Thirty five ETEC strains were included in the study, each of which were considered negative for known adhesins (AN-ETEC), because no product was obtained in PCR analyses aimed at detecting loci encoding CFA/I, CS1, CS2, CS3, CS4, CS5, CS6, CS7, CS8, CS12, CS13, CS14, CS15, CS17, CS18, CS19, CS20, CS21, CS22, CS23, Tia, TibA, or EtpA. The complete list of primers used in this work is included in Table S2 (Supporting Information). All AN-ETEC strains had been isolated from children under five years of age with watery diarrhea: 14 of the 35 strains were obtained in Chile and had been previously characterized according to O serogroup and enterotoxin genes (Del Canto et al., 2011). The other 21 isolates were obtained as part of the Global Enterics Multicenter Study (GEMS) in Kenya, Mali, Mozambique, The Gambia, Pakistan, India, and Bangladesh (Kotloff et al., 2013). Enterotoxin gene repertoires were determined using standard procedures as described previously by our research group (Del Canto et al., 2011; Panchalingam et al., 2012). For this group of strains (GEMS' strains), serotyping of the O antigen was carried out by seroagglutination at the Programa de Microbiología y Micología of the Universidad de Chile.

To obtain draft genomic sequences for the 35 AN-ETEC strains, genomic DNA was purified using the Wizard Genomic DNA purification kit (Promega), processed according to the Illumina Paired End Protocol with inserts of 400 bp, and sequenced in a HiSeq 2000 platform at the Institute for Genome Sciences (Baltimore, MD). Reads were analyzed using FastQC v0.10.1 (Babraham Bioinformatics, Babraham Institute, Cambridge, UK. Available: http://www.bioinformatics.bbsrc.ac.uk/projects/fastqc/) and assembled using MaSuRCA 2.2.1 or SPAdes 3.1.0 (Bankevich et al., 2012; Zimin et al., 2013). The best of the two assemblies was finally chosen. Best kmer length for assembly was predicted with KmerGenie v.16476 prior to assembly with SPAdes (Chikhi and Medvedev, 2014). Assembly statistics were obtained with Quast v2.3 and the completeness/contamination report with CheckM (Gurevich et al., 2013; Parks et al., 2015). Sequence annotation was first performed using the Rapid Annotation Subsystem Technology (RAST; Aziz et al., 2008) and then through the NCBI Prokaryotic Genome Annotation Pipeline (National Center for Biotechnology Information, Bethesda MD, USA. Available: http://www.ncbi.nlm.nih.gov/genome/annotation_prok/).

A phylogenetic tree was inferred based on single nucleotide polymorphisms in draft genomes of our AN-ETEC, by using the CSI phylogeny 1.1, a tool on the Center for Genomic Epidemiology server (Kaas et al., 2014). The following genomic sequences, containing their respective plasmids, were included in the analysis: E. coli B, E. coli K-12 MG1655, E. coli HS, E. coli W, E. coli ED1a, E. coli IAI-1, E. coli SE11, E. coli SE15 and ETEC prototype strains H10407, E24377A, and B7A. The accession numbers of these sequences are included in Table S3 of the Supporting Information. The genomic sequence of E. coli K-12 MG1655 was set as the reference. A total of 2,481,579 positions were found in all analyzed genomes (53.46% of the reference). The tree was drawn using Mega 6.06 (Tamura et al., 2013). Phylogenetic groups were assigned according to the PCR protocol described by Clermont et al. and complemented with manual analysis of the target genes (Clermont et al., 2013). Sequence types were assigned using the MLST 1.7 of the CGE, according to the scheme proposed by Wirth et al. based on genes adk, fumC, gyrB, icd, mdh, purA, and recA (Wirth et al., 2006; Larsen et al., 2012).

