filter-sterilized or the effluent was untreated thereby harbouring its microbial community developed in the first dark fermentation stage (see Table 1 for sample specifications). Thus, Ch. vulgaris showed not only a stable survival rate in the effluent, but clear photosynthetic growth was observed regardless the addition of extra enriched microbial inoculum (Figure 1). Nevertheless, the algal growth rate was higher in the diluted dark fermentation effluent compared to the original, non-diluted effluent.

Hydrogen production was monitored during the photoheterotrophic algae cultivation, the headspace was sampled and analyzed in every 24 h throughout the experiments (Figure 2). Significant differences in daily bio-H₂ production rates were observed under the experimental conditions applied (Supplementary Table 1). The experimental conditions employing microalgae in the filter-sterilized effluent generally resulted in moderate biohydrogen yields with a maximum of 52 mL H₂ L⁻¹ d⁻¹ when grown in the non-diluted effluent (S-EFF+A 100). No significant H₂ consumption was detected in the 72 h experimental period. When growing in the diluted, filter-sterilized dark fermentation effluent (S-EFF+A 33), only minor H₂ production was observed. The addition of enriched microbial inoculum together with the microalgae (EFF+M+A 100 and EFF+M+A 33) led to a considerable increase in biohydrogen production. The highest maximum yields (154 mL H₂ L⁻¹ d⁻¹) were observed on the non-diluted dark fermentation effluent as a substrate (EFF+M+A 100). The dark fermentation effluent with enriched microbial inoculum but without Ch. vulgaris addition (EFF+M 100 and EFF+M 33) resulted in lower levels of biohydrogen with a maximum of 88 mL H₂ L⁻¹ d⁻¹ when non-diluted effluent was used as substrate (EFF+M 100). The effluents supplemented with Ch. vulgaris but lacking additional microbial inoculum (EFF+A 33 and 100) showed comparable biohydrogen generation to that observed in the sterilized effluents with added algae (maximum of 56 mL H₂ L⁻¹ d⁻¹ on the non-diluted effluent EFF+A 100). However, the observed strong H₂ consumption in these samples (EFF+A 33 and 100) as revealed by the comparison of the daily and the accumulated hydrogen amounts made an important difference (Figure 2).

In order to assess the effluent treatment potential of the added microalgae in the photoheterotrophic system, biological oxygen demand (BOD) measurements as well as total nitrogen (N) and total phosphorous (P) investigations were performed. The results showed that green algae significantly contributed to the consumption of total N, P, and BOD under all applied conditions irrespective of the presence of enriched microbial inoculum (Figure 3). Similar N and P metabolizing efficiency as well as BOD decrease were achieved by the green algae under experimental conditions using either diluted or non-diluted dark fermentation effluent as substrate (Figure 3). Significantly lower biodegradation efficiency was observed in the samples lacking green algae indicating the primary role of algae in N and P accumulation in the photoheterotrophic stage. Interestingly, the addition of enriched microbial inoculum resulted in similar biodegradation rate compared to other conditions at the photoheterotrophic stage indicating that the microbial community originated from the dark fermentation stage was not able to exert further biodegradation of the applied wastewater substrate. However, green algae proved to be highly efficient, ~75% of the remaining total N, P, and BOD could be metabolized by photosynthetic eukaryotic algae under illumination. Ch. vulgaris green algae showed survival, photosynthetic growth, and metabolic activity (N and P utilization) both in the sterilized effluent and in the effluent harbouring its natural microbial community (Figure 3).

An average of 232,160 sequencing reads was generated for each sample, with a mean read length of 227 nucleotides. To estimate how efficiently this sequencing approach captured the microbiome information, sample datasets were examined using Nonpareil. On an average, all samples had a coverage of 66.01%, with the exemption of the EMI (enriched microbial inoculum) which had a lower coverage of 24.02%. This is most probably due to the high diversity of microbes present in this sample (Table 2).

