We studied six SPM-1-producing P. aeruginosa isolates, including the index isolate PA1088 (previously named 48-1997A), which was the first reported clinical isolate to carry bla_SPM−1 (Toleman et al., 2002). The remaining five isolates were recovered from distinct patients admitted to a single intensive care unit between the years 2003 and 2012 (Table 1). All isolates were collected from the same tertiary teaching hospital located in the city of São Paulo, Brazil. The presence of bla_SPM−1 was initially confirmed by PCR and DNA sequencing (BigDye Terminator Cycle Sequencing, Applied Biosystems, Foster City, USA) using primers previously described (Mendes et al., 2004). For whole genome sequencing, the bacteria were grown overnight in LB broth (Oxoid, Basingstoke, England) at 37°C. Total DNA was extracted using the Qiamp DNA Stool Kit (Qiagen, Hilden, Germany) according to the manufacturer's instructions. The DNA concentration was measured in a NanoVue digital spectrophotometer (GE Healthcare Life Sciences, New Jersey, USA) and submitted to the Unidade de Genômica Computacional Darcy Fontoura de Almeida (UGCDFA) of Laboratório Nacional de Computação Científica (LNCC) for further analysis.

Six whole genome sequencing libraries were generated using the Illumina TruSeq DNA PCR-free sample preparation kit with a median insert size of 550 bp according to the manufacturer's protocols. Briefly, 2 μg of genomic DNA was sheared using a Covaris M220 Focused-ultrasonicator, end-repaired, A-tailed, and adapter ligated. Library quantification was carried out by real-time PCR. Libraries were pooled together in equimolar amounts and sequenced by an Illumina MiSeq instrument in one 2 × 300 bp paired-end run. Genome assembly was performed using Newbler version 3.0. In addition, Celera assembler version 8.2 was used to close eventual gaps. Gaps within scaffolds resulting from repetitive sequences were resolved by in silico gap filling. We achieved mean sequence coverage of 170-fold for each of the six genomes. Mauve-based alignment of contigs revealed extensive synteny between the genomes of the six isolates and the reference genome of P. aeruginosa PAO1. However, two contigs of 49 kb did not align with chromosomal sequences. Notably, the mean sequence coverage for these putative extrachromosomal contigs was 3-fold higher than that observed for the syntenic contigs. Moreover, the two contigs showed different start points in different assemblies, indicating a circular sequence. The System for Automated Bacterial Integrated Annotation (SABIA) pipeline was used for gene prediction and automatic annotation followed by manual validation (Almeida et al., 2004). After annotation, the genomes were analyzed by the Bidirectional Best-Hits (BBH) clustering method (Overbeek et al., 1999), which compares different genomes with each other using the BLAST program (Altschul et al., 1997) to identify pairs of corresponding genes (clusters) and to recognize the best hit in other genomes. The parameters applied were 90% coverage, 90% of similarity and an e < 10⁻⁵. The GView Server (Petkau et al., 2010) was used to obtain the sequence of pan, core and unique genome of the six isolates using the P. aeruginosa PAO1 strain as a reference when necessary, with a minimum identity of 90% and an e < 10⁻⁵. The genomes of the six SPM-1-producing P. aeruginosa isolates were deposited in Genbank repository under the accession numbers: CP015001 (PA1088); LVWC01000000 (PA3448 contigs and plasmid); CP014999 (PA7790); CP015000 (PA7790 plasmid); CP015002 (PA8281); CP015003 (PA11803) and LVXB01000000 (PA12117).

We used all conserved open reading frames (ORFs) among our six strains, the reference genome PAO1 (Stover et al., 2000), 11 ST277 strains with genome available:19BR (GCA_000223945.2), 213BR (GCA_000223965.2), 9BR (GCA_000223925.2), BWHPSA041 (GCA_000520375.1), AZPAE12409 (GCA_000797005.1), AZPAE14819 (GCA_000795205.1), AZPAE14821 (GCA_000795235.1), AZPAE14822 (GCA_000795265.1), AZPAE14853 (GCA_000789905.1), AZPAE14923 (GCA_000791205.1), CCBH4851 (GCA_000763245.1) and a single MLST locus variant strain: BWHPSA007 (GCA_000481565.1) to reconstruct the phylogeny. A total of 5 042 protein sequences were concatenated applying neighbor joining, minimum evolution, UPGMA and maximum parsimony methods for reconstruct initial trees using Poisson method, uniform rates among sites, complete deletion treatment to gaps and bootstrap 100 with MEGA software (Kumar et al., 2016). The tree was visualized using the TreeView tool (Page, 1996).

