Targeted depletion of pks+ bacteria from a fecal microbiota using specific antibodies

ABSTRACT The pks island is one of the most prevalent pathogenicity islands among the Escherichia coli strains that colonize the colon of colorectal carcinoma (CRC) patients. This pathogenic island encodes the production of a nonribosomal polyketide-peptide named colibactin, which induces double-strand breaks in DNA molecules. Detection or even depletion of this pks-producing bacteria could help to understand the role of these strains in the context of CRC. In this work, we performed a large-scale in silico screening of the pks cluster in more than 6,000 isolates of E. coli. The results obtained reveal that not all the pks-detected strains could produce a functional genotoxin and, using antibodies against pks-specific peptides from surface cell proteins, a methodology for detection and depletion of pks+ bacteria in gut microbiotas was proposed. With our method, we were able to deplete a human gut microbiota of this pks+ strains, opening the door to strain-directed microbiota modification and intervention studies that allow us to understand the relation between these genotoxic strains and some gastrointestinal diseases. IMPORTANCE The human gut microbiome has also been hypothesized to play a crucial role in the development and progression of colorectal carcinoma (CRC). Between the microorganisms of this community, the Escherichia coli strains carrying the pks genomic island were shown to be capable of promoting colon tumorigenesis in a colorectal cancer mouse model, and their presence seems to be directly related to a distinct mutational signature in patients suffering CRC. This work proposes a novel method for the detection and depletion of pks-carrying bacteria in human gut microbiotas. In contrast to methods based on probes, this methodology allows the depletion of low-abundance bacterial strains maintaining the viability of both targeted and non-targeted fractions of the microbiota, allowing the study of the contribution of these pks-carrying strains to different diseases, such as CRC, and their role in other physiological, metabolic or immune processes.

monomers for PKS or amino acid monomers; and proteinogenic or non-proteinogenic amino acids for NRPS (8,9). The NRPSs and PKSs then function as a multimodular assembly line that synthesizes an inactive precolibactin. Precolibactin biosynthetic intermediates are offloaded from the assembly line by the ClbQ thioesterase, thus possibly regulating colibactin synthesis and genotoxic activity (10,11). Once precolibactin synthesis is finished, the prodrug is taken in charge by ClbM, a multidrug and toxic compound extrusion (MATE) transporter, and released into the periplasmic space (12). Once into the periplasmic space, precolibactin is matured by the ClbP peptidase, which will generate the mature colibactin (13,14) via removal of the N-myristoyl-D-Asn side chain (15,16). However, the mechanism by which colibactin is exported outside the bacteria is still unexplained (17). As a supplemental self-protection mechanism, the pks island encodes the ClbS resistance protein, a cyclopropane hydrolase that inactivates colibactin in the producing bacteria (18,19).
The pks island was detected predominantly in E. coli strains of phylogenetic group B2 (2,4,20), and the prevalence of these strains has increased significantly in developed countries (21,22). Interestingly, E. coli strains carrying the pks island (pks+ strains) are present in about 20% of healthy individuals, about 40% of patients with inflammatory bowel disease, and about 60% of patients with familial adenomatous polyposis or CRC (23)(24)(25)(26)(27). Moreover, one such commensal, the probiotic strain Nissle 1917, commercialized as Mutaflor and efficacy and safety in maintaining remission of intestinal inflammation (28), contains the pks island and produces a functional genotoxin (2). Several studies have also found this pathogenic island in other members of the Enterobacteriaceae family, such as Citrobacter, Klebsiella, and Enterobacter genus (29), in commensal species, such as Frischella perrara (30), Pseudovibrio (31) and, more recently, in other families from the Enterobacteriales order (32), suggesting that its transmission is carried out via horizontal gene transfer (29,33). Whether harboring pks+ strains in the long term is a risk factor for developing colorectal tumors is left for further epidemiological studies. However, it has already been reported a distinct mutational signature in patients suffering CRC that seem to be directly related to the exposure to bacteria carrying the pks pathogenicity island (34).
Flow cytometry (FC) has been used for the analysis of bacteria since the 1970s (35)(36)(37). FC allows individual cell analysis and presents some advantages such as rapid data acquisition and multiparameter analysis. There are a lot of applications for FC in the field of microbiology (38,39), and recent studies have used this technique for the detection, isolation, and cultivation of specific bacteria targeting predicted cell surface proteins (40,41).
In the present article, we developed an in silico method for the detection of the pks island and we applied it on more than 6,000 E. coli human isolates from the PATRIC database. Results showed that several genes from this cluster are exclusive of these colibactin-producing strains, and we discovered pks-specific peptides derived from these pks-associated genes. Using antibodies against four of these pks-specific peptides, an experimental methodology for detection and depletion of pks+ bacteria in gut microbiotas was proposed. This methodology can be applied for the modification of a microbiota in a target manner, allowing the study of the contribution of these pks-carrying strains to different diseases, such as CRC, or their role in other physiological, metabolic, and immune processes. The main results are presented next.

