Non-vegetated and root-associated soil fractions were collected in the SIN Caffaro, a site located in Northern Italy (Brescia municipality), within the sampling area A, a former agricultural grassland field that was previously chemically characterized and contained PCBs and other chlorinated pollutants in concentrations often exceeding the safety limits (Di Guardo et al., 2017). The root system of three spontaneous plant species was collected from triplicate specimens for each plant species. Plants were identified basing on their morphological traits as M. sativa L. (MS), C. nigrescens Willd. (CN), and D. glomerata L. (DG). These plant species were among the most widely widespread in the meadows present in the agricultural areas of the SIN Caffaro, according to a recent survey (Armiraglio et al., 2009). Non-vegetated bulk soil (B) was collected in triplicate in the same sampling station. Samples were transferred to the laboratories and the different soil fractions (as defined below) were separated within 4 h from collection as described in Marasco et al. (2012). Soil fractions were stored at -20°C for molecular analyses and at 4°C for bacterial isolation.

Total DNA extraction was performed from 0.25 g of each of the three non-vegetated bulk soil (B), root surrounding soil (S) and rhizosphere soil (R) replicates, using the PowerSoil DNA kit (MoBio) according to the manufacturer’s protocol. Illumina tag screening of the V3–V4 hypervariable regions of the 16S rRNA gene was applied on DNA of R, S, and B triplicate samples using the primers 341F and 785R (Klindworth et al., 2013). The obtained sequences were analyzed using a combination of the UPARSE v8 (Edgar, 2013) and the QIIME v1.8 (Caporaso et al., 2010b) softwares. Briefly, raw forward and reverse reads for each sample were assembled into paired-end reads considering a minimum overlapping of 50 nucleotides and a maximum of one mismatch within the region using the fastq-join algorithm¹. The paired reads were then quality filtered, the primer sequences were removed and the individual sample files were merged in a single fasta file. This file was imported in UPARSE where operational taxonomic units (OTUs) of 97% sequence similarity were formed and chimeras were removed using both de novo and reference-based detection. For reference chimera detection, the “Gold” database containing the chimera-checked reference database in the Broad Microbiome Utilities² was used. Taxonomy was assigned to the representative sequences of the OTUs in QIIME using UClust (Edgar, 2010) and searching against the latest version of the Greengenes database (McDonald et al., 2012). After the removal of reads affiliated to chloroplast, Archaea and unassigned sequences, a total of 1319653 high-quality merged paired-end reads with an average length of 450 bp were obtained. All the samples analyzed presented Good’s coverage values ranging from 90 to 99 capturing sufficient diversity with an adequate sequencing depth (Supplementary Table 1).

The OTU table, composed by 3587 OTUs, and the phylogenetic tree were calculated with FastTree2 (Price et al., 2010) using default parameters and the PyNast-aligned (Caporaso et al., 2010a) representative sequences as an input. The OTU table and the phylogenetic tree were used as inputs for the subsequent analyses of alpha- and beta-diversity. Bray–Curtis distance matrix on the log transformed OTU table was used to perform a Principal Coordinates Analysis, Canonical Analysis of Principal coordinates (CAP) and to conduct a permutational multivariate analyses of variance (PERMANOVA). Statistical analyses were conducted in PRIMER v. 6.1, PERMANOVA++ for PRIMER routines (Anderson et al., 2008) to test differences in bacterial community composition among the three soil fractions and among the three plant species. A first analysis was conducted in a one-way anova to explore difference among the fraction considering as a factor “Fraction” as fixed and orthogonal (three levels: rhizosphere/root surrounding soil/non-vegetated bulk). A second analysis was conducted exploring difference of the microbial assemblage among plants and fractions using the factor “Plant” (three levels: M. sativa, C. nigrescens and D. glomerata) and the factor “Fraction” (two levels: rhizosphere and root surrounding soil) both as a fixed and orthogonal, and their interaction (Fraction × Plant). Prior to perform the statistical analysis we verify that the data were not over-dispersed using PERMDISP for the factor “Fraction” (F_1,18 = 3.61, p = 0.21) and the factor “Plant” (F_2,15 = 9.36, p = 0.06). The shared OTUs among different fractions and plant species have been defined by Venn-diagram analysis using the software available at http://bioinformatics.psb.ugent.be. Diversity indexes were calculated using PAST (Hammer et al., 2001) and their statistical difference was evaluated with the analysis of variance considering the index as response variable and “Fraction” and “Plant” as explanatory categorical variable (see above for details). Raw sequences have been deposited at the ENA European Read Archive under accession number from SAMN07167786 to SAMN07167806, BioProject PRJNA388028, Submission ID SUB2728610.

16S rRNA gene quantitative PCR (qPCR) was performed in a reaction volume of 15 μl with 1 μl of template using universal primers 27F and 1492R and the SsoAdvanced^TM Universal SYBR^® Green Supermix (Bio-Rad). Thermal protocol was set up as follow: 98°C (3 min), then 40 cycles at 98°C (15 s), 58°C (30 s) and 72°C (1 min). Starting DNA concentration (ng/μl) was measured using a PowerWave HT Microplate Spectrophotometer (BioTek) and the number of 16S rRNA copies obtained with qPCR was normalized with the DNA concentration.

For bacterial isolation, the R samples obtained from the triplicate plants of the same species were pooled and homogenized. One gram of the resulting soil was suspended in 9 ml of physiological solution (0.9% NaCl), diluted in 10-fold series and plated onto mineral medium (Uhlik et al., 2011), adding biphenyl crystals on the plate lid as unique carbon source. Colonies were randomly picked after 1 week of incubation at 30°C, after the appearance of a stable number of colonies, and were spread three times on the same medium to obtain pure bacterial cultures. A collection of 165 biphenyl-utilizing rhizobacterial strains was established (52 isolates from MS, 56 from DG and 57 from CN) and cryopreserved in 25% glycerol at -80°C. Strain code includes a different number according to the plant of origin (1: MS, 2: CN, 3: DG).

The genomic DNA of each isolate was extracted through boiling cell lysis (Ferjani et al., 2015). Bacterial strains were identified through 16S rRNA gene amplification and partial sequencing (Macrogen, Rep. of South Korea) as described by Mapelli et al. (2013). 16S rRNA nucleotide sequences were subjected to BLAST search (using the blastn suite) and were deposited in the ENA database under accession numbers LT838007–LT838169.

The presence of the genes encoding for biphenyl dioxygenase α subunit (bphA) was assessed in the metagenome of soil samples (B, S, and R) through PCR as described by Iwai et al. (2010), using the primers BPHD-F3/R1, and further confirmed using 512F and 674R primer set (Leewis et al., 2016b). The latter primer set was also used for a bphA gene qPCR assay as described by Leewis et al. (2016b) and the relative abundance of bphA gene copies was expressed as a ratio over the total community 16S rRNA gene copy number. Bacterial isolates were grown overnight in Tryptic Soy Broth medium and then subjected to CTAB – phenol chloroform DNA extraction (Chouaia et al., 2010). BphA gene amplification was performed with 2 μl of DNA template in a final volume of 30 μl with primers 463F/674R (Petric et al., 2011), at the following conditions: buffer 1X, MgCl₂ 1.8 mM, dNTPs 0.2 mM, primers 1 μM, Taq 1.5 U per reaction. PCR thermal protocol was set up as follows: 10 min at 95°C, then 40 cycles of 95°C (15 s), 65°C (1 min), 72°C (2 min) and a final elongation step of 10 min at 72°C. All PCR reactions were performed utilizing FastStart^TM High Fidelity PCR System (Roche). Genomic DNA of the model PCB-degrading strain Paraburkholderia xenovorans LB400 (DSMZ, Germany), was used as positive control for all bphA PCR reactions. PCR results were visualized on 1.2% agarose gel and PCR products that did not show the presence of aspecific bands were sequenced at Eurofins Genomics (Italy). Sequences were then identified using the BLASTn suite of the NCBI website³ and a Neighbor-Joining phylogenetic tree was then built using MEGA 5.1 (Tamura et al., 2011), computing the evolutionary distances using the Jukes–Cantor method. Nucleotide sequences were deposited in the ENA database under accession numbers LT840193– LT840239.

In vitro screening for the presence of activities related to plant growth promotion (PGP) was performed for the entire bacteria collection. Inorganic phosphate solubilization and the production of indole-3-acetic acid (IAA), ammonia, protease and exopolysaccharides (EPS) were assessed as described by Cherif et al. (2015); 1-aminocyclopropane-1-carboxylic acid (ACC) deaminase activity was evaluated according to Belimov et al. (2005). Strains were also tested for abiotic stress resistance, namely the ability to grow at 4 and 42°C, and in presence of 5% NaCl and 20% polyethylene glycol (PEG) supplemented to the medium (Cherif et al., 2015). Catechol 2,3-dioxygenase activity was tested following the protocol described by Margesin et al. (2003).