The seven housekeeping genes acs, aro, gua, mut, nuo, pps, and trp were selected according to the multilocus sequence typing (MLST) scheme for P. aeruginosa² to confirm the allelic profiles of the six isolates.

Genomic islands (GIs) are segments of DNA mostly acquired by horizontal gene transfer. IslandViewer 3 (Dhillon et al., 2015) and PIPS (Soares et al., 2012) software were applied to detect genomic islands. IslandViewer 3 is based on sequence composition and comparative genomic methods, which may result in a prediction of false positives GIs, whereas PIPS includes the detection of virulence factors, hypothetical proteins, and flanking tRNAs in its analysis, which may exclude regions that do not meet these parameters. Both outputs were validated manually by observing the following criteria: (i) atypical G+C content; (ii) presence of mobile elements; (iii) adjacency to tRNA genes; (iv) size above 5 000 bp; and (v) comparison of the boundaries to the genomic context of the PAO1 reference strain. All criteria should be fulfilled, except for items (ii) and (iii), which were not mandatory to characterize a genomic island. We also used the IS Finder database and tools (Siguier et al., 2006) to identify the insertion sequences (ISs) from P. aeruginosa genomes. We considered full elements or fragments with e < 10⁻⁶ in BLASTn searches (Altschul et al., 1997). We also incorporated ORF regions sharing similarities with transposases into genome annotation with the SABIA platform (Almeida et al., 2004).

To identify possible polymorphisms in the six P. aeruginosa samples, we performed a single nucleotide polymorphism (SNP) calling using the PAO1 genome (NC_002516) as a reference. Briefly, the FASTA genome and GTF gene coordinates files were retrieved from NCBI. The deep sequencing libraries files were quality checked with the FastQC tool³ and trimmed with the FASTX_Toolkit⁴. The trimmed reads from the six samples were mapped separately against the PAO1 genome with the Bowtie 2 (Langmead and Salzberg, 2012) mapper with one mismatch per seed region (20 nucleotides in length), using three different seed regions for each read with repetitive regions and trying to extend 20 nucleotide after mapped seed region (very sensitive preset). The resulting mapping files were treated with the SAMtools program (Li et al., 2009); only mapping reads with a map quality (mapQ) greater than 30 were kept. The Picard mark duplicates tool⁵ was used to flag putative sequencing artifacts, such as optical duplicates. The Genome Analysis Toolkit (GATK) (McKenna et al., 2010) was used to call the variants using default parameters. The SNPs were annotated with the SnpEff tool (Cingolani et al., 2012) and custom Python scripts. Only SNPs with coverage larger than 10 reads were considered in further analyses. To find SNPs among the six P. aeruginosa isolates, the PA1088 genome was used as a reference, and the SNP call was performed as previously mentioned. In addition to the SNP call analysis, we performed a BLASTp search (Altschul et al., 1997), using the protein sequences of the six isolates against the PAO1 encoded proteins. This approach enabled us to find any polymorphism that was not detected by the previous method.