Bacterial genomes and proteomes
The proteome sequence data of the E. coli strains considered in this study were downloaded from the PATRIC database (42). In total, 6,212 strains isolated from humans (tagged Homo sapiens at PATRIC) were retrieved from their public FTP site (ftp:// ftp.patricbrc.org/genomes/). Accession numbers are listed in Table S1.

Distribution of the colibactin genome island
In total, 6,212 E. coli strains isolated from humans and deposited in the PATRIC database were selected for the study of the distribution and structure of the colibactin genome island (pks+ strains). For colibactin island detection, proteomes from the different strains were aligned against the published sequence of the colibactin genome island of the E. coli strain IHE3034 (accession no. AM229678) using BLASTp. For positive detection, 20 out of 23 genes must be detected with less than five gaps between consecutive genes. For each gene, a cutoff value was set based on the alignment expected value (E-value < 1e −5 ), sequence coverage (coverage > 80%), and identity (identity > 80%). The complete list of pks+ strains is listed in Table S1.

Selection and synthesis of pks+ unique peptides
The proteins of the colibactin genomic island of the E. coli strain IHE3034 were split into peptides of 20 amino acids. Later, the proteomes of the 6,212 E. coli strains from the PATRIC database were aligned against these peptides using BLASTp. For positive detection, a cutoff value was set based on the alignment expected value (E-value < 1e −5 ), sequence coverage (coverage > 80%), and identity (identity > 80%). The four peptides with the biggest detection score on the pks+ strains and the lowest detection score on the pks− strains were selected. These 20 amino acid length peptides were located, according to PSORTb v3.0 subcellular predictions (43), on periplasmic loops of three membrane proteins, that is, clbH (NRPS), clbC (PKS), and clbD (NRPS/PKS).
The selected four peptides were further aligned against the NCBI RefSeq database (accessed at 16 March 2023) (44) to screen for positive hits against other bacterial taxa. For positive detection, a cutoff value was set based on the alignment expected value (E-value < 1e −5 ), sequence coverage (coverage > 80%), and identity (identity > 80%). Non-E. coli positive hits are reported in Table S3.
About 5 mg of the four pks-specific peptides was synthesized at the Genecust Europe facilities (Laboratoire de Biotechnologie du Luxemburg S.A., Dudelange, Luxembourg), with a minimum purity of 95%. Peptides were conjugated to the keyhole limpet hemocyanin as carrier protein to favor the generation of peptide-specific antibodies.