Putative adhesin loci in AN-ETEC genomes were identified first by screening of major CF subunits and then by screening of usher genes using Blast Ring Image Generator (BRIG) v0.95 and/or Large-scale Blast Score Ratio (LS-BSR) (Alikhan et al., 2011; Sahl et al., 2014). In LS-BSR, a blast score ratio (BSR) between 0.00 and 1.00 was obtained for every screened gene in an individual genome. BSR values of 1.00 indicate the maximum sequence identity. For CF screening, genes encoding the following structural subunits were considered: CFA/I, CS1, CS2, CS3, CS4, CS5, CS6, CS7, CS8, CS12, CS13, CS14, CS15, CS17, CS18, CS19, CS20, CS21, CS22, and CS23. For the usher screening, a set of 162 genes of CU pili belonging to the α (16 genes), β (5 genes), γ₁ (25 genes), γ₂ (7 genes), γ₃ (15 genes), γ₄ (40 genes), κ (9 genes), π (25 genes), and σ families (20 genes), were selected based primarily on a previous review (Nuccio and Bäumler, 2007). The complete list of genes used for screening is included in Table S1 (supporting information). Homolog genes were located within AN-ETEC genomes and their genetic context was analyzed to look for open reading frames (ORFs) encoding putative structural subunits, chaperones, transcriptional regulators, and others that could be part of a CU pili locus. Gene clusters were characterized and compared to known CU pili loci using the NCBI ORF finder (National Center for Biotechnology Information, Bethesda MD, USA. Available: http://www.ncbi.nlm.nih.gov/projects/gorf/), Clustal Omega and/or Unipro UGENE (Sievers et al., 2011; Okonechnikov et al., 2012). In most cases, phylogenetic trees were inferred according to amino acid sequence alignment using Mega 6 (Tamura et al., 2013). Heat maps for screening with LS-BSR or percentage of identity were draw with the Multiple experiment Viewer (MeV) v4.9.0, and loci comparison graphics were drawn with EasyFig v2.1 (Saeed et al., 2006; Sullivan et al., 2011).

The 35 AN-ETEC genomes were partially sequenced as part of an ETEC whole genome shotgun project and deposited in GenBank under the accession BioProject PRJNA287625. In addition, a file containing the amino acid sequences of the major structural subunits was included as Supplementary Material (Supplementary Data Sheet 1). Chilean bacterial isolates may be requested by contacting corresponding authors and GEMS isolates may be requested at http://www.medschool.umaryland.edu/GEMS/GEMS-Data--Specimen-Requests/.

Strains in this study had been previously characterized by PCR and all of them had tested negative for known ETEC CFs genes. We used the same genes used in the PCR screening to probe the draft genomes using LS-BSR with the tblastn algorithm (Sahl et al., 2014). As seen in Figure 3, CF genes were not detected in non-pathogenic E. coli strains, with blast-score ratio (BSR) values ranging between 0.00 and 0.32. In contrast, they were correctly detected in ETEC prototype strains H10407 (CFA/I), ETEC E24377A (CS1 and CS3), and ETEC B7A (CS6) with BSR values ranging between 0.97 and 1.00. Although most of these AN-ETEC genomes were confirmed negative for known CF genes by this second screening method, the blast-score ratio (BSR) values obtained in some cases (between 0.57 and 0.79) for genes encoding CS8/CS21 (ETEC 401909), CS20 (ETEC 8255a, ETEC 8350a-1, ETEC200144, ETEC200144, ETEC 200617, ETEC 702251 and ETEC 505077), and CS23 (ETEC 100664, ETEC 402594, ETEC 702251 and ETEC 703098), suggest the presence of homologs to those genes.

CS8 and CS21 are homologous structures belonging to the type IV pili family (Madhavan and Sakellaris, 2015). Interestingly, a gene homologous to cofA and lngA, encoding CS8 and CS21 pilins, respectively, was localized in the ETEC 401909 strain (Kenya, O9, phylogroup C, ST-410, LT), the only AN-ETEC strain with relatively high BSR values (0.79 for CS8 and 0.75 for CS21). Analysis of the PCR primers' target sequences for cofA and lngA showed discrepancies that may explain negative PCR results (Figure S1A in Supporting Information). The full-length gene nucleotide sequence shared 77% identity with cofA and 60% identity with lngA, but was 100% identical with the recently discovered CS8b variant, which has yet to be functionally characterized (Njoroge et al., 2015). The alignment of the derived amino acid sequences allowed us to probe this identity (Figure S1B). Therefore, even though the ETEC 401909 strain was negative for detection of cofA and lngA, it harbored a recently discovered and related variant.