Antimicrobial susceptibility testing was performed by broth microdilution, and the results were interpreted according to the criteria of the European Committee on Antimicrobial Susceptibility Testing⁶. P. aeruginosa ATCC 29853 and Escherichia coli ATCC 25922 were tested as quality control strains. The genetic relatedness was initially determined by pulsed-field gel electrophoresis (PFGE) using SpeI enzyme (200 V [6 V/cm]; 13°C; switch time initial 5.0 and final, 60.0; 23 h) (Pfaller et al., 1992), and the results (Figure S1) were interpreted as previously recommended (Tenover et al., 1995). Two copies of bla_SPM−1 were identified in the isolates PA3448 and PA8281. To confirm whether the bla_SPM−1 multiple copies had led to an increase in transcriptional levels, qRT-PCR experiments were carried out. Total RNA was collected from SPM-1-producing P. aeruginosa isolates using the RNeasy Mini Kit (Qiagen, Hilden, Germany) with addition of RNase-free DNase (Qiagen, Hilden, Germany). Reverse transcription of the extracted RNA was performed using the High Capacity cDNA Reverse Transcription Kit (Life Technologies, Carlsbad, CA, USA). The pair of primers used for the amplification of the bla_SPM−1 and 16S rDNA genes were as previously reported (Mendes et al., 2007). Transcripts were quantified in triplicate using SYBR® Green PCR Master Mix (Life Technologies, Carlsbad, CA, USA) and the 7500 Real Time system (Life Technologies, Carlsbad, CA, USA). The 16S rDNA gene was used as a reference to normalize the relative amount of mRNA. The bla_SPM−1 transcriptional levels were compared using PA1088 as the reference strain because it has been known to carry a single copy of bla_SPM−1. The transcriptional level of genes encoding efflux pumps (mexB, mexD, mexF, and mexY), and the OprD porin (oprD) was also studied but using the PAO1 as the reference strain. Mean values (± standard deviations) of mRNA levels obtained in triplicate were calculated. Strains showing mRNA values of >5-fold for mexD, mexF, or mexY or >2-fold for mexB were considered to overexpress these genes (Cabot et al., 2011).

A summary of the genomic features of the six newly sequenced genomes of P. aeruginosa isolates is provided in Table 2. The whole genome size ranged from 6,643,783 to 7,018,690 bp, with only the PA3448 and PA7790 isolates observed to carry a plasmid. Although the isolates had similar G+C contents, it was possible to note some differences in the total number of coding sequences (CDSs) according to the variation in the chromosome size. An overview of the whole genome homology among the P. aeruginosa isolates, core genome and unique regions is presented in Figure 1.

A functional classification based on KEGG analysis assigned the CDSs into the 19 main categories was performed (Table S1). All isolates showed a conserved distribution of CDSs among the categories relative to the reference strain PAO1; one exception was the replication and repair category, in which 10 additional CDS were found (Table S2), such as the subunits A and B of excinuclease ABC, a single-stranded DNA-binding (SSB) protein and a DNA methyltransferase present in all six isolates. Interestingly, these additional CDSs involved in replication and repair were located in the genomic islands found among the six isolates.

The MLST allelic profiles confirmed the expected relationship showing that all isolates were grouped under the same sequence type, ST277. The concatenated ORFs were used to build a phylogenetic tree representing approximately 70% of the size of each genome. Our phylogenic reconstruction showed that the six P. aeruginosa isolates formed a monophyletic group with other ST277 strains, which supports the high similarity observed among these isolates (Figure 2). All methods used showed the same result, corroborating the validity of the groups. Among the ST277 strains were found two main groups, one including the PA3448 and PA12117 isolates, and another with PA11803, PA7790, and PA8281 isolates. Only the PA1088 isolate showed an unclear relationship due to its low branch support.

The overall chromosome organization of the P. aeruginosa isolates was compared with that of the reference strain PAO1 using Mauve software (Darling et al., 2004). This analysis revealed a conserved structure among the chromosomes of the SPM-1-producing P. aeruginosa isolates. The multiple alignments showed the existence of 11 conserved blocks; however, it was observed that unique regions were also present (Figure 3). Compared with the PAO1 strain, a major rearrangement was observed in all six P. aeruginosa clones. This rearrangement is an inversion that could be a result of a homologous recombination between genes encoding a 23S ribosomal RNA (PA0668.4 and PA4280.2 relative to PAO1 strain), which was orientated in opposite directions, and share 99% identity.

