The multicellular morphotypes of Salmonella typhimurium and Escherichia coli produce cellulose as the second component of the extracellular matrix



Production of cellulose has been thought to be restricted to a few bacterial species such as the model organism Acetobacter xylinus. We show by enzymatic analysis and mass spectrometry that, besides thin aggregative fimbriae, the second component of the extracellular matrix of the multicellular morphotype (rdar) of Salmonella typhimurium and Escherichia coli is cellulose. The bcsA, bcsB, bcsZ and bcsC genes responsible for cellulose biosynthesis are not regulated by AgfD, the positive transcriptional regulator of the rdar morphotype. Transcription of the bcs genes was not co-expressed with the rdar morphotype under any of the environmental conditions examined. However, cellulose biosynthesis was turned on by the sole expression of adrA, a gene encoding a putative transmembrane protein regulated by agfD, indicating a novel pathway for the activation of cellulose synthesis. The co-expression of cellulose and thin aggregative fimbriae leads to the formation of a highly hydrophobic network with tightly packed cells aligned in parallel in a rigid matrix. As the production of cellulose would now appear to be a property widely distributed among bacteria, the function of the cellulose polymer in bacteria will have to be considered in a new light.


Cellulose is the most abundant organic polymer found in nature. It is produced by plants, some animals, algae, fungi, flagellates and bacteria such as Acetobacter xylinus, Agrobacterium tumefaciens and Rhizobium spp. (Ross et al., 1991). Cellulose is an intrinsic structural component of plant primary and secondary cell walls, building a network with other carbohydrate polymers. Bacteria generally produce cellulose as an extracellular component for mechanical and chemical protection, but it is also used by A. tumefaciens and Rhizobium spp. to facilitate cell adhesion to host tissue (Matthysse et al., 1981).

Although its chemical structure of (1 → 4)-β-linked linear glucose chains is quite simple, analysis of the biosynthesis of cellulose has proved difficult for a number of reasons (Cutler and Somerville, 1997; Kawagoe and Delmer, 1997). The bacterial organism A. xylinus has been used as a model to study the process and the genetic background of cellulose biosynthesis. The bcs (bacterial cellulose synthesis) operon encodes four proteins essential for cellulose biosynthesis: bcsA, the cellulose synthase; bcsB, the cylic-di-GMP (c-di-GMP) binding protein; bcsC and bcsD (Ross et al., 1991). The precursor molecule UDP-glucose is polymerized to cellulose by the cellulose synthase complex in the bacterial membrane. The activator c-di-GMP binds to the c-di-GMP binding protein, which is structurally associated with cellulose synthase and regulates its activity. The c-di-GMP pool is controlled by the activities of two enzymes, diguanylate cyclase and phosphodiesterase A. These are encoded on the chromosome in three highly homologous operons for cyclic diguanylate (Tal et al., 1998). The structures of the biosynthesis operons in A. tumefaciens and Rhizobium leguminosarum differ considerably, with only the cellulose synthase gene being conserved within the operon. The endo-1,4-β-glucanase is encoded outside the cellulose synthase operon in A. xylinus, whereas it is part of the operon in A. tumefaciens and R. leguminosarum (Matthysse et al., 1995; Ausmees et al., 1999). Cellulose biosynthesis in A. tumefaciens seems to occur via a different pathway, and its stimulation by c-di-GMP is debated (Amikan and Benziman, 1989; Matthysse et al., 1995).

Multicellularity is a common behaviour of bacteria in certain developmental stages, offering strategic advantages in comparison with single cells (Shapiro, 1998). Multicellular behaviour in Salmonella typhimurium (rdar morphotype) is positively regulated at the onset of the stationary phase by agfD, a gene encoding a putative response regulator of the FixJ family (Römling et al., 2000). At least two extracellular matrix components are produced by the rdar morphotype, thin aggregative fimbriae (agf) and an unknown extracellular substance. Recently, adrA (agfD regulated gene) was shown to be involved in the regulation of production of the unknown extracellular substance (Römling et al., 2000).

The expression of matrix components of the rdar morphotype can easily be followed by the absorption of the dye Congo red (CR). Expression of thin aggregative fimbriae leads to a brown colony (bdar morphotype), whereas expression of the unknown extracellular substance yields a pink colony (pdar morphotype). Bacteria of the pdar morphotype also bind Calcofluor (Römling et al., 2000). Certain aspects of multicellular behaviour can be attributed to the specific components of the extracellular matrix. Thin aggregative fimbriae mediate adhesion to abiotic surfaces (biofilm formation) and matrix molecules and fragile cell–cell interactions. Expression of the unknown extracellular substance leads to a reduced biofilm formation and elastic cell–cell interactions.

In this communication, we show that the unknown extracellular substance is cellulose, which is produced by the multicellular morphotypes of Salmonella spp. and Escherichia coli. Although the cellulose biosynthesis genes bcsA, bcsB, bcsZ and bcsC (formerly yhjO, yhjN, yhjM and yhjL) are constitutively transcribed, cellulose synthesis occurs only when adrA, a gene encoding a putative transmembrane protein regulated by agfD, is expressed. Co-expression of cellulose and thin aggregative fimbriae leads to a hydrophobic network with tightly packed cells embedded in a highly inert matrix. In addition, we show that cellulose is also produced by Klebsiella pneumoniae and, most probably, by pseudomonads such as Pseudomonas putida. Identification of cellulose as a component in biofilm formation may therefore lead to new approaches to fight adhesive bacteria that contaminate medical settings as well as industrial plants.


Mutant isolation and characterization of mutant loci

Our interest was to identify additional components of the extracellular matrix of the rdar morphotype of S. typhimurium strain MAE52. Recently, the adrA mutant (strain ