Three soil types on which bananas were grown were collected in Guadeloupe in April 2010 and their chlordecone concentration was estimated: Andosol (a, 30 mg chlordecone/kg Dry Soil, DS), Fluvisol (b, 60 μg chlordecone/kgDS) and Nitisol (c, 1.5 mg chlordecone/kgDS) by Cabidoche Y.M. (personal communication).

In addition, organochlorines contaminated soils and sediments (dichloropropane, Bis(2-chloroisopropyl) ether (d); tetrachloroethylene and trichloroethylene (e) were used as well as sludge samples from the aerobic (f) or anoxic (g) basins of the wastewater treatment plant (WWTP) of Evry (Chouari et al., 2003).

The mineral medium (MM) used was as described (Löffler et al., 1996) with the following modification: KH₂PO₄ was replaced by a 10 mM (KH₂PO₄ and K₂HPO₄) phosphate buffer, pH 7.5. When indicated this medium was supplemented with 10 mM pyruvate as carbon source, 2 g/L yeast extract and 2 g/L tryptone (MM+). 0.4 g/L Na₂S, 0.05 g/L cysteine-HCl, 0.5 mM dithiothreitol (DTT) were used as reductants and 0.1% resazurin as an indicator of anaerobiosis.

Liquid cultures in MM or MM+ medium were carried out in 100 mL glass serum vials, sealed with butyl rubber septa and incubated in room temperature.

Analytical standard chlordecone (Ehrenstorfer GmbH, Fluka 45379, Pestanal® and Supelco 49046) was solubilized in dimethylformamide (DMF) to a 200 mg/mL stock solution.

Enrichment cultures were conducted following the three strategies described below (Supplementary Figure 2).

Four successive extinction dilution liquid cultures experiments were carried out (v/v, 10⁻²–10⁻¹¹) on MM+ (50 μg/mL chlordecone) under H₂, N₂ (5, 95%) atmosphere. Last dilution growing cultures were spread on 2% agar plates (MM+, 50 μg/mL chlordecone) and isolated colonies were purified three times in the same medium.

16S rRNA gene amplification and sequence analyses were performed as described (Riviere et al., 2009).

Genomic DNA was extracted from 50 mL cultures from the two bacterial consortia 86 and 92 (Bertrand et al., 2005). Sequencing of the two consortia 86 and 92 was performed on Illumina instruments (http://www.illumina.com). Genomic DNA of these metagenomes was extracted and fragmented according to the targeted library fragment size. Inserts from 150 to 300 bp and from 8 to 10 kb were selected to construct paired-end (PE) and mate-pair (MP) libraries respectively. These libraries were loaded on HiSeq2500 sequencing device flowcells (150 nt were sequenced at each extremity) producing 6–10 Gb (PE data) and 23 Gb (MP data). The PE reads of a same fragment were merged (leading to ~ 180 nt read length on average), assembled with Newbler (http://www.roche.com) and contigs larger than 2 kb were ordered by SSPACE (Boetzer et al., 2011). Based on SSPACE output files, we constructed 10 and 7 scaffold pools for consortia 86 and 92 respectively, most of them corresponding to almost complete genome (Table 1). Paired-end and mate-pair data were then individually mapped with BWA (Burrows-Wheeler Aligner, http://bio-bwa.sourceforge.net) on each pool resulting on batch reads that were assembled using a mix of Newbler/SSpace or only Newbler. GapCloser (http://soap.genomics.org.cn/soapdenovo.html) was run successively with the PE and MP data to reduce the number of undetermined bases (Supplementary Table 1). Completeness and representativeness of the assembly was assessed using sequence clustering (binning) software described in Gkanogiannis et al., (2016). The assembly data were integrated for automatic annotation into the Microscope platform (Vallenet et al., 2013), http://www.genoscope.cns.fr/agc/microscope/home. This platform has been used for subsequent manual annotation. Accessions numbers are available in supplementary material.