We identified 26 genomic islands in the six studied P. aeruginosa isolates. They were named PAGI (P. aeruginosa genomic island) and numbered accordingly from 15 to 40, i.e., PAGI-15 to PAGI-40. The size of the smallest region found was 5 914 bp, whereas the largest was 132,631 bp. A total of 14 islands were common to all six isolates, whereas other 6 were unique, 3 of which were present only in PA11803 (Figure 4; Table S6). Most CDSs observed in PAGIs were predicted to encode hypothetical proteins in addition to transposases, integrases and phage-related proteins (Table S7). Genes conferring antibiotic resistance were mainly located in PAGI-15 and -25, such as bla_SPM−1, bla_OXA−56, rmtD, cmx, and sul1. Genes homologous to those encoding proteins involved in the cell response to a stress condition, such as hicAB, hipAB, and higAB, were identified in PAGI-17, -24, and -31, respectively. PA1088, PA34448, PA7790, and PA8281 isolates acquired a gene predicted to encode a pyocin, a potent bacteriocin implicated in intraspecific microbial competition, homologous to the S5 type. This gene was carried by PAGI-34. Additionally, PAGI-20, -21, -32, -33, and -38 harbor genes homologous to prtN, ptrB, and prtR, which are involved in the regulation of pyocin production. Interestingly, PAGI-34 was 63% similar to a previously described island, PAPI-1; this island, in contrast to PAGI-34, did not carry a CRISPR-Cas system. This system, which possesses cas3, cas5, cas8c, cas7, cas4, cas1, and cas2 genes (type I-C), was present in four (PA1088, PA3448, PA7790, PA8281) of the six sequenced isolates, but it was absent in the remaining two isolates. Whereas the CRISPR-Cas system was intact in the PA1088 and PA3448 isolates, it was interrupted by the insertion of another island (PAGI-35) in the PA7790 and PA8281 isolates (Figure S3). PAGI-34 also shared a higher coverage (99%) and similarity (99%) to other mobile elements, the pKLC102-like ICEs, which have been previously reported to carry a type I-C CRISPR-Cas system (van Belkum et al., 2015). PAGI-28, an island present only in PA11803, carried a gene locus predicted to encode proteins homologous to type I-E and type I-F anti-CRISPR systems, namely JBD5-gp34, -gp35, -gp36, and -gp37 phage proteins (Bondy-Denomy et al., 2013; Pawluk et al., 2014).

We identified 20 different types of insertion sequences in the six P. aeruginosa isolates; some ISs were found more than one time in the genome, such as IS222, CR4, TPAse5. The overall distribution of ISs was quite similar among the SPM-1 isolates (25–32) (Figure 5). We observed 21 IS sites conserved between all six isolates. Regarding the variable sites, mobile elements such as TPAse4 were observed at different locations in the genome inserted within PAGI-19, which was present in all six isolates at different chromosomal positions (Figure 5; Table S6). Other variable sites comprised the CR4 elements. Two of them were located inside of PAGI-15 but suffered duplication in PA3448 and PA8281 isolates, resulting in an additional copy of the bla_SPM−1 gene. Additional copies of CR4 elements were found in PAGI-25; in PA1088, PA11803, and PA12117, we observed these copies occur next to one copy of the sul1 gene (absent in PA3448) and another next to the rmtD gene (absent in PA3448, PA7790, and PA8281) (Figure 6). The complete list of IS elements identified is presented in Table S8.

To find SNPs in the six P. aeruginosa isolates, the complete genome of P. aeruginosa PAO1 was used as a reference to perform an SNP call. The same strategy was used to find SNPs among the six isolates using the PA1088 genome as reference.

The overall number of SNPs was very similar among the six P. aeruginosa isolates in comparison to PAO1, demonstrating that most SNPs were commonly shared by all SPM-1-producing isolates (25%), except by PA11803, which exhibited approximately 400 more SNPs than the other five isolates (37%) (Table S9). The SNPs found in PA11803 were not homogeneously dispersed in the genome; instead, they were concentrated in a region located between 2,484,263 and 2,737,996 bp, a SNP hot spot measuring approximately 250 Kb in length (Figure 7). It is not uncommon to find these hot spot characteristics, as previously reported for other P. aeruginosa strains when using PAO1 as a reference genome (Bezuidt et al., 2013). Approximately 63% of SNPs in this region were classified as synonymous coding, followed by 22% classified as non-synonymous coding and 15% classified as intergenic. Most non-synonymous coding SNPs were found in genes predicted to encode hypothetical proteins (45%), but some were detected in the putative operon ambB-E and the pvd gene cluster (Table S10).

Among the SNPs shared by all P. aeruginosa isolates, the majority were located in the intergenic regions, followed by SNPs in CDSs, but without an amino acid change (synonymous coding). SNPs causing amino acid changes were also found in all six P. aeruginosa isolates (Table S9). We focused on this class of SNPs that could induce changes in phenotype. The remaining SNP classes were mostly found in hypothetical genes, except for a hydrolase, an amidase and an ABC transporter gene that harbored a lost stop codon (Table S10).