Most of the CFs are CU pili. By screening of the CF structural subunit genes that were targets for PCR detection in AN-ETEC strains, we were able to identify sequences with high similarity (Figure 3). In order to complement this analysis and look for additional potential loci homologous to known CU pili loci, we bioinformatically screened a set of 162 usher genes from AN-ETEC genomes, using the blast- ring image generator (BRIG) with the blastn option and LS-BSR with tblastn (Alikhan et al., 2011; Sahl et al., 2014). Usher genes were chosen because they are present only in a single copy within the CU loci. This analysis included the genomes of 13 non-pathogenic E. coli strains and three ETEC prototype strains. As shown in Figure 4, the results derived from both approaches were similar. As previously reported, there is a set of core CU pili loci in E. coli including mat (α family), fim (γ₁), yde (γ₁), yad (γ₄), yeh (γ₄), yfc (π), and ybg (π) (Wurpel et al., 2013). Usher genes from all of these loci were detected in both non-pathogenic and pathogenic ETEC genomes. Usher genes from loci encoding known CFs of ETEC prototype strains were detected, namely CFA/I of ETEC H10407 and CS1 of ETEC E24377A in the α family, along with CS3 of ETEC E24377A and CS6 of ETEC B7A in the γ₃ family. Consistent with CF gene screening, homologous sequences encoding ushers from the γ₂ and κ families were found in several AN-ETEC strains but were not detected in the majority of the non-pathogenic strains. Similar results were observed for usher genes from the β family and from some representatives of the π family. No homologs of usher genes from the σ family were detected in our set of AN-ETEC genomes. Our results suggest that AN-ETEC strains may contain CU-pili loci encoding structures from the β, γ₂, κ, and π families.

According to our CF gene screening, genes homologous to aalE, which encodes the CS23 major subunit, would be harbored by four AN-ETEC strains (and were not restricted to any particular phylogroup or ST): ETEC 100664, ETEC 402594, ETEC 702251, and ETEC 703098. This observation was consistent with the fact that sequences homologous to usher genes belonging to the kappa family were detected in these strains (BSR values between 0.41 and 0.99, Figure 4B). However, the aalE primer did not match perfectly in any of these genomes, which explains the negative PCR results (Figure 5A). Based on the organization of the genetic cluster encoding CS23 and CS13 (Nuccio and Bäumler, 2007; Madhavan and Sakellaris, 2015), complete loci were identified in three of the four mentioned AN-ETEC strains (ETEC 100664, ETEC 402594, and ETEC 702251). This includes six genes encoding structural subunits, one for a chaperone, and one for the usher (Nuccio and Bäumler, 2007; Madhavan and Sakellaris, 2015). In the case of the ETEC 703098 strain, the sequencing did not cover the complete cluster and only five structural subunits were found: the chaperone and the 3′ half portion of the usher gene. Alignment of the amino acid sequences of the putative major structural subunits (PMSs) indicated that these identified loci would encode pili more closely-related to CS23 than to CS13 (Figure 5B). Identity percentages between AalE and the novel PMSs were 68% for ETEC 402594, 67% for ETEC 100664 and ETEC 702251, and 59% for ETEC 703098. As κ-CU pili loci are also present in the genomes of non-pathogenic E. coli W and E. coli SE11, we included these in the analysis. However, the κ-PMS sequences derived from both strains are different from those found in AN-ETEC strains, sharing only 26 and 25% amino acid identity with AalE (Figure 5B). The regions of identity among all the κ-PMSs resided mainly in the first 30 amino acids, which corresponds with the predicted AalE signal sequence for secretion to the periplasm (Del Canto et al., 2012).

