Artificial selection of microbial communities to enhance degradation of recalcitrant polymers

Recalcitrant polymers are widely distributed in the environment. This includes natural polymers, such as chitin, but synthetic polymers are becoming increasingly abundant, for which biodegradation is uncertain. Distribution of labour in microbial communities commonly evolves in nature, particularly for arduous processes, suggesting a community may be better at degrading recalcitrant compounds than individual microorganisms. Artificial selection of microbial communities with better degradation potential has seduced scientists for over a decade, but the method has not been systematically optimised nor applied to polymer degradation. Using chitin as a case study, we successfully selected for microbial communities with enhanced chitinase activities but found that continuous optimisation of incubation times between selective generations was of utmost importance. The analysis of the community composition over the entire selection process revealed fundamental aspects in microbial ecology: when incubation times between generations were optimal, the system was dominated by Gammaproteobacteria, main bearers of chitinase enzymes and drivers of chitin degradation, before being succeeded by cheating, cross-feeding and grazing organisms. Importance Artificial selection is a powerful and atractive technique that can enhance the biodegradation of a recalcitrant polymer and other pollutants by microbial communities. We show, for the first time, that the success of artificially selecting microbial communities requires an optimisation of the incubation times between generations when implementing this method. Hence, communities need to be transferred at the peak of the desired activity in order to avoid community drift and replacement of the efficient biodegrading community by cheaters, cross-feeders and grazers.


60
Recalcitrant compounds are widely distributed in the environment (1-6). These include natural 61 polymers, such as cellulose, (7) and chitin (1), and, more recently, xenobiotic compounds like plastics 62 (2, 3, 5, 8), pesticides and detergents (9). Whilst processes to degrade natural compounds have had 63 time to evolve and adapt, these processes may still require the participation of a consortia of 64 organisms, each specialised in one of the multiple steps involved in the breakdown of the compound 65 (10, 11). Laborious biodegradation processes are therefore rarely carried out entirely by a single 66 microorganism in nature, and it is now well documented that a distribution of labour is favoured in 67 natural microbial communities (12-16). Hence, although this enzyme shares considerable sequence homology with other enzymes capable 75 of PET degradation (17)(18)(19), it has developed a higher hydrolytic activity against this polymer than 76 any other tested esterase but, still, there is room for evolutionary improvement (18). The bacterium 77 encoding this enzyme was isolated from a PET-degrading consortia of microorganisms and is capable 78 The daily microbial community analysis over the four days at generation 20 showed a progressive 148 increase in prokaryotic diversity (from 0.83 to 0.93, according to Simpsons index of diversity) whereas 149 a strong decrease in diversity was observed amongst the eukaryotic community (from 0.93 to 0.38; 150 ASVs in these analyses were responsible for 50% and 60% of variation for the 16S and 18S rRNA 153 genes, respectively (Fig. 3C). 154

155
For the 16S rRNA gene, the most important ASVs were: ASV3 (Thalassotalea, contributing to 16% of 156 the community variation between the four days, p=0.025), ASV4 (Cellvibrionaceae, 15% variation, 9 p=0.033), ASV5 (Crocinitomix, 8% variation, p=0.033), ASV7 (Terasakiella, 6% variation, p=0.094) and 158 ASV2 (Spirochaeta, 5% variation, p=0.022) (Fig. 3C). ASVs 3 and 4 (both Gammaproteobacteria) 159 represented over 50% of the prokaryotic community abundance on day 2, when chitinase activity 160 was highest, and their abundances followed a similar pattern to the chitinase activity over the four 161 days (Fig. 2C), suggesting that these ASVs may be the main drivers of chitin hydrolysis. On the other 162 hand, ASVs 7 and 2 both showed a progressive increase over time (i.e. from a combined relative 163 abundance of 5% on day 1 to 23% on day 4; Fig. 3C), suggesting that these ASVs could be cross-164 feeding organisms that benefit from the primary degradation of chitin. Interestingly, the overall 16S 165 rRNA gene analysis also showed a strong succession over time at higher taxonomic levels (Fig. 4). 166 While Gammaproteobacteria pioneered and dominated the initial colonisation and growth, 167 presumably, via the degradation of chitin (i.e. with 73% relative abundance during the first two days), 168 all other taxonomic groups became more abundant towards the end of the incubation period (e.g. 169 Clostridia, Bacteroidia and Alphaproteobacteria increased from an initial relative abundance of 0.1, 170 2.8 and 12% on day one to 13.5, 22 and 21% on day four, respectively; Fig. 4). Microbial isolates 171 confirmed Gammaproteobacteria as the main contributors of chitin-biodegradation (as discussed 172 below). 173