Genomic DNA was extracted from 50 mL cultures of the two isolated Citrobacter 86-1 and Citrobacter 92-1 as described (Bertrand et al., 2005). Sequencing was performed on Illumina instruments. For both genomes, an overlapping paired-end and a 8 kb mate-pair libraries were constructed out and loaded on MiSeq (2 × 300 nt) and HiSeq2500 (2 × 150 nt) sequencing device flowcells respectively. The PE (corresponding to 38 fold coverage for Citrobacter 86-1 and 25 fold coverage for Citrobacter 92-1) and MP (corresponding to 50 fold coverage for Citrobacter 86-1, 42-fold coverage for Citrobacter 92-1) data were assembled with Newbler. To validate the Citrobacter_86-1 and Citrobacter 92-1 genome assemblies, optical maps were obtained from the Argus Whole-Genome Mapping System (www.opgen.com) for both genomes. The average size of the single-molecule restriction maps was 270 kb and 230 kb for Citrobacter 86-1 and Citrobacter 92-1 respectively. For each genome, we obtained one assembly map with a total size of ~ 4.9 Mb. No misassembly was detected after analysis but variations (essentially due to displacement of insertion sequences) were observed between Citrobacter_86-1 and Citrobacter_92-1 genomes (Supplementary Figure 3). Accessions numbers are available in supplementary material.

Samples were usually extracted from cultures twice a month. After homogenization of the liquid culture, 500 μL of the turbid solution were collected and extracted twice using 250 μL isooctane.

GC-MS analysis was performed on a gas chromatograph (GC), (Thermo Fisher Focus GC) coupled to a single-quadrupole mass spectrometer (Thermo Fisher DSQ II). The instrument was equipped with a non-polar 30 m × 0.25 mm × 0.25 μm DB-5MS column (Agilent J&W) and a split/splitless injector. Carrier gas was helium at a constant flow rate of 1 mL min⁻¹. GC program started at 80°C (hold time 1 min), continued with 50°C min⁻¹ to 140°C, followed by 6°C min⁻¹ to 280°C (hold time 5 min). Injection and transfer line temperatures were set up at 200°C, respectively 280°C. For mass spectrometry (MS) analyses, the following standard working conditions were applied: electronic impact ionization, positive mode detection, ion source at 220°C, detector voltage 70 eV, full scan mode m/z 50–650 (scan time 0.26 sec).

Analyses were conducted using a Dionex Ultimate 3000 LC system (Thermo Fisher Scientific) coupled to a LTQ-OrbiElite mass spectrometer (Thermo Fisher Scientific) fitted with a heated electrospray ionization source (HESI) operating in negative ionization mode. Mass spectra were acquired over an m/z range from m/z 50 up to m/z 2000 with the mass resolution set to 30.000 FWHM at m/z 400 in the Orbitrap analyzer.

For metabolite A1 and chlordecone, the ion spray voltage was set to 4.5 kV, the sLens RF level to 67.9%, the heater temperature to 50°C and the capillary temperature to 275°C. The sheath and auxiliary gas flows (both nitrogen) were respectively optimized at 60 and 50 (arbitrary units), and the sweep gas was set to 0 (arbitrary unit).

For metabolites B1 and B3, the ion spray voltage was set to 4.5 kV, the sLens RF level to 60%, the heater temperature to 275°C and the capillary temperature to 275°C. The sheath and auxiliary gas flows (both nitrogen) were respectively optimized at 60 and 50 (arbitrary units), and the sweep gas was set to 0 (arbitrary unit).

The chromatographic separation was performed using an Acquity^®; C18 column (150 x 2.1 mm2; 1.7 μm; Waters) and carried out at 40°C as follow: a mobile phase gradient was used with a flow rate of 0.3 mL min⁻¹ in which mobile phase A consisted of 10 mM ammonium carbonate with pH adjusted to 9 and mobile phase B consisted of acetonitrile. The gradient started at 60% A for 2 min, followed by a linear gradient at 100% B for 18 min, and remained 10 min at 100% B. The system returned to the initial solvent composition in 5 min and was re-equilibrated under these conditions for 15 min.

We initiated a large number of enrichment cultures (in MM medium supplemented with different chlordecone concentrations) under anaerobic conditions using as inoculum different soils or sediments contaminated with chlordecone or organochlorines (Supplementary Figure 2, Strategy 1). In addition, extracted bacterial fractions using a Nycodenz gradient separation method were used as inoculum (Supplementary Figure 2, Strategy 2). Most of these cultures did not grow and never produce detectable chlordecone metabolites (data not shown).

In parallel, six microcosms were established anaerobically with microorganisms of diverse origins and perfused with a mineral medium containing 2 mg/mL chlordecone (Supplementary Figure 2, Strategy 3). After 1 year of incubation, sediment-free liquid cultures were set up with the perfusing suspensions and renewed once a month by dilution (1/10, v/v) using fresh mineral medium (MM) containing decreasing chlordecone concentrations (from 2 mg/mL to 50 μg/mL).