Reads were uploaded to MG-RAST for taxonomic and functional classification. MG-RAST classified at least 50% of the reads on average for all samples with the exception of the enriched microbial inoculum sample (35%) (Table 2). It was able to classify fewer reads for both the sterilized samples (~15%). Read counts for all samples were downloaded from MG-RAST and filtered to remove classifications that had <5 reads assigned to them. This ensured high confidence of classifications. Since an axenic Ch. vulgaris culture was used as algae inoculum, and the effluent as well as the EMI did not contain any algae (shown by their identified composition), the observed minor presence of other algae hits must be caused by the mis-assignment of the MG-RAST software. Thus, all classifications automatically assigned to genera Actinastrum, Parachlorella, Chlamydomonas, Micractinium, Volvox, Prototheca, and Helicosporidium were manually re-assigned to Chlorella sp. Alpha diversity was measured by both MG-RAST and Fisher's alpha in R (Table 2). The samples with enriched microbial inoculum (EMI) showed the highest alpha diversity, while the filter-sterilized samples had the lowest alpha diversity. The low coverage estimated by Nonpareil for the samples with added enriched microbial inoculum can be attributed to the high diversity. Further, discrepancy was observed in alpha diversity metrics between MG-RAST values and Fisher's alpha. This is due to the fact that low count assignments were filtered out. The observed high diversity of the dark fermentation effluent indicated that the raw wastewater (RIW) was dominated by only a few highly abundant strains, which reduced in abundance during the dark fermentation and instead other bacterial members originally with low relative abundance increased (Figure 4). Finally, higher diversity was found in samples with non-diluted effluents compared to the diluted effluents. This can most probably be attributed to the fact that certain rare species were lost when samples were diluted. After filtering sample counts for low abundance genera, read counts were also proportion normalized. This allowed us to calculate distance matrix downstream. The raw initial wastewater (RIW) sample had lower diversity compared to all other samples. Bacterial members of the genera Lactobacillus and Saccharomyces were present in high abundance (Figure 4). However, after dark fermentation (EFF) this profile clearly changed to a more diverse microbiome. There was an increase in the abundance of Prevotella, Clostridium, Bacterioides, Anaerolinea, Candidatus Cloacimonas, and Megasphera genera along with a concomitant decrease in Saccharomyces sp. abundance. All photoheterotrophic fermentation samples had similar profiles with increases in Prevotella, Veillonella, and Dialister genera with a concomitant sharp decrease in Lactobacillus abundance. Higher relative abundance of Acidaminococcus sp. was detected in the non-diluted wastewater effluent. However, higher abundance of Escherichia sp. was detected in the diluted wastewater effluent. Filter-sterilized samples were not completely sterile, Prevotella, and Lactobacillus genera were detected in low abundance in both samples along with a few other bacterial genera (Figure 4).

Proportion normalized read counts were used to calculate a “bray-curtis” distance matrix. Sample clustering was carried out using this distance matrix (Figure 5). Samples primarily clustered by wastewater concentration and then by the addition of the enriched microbial inoculum. This implied that some rare microbes have driven the clustering between samples and the overall microbiome composition changed relatively little during the photoheterotrophic stage. A PCA plot was built using the same distance matrix. The PCA plot showed that the samples with the enriched microbial inoculum were highly divergent from other samples (Figure 6A). This is understandable as the microbial inoculum came from a different source. Further, when dropping the microbial inoculum sample, there is a great degree of variation between all photoheterotrophic samples and the initial raw wastewater sample (Figure 6B). All samples throughout the photoheterotrophic treatment stage clustered together, specifically driven by the effluent concentration. Adonis-test was carried out on the distance matrix of only the photoheterotrophic samples with the presence or absence of either the microbial or the algal inoculum. This allowed us to test which inoculum had a stronger role in shaping the microbial community structure. The addition of microbial inoculum changed the community structure significantly while the algal inoculum had only minor effect (Table 3).

Metagenome assembly was carried out by Megahit. A total of 50,165 contigs were generated. These contigs were then binned together using three different programs within MetaWrap. The generated bins were further refined and checked for contamination and completion using CheckM database (Table 4). A total of four different bins were generated, three of them showed a high degree of completeness (>70%). These three bins belonged to Lactobacillus, Prevotella, and Veillonella genera (Supplementary Image 1). These bins were uploaded to RAST to get a complete annotation. Gene and protein predictions for these bins were carried out using PROKKA. The predicted proteins were then uploaded to KEGG to identify completion of specific co-factor pathways (Table 5). The Veillonella bin genome had a complete pathway for thiamine biosynthesis and nearly complete pathways for biotin and cobalamine biosynthesis. The Prevotella bin also had a nearly complete pathway for thiamine biosynthesis. No biosynthetic pathways were found for these particular cofactors in the generated Lactobacillus bin genome.

RW was employed by company RAW Molecular Systems LLC. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.