Analysis of the κ-CU pili loci of AN-ETEC strains showed that they share the same genetic organization (Figure 5C). A gene encoding a putative transcriptional regulator, similar to that harbored in the CS23 locus, was found in the three AN-ETEC strains having a complete κ-CU pili locus. In addition, a putative seventh structural subunit gene was found in the ETEC 100664 and ETEC 702251 κ-CU pili loci, in which the contig length allowed for recognition of downstream sequences. This was located downstream a gene encoding a putative phosphodiesterase (EAL domain-containing protein). Alignment using tblastx indicated that κ-CU pili loci carried by ETEC 100664 and ETEC 702251 are nearly identical, and differ from loci harbored by ETEC 402594 and the aal locus, mainly in the PMS sequence. A higher level of dissimilarity for all ORFs was observed between the κ-CU pili loci of AN-ETEC strains, the csh locus (CS13), and κ-CU pili loci of E. coli W and E. coli SE11.

Screening of usher genes also allowed us to identify a locus similar to the κ-CU pili loci, encoding fimbriae AF/R1 of rabbit enteric pathogen E. coli RDEC-1 and F18 of porcine ETEC in the AN-ETEC 602687 strain. However, as sequencing did not allow for assembly of a complete locus, we did not analyze this cluster.

Genomes of six AN-ETEC strains showed high BSR scores for a gene encoding the major structural subunit of CS20, suggesting the presence of γ₂-CU pili loci in AN-ETEC strains (Figure 3). After complementing this information with the results from our usher genes screening, γ₂-CU pili loci were identified in 23 AN-ETEC genomes (BSR values ranging between 0.46 and 1.00, Figure 4B). According to the genetic organization of loci encoding γ₂-CFs CS12, CS18, and CS20, the loci identified in AN-ETEC strains do have the minimum number of genes required to direct synthesis of a functional structure, including three encoding structural subunits, three for chaperones, and one for an usher (Nuccio and Bäumler, 2007; Madhavan and Sakellaris, 2015). However, alignment of amino acid sequences of the PMSs proved that none were identical to the currently known major structural subunits CswA (CS12), FotA (CS18), or CsnA (CS20) and that most of these were closely related to the last two (Figure 6A; Nuccio and Bäumler, 2007; Madhavan and Sakellaris, 2015). Overall, there were six groups derived from the alignment, containing between two and five 100% identical PMS sequences. Putative γ₂-PMS from two of these groups (PMS harbored by ETEC 200617, ETEC 505707, ETEC 8255a and ETEC 200144, ETEC 702251) plus that harbored by ETEC 8350a-1, shared about 76% identity with CsnA. Five of these six strains belong to the phylogroup A and one to the phylogroup D, representing four different serogroups and five sequence types.

Eight AN-ETEC strains (the group of ETEC 4155a, ETEC 603388, ETEC 1030c-1, ETEC 963a-1, ETEC 703200, the group of ETEC 3693, ETEC 503210, plus ETEC 100664) harbor γ₂-PMS with 58–59% identity to FotA (Figure 6A). This group includes five phylogroup A strains and three phylogroup B₁ strains. Other five strains (ETEC 10380a and the group of ETEC 9312a, ETEC 9343a, ETEC 9788a, ETEC 401900) harbor γ₂-PMS that share 59% identity with CsnA. Four of these strains belong to phylogroup A and three are closely related to ST-100 strains isolated in Chile. Within the other four AN-ETEC strains carrying γ₂-CU pili loci, there are two identical PMS sharing 51% identity with CsnA (ETEC 1517a and ETEC 203518) and two different sequences with 50 and 49% identity to CsnA (ETEC 1241a and ETEC 702332). Alignment revealed a block of eight amino acids conserved across CS20 and all the γ₂-PMS of the AN-ETEC strains between positions 50 and 57 (Figure 6A). This region corresponds to the residues 18–25 of the mature CsnA protein (Valvatne et al., 1996). Other 33 conserved residues among all the γ₂-PMS and the known γ₂-CFs major subunits were found distributed along the whole extension of the sequences and including single residues, as stretches ranging between two and four amino acids (Figure 6A).

Analysis of the genetic structure and alignment of the complete γ₂-CU pili loci using tblastx showed that the structure is maintained across AN-ETEC strains and those identity-percentages are consistent with groups described based on γ₂-PMS sequences (Figure 6B). Differences between CS20 loci and those γ₂-CU pili loci included in the first group of six AN-ETEC strains resided mainly in the PMS, while, in the rest of the γ₂-CU pili, it also depends on the putative chaperones, minor structural subunits and even usher genes.