Cultures were regularly monitored by (i) microscopic observations showing slow growth and microbial diversity, (ii) GC-MS analysis allowing the detection of more than 40 chlorinated metabolites as trace compounds (Figure 2 and Table 2). Chlordecone metabolites (exempt 5b-monohydrochlordecone present as contaminant in CLD commercial preparations) were never detected by GC-MS analyses of control cultures.

After 1 year cultivation of these bacterial consortia, we attempted to isolate individual consortium members on agar plates on a MM⁺ medium containing 50 μg/mL chlordecone. 250 colonies were picked and analyzed through their 16S rRNA gene sequences. Twenty colonies representative of the taxonomic diversity were grown further in liquid MM⁺ medium (with 50 μg/mL chlordecone). Growth was observed for seven colonies only, among which five selectively accumulated metabolite B1 (C₉Cl₅H₃) (Figure 3B). The two most active colonies were selected on the basis of chlordecone degradation, followed by GC-MS. 16S rRNA gene sequences analysis showed that these colonies were bacterial consortia and not pure cultures (hereafter called consortium 86 and consortium 92). Consortium 86 was obtained from the pool of the six microcosms (M1 to M6). Consortium 92 was obtained from microcosm M3 (aerobic basin of the WWTP of Evry). Since soil autoclaving does not ensure sterility, the exact origin of these colonies remains unknown. DNAs extracted from these two consortia were sequenced. Analyses of their metagenomes showed the presence of a dozen of bacterial organisms, five of which being shared between the two consortia (Table 1). To highlight bacteria or bacterium responsible of chlordecone transformation serial dilutions experiments were performed. Thus, consortia cultures were diluted until extinction and ability to produce metabolite B1 (C₉Cl₅H₃) was checked for each dilution where growth was observed. The most diluted (between 10⁻⁶ and 10⁻⁸) sample producing B1 (C₉Cl₅H₃) was used to inoculate the next dilution experiment. After four successive rounds of serial dilutions only one bacterial morphology was detected and these cultures were plated. Optical mapping and genome sequencing confirmed that colonies were two pure and very similar Citrobacter strains and correspond to Citrobacter_86-1 and Citrobacter_92-1 (Table 1). These two Citrobacter strains showed accumulation of an apparent major metabolite B1 (C₉Cl₅H₃) concomitant to chlordecone decrease while minor metabolites A1 (C₁₀Cl₉HO) and B3 (C₉Cl₄H₄) were also formed (Figures 3, 4). Metabolite B1 usually becomes detectable 7–20 days after inoculation. It is worth noticing that, beyond the strict requirement of anaerobic conditions, chlordecone degradation strikingly occurs during the lysis phase, not in the exponential growth phase as one could have expected. This behavior is reminiscent of older observations about the dechlorination of hexachlorocyclohexane (HCH) isomers by Citrobacter freundii (Jagnow et al., 1977). The authors showed indeed that dechlorination of γ-HCH may not be related to Citrobacter freundii growth since release of ³⁶Cl was maximal after a 3 days period of incubation in a complex medium. Therefore, one would hypothesize that dechlorination of chlordecone and γ-HCH are not related to halorespiration.

The high chlordecone concentrations (from 2 mg/mL to 50 μg/mL) prevented quantitative measurements of chlordecone consumption, forcing us to mainly focus on the detection of its potential metabolites. The anticipated high number of samples (roughly thousand per year) to be analyzed guided us to a simplified qualitative analytical procedure. LC-MS technique was discarded due to solubility limitation caused by aqueous mobile phase and low MS-response for apolar chlorinated species. A liquid-liquid micro-extraction with pure isooctane followed by direct GC-MS analysis in full scan mode was developed. Contrary to polar organic solvents previously used for chlordecone extraction (Bristeau et al., 2014; Martin-Laurent et al., 2014), isooctane avoided the insertion of additional step (eg. solid phase extraction or solvent switch) prior to GC-MS analysis. Both electronic impact and chemical ionizations have been previously used to detect chlordecone and its two known metabolites (Harless et al., 1978). Since chemical ionization gave rise to broader chromatographic peaks, electronic impact ionization was finally retained. Even if a majority of microbiological experiments did not lead to any new chlorinated compounds, the fast analytical procedure appeared to be successful in a number of cases. Based on the analysis of several thousands of samples, a library associating MS data and retention time (RT) for more than 40 chlorinated metabolites has been established over the years.