Control bacterial strains and growth conditions
E. coli LMG2092 and Nissle 1917 strains were selected as negative and positive control for pks+ bacteria, respectively. Both strains were grown in Luria-Bertani (LB) broth containing 10 g/L tryptone (Biokar Diagnostics, France), 5 g/L yeast extract (Biokar Diagnostics, France), 10 g/L of NaCl (Merck, KGaA, Darmstadt, Germany), and 1 L of deionized H 2 O. All cultures were grown on the surface of agar plates at 37°C in an MG500 anaerobic chamber (Don Whitley Scientific, West Yorkshire 100, UK) with an atmosphere of 10% (v/v) H 2 , 10% CO 2 , and 80% N 2 for 48 h. After that, 1 single colony was inoculated in 4 mL of fresh liquid media and all cultures were incubated at 37°C for 24 h in anaerobic, aerobic, and aerobic with shaking conditions. The next day, 100 µL of the bacteria suspension was inoculated into 4 mL of fresh medium and incubated at 37°C until an optical density (OD 600 ) of about 0.6 after approximately 3 h. After that, cultures were harvested by centrifugation, washed once with bacterial FC buffer (Miltenyi, Bergisch Gladbach, Germany) and resuspended in the same buffer to an OD 600 = 0.2, which represents around 1 × 10 8 CFUs/mL.
The bacteria were also grown in fresh liquid media of Nutrient broth (Oxoid, Ltd., Basingstoke, Hampshire, UK) and nutrient broth supplemented with 2% of glucose (Sigma-Aldrich, St. Louis, MO, USA).

Fecal sample collection and microbiota extraction
The study sample comprised of 13 fecal samples from healthy donors. Fresh fecal material from healthy donors was collected in a sterile container and immediately manipulated and homogenized within a maximum of 2 h from defecation. About 9 Research Article mSystems mL of sterile NaCl 0.9% (w/v) was added to 1 g of the sample, and the mixture was homogenized in a sterile bag using a laboratory paddle blender (Stomacher Lab Blender 400, Seward Ltd., UK). Microbiota extraction was then performed following the protocol described by Hevia et al. (45). A solution of Nycodenz 80% (w/v) (PROGEN Biotechnik GmbH, Heidelberg, Denmark) was prepared in ultrapure water, and sterilized at 121°C for 15 min. A volume of 3 mL of the diluted, homogenized fecal sample was placed on top of 1 mL of the Nycodenz solution and centrifuged for 40 min at 4°C (9,000g, TST41.14 rotor, Kontron, Milan, Italy). The upper phase (soluble debris) was discarded after centrifugation, and the layer corresponding to the microbiota was collected, washed once, and resuspended in 1 mL of FC buffer (1× MACSQuant Running Buffer, Miltenyi Biotec, Germany).

Polyclonal antibody generation
Polyclonal sera against the purified pks-specific peptides were generated in the Central Facilities of the University of Oviedo (Spain). A rabbit was immunized five times, with an interval of 15 days between immunizations, with 500 µg of peptide dissolved in 1 mL of phosphate-buffered saline (PBS) and mixed with 1 mL of Freund's Incomplete Adjuvant.
The rabbit was finally sacrificed by intracardiac puncture and blood was let to coagulate at 37°C for 4 h and subsequent overnight incubation at 4°C. Serum was separated by centrifugation (30 min, 2,000g), and used for purifying the IgG. First, ammonium sulfate was added to a final concentration of 45% (w/v), and the mix was incubated overnight at 4°C. After centrifugation (1 h, 10,000g, 4°C), the pellet was resuspended in 30 mL of PBS. This was extensively dialysed against PBS, and loaded in a ProteinA Sepharose 4 Fast Flow, previously equilibrated with 10 column volumes of PBS (50 mL). The column was washed with six column volumes of PBS, and five fractions of 5 mL were eluted with citric acid 100 mM pH 3.0. pH was corrected in each aliquot by adding 1 mL of 1M Tris-HCl pH 9.0. Fractions were mixed, centrifuged in a Vivaspin 20 device (3,000g, molecular weight cutoff of 10 kDa) and washed with 20 mL of PBS. Protein concentration was estimated by measuring the A 280 and samples were aliquoted and stored at −80°C.