The SNP call performed using the PA1088 genome as a reference showed a number of SNPs lower than that detected in the previous analysis because of the high similarity among the isolates. Most of the non-synonymous coding SNPs found in genomic islands were located in genes predicted to encode hypothetical proteins, except for those encoding an ATP-dependent CLP protease found in PAGI-23 (PA3448), a zonula occludens toxin gene found in PAGI-29 (PA3448, PA7790, PA11803) and a lytic enzyme found in PAGI-32 (PA7790, PA8281) (Table S11).

The six SPM-1-producing P. aeruginosa isolates were fully susceptible to polymyxin B (MICs, 0.25–0.5 μg/mL) but were resistant to ciprofloxacin (MICs, >32 μg/mL) and all β-lactams tested, except for aztreonam (Table 3). P. aeruginosa isolates exhibited intermediate susceptibility to this compound, which is not surprising because aztreonam is not recognized as a substrate by β-lactamases such as SPM-1, AmpC, OXA-50, and OXA-56 (Toleman et al., 2002; Lister et al., 2009; Leonard et al., 2013). All P. aeruginosa isolates were resistant to both amikacin and gentamicin, except PA3448. This isolate was susceptible to amikacin but resistant to gentamicin.

The SPM-1 encoding gene, bla_SPM−1, was found in all six P. aeruginosa isolates within two distinct genetic contexts (Figure 8). The bla_SPM−1 gene was carried by a transposon with two CR4 elements in PAGI-15. The isolates PA1088, PA7790, PA11803, and PA12117 showed a duplication of a 4.2 Kb flanking sequence with two directly oriented copies of a region carrying one gene coding a hypothetical protein, one traR, one bcr1, and one virD2 genes. This context is similar to that described for ICE_Tn43716061 (Fonseca et al., 2015), except that the transcription direction was inversely oriented since PAGI-15 is located in a genomic region that suffered a major chromosomal inversion. In the remaining two SPM-1-producing P. aeruginosa, PA3448, and PA8281, we observed a duplication of a region measuring approximately 10 Kb possibly caused by recombination of the directly oriented repeats aforementioned, between which bla_SPM−1 was inserted, resulting in one additional copy of this gene (Wozniak and Waldor, 2010; Reams et al., 2012). The bla_SPM−1 transcriptional levels were 2.6 and 1.6 times higher in the PA3448 and PA8281 isolates, respectively, compared with the transcription level of PA1088, suggesting that both bla_SPM−1 copies were expressed (Figure 9A).

To verify any difference in genes involved in multidrug resistance mechanisms in all six isolates relative to the PAO1 reference strain, we performed BLASTp searches using default parameters. The chromosomal β-lactamases bla_OXA−50h and the cephalosporinase AmpC were detected among all six P. aeruginosa isolates. AmpC carried substitutions at R79Q and T105A that were identical to those described in the PDC-5 variant previously described (Rodríguez-Martínez et al., 2009). In addition, substitutions in AmpC regulators such as DacB (PBP4; A394P) and AmpD (G148A and S175L) were also observed in our study. However, no alterations in AmpD homologous proteins AmpDh2 and ApmDh3 were detected. Among the six P. aeruginosa isolates evaluated in this study, we also observed a substitution (A104P) in PbpC, a penicillin binding protein (PBP3A), which was detected by our SNP call analysis but only in the PA11803 isolate.

Other antimicrobial resistance genes were detected among P. aeruginosa isolates. The β-lactamase bla_OXA−56, a narrow-spectrum oxacilinase; the aminoglycoside-modifying enzymes (AMEs) aadA7 and aac(6′)-Ib-cr; the sulphonamide resistance gene sul1; and the chloramphenicol-related resistance gene cmx were present in three distinct mobile genetic contexts carried by a TnAs3 transposon, which were inserted into the PAGI-25 (Figure 6). aac(6′)-Ib-cr, bla_OXA−56, and aadA7 were harbored as gene cassettes of In163, a class 1 integron. These mobile genetic elements were integrated into the chromosome of all six P. aeruginosa isolates evaluated. We also observed two copies of sul1 in all isolates, except in PA3448.

The six P. aeruginosa isolates showed a deletion in oprD (coding the main porin for the uptake of carbapenems), as outlined before, and an important reduction in the oprD transcriptional levels by qRT-PCR when compared to that of PAO1 strain (Figure 9B).