MS data were interpreted using the extensive fragmentation work previously published for polychlorinated bishomocubanes including mirex, chlordecone and its two known partially dechlorinated derivatives (Uk et al., 1972; Alley et al., 1974; Harless et al., 1978). Finally, a chemical formula was postulated for 14 metabolites divided into two families: C₁₀-compounds Ai (1 ≤ i ≤ 11) and C₉-compounds Bi (1 ≤ i ≤ 3) (Figure 3 and Table 2). Compared to chlordecone fragments, mass spectra of metabolites Ai (1 ≤ i ≤ 11) showed similar isotopic patterns with one or two Cl atoms being replaced by H atoms: (i) fragments of generic formula C₁₀Cl_10−nH_nO⁺, C₁₀Cl_9−nH_nO⁺, C₁₀Cl_8−nH_nO⁺, and C₁₀Cl_7−nH_nO⁺ (n = 1 or 2); (ii) fragments of generic formula C₉Cl_9−nHn+, C₉Cl_8−nHn+, C₉Cl_7−nHn+, C₉Cl_6−nHn+, and C₉Cl_5−nHn+ (n = 1 or 2); (iii) fragments of generic formula C₅Cl_6−nHn+, C₅Cl_5−nHn+, and C₅Cl_4−nH_nO⁺ (0 ≤ n ≤ 2); (iv) fragments of formula C₉Cl₇H²⁺, C₉Cl₆H²⁺ and C₉Cl₅H²⁺(Supplementary Figure 4).

Based on these fragments, identification (i.e., mono-, di-or tri-hydrochlordecone) was made for metabolites Ai (1 ≤ i ≤ 11) (Table 2). Schrauzer and Katz reported in 1978 the formation of C₉Cl_5−nH_n compounds (0 ≤ n ≤ 2) in free-corrinoid degradation of chlordecone (Schrauzer and Katz, 1978). Based on MS and UV data, the authors assigned them as polychloroindenes (Figure 1D). Metabolites Bi (1 ≤ i ≤ 3) turned out to show same type of fragments indicating a probable similar chemical structure: (i) C₉Cl₅H3+, C₉Cl₄H3+, C₉Cl₃H3+, C₉Cl₃H2+, and C₉Cl₂H3+ for metabolite B1; (ii) C₉Cl₄H4+, C₉Cl₃H4+, C₉Cl₂H4+, C₉Cl₂H3+, and C₉ClH4+ for metabolites B2 and B3 (Supplementary Figure 4). LC-MS analysis confirmed the proposed chemical formula for the three main metabolites produced by Citrobacter strains (Table 3). Even if neither GC-MS nor LC-MS analysis can, at that stage, quantify the ratio between A1, B1 and B3 in the Citrobacter degradations, it is worth noticing that a loss of five and six chlorine atoms (for metabolites B1 and B3, respectively) has never been observed so far in presence of microbes nor fungi (Orndorff and Colwell, 1980; George and Claxton, 1988; Jablonski et al., 1996; Fernández-Bayo et al., 2013; Merlin et al., 2014).

Consortia 86 and 92 are composed of 10 and 7 scaffolds, respectively, and represent almost complete bacterial genomes (Table 1). Out of them, five are shared by the two consortia and correspond mainly to fermenter (Citrobacter sp., Dysgonomonas sp.), sulfate reducing (Desulfovibrio sp.), acetogenic (Sporomusa sp.) and a Pleomorphomonas species. The latter belongs to the order of Rhizobiales which includes among others plant endosymbionts and a large number of nitrogen fixing species. Isolated Citrobacter sp. have the highest sequence coverage, which is consistent with an important role in the two consortia. None of these species are known to be halorespiring bacteria, although a Desulfovibrio dechloracetivorans isolated from marine sediments is able to grow by coupling the oxidation of acetate to the reductive dechlorination of 2-chlorophenol (Sun et al., 2000). However any protein sequence involved in this dehalogenation has been described. Otherwise, no gene sequences related to known reductive dehalogenases (PFAM: PF13486; TIGRFAM: TIGR02486), the key enzymes in halorespiraton (Maymó-Gatell et al., 1997; Hug et al., 2013) have been found in these genomes. Interestingly, Citrobacter and Desulfovibrio spp. were previously reported to be involved in the dehalogenation of hexachlorocyclohexane isomers by a still unknown mechanism (Jagnow et al., 1977; Badea et al., 2009).