Antibody conjugation
Polyclonal of antipeptide 1, antipeptide 2, antipeptide 3, and antipeptide 4 serum IgG fractions were conjugated with fluorescein isothiocyanate (FITC) and allophycocyanin (APC) using the commercial kits (Abcam, Cambridge, MA, USA) and following the manufacturer's instructions. For FITC and APC conjugation, the antipeptide polyclonal antibodies were reconstituted in amine-free PBS. About 100 mL of each 1.5 mg/mL antibody was added to the reactive dye for each conjugation. Before adding the antibody to the FITC/APC mix, 10 µL of FITC-Modifier or APC-Modifier reagent was added to the antibody. The antibody-dye mixtures were incubated in the dark at room temperature (RT; 20-25°C) for 3 h. After incubation, 10 µL of FITC-Quencher or APC-Quencher reagent was added and gently mixed. The concentration of antibodies in the final sample was 20 µg/mL.

FC analysis
Labeled cells with polyclonal antibodies were acquired and analyzed in a MACSQuant Flow Cytometer device (Miltenyi Biotec, Germany) using the following acquisition parameters: flow rate set to "low, " uptake volume of 10 µL, FSC set to hyperlogarithmic amplification (370 V), SSC set to hyperlogarithmic amplification (440 V), channel B1 corresponding to the FITC detection set to hyperlogarithmic amplification (370 V) and channel R1 corresponding to the APC detection set to hyperlogarithmic amplification (360 V). At least 10,000 events were acquired in each run.

Detection of E. coli Nissle 1917 and LMG2092 by FC
The binding capability of the four antipeptides antibodies was tested over a pks+ (E. coli Nissle 1977) and a pks− (E. coli LMG2092) strains. About 25 µL of the bacterial suspension was mixed with 25 µL of the FITC-conjugated antibody at a final concentration of 20 µg/mL. The samples were incubated for 15 min at RT and were then washed with FC buffer at 13,000 rpm for 5 min. Finally, bacteria were resuspended in 150 µL of FC buffer and samples were acquired using a MACSQuant Flow Cytometry (Miltenyi Biotec, Germany). Bacteria were labeled and analyzed at a different optical density (i.e., OD 600 =0.3, 0.6, 1, and 1.5) and with a different medium (i.e., LB, Nutrient Broth, and Nutrient Broth supplemented with 2% of glucose).

Detection and depletion of pks+ bacteria from gut microbiotas
The detection and depletion of pks+ strains were investigated over 13 gut microbiotas from healthy donors. Microbiotas were labeled with antipeptide 2 polyclonal antibodies conjugated to APC and incubated for 15 min at RT. Then, they were washed with FC buffer at 13,000 rpm for 5 min and the supernatant was removed. Finally, microbiotas were resuspended in 150 µL of FC buffer and data were acquired and analyzed using a MACSQuant flow cytometer as indicated previously.
For depletion, microbiotas were labeled with the polyclonal antipeptide 2 serum IgG fraction conjugated to APC and incubated for 15 min at RT. Then, they were washed with FC buffer at 13,000 rpm for 5 min and the supernatant was removed. For the positive selection of E. coli pks+, microbiotas were resuspended in 100 µL of resuspension buffer (PBS supplemented with 2 mM EDTA and 3% (v/v) of de-complemented bovine fetal serum). To this mix, 20 µL of magnetic anti-APC microbeads were added (BD Biosciences, San José, CA, USA). The mixture was incubated for 15 min at RT with shaking. Then, the microbiotas were washed at 13,000 rpm for 5 min with 1× BD IMag buffer (BD Biosciences, San José, CA, USA) and the column was conditioned with 1× BD IMag buffer (BD Biosciences, San José, CA, USA). The positive fraction was retained in the magnetic column after three washes with 1× BD IMag buffer (BD Biosciences). Finally, the positive fraction was eluted through the 1× BD IMag buffer. Both positive and negative fractions were further analyzed by FC using a MACSQuant flow cytometer as previously indicated.