Tomato seeds were sown in commercial non-sterile soil placed in trays and, after 1 week, uniform-sized seedlings were selected, each transplanted in separated 0.3 kg commercial soil pots and maintained under greenhouse conditions (≈110 photons m^-2 s of light for 12 h during the day and average temperature of 25°C). A subset of 11 rhizobacteria strains, selected based on a cluster analysis performed using MVSP (Kovach, 1999) combining the PGP activities and the abiotic stress tolerance traits, were inoculated separately on tomato plants. One week after transplantation, the tomato plants (n = 5 for each treatment) were fertilized once with a bacterial suspension of the selected strains at a final concentration of 10⁸ cells g^-1 of soil (fresh weight). The inocula have been prepared according to Rolli et al. (2015) resuspending the bacterial cells in sterile tap water. Non-inoculated plants (n = 5) were irrigated with the same amounts of sterile tap water and used as control. After 30 days, plants were harvested for the measurement of shoot/root length and weight. Statistical analysis was performed by a pairwise comparison between each bacterial strain inoculated with the negative control, using Student’s t-test.

The β-diversity of bacterial communities colonizing non-vegetated bulk soil (B) and root-associated soil fractions (rhizosphere, R, and root surrounding soil, S) in the SIN Caffaro was significantly different (Figure 1A, PERMANOVA, F_2,18:3.48, p = 0.0007, Supplementary Table 2A). Considering only the R and S soil fractions associated to the three plant species M. sativa (MS), C. nigrescens (CN) and D. glomerata (DG), the bacterial population assemblage was significantly driven by both plant species and soil fraction (Figure 1B, PERMANOVA, F_1,12 = 4.72, p < 0.001 and PERMANOVA, F_2,12 = 3.21, p < 0.05, respectively, Supplementary Table 2B) but not by their interaction (PERMANOVA, F_2,17 = 1.67, p > 0.05, Supplementary Table 2B). Quantification of the individual factors’ contributions to the observed bacterial community variations, determined by PERMANOVA of Bray–Curtis, showed that the two factors “Fraction” and “Plant species” equally contributed to determine the observed bacterial microbiome variation, explaining 21 and 18% of the variation, respectively (Supplementary Table 3), followed by the interaction of these two factors (11%). For the three fractions, not significant differences have been retrieved considering the OTU richness, but among the rhizosphere the MS and CN plants presented a lower number of OTUs (Figure 1C). OTU evenness was instead significantly different among fractions (ANOVA, F_2,20 = 21.63, p < 0.001) but not between plants (Figure 1D). OTU diversity, calculated as inverse Simpson diversity index, was significantly lower in R compared to B and S (ANOVA, F_2,20 = 4.7, p < 0.05). While not significant differences have been observed among the different plants in S fraction, plant species determined a strong effect in R fraction where DG hosted the significantly highest diversity, followed by CN and MS (Figure 1E).

The three soil fractions collected in the SIN Caffaro showed (i) low number of specific OTUs typical of each fraction, and (ii) a consistent group of shared OTUs (2831/3587) accounting the 97% of the 16S rRNA sequences (Figure 2A and Supplementary Table 4A). The rhizosphere and root surrounding soil associated to the three plant species shared a total of 1922/3436 and 2558/3436 OTUs, respectively, accounting up to 98% of relative abundance in S (Figures 2B,C and Supplementary Tables 4B,C). The high number of shared OTUs in non-vegetated and plant-associated soils indicated that all the fractions hosted bacterial communities with similar phylogenetic composition and different in the structure, as demonstrated by the analysis of Beta and Alpha-diversity (Figure 1).

The taxonomic affiliation of OTUs (Figure 2D and Supplementary Tables 5, 6) revealed that the soil fractions hosted 37 bacterial phyla, 105 classes (99.3% sequences classified), 153 orders (89% classified), 167 families (67% sequences classified) and 162 genera (23% sequences classified). The most represented phyla were present in all the soil fractions (R, S, and B): Proteobacteria 34% (among these Gammaproteobacteria 16% and Alphaproteobacteria 10%), Actinobacteria 30%, Chloroflexi (8%), and Acidobacteria (8%). The three soil fractions were nevertheless dominated by different phyla/classes. In the non-vegetated soils Proteobacteria (on average 35%, mainly represented by Alphaproteobacteria and Betaproteobacteria) was the prevalent phylum followed by Acidobacteria (22%) and Bacteroidetes (12%) (Figure 2D). Proteobacteria dominated also the rhizospheres with Gammaproteobacteria (26%) and Alphaproteobacteria (9%) as the main Proteobacteria-classes, while in S fraction the Actinobacteria was the dominant phylum (30%).