Comparisons with known bacteria were based on 16S rRNA gene sequences alignments and comparative analyses with sequenced genomes sharing a high fraction of syntenic genes. Two different Citrobacter species were found in consortium 86: Citrobacter_86-1 is closely related to Citrobacter amalonaticus and Citrobacter_86-2 to Citrobacter freundii (Table 1). A Desulfovibrio sp. was closely related to D. desulfuricans whereas the affiliation of the other Deltaproteobacterium was not clear and may represent a new genus (the 16S rRNA sequence shares only 89% identity with the closest cultivable bacterium). On the basis of its 16S rRNA gene sequence, the Sporomusa acetogenic bacterium was closely related to Sporomusa sphaeroides (not sequenced so far). The genome of this bacterium is related to the KB1 strain isolated from a trichloroethene dechlorinating consortium (Hug et al., 2012). Two Bacteroidetes belonging to the Dysgonomonas genus as well as two Alphaproteobacteria (one being close to Pleomorphomonas oryzae) are found in consortium 86. Finally, a bacterium belonging to the Clostridiales order was only found in consortium 86.

Most of the bacteria from consortium 86 appear to be facultative anaerobes, e.g., the Citrobacter genomes harbor a large number of genes involved in response to molecular oxygen (Table 1).

Since vitamin B₁₂ is an important compound in the reductive dehalogenation, as cofactor of reductive dehalogenases but also as protein-free corrinoid (Bommer et al., 2014; Renpenning et al., 2014; Payne et al., 2015), in both biotic (Fetzner and Lingens, 1994) and abiotic settings (Schrauzer and Katz, 1978), we searched for genes involved in the biosynthesis of this compound. Complete anaerobic pathways for vitamin B₁₂ biosynthesis were actually identified in the genomes of Citrobacter, Desulfovibrio, Sporomusa and Pleomorphomonas detected in consortia. In addition, genes coding for vitamin B₁₂-dependent proteins were found, such as a large number of dimethylamine and trimethylamine methyltransferases present in the Sporomusa genome, which are likely to be involved in growth on N-methylated compounds. Citrobacter_86-1 genome possesses five vitamin B₁₂-dependent proteins involved in different metabolic pathways (see below).

The reductive dehalogenases being iron-sulfur proteins, special interest has been focused on these proteins of unknown function and on their genomic context (Table 1).

Because no candidate genes for reductive dehalogenases were found in these consortia genomes, we searched for genes encoding other enzymes that could be involved in dehalogenation of chlordecone such as genes linA, linB (coding respectively for a dehydrochlorinase and a haloalkane dehalogenase) and haloacid dehalogenases (van der Ploeg et al., 1991; Nagata et al., 1999). Although these genes were involved in aerobic dehalogenation, we cannot exclude that homologs of these proteins could act on some CLD dehalogenation steps. In summary, no evidence of proteins involved in CLD transformation has been found through analyses of genomes of consortium 86 and 92.

A special interest was for the isolated Citrobacter_86-1 and Citrobacter_92-1 which produce A1, B1 and B3 metabolites. Genomes from these two bacteria are nearly identical and are closely related to Citrobacter amalonaticus strain L8A (RefSeq assembly accession: GCF_000731055.1).

Analysis of Citrobacter_86-1 genome shed light on the anaerobic metabolism of this bacterium. Thus, genes involved in lactate, acetate, CO₂ and H₂ production from glucose and from pyruvate were found as well as a number of genes coding for proteins involved in anaerobic respiration (Supplementary Table 2). Among them, two DMSO reductases-like for which the substrate is unknown. The presence of a putative selenate reductase could indicate that this Citrobacter could use selenium in anaerobic respiration. Genes involved in the anaerobic glycerol utilization, including oxidative and reductive branches of fermentation were found. The vitamin B₁₂-dependent glycerol dehydratase is the key enzyme in the reductive pathway of this fermentation. Another alcohol metabolized by Citrobacter_86-1 is the 1,2-propanediol (1,2-PD). A gene cluster involved in the degradation of 1,2-PD has been found including genes encoding 1,2-propanediol dehydratase (a vitamin B₁₂-dependent protein) as well as shell protein of the micro compartments. Besides the ability to ferment sugars, Citrobacter_86-1 could ferment several amino acids, notably through vitamin B₁₂-dependent glutamate mutase (glutamate fermentation), threonine and serine deaminases (threonine/serine fermentations). Finally, a gene encoding a vitamin B₁₂ dependent methylmalonyl-CoA mutase has been found in a cluster involved in a pathway allowing the conversion of succinate to propionate in Escherichia coli, although the metabolic context of this pathway is unknown (Haller et al., 2000). Taken together these genome sequence data indicate that Citrobacter_86-1 has a versatile anaerobic metabolism and is able to use a large number of electron acceptors.

The 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.