Immunofluorescence microscopy
The binding ability of the antipeptide 1, 2, and 3 antibodies contained in the polyclonal serum was also determined by a confocal scanning laser microscope equipped with a Leica DFC365FX digital camera (DMi8, Leica Microsystems). The bacterial species were grown until exponential phase OD 600 of 0.7, were then washed with FC buffer, fixated with paraformaldehyde 4%, permeabilized with FC buffer supplemented with 2% of Tween 20 and incubated for 15 min with the polyclonal antibody antipeptides conjugated with FITC (200 µg/mL). Finally, bacteria were washed once with FC buffer, resuspended in 10 µL of flow cytometer buffer and were analyzed with confocal microscopy using a 100× oil objective. The images were acquired with software LasX (Leica Microsystems). The FITC filter cube (excitation 480/40 and emission 527/30) was used.

Evaluation of the pks+ detection and depletion method by qualitative PCR
Conventional qualitative PCR was performed to evaluate the proposed method on four human gut microbiotas from healthy subjects. The four samples were supplemented with 0.1%, 1%, and 10% of E. coli Nissle 1917 cells, and the positive and negative fractions were assessed. Primers amplifying a region of the clbB gene were used following previously published protocols and primers (forward primer 5′-GATTTGGA-TACTGGCGATAACCG-3′ and reverse primer 5′-CCATTTCCGTTTGAGCACAC-3′) (4). About 20 mL of reaction mixture for conventional qualitative PCR consisted of 1 µL of DNA (DNA concentrations are summarized in Table S5), each 0.6 µL of 10 µmol/L forward and reverse primer solutions, deoxynucleotide mixture, the DNA polymerase Master Mix RED (Ampliqon), and H 2 O. The PCR conditions were 10 min at 95°C, 30 cycles of 45 s at 95°C, 45 s at 54°C, and 1 min at 72°C, and a final step of 7 min at 72°C. E. coli Nissle 1917 was selected as a positive control for pks+ detection.
One of the experiments that was carried out consisted of passing a mixture of E. coli Nissle 1917 and E. coli LMG2092 labeled with polyclonal antibody antipeptide 2 several times through the magnetic column in order to check if we were able to increase the recovery of E. coli Nissle and decrease by the positive fraction the recovery of E. coli LMG2092. The mixture contained 50% Nissle and 50% LMG2092. About 250 mL of magnetic beads was added and it was tried to pass one, three, and six times through the column. Results for checking enrichment and depletion of Nissle 1917 and LMG2092 were acquired by FC.

Isolation and identification of the pks+ bacteria
The positive fractions were streaked in agar plates of LB and EC with reduced bile salt (Oxoid) and incubated at 37°C. Random colonies were isolated and grown in fresh liquid LB broth and incubated at 37°C. DNA from isolates was extracted using the GenElute Bacterial Genomic DNA kit (Sigma-Aldrich) and subjected to PCR amplification with primers 27f (5′-AGAGTTTGATCMTGGCTCAG) and 1492r (5′-TACCTTGTTACGACTT). 16S identification was performed, and sequences were aligned against the NCBI nonredundant database using BLAST (46).

Validation of the pks+ isolates by qualitative PCR analysis
Conventional qualitative PCR was performed to evaluate the isolates of the fecal samples retrieved in the positive fraction of the magnetic separation. To detect pks+ bacterial DNA, primers amplifying a region of the clbB gene were used following the previously published protocols and primers (forward primer 5′-GATTTGGATACTGGCGATAACCG-3′ and reverse primer 5′-CCATTTCCCGTTTGAGCACAC-3′) (2,47). About 20 mL of reaction mixture for conventional qualitative PCR consisted of 1 µL of DNA (DNA concentrations are summarized in Table S5)

Genome sequencing, assembly, and annotation
Fourteen isolates from the positive fraction of the fecal samples were selected to sequence their entire genome. DNA library preparation was performed using the Nextera XT DNA sample preparation kit (Illumina, San Diego, CA, USA) according to the manufacturer's instructions. About 1 ng input DNA from each sample was used for library preparation. The isolated DNA underwent fragmentation, adapter ligation, and amplification. The ready-to-go libraries were pooled equimolarity, denatured and diluted to a sequencing concentration of 2 pM. Sequencing was performed on NextSeq 550 instruments (Illumina, San Diego, CA, USA), according to the manufacturer's instructions, using the 2 × 150 bp High Output sequencing kit, and spike-in of 1% PhiX control library. Raw sequence reads were preprocessed to remove bad quality and sort reads and assembled using Spades 3.13.0 (48). Contigs with less than 1,000 nt were discarded. The quality of the assembled genomes was checked using CheckM 1.1.2 (49) and sequentially annotated using Prokka 1.12 (50). The assembled genomes are available at https:// www.ncbi.nlm.nih.gov/bioproject/PRJNA668898.

Genome classification and pks cluster detection
The assembled genomes were taxonomically classified using two independent approaches. On the one hand, the 16S sequences were retrieved from the assembled genomes using barrnap 0.9 (https://github.com/tseemann/barrnap) and analyzed with Research Article mSystems RDP classifier (51). On the other hand, the assembled genomes were processed with the phylophlan_metagenomic script of PhyloPhlAn 3.0 (52) over the SGB Jan19 database (53). Moreover, the phylogenetic group of the E. coli strains was predicted using EzClermont 0.4.5 (54).
For pks island detection, the annotated genomes were aligned against the published sequence of the colibactin genome island of the E. coli strain IHE3034 (accession no. AM229678) using BLASTp. For each gene, a cutoff value was set based on the alignment expected value (E-value < 1e −5 ), sequence coverage (coverage > 80%), and identity (identity > 80%).

In silico analysis of the pks island distribution
We obtained 6,212 E. coli human isolates from the PATRIC database to study the distribution and structure of the pks island. After screening, we detected the pks island in 1,226 of them. However, not all harbored all the genes present in the reference pks island described for strain IHE3034 (Table S1). The intP4, clbS, clbQ, clbK, clbJ, trpA, and trpB genes were detected with a percentage of distribution between 97% and 99%, while for the clbR, clbA, and trpC genes their percentage was lower, between 70% and 78%, with clbR being the gene detected in only 70% of the strains. In addition to the fact that not all the genes are present in most of the distributions of the island, there are genes not belonging to the island distributed in the inner part of the cluster (Fig. 1).

Selection of the pks-specific peptides
We further screened the 6,212 E. coli strains to select pks-specific peptides for antibody generation. We split the proteins of the colibactin genomic island of the IHE3034 strain into peptides of 20 amino acids and mapped them against the 6,212 E. coli strains (see Materials and methods). We selected the four peptides with the biggest detection score on the pks+ strains and the lowest detection score on the pks− strains (Table S2). These 20 amino acid length peptides were predicted on three different proteins of the pks cluster, that is, clbH (NRPS), clbC (PKS), and clbB (NRPS/PKS).
We further screened the four pks-specific peptides against the NCBI RefSeq database (44) to assess the specificity of the selected peptides in other bacterial species (see Materials and methods). With the exception of peptide #3, the other peptides appear to be specific from the colibactin-producing machinery, primarily showing significant hits against species from the Enterobacteriaceae family previously reported carrying the pks genomic island such as Klebsiella, Citrobacter, or Serratia spp. (Table S3).

Selection of the best labeling conditions
We evaluated the labeling of a pks+ (E. coli Nissle 1917) and a pks− (E. coli LMG2092) strains with antipeptide 1, 2, 3, and 4 polyclonal antibodies by FC. FC analyses showed the capability of the antibodies against three of these peptides to detect the pks+ strains, labeling approximately a 50% of the E. coli pks+ cells analyzed (antibodies against peptide #4 were less effective and, therefore, it was discarded for the following experiments; Fig. 2a). We also evaluated the recognition of the pks+ strain with polyclonal antipeptide 1, 2, and 3 antibodies conjugated with FITC using a confocal laser scanning microscope (Fig. 2b, Fig. S1).
In an attempt to improve these results, different growth conditions and optical density were taken into account to discover their influence in the expression of the pks machinery (Table S4). Best results (labeling ~76% of the pks+ and ~1% of pks− cells) were obtained using the antibody directed to the peptide #2 over cells grown on Nutrient Broth medium at optical density of 1 OD 600 . Negative controls, using the preimmune serum conjugated with APC, further validate the specificity of the selected antibody (Fig.  S2). The higher emission of E. coli Nissle and the lower emission of the E. coli LMG2092 can be deduced by comparing the scatter and histogram plots (Fig. 2c).