Moving from the root-associated fractions (R and S) to the non-vegetated bulk soil, a strong reduction of the Actinobacteria phylum was observed in the bacterial communities, from an average of 33% in R to an average of 6% in B (Supplementary Table 5). An opposite trend has been observed for Acidobacteria, Bacteroidetes, Verrucomicrobia, and Gemmatimonadetes phyla that were enriched in the non-vegetated bulk soil and strongly limited in R and S fractions (Supplementary Table 5). Finally, while Firmicutes bacteria were enriched in R (7%) respect to the other two fractions (1%), Chloroflexi were more prevalent in S (10%) respect to R (7%) and B (4%). Difference in taxa enrichment and selection were observed also according to the plant species in both R and S fractions (Supplementary Table 7). The MS and CN bacterial communities in R fractions were dominated by Gammaproteobacteria (52%) and Actinobacteria (50%) respectively, determining a higher dominance index value (0.32 and 0.28) in these communities respect to the DG one (0.15), while DG hosted a more equally distributed bacterial community with Evenness index of 0.62. An opposite trend has been observed in the S fraction, where DG was dominated by Actinobacteria (37%) while MS and CN presented several taxa equally distributed within the bacterial communities (Evenness 0.72 and 0.69, respectively, Supplementary Table 7).

The bphA gene was detected, by means of qualitative PCR using different primer sets, in the metagenomes of all B, S and R samples. Despite that, the quantification of bphA gene with primers 512F and 674R was possible only for two replica of the CN rhizosphere (bphA/16S rRNA ratio, CN R2: 0.56 ± 0.03; CN R3: 0.50 ± 0.08) and one replica of the DG S (bphA/16S rRNA ratio, DG S2: 0.08 ± 0.004), showing the highest bphA/16S rRNA gene ratio in the CN rhizosphere.

One-hundred sixty-five bacterial strains have been isolated from the rhizosphere of the three plant species MS, CN, DG in mineral medium supplemented by biphenyl as unique carbon source, hence considered as potential biphenyl degraders. The strains belonged only to 3 phyla, Actinobacteria (75, 51, and 52% in MS, CN, and DG collections, respectively), Proteobacteria (17, 30, and 32% in MS, CN, and DG collections, respectively) and Bacilli (8, 11, and 13%, in MS, CN, and DG collections, respectively). All the isolates belonged to phyla that were present in all the metagenome 16S rRNA high-throughput sequencing libraries, even though with a different relative abundance. All the isolates exhibited sequence identity with the closest described species in NCBI database higher than 97% (Supplementary Table 8) and their phylogenetic affiliation at the genus level is represented in Supplementary Figure 1. Among Gammaproteobacteria, Pseudomonas species were widespread in the three sub-collections, and constituted 15, 23 and 13% of MS, CN, DG rhizosphere, respectively. Acinetobacter was isolated only from CN and DG rhizospheres (5 and 16%, respectively). Among the phylum Bacilli, the genus Bacillus was the most represented, isolated from the rhizosphere of all plant species (7, 13, and 6% in CN, DG, and MS, respectively). Seventy-three percent of all the isolates from the rhizosphere of the three plant species belonged to three genera: Arthrobacter (18, 14, 10 strains from DG, MS, CN rhizospheres, respectively), Microbacterium (7, 24, 11 strains from DG, MS, CN rhizospheres, respectively), Pseudomonas (7, 8, 13 strains from DG, MS, CN rhizospheres, respectively). MS rhizosphere showed the lowest number (6) of cultivable genera within the biphenyl-utilizing bacteria, widely dominated by Microbacterium (44% of the isolates) and Arthrobacter (29% of the isolates), compared with CN and DG rhizosphere that included 11 genera.

The isolate collection has been screened in vitro for PGP activities, bioremediation potential and abiotic stress resistance (results are detailed for each isolate in Supplementary Table 8). Production of IAA and proteases were the PGP related activities most represented among the collection, present in more than 50% of the strains, in similar percentages among sub-collections isolated from each plant species (Figure 3). A large fraction of the collection produced ammonia (21, 33, and 39% of the MS, CN, and DG strains, respectively), EPS (13, 25, and 14%, respectively) and siderophores (10, 24, and 24%, respectively). ACC deaminase activity was detected in higher percentage among CN (65%) than MS (37%) and DG (36%) sub-collections. Phosphate solubilization activity was present in a small proportion among CN (4%) and DG (11%) isolates, while it was absent in MS sub-collection. Concerning the biodegradation potential, catechol 2,3-dioxygenase activity was widespread in the entire collection with 83, 70, and 63% of the MS, CN, and DG active isolates, respectively. Biphenyl dioxygenase α subunit gene (bphA) was detected in the genomic DNA of 65, 81, and 68% of the MS, CN, and DG isolates, respectively. PCR products presenting one single band of the expected size (211 bp) in agarose gel electrophoresis were sequenced, obtaining 45 partial bphA gene sequences (9, 19, and 17 from MS, CN, and DG isolates, respectively). Eighty percent of the sequences showed high nucleotide identity (>99%) with Pseudomonas pseudoalcaligenes KF707 bphA, and the remaining ones showed >99% nucleotide identity with Rhodococcus wratislaviensis strain P13 bphA1 and R. opacus bphA1 genes (Supplementary Table 9). The bphA sequences were clustered in a phylogenetic tree with bphA sequences of reference strains having demonstrated PCB degradation ability. The sequences were clustered in two groups, both including known PCB-degraders (Supplementary Figure 2): Pseudomonas KF707-like sequences clustered together with Paraburkholderia xenovorans LB400, while Rhodococcus-like sequences grouped separately.

All strains were able to cope with the osmotic stress induced by the addition of 20% of PEG in the growth medium (Supplementary Table 2). The majority of the strains (71% on average) from each plant species proved to grow in presence of salt in the growth medium, while a percentage comprised between 58 and 71% tolerated high (42°C) and low (4°C) temperatures for growth (Figure 3 and Supplementary Table 8).

A total of 11 strains has been selected basing on PGP activities and abiotic stress tolerance traits (Supplementary Figure 3) and have been tested in vivo on tomato as a model plant. The strains were affiliated to different species belonging to the genera Pseudomonas, Acinetobacter, Arthrobacter, and Curtobacterium, and presented an array of different in vitro PGP traits, abiotic stress tolerance and bioremediation potential (Supplementary Table 10).

Despite showing only 1 and 2 in vitro PGP traits, respectively, strains 2–30 and 2–50, both isolated from CN plant and affiliated to different species of the genus Arthrobacter (Supplementary Table 10), significantly promoted plant growth compared to the non-inoculated control (Figure 4). Both strains mainly affected shoot development, significantly increasing shoot length (p < 0.01) and fresh weight (p < 0.001). While 2–30 did not showed effect on the root development, 2–50 significantly promote also the root length (p < 0.05). Strain 2–50 possesses, moreover, both the tested bioremediation potential traits (Supplementary Table 10).

In vegetated soils physico-chemical properties, together with plant root exudation and turnover, shape the structure of the microbial community. Resulting in the so-called “rhizosphere effect,” every plant species presents a specific pattern of root exudation and interacts with the soil community selecting a specific microbiota recruited from the surrounding soil (Haichar et al., 2008; Berg and Smalla, 2009). Here we compare the bacterial communities inhabiting the rhizosphere and root surrounding soils of three spontaneous plant species with a non-vegetated bulk soil from a site heavily contaminated with chlorinated persistent organic pollutants and heavy metals, demonstrating that both the soil fraction and the plant species cooperated in the selection of a specific bacterial assemblage in the root-associated soils. The bacterial communities inhabiting all soil fractions hosted the typical phyla that dominate soil ecosystems, i.e., Acidobacteria, Actinobacteria, Bacteroidetes, Chloroflexi, and Proteobacteria (Bulgarelli et al., 2013). R and S fractions, that are directly under plant root influence, showed an increase of Actinobacteria and Proteobacteria. Within the Actinobacteria, the order Gaiellales was retrieved at higher percentage in all the plant-associated fractions and in particular in the CN rhizosphere. This bacterial order was previously detected in the root system of rice (Hernández et al., 2015) and maize planted on heavy metals polluted soil (Touceda-Gonzalez et al., 2015), however, it is still poorly studied and includes only one described species (Albuquerque et al., 2011). Plant exudates enriched also the orders Sphingomonadales and Pseudomonadales in the DG and MS rhizospheres, respectively. Members of these orders have already been retrieved in PCB polluted soils, involved in biphenyl degradation and for some genera also in PCB degradation (Leigh et al., 2007; Hu et al., 2015; Jayanna and Gayathri, 2015; Leewis et al., 2016b). The phylum Chloroflexi, comprising a well known group of anaerobic PCB degraders (Jugder et al., 2015), was retrieved at relative abundance higher than 1% exclusively in the S fractions, being particularly enriched in MS1. It is possible that an anoxic microniche occurred in this sample, allowing the selection of Chloroflexi members, in agreement with previous finding of Chloroflexi enrichment in the rhizosphere of M. sativa during a PCB polluted soil phytoremediation trial (Tu et al., 2011) and of Sparganium sp. during biostimulation of the autochthonous microbiota in historically PCB polluted sediments (Di Gregorio et al., 2013).