We aligned sequences with those of five, previously described, partial γ₂-PMS gene sequences: CS26, CS27A, CS27B, CS28A, and CS28B (Nada et al., 2011), using blastn, and found both high similarities and identities. The gamma-2 PMS gene found in ETEC 100664 shared 91% identity with CS26 (Figure 6C). CS27A did not share high identity (≥90%) with any gene sequences in our study. In contrast, a group of seven AN-ETEC strains harbored γ₂-PMS genes containing sequences highly similar (94–96%) to CS27B. This group also includes strains whose γ₂-PMS shared 58–59% identity with FotA. Three AN-ETEC bear γ₂-PMS genes sharing high identity with CS28A (96% identity), as was other three AN-ETEC bear γ₂-PMS genes almost identical to CS28B with (97% identity). Together, these results suggest that γ₂-CU pili, other than CS12, CS18, and CS20, might be part of the ETEC adhesin repertoire.

None of the known ETEC CFs exhibit π-CU pili, but in our screening of usher genes, we identified three strains from the B₁ phylogroup that were positive for homologs of the π-CU pili family (BSR values ranging between 0.70 and 0.82, Figure 4B), namely sfpC, pixC, prfC, and papC. These usher genes represent loci that differ from the π-CU pili loci yfc, ybg, and yqi, which were found to be widely distributed among E. coli strains (Wurpel et al., 2013). Two of these strains, ETEC 700377 and ETEC 300659, were closely related and were classified as ST-328, while the third, ETEC 302025, was classified as ST-155. Localization of the loci in these three genomes allowed for identification of clusters composed of seven ORFs that encoded four structural subunits, two chaperones and the usher. Alignment of the amino acid sequences derived from the PMS gene using known π-major structural subunits PapA, PrfA, PixA, and SpfA revealed a higher degree of identity with SpfA (67% identity with π-PMS of ETEC 700377 and ETEC 300659, and 65% with π-PMS of ETEC 302025; Figure 7A). Four stretches of four amino acids in the N-terminal region of the predicted mature proteins (Brunder et al., 2001), and one of seven amino acids in the C-terminal region found to be conserved across all the sequences included in the analysis (Figure 7A). Aligning the genetic clusters using tblastx showed similarities to known loci residing mainly in genes encoding the PMS, the adjacent gene encoding a putative structural subunit gene, the usher, and the two chaperones, although differing in the downstream putative structural subunit genes. In contrast, among the AN-ETEC π-CU pili loci (ETEC 700377/ETEC 300659 v/s ETEC 302025), dissimilarity resided mainly in the PMS and the adjacent putative subunit gene (Figure 7B).

A recently described hybrid enteropathogenic E. coli (EPEC)/ETEC strain harbors a π-CU pili locus, identical to the locus harbored by ETEC 700377 and ETEC 300659 (Dutta et al., 2015). Therefore, we computationally screened a set of EPEC genes in the three AN-ETEC strains in which we found π-CU pili loci; these genes were typically contained in the locus of enterocyte effacement (LEE), a pathogenicity island carried by both EPEC and enterohemorrhagic E. coli (Croxen et al., 2013). According to LS-BSR screening using tblastn, almost all of the EPEC genes are present in ETEC 700377 and ETEC 300659, except for bfp (Figure 7C). In contrast, none of the EPEC genes were found in ETEC 302025, suggesting that the presence of π-CU pili loci is not restricted to EPEC/ETEC hybrids.

To our knowledge, CU pili belonging to the β family have never been characterized, and the denomination as an actual pili family has been only been supported by sequence data (Nuccio and Bäumler, 2007). Although the presence of a β-CU pili locus in E. coli K-12 was reported, it is disrupted by an insertion sequence (IS), which may be why it has not received more attention (Korea et al., 2010). Our usher gene screening shows that 15 AN-ETEC genomes, in addition to the three E. coli K-12 substrains, also carry β-CU pili loci (BSR values ranging between 0.90 and 1.00, Figure 4B). All these strains belong to the phylogroup A. Based on the structure of the locus in E. coli K-12 substrain MG1655, 13 AN-ETEC genomes carry a complete cluster, lacking the IS and including five genes: gltF, a putative structural subunit; yhcA, a putative chaperone; yhcD, the putative usher; yhcE, whose function is unclear; and yhcF, another putative structural subunit (Nuccio and Bäumler, 2007; Korea et al., 2010). In one of the AN-ETEC strains (ETEC 401900) a complete locus was found, but a premature stop codon disrupts the yhcD gene. For this reason, this strain was not considered in further analyses.