Detection and isolation of pks+ bacteria from human gut microbiotas
Four stool samples from healthy subjects (V-HD1 to 4) were collected to validate the suitability of the proposed method to detect, isolate and eventually deplete pks+ cells from a complex bacteria community. This method is based on the recognition of a specific molecule on the surface of pks+ bacteria. In this case, the chosen target was a peptide loop inside one of the proteins encoded in the pks operon. Polyclonal antibodies developed against these peptides allowed to obtain, after an immunomagnetic step, two bacteria fractions, denominated positive and negative. Positive fractions contained bacteria retained by the antibodies, which were expected to be enriched in pks+ bacteria. On the contrary, negative fractions contained the rest of microorganisms with an expected depletion of pks+ bacteria. We applied this procedure on the microbiotas extracted from the four stool samples, supplemented with a proportion of 0.1%, 1%, and 10% of E. coli Nissle 1917 cells. Then, we assessed the pks+ presence in the positive and negative fractions by qualitative PCR (Fig. S3; see Materials and methods). The results showed that it is possible to deplete a portion of the E. coli pks+ cell population from a human gut microbiota, but only when they are present above a certain abundance. In fact, with 0.1% and 1% supplementation, it was possible to detect pks+ cells only in the positive fraction of one donor (V-HD1), while in all the positive fractions supplemented with a proportion above 10% a positive detection was observed. However, pks+ cells In order to test whether repeated immunomagnetic steps could improve the depletion results, we repeated the experiment using a synthetic community containing both a pks+ (E. coli Nissle 1917) and a pks− (E. coli LMG2092) strain, and measured the proportion of pks+ cells in the positive and negative fractions after one, three, and six steps using FC. According to the number of passes, the amount of pks+ cells detected in the negative fraction decreased from 0.59% after the first step to 0.29% after six steps (Fig. S4).
After validating the suitability of the methodology, the same stool samples (HD1 to 4; in this case without any pks+ supplementation) together with additional nine stool samples from healthy donors (HD5 to 13), were processed to detect (Fig. 3a, Fig. S5) and Research Article mSystems deplete from pks+ strains. After depletion, the positive fractions from these 13 samples were cultivated to validate the viability of the depleted cells. From these cultures, 39 isolates were retrieved, their 16S region was sequenced, and subsequently validated of the clbB gene by PCR analysis (Fig. 3b). Moreover, the complete genome of 14 of these 39 isolates was sequenced, assembled, taxonomically classified, and analyzed to detect the pks island (Table 1, Table S5). The results obtained with the PCR detection of the pks cluster were in agreement with the results obtained by analyzing the assembled genomes. Four out of the 14 analyzed isolates were detected as pks+ bacteria, all of them belonging to the phylogroup B2 of the E. coli species (Table 1, Table S5).