174
The SIMPER analysis of the 18S rRNA gene highlighted ASV2 (Cafeteria sp., contributing to 34% of the 175 community variation between the four days, p=0.016), ASV4 (Paraphysomonas, 10% variation, 176 p=0.023), ASV1 (Cafeteria sp., 6% variation, p=0.392), ASV6 (Apsidica, 5% variation, p=0.040) and 177 ASV3 (Incertae Sedis, 5% variation, p=0.059) as the five main ASVs contributing to 60% of the 178 community variation over the four days ( Fig 3C). ASV2, which was 96% similar to the bactivorous 179 marine flagellate Cafeteria sp., was by far the most striking Eukaryotic organism, showing an increase 180 in relative abundance from 2% on day 1 up to over 76% on day 4 ( Fig. 3B and 3C). As observed in 181 prokaryotes, Eukaryotic phylogenetic groups also showed a large variation between the beginning 182 and the end of the incubation period, mainly due to the increase of Bicosoecophyceae over time (i.e. 183 from 2.6 to 89% relative abundance driven by both ASV1 and ASV2; Supplementary Fig. S4). contributed to 35% of the community variation, while for the 18S rRNA gene, they accounted for 61% 203 (Fig. 5B). The 16S rRNA gene ASVs 5, 7 and 11 (Crocinitomix, Terasakiella and Carboxylicivirga flava, 204 respectively) presented a much higher abundance in the four-day positive selection than in any other 205 selection (13%, 11% and 8%, respectively), suggesting that these species were the major contributors 206 to the differentiation of these communities, as seen in Fig. 5A. As observed above for the four-day 207 incubation analysis, Cafeteria sp. (18S rRNA gene ASV1 and ASV2, both 96% similar) was again the 208 most conspicuous Eukaryotic organism. ASV2 was more abundant in the positive four-day selection 209 (32% of the relative abundance), while ASV1 was highest in the three other selections (70% and 82% 210 in the positive and random nine-day selection, respectively, and 16% in the random four-day 211 selection; Fig. 5B  incubation experiment showed over 30 times more chitinase (K01183) gene copies than the nine-day 220 incubation experiment (i.e. an average of 0.66 copies per bacterium were observed in the four-day 221 incubation experiment while only 0.025 copies per bacterium were observed over the same 222 generations in the nine-day experiment). Also, from the daily analysis of generation 20, the chitinase 223 activity was positively correlated with the normalised chitinase gene copy number (r 2 =0.57), with a 224 peak in chitinase activity and chitinase gene copies on day 2 (i.e. over one chitinase gene copy per 225 bacterium). The most striking result from this analysis was the strong bias of taxonomic groups that 226 contributed to the chitinase and chitin deacetylase genes; chitinase genes were mainly detected in 227 Gammaproteobacteria and some Bacteroidia, whereas the chitin deacetylase genes were almost 228 exclusively present in Alphaproteobacteria. It is worth highlighting that the chitosanase gene 229 (K01233), the enzyme required to hydrolyse the product from chitin deacetylation, chitosan, was not 230 detected in any of the artificial metagenomes. Chitobiosidases (K01207 and K12373) and enzymes 231 involved in the conversion of GlcNAc to Fructose-6 Phosphate (K00884, K01443, K18676 and K02564) 232 were more widespread. Nevertheless, this data needs to be taken with caution as these were not 233 real metagenomes. 234 235

Isolation and identification of chitin degraders 236
Bacterial isolates were obtained from the end of the artificial selection experiments to confirm the 237 ability of the identified groups to degrade chitin. From the 50 isolates obtained, 20 were unique 238 according to their 16S rRNA gene sequences. From these, 18 showed at least 98% similarity with one 239 or more of the MiSeq ASVs (Supplementary Table S3) although, unfortunately, none belonged to the 240 most abundant ASVs detected during the community analysis. The ability for chitin and GlcNAc 241 degradation by each one of the isolates was assessed. We found that 16 of these isolates could grow 242 using GlcNAc as the sole carbon source, but only 11 of these strains could grow on chitin (Fig. 4). The the first experiment was 0.9 µM day -1 (Fig. 2C), demonstrating that implementing an optimised 252 incubation time between generations largely enhances the selection of a desired trait. Chitinase 253 13 activity was measured daily until a peak in chitinase activity was observed. The communities with 254 highest chitinase activity on this day were used to start the next generation. 255