All of the β-CU pili loci identified share the same organization, with the exception that the insertion sequence disrupting yhcE in E. coli K-12 is absent in all AN-ETEC genomes (Figure 8A). According to the alignment performed with tblastx, the gltF, yhcA, yhcD, and yhcF sequences are highly conserved among E. coli K-12 and AN-ETEC strains (Figure 8A). Only in the ETEC 963a-1 strain, in which sequencing did not allow for the identification of a complete β-CU pili locus, a longer form of yhcE was found. This variant likely encodes a protein with additional 57 amino acids in the C-term, unlike the form found in the other AN-ETEC strains. The variability among the identified β-CU pili loci can be almost exclusively attributed to yhcF, which was found in three different forms. One of them, of 717 bp, was found in nine AN-ETEC strains and corresponds to the form present in E. coli K-12. Two other shorter forms, of 633 and 405 bp, were identified in two and three AN-ETEC genomes, respectively (Figure 8A). Alignment of the predicted protein products is consistent with such an observation (Figure 8B). The three forms have an identical block of 118 amino acids and the divergence is determined by the length of the C-terminal regions.

In summary, at least one pili locus, different from those that seem to be widely distributed among E. coli, was identified in 33 of the 35 AN-ETEC strains included in this study. No loci were identified in strains ETEC 9571a (ONT, phylogroup A, ST-731, LT) and ETEC 603936 (O64, phylogroup B₁, ST-155, STp-LT). We identified 10 different CU pili loci among the 35 AN-ETEC strains analyzed, which, according to the number of ORFs they contain, could encode functional pili structures. A list of these loci, along with their organization and positive strains, is provided in Table 2. None of the AN-ETEC strains seems to bear two β-, γ₂-, κ-, or π-CU pili loci simultaneously, but they do harbor representatives of different families. In both, ETEC 100664 and ETEC 702251, β-, γ₂-, and κ-CU pili loci were identified; while seven strains carry β-, and γ₂-CU pili loci simultaneously (ETEC 8350a, ETEC 200144, ETEC 702332, ETEC 4155a, ETEC 9312a, ETEC 9343a, and ETEC 9788a). No β-, γ₂-, nor κ-CU pili loci were identified among the AN-ETEC strains carrying π-CU pili loci.

In the case of previously described loci, we maintained the established names; for novel gene variants, these names were complemented with subscripts numbers or letters. We now propose five novel loci names. Thus, the locus name yhc was maintained for identified β-CU pili loci, with novel variants of yhcF being named yhcF₂ and yhcF₃ (Korea et al., 2010). For the γ₂-CU pili loci, the previous names crs/CS26, cma/CS27b, cnm/CS28a, and cnm/CS28b were kept in the case of loci found in 14 AN-ETEC strains; and a novel variant of crs/CS26 was identified (crs/CS26b) (Nada et al., 2011). Four novel names were given to the γ₂-CU pili loci found in nine strains: ctg (CU ETEC gamma-2 pilus locus), gep (gamma-2 ETEC pilus locus), gte (gamma-two pilus locus of ETEC) and gtt (gamma-two pilus locus of toxigenic E. coli). As κ-CU pili loci were found in three AN-ETEC strains and displayed high identity with the aal locus along their entire extension, the name aal was maintained and variants for aalE were introduced (aalE₂ and aalE₃; Del Canto et al., 2012). Finally, the acronym ctp (CU ETEC pi pilus locus) was chosen for the π-CU pili loci found in three AN-ETEC strains, with variants for the putative structural subunit genes ctpA and ctpH (ctpA₂ and ctpH₂).

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.