Substitutions at quinolone resistance determining regions of GyrA (T83I) and ParC (S87L) were observed, justifying the ciprofloxacin resistance exhibited by all P. aeruginosa isolates. In addition, an unknown substitution H262Q in ParC was observed.

Among the RND-type efflux systems present in P. aeruginosa, a Q25L substitution in the MexR was only observed in the PA1088 isolate suggesting that MexAB-OprM was not overexpressed in most isolates. In all six clones, the mexZ sequence was incomplete likely leading to the overexpression of MexXY-OprM system. In contrast, substitutions (F172I, I341F and D345E for MexT; D249N for MexS) not related to date to the MexEF-OprN system overexpression were observed in both transcriptional regulators, MexT and MexS, in all six isolates. By qRT-PCR, comparing the results to those obtained for PAO1 strain, increased mexY transcriptional levels were observed for PA8281 (5.7-fold), PA3448 (5.2-fold), PA11803 (2.9-fold), PA12117 (1.8-fold), and PA7790 (1.6-fold). On the other hand, a not significant increase in the mexB transcriptional levels was observed only for PA8281 (1.4-fold, Figure 9B).

A BLASTn search using the Virulence Factors of Pathogenic Bacteria (VFDB) database (Chen et al., 2016) to find homology between sequences of the genomic islands found in SPM-1-producing P. aeruginosa isolates showed no significant results. Moreover, the comparative analysis did not yield any results related to genes encoding the main known virulence-related factors (Table S4).

The six SPM-1-producing P. aeruginosa showed several SNPs in genes encoding virulence factors (Table S12). We found 19, 20, and 19 non-synonymous SNPs in cupC1, 2, and 3, respectively. These genes participate in the chaperone-usher pathway involved in biofilm formation (Vallet et al., 2004). Moreover, the S → Y SNP, previously reported in CupC2 (Bezuidt et al., 2013), was also present in our SNP call analysis only in PA3448. Our analysis also identified polymorphisms in genes related to virulence, such as clpV3, encoding a protease related to type VI secretion system (T6SS). We observed a non-synonymous SNP in clpV3 that has not been related until now; however, it is known that mutations in this gene can cause inactivation of T6SS (Hachani et al., 2011). In addition, the PA11803 isolate has one ClpV3 SNP (I → T) that is different from another isolates (L → F), but it is also not clear whether this SNP could affect T6SS. The vgrG1 gene is also related to T6SS, and it is responsible for encoding a protein that acts as a puncturing device (Hachani et al., 2014). Four isolates (PA3448, PA8281, PA11803, and PA12117) showed a non-synonymous coding SNP (N → D) that was previously identified (Bezuidt et al., 2013). The pvd operon plays a role in the pyoverdine pathway and is related to iron acquisition. Mutations in this operon could decrease the virulence of P. aeruginosa (Lehoux et al., 2000). Our analysis revealed polymorphisms in pvdA, pvdQ, pvdR, and pvdT, but only pvdR and pvdT showed non-synonymous SNPs, in contrast to a previous report (Bezuidt et al., 2013). The ptxR gene is another gene related to the virulence process and is involved in quorum sensing in P. aeruginosa (Carty et al., 2006). We found one non-synonymous coding SNP in this gene in PA11803 (S → G). The pilY1 gene is involved in pilus assembly, twitching motility and adhesion to host cells. All six isolates showed a non-synonymous SNP (F → Y) in PilY1 that was previously described (Bezuidt et al., 2013).

We found SNPs in the amb operon, which produces l-2-amino-4-methoxy-trans-3-butenoic acid (AMB), a potent antibiotic and toxin (Lee et al., 2010). PA11803 showed one non-synonymous SNP in ambB and ambD, three in ambC and eight in the ambE. It has been shown that some site mutations abolish AMB production (Lee et al., 2010), but in our case, more studies are necessary to elucidate whether these SNPs could cease the toxin production.

We identified polymorphisms among the six isolates located in genomic islands when using PA1088 as a reference genome. Our analysis identified a non-synonymous coding SNP (V → D) in a gene encoding a protein homologous to zonula occludens in PA3448, PA7790 and PA11803. The zonula occludens is an enterotoxin elaborated by Vibrio cholerae that increases intestinal permeability by interacting with a mammalian cell receptor with subsequent activation of intracellular signaling leading to the disassembly of intercellular tight junctions (Di Pierro et al., 2001).

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.