257
Artificial selection of microbial communities is, in principle, a powerful and atractive technique which 258 has surprisingly been used in only a limited number of studies to date (28, 29, 31), possibly due to 259 the lack of success as a consequence of poor process optimisation. Here, using chitin degradation as 260 a case study and a detailed analysis of the community succession, we show that artificial selection of 261 microbial communities can be largely improved by controlling the incubation times between 262 generations. The rapid succession of microbial community structure means generations need to be 263 transferred at the peak of the selected phenotypic activity (e.g. chitinase activity) or these get rapidly 264 replaced by less efficient communities of cross-feeding microorganisms (i.e. bacteria and grazers). 265 Previous studies that have artificially selected microbial communities for a particular phenotype did 266 not report optimisation of the incubation time between generations (28, 29, 31) which, in our hands, 267 would have resulted in a negative selection (Fig. 2). In agreement with our results, Penn and Harvey 268 to benefit from others that could, i.e. "cheaters" and cross-feeders (43, 44). During our first 277 experiment, as communities become better and faster at degrading chitin, we were measuring the 278 chitinase actvity when the communities were in the succession rather than in the selection stage and, 279 therefore, the active chitinolytic community had decayed and was dominated by cheaters and cross-280 feeders (Figs. 3 and 4). Hence, it was only when selecting at phenotypic time optima when chitinase 281 activity improved and the overall community differentiated from the random control communities 282 ( Fig. 5 and 6). It is also interesting to note the selection of the grazer Cafeteria sp. ( Gammaproteobacteria isolates obtained from the end of the experiments were able to grow using 296 chitin as the only source of carbon and energy (Fig. 4) confirming that this class is likely responsible 297 for most of the chitinase activity observed. On the other hand, Alphaproteobacteria, the numerically 298 dominant class of heterotrophic bacteria in surface oceans (54, 55), follow a cross-feeding and/or 299 cheating life-strategy as five out of eight Alphaproteobacterial isolates could only use N-acetyl-D-300 glucosamine (GlcNAc) and only one could use chitin. This was confirmed by the PICRUSt metagenome 301 analysis (Fig. S6), where almost all chitinase enzymes copies were encoded by Gammaproteobacteria 302 (i.e. 90%; encoding almost one gene copy per bacterium) and, to a lesser extent, by some Bacteroidia. 303 Chitin is made up of molecules of GlcNAc linked by (1,4)--glycosidic bonds, and it has previously 304 been found that initial degradation of chitin takes place predominantly by: i) chitinases which 305 depolymerise the (1,4)--glycosidic bonds either at the ends or in the middle of chains, or ii) 306 chitobiosidase enzymes which also hydrolyse (1,4)--glycosidic bonds but only at the ends of chitin 307 chains. Genes for the intracellular enzymes involved in GlcNAc utilisation (i.e. transformation of 308 GlcNAc to Fructose-6-phosphate) were much more widespread amongst different taxonomic groups, 309 highlighting the broader distribution of cross-feeding or cheating organisms which can benefit from 310 the extracellular depolymerisation of chitin which generates freely available GlcNAc to the 311 community. Alternative degradation of chitin may also occur by deacetylation and deamination of 312 the GlcNAc amino sugar, transforming chitin into chitosan and cellulose, respectively, after which 313 they can be depolymerised by a range of other enzymes (e.g. chitosanases or cellulases) (10, 56, 57). 314 While Alphaproteobacteria did not contribute to chitinase enzymes, it did potentially encode for 315 most of the chitin deacetylases in the system, although no chitosanases were detected. Here we have proven the validity of artificially selecting a natural microbial community to better 327 degrade a recalcitrant polymer, but have highlighted the caveats for achieving this goal, which 328 require a better understanding of the ecology of the system. We found that optimisation of 329 incubation times is essential in order to successfully implement this process, as optimal communities 330 enter rapid decay due to their replacement by cheaters and cross-feeders, as well as the increase of 331 potential predators such as grazers and, although not tested here, viruses. Hence, future artificial 332 selection experiments should adjust generation incubation times to activity maxima to successfully 333 evolve enhanced community phenotypes.

Artificial selection 352
The process for artificial selection is depicted in Figure 1 The average number of reads per sample was approximately 12,500 for the 16S rRNA gene and 428 20,000 (Mothur) or 34,000 (DADA2) for the 18S rRNA gene. Samples with less than 1,000 total reads 429 os, pandas, random, scipy, scikit-bio, sklearn (71), and statsmodels. SIMPER analyses and plotting of 454 phylogenetic trees were performed in R (R version 3.3.3) (72) using the following packages: ape (73), 455 dplyr, ggplot2, gplots, ggtree (74), lme4, phangorn (75), plotly, tidyr, vegan (76), phyloseq (77). The 456 top 5 ASVs identified in each SIMPER analyses were classified to their closest relative using a BLAST 457 search of the GenBank database with a representative sequence. Hypothetical community functions 458 were obtained using PICRUSt in QIIME1 (36, 78). Sequences used for phylogenetic trees were aligned 459 using the SILVA Incremental Alignment (www.arb-silva.de) (79) and mid-point rooted maximum 460 likelihood trees were constructed using QIIME1 (78). All scripts can be found at https://github.com/R-  The three microcosms with the highest enzymatic activities are selected and pooled (3), and used to inoculate the next generation (4). This process is repeated over n generations (5).   showing the day at which the ASV showed maximum abundance. Black circles on the right of the heatmap represent the maximum relative abundance for that ASV amongst the entire community.
The 20 isolates are coloured depending on their ability to grow on chitin and the monomer, GlcNAc (green), the GlcNAc only (orange), or neither (red).