Studies on a hydrocarbon-contaminated soil proved that the pollutant concentration rather than rhizosphere effect of planted willows had a major role in shaping the bacterial community (Bell et al., 2014; Yergeau et al., 2014). In a previous work (Di Guardo et al., 2017) we suggested that soil pollutant profiles of different former agricultural fields in the SIN Caffaro acted as drivers of the bacterial community composition. Here we demonstrated that in one of the previously analyzed fields, the rhizosphere effect of spontaneous plants adapted to the occurring soil contamination significantly influenced bacterial assemblages in the root-associated soil fraction, R and S. We speculate that in the rhizosphere an enhanced pollutant degradation, according to the higher abundance of taxa previously associated to PCB metabolism, contributed to shape bacterial diversity (Leewis et al., 2016a; Ridl et al., 2016). Our results indicated that autochthonous vegetation, not specifically selected for rhizoremediation efficiency, but rather naturally adapted to counteract phytotoxicity and the specific polluted soil conditions, can establish a strong relationship with the soil bacterial community, potentially sustaining PCB detoxification and in turn the natural attenuation process. The diversity of R bacterial communities was significantly influenced by the three plant species, which selected a root-associated microbiome having a peculiar structure. Despite this, the occurrence of a high percentage of OTUs shared among the different soil fractions, may suggest a strong selection force toward specific taxa constituted by the high pollution level, and possibly related to contaminant detoxification.

Different plant species, including MS, were previously demonstrated to enrich and stimulate the PCB-degrading bacterial communities, inducing an increase of bphA gene copy number and its expression levels in the rhizosphere compared to the non-vegetated bulk soil (Tu et al., 2011; Li et al., 2013; Pagé et al., 2015). In this work, the bphA gene was detected in the metagenome of all the soil fractions and could be quantified in CN rhizosphere and DG soil surrounding roots. The copy number ratio between this gene and the 16S rRNA bacterial gene was higher in R than in S samples, leading to speculate that bacterial populations harboring degrading potential were enriched in the rhizosphere fraction. However, the spread of bphA gene in the bacterial microbiota of the SIN Caffaro soil indicated an intrinsic PCB attenuation potential in this site, regardless of soil fraction or plant species, suggesting a stronger role of edaphic conditions rather than vegetation.

The root system of plant growing on other PCB polluted sites, was shown to host PCB metabolizing bacteria (Leigh et al., 2006; Ionescu et al., 2009). Hence, we applied a cultivation approach to identify indigenous bacteria in the rhizosphere of plants naturally able to cope with the high level of pollution in the SIN Caffaro, having PGP and PCB degrading potentials. The synergistic effect of these traits has indeed the potential to sustain rhizoremediation approaches, in which both plants and bacteria are involved in remediation (Vergani et al., 2017). The cultivation approach applied to MS, CN, DG rhizospheres, due to the selective conditions determined by biphenyl as unique carbon source, led to isolate only bacteria species belonging to three of the thirty-seven phyla identified in the soil metagenome by 16S rRNA high-throughput sequencing (Actinobacteria, Proteobacteria, and Firmicutes). The high taxonomic similarity shown by the three sub-collections, in which four species over the 21 detected accounted for the 78% of the whole collection, suggests that the cultivable biphenyl-utilizing bacteria likely were mainly selected by the soil characteristics occurring at the SIN Caffaro site rather than the plant species. The detection of bphA gene in 72% of the isolates indicates that these strains harbor the genetic information to initiate the upper pathway of PCB degradation. Moreover, the large majority of the isolates belonged to the genera Arthrobacter, Microbacterium, and Pseudomonas, taxa previously isolated from the rhizosphere of plants growing in PCB contaminated soils and known for their capacity to grow on biphenyl and, for some strains, to degrade PCB (Gilbert and Crowley, 1997; Leigh et al., 2007; Uhlik et al., 2011; Kurzawova et al., 2012). Most of the bphA sequences amplified from the rhizobacteria collection showed higher similarity with the biphenyl dioxygenase α subunit of Pseudomonas pseudoalcaligenes KF707, a model strain studied for its ability to metabolize PCBs through 2,3-dioxidation (Furukawa, 1994). This large group of sequences clusters together with the bphA sequence of the reference strain Paraburkholderia xenovorans LB400. The functional diversity between Paraburkholderia xenovorans LB400 and Pseudomonas pseudoalcaligenes KF707 is determined by minor differences in the gene sequence (Furukawa and Fujihara, 2008), whose detection was not possible in our analysis due to the short sequence coverage of the used primer set. Functional redundancy for PCB degradation in the SIN Caffaro soil is supported by bphA sequences displaying sequence divergence. A second cluster of sequences was affiliated with a gene previously sequenced in Rhodococcus spp., including the reference strain R. jostii RHA1, which shows a different range of substrates and a low degree of homology compared with the bphA proteins produced by Pseudomonas KF707 and Paraburkholderia LB400 (Masai et al., 1995). The bphA sequences of the SIN Caffaro isolates were not related to the phylogenetic identification of the isolates, reflecting that bph genes are frequently associated to mobile elements and can be spread through horizontal gene transfer (Pieper and Seeger, 2005). Catechol 2,3-dioxygenase activity was widespread throughout the collection and all the isolated strains harboring bphA gene also presented this activity. Biphenyl dioxygenase and catechol 2,3-dioxygenase are enzymes involved in the metabolism of several root-derived and xenobiotic aromatic compounds, so these results support the hypothesis that plants foster organic contaminant degradation by their root-associated bacteria (Fuchs et al., 2011). In particular, since catechol is a toxic metabolite produced by the degradation of biphenyl in the benzoate lower pathway, its degradation capability is essential for a complete mineralization of PCBs that can occur through co-metabolism by bacteria harboring the upper and/or lower degradation pathways (Leewis et al., 2016b).

Plant growth promotion activities have been frequently reported in bacteria from polluted soils (Croes et al., 2013; Thijs et al., 2014; Franchi et al., 2016), according to the ability of plants to select beneficial bacteria when growing under phytotoxic and stress conditions. Likewise, IAA production and ACC deaminase activity are PGP traits well represented in the cultivable biphenyl-utilizing rhizobacteria hosted by all the three plant species. Bacterial IAA production can influence root proliferation and elongation, thereby affecting nutrient and water uptake by plants (Lambrecht et al., 2000) besides phytoextraction and phytostabilization in soils contaminated by heavy metals (Ma et al., 2011). Moreover, indole and its derivatives are considered inter-kingdom signal molecules and play a role as biofilm regulators, a further feature in PGP bacteria-plant interactions (Lee et al., 2015). ACC deaminase activity is known to lower ACC level in plant cells interfering with ethylene biosynthesis and thereby decreasing plant stress response potentially deriving from chemical phytotoxicity (Glick, 2010). Therefore, the isolated strains showed a notable potential in supporting plant adaptation and growth in the highly polluted soil of the SIN Caffaro. Moreover, a significant fraction of isolates tolerated moderate saline, osmotic and temperature stresses. These phenotypes are not directly related to PGP or biodegradation activity, but could confer to the strains a higher fitness in soils with complex and uneven pollutant fingerprints and subjected to seasonal changes (Di Guardo et al., 2017).

Two strains belonging to the genus Arthrobacter, one of the most abundant in our collection, also proved to promote the growth of a model plant (tomato) under greenhouse conditions. Since in vitro screening alone is not always sufficient to evaluate the actual PGP potential of bacterial strains (Cardinale et al., 2015), this result is encouraging for further in vivo tests with plant species selected for rhizoremediation trials, considering that these bacteria are also well adapted to the heavy pollution of the SIN Caffaro soil.