DISCUSSION
The human gut microbiome has been hypothesized to play a crucial role in the development and progression of CRC (27,(55)(56)(57). Among the different microorganisms potentially involved, the E. coli strains carrying the pks genomic island were shown to be capable of in producing genomic aberrations on colon epithelial cells (58), colon tumorigenesis in mice models of chronic inflammation and colorectal cancer (23,59), and their presence seems to be directly related to a distinct mutational signature in patients suffering CRC (34). Acknowledging this relation and therefore the carcinogenic potential of these strains, identification, isolation, and depletion of these bacteria could help to understand the role of these strains not only in the context of CRC, but also in other diseases or physiological, metabolic, and immune processes.
Previous studies determined which strains were colibactin producers based on the detection of the clbB gene by PCR (25). However, to our knowledge, studies using antibodies against specific peptides of the pks island have never been designed. For this reason, we selected four pks-exclusive peptides from genes presented in all the pks distributions to generate polyclonal antibodies against them, and different growth conditions and optical densities were tested to determine their influence in the expression of the pks machinery.
These results showed that our methodology was able to detect pks+ strains in simple bacteria mixes. The next step was to assess whether it was also suitable for detecting, isolating, and depleting pks+ bacteria in human gut microbiotas. Therefore, we enriched four human gut microbiotas obtained from healthy donors with a proportion of 0.1%, 1%, and 10% of E. coli Nissle 1917 cells, and positive/negative fractions were obtained using our immunomagnetic protocol. In general, presence of pks+ bacteria was only detected in positive fractions obtained from gut microbiotas supplemented with 10% pks+ bacteria. Further investigation is therefore necessary to establish the appropriate conditions for optimized labeling of colibactin-producing bacteria in complex microbial communities.
In contrast to methods based on probes, our methodology allowed identification and certain level of depletion of pks+ bacterial strains. Further culture of depleted pks+ bacteria and t positive fractions from 13 gut microbiotas from healthy donors, allowed isolation of 39 potential pks+ isolates belonging to the Enterobacteriaceae family, for example, Escherichia, Klebsiella, Enterobacter, and Citrobacter genus. However, after PCR and WGS validation, we discovered that not all the bacteria retrieved harbored the pks island, and only 4 out of 14 isolates were detected as pks+ and were identified as belonging to E. coli species. The depletion of pks− bacteria from other not yet described potential pks+ genus, such as Enterococcus and Serratia, may be related to the polyclonal nature of the antibody, raised in an animal that contains Firmicutes and y-Proteobacteria as part of the normal microbiota.

Conclusions
To our knowledge, this represents the first method for the detection and depletion of pks+ bacteria in human gut microbiotas based on in silico predicted pks-specific anti-peptide antibodies. In contrast to methods based on probes, our methodology allows the depletion of low-abundance bacterial strains maintaining the viability of both targeted and non-targeted fractions of the microbiota, allowing the study of the contribution of these pks-carrying strains to different diseases, such as CRC, and their role in other physiological, metabolic, or immune processes.
The methodology presented in this work allows detection and isolation of pks+ bacteria using specific antibodies targeting extracellular loops of the pks machinery.

Research Article mSystems
However, care should be taken with this experimental approach as it would require further developments in order to obtain complete colibactin-producing bacteria depletion, as cells with no or low level of pks+ gene expression are not completely eliminated. Future developments of this methodology, such as implementation of fluorescence-activated cell sorting and monoclonal antibodies with high affinity will render a more advanced and efficient protocol that, coupled with other methodologies such as microbiota transplant, might be capable of modifying the gut microbiota in an animal model after modified-microbiota reimplantation.

ETHICS APPROVAL
Ethics approval for this study (2016/401) was obtained from the Bioethics Committee of CSIC (Consejo Superior de Investigaciones Científicas) and from the Regional Ethics Committee for Clinical Research (Consellería de Sanidades, Xunta de Galicia) in compliance with the Declaration of Helsinki. All determinations were performed with fully informed written consent from all participants involved in the study.

ADDITIONAL FILES
The following material is available online.   FIG S1 (299416_1_supp_6698423_rslh5m.tif). Immunofluorescence microscopy photography showing the binding of the FITC-anti peptide 1 (a) and 3 (b) antibodies to E. coli Nissle 1917 .  FIG S2 (299416_1_supp_6698424_rslh66.tif). Labeling results obtained using antipeptide polyclonal antibodies and the preimmune serum on E. coli LMG2092 and Nissle 1917.  FIG S5 (299416_1_supp_6698429_rslhpp.pdf). Scatter plots and histograms showing the flow cytometry acquisition using the anti-peptides 2 polyclonal antibody on 13 microbiotas from healthy donors (HD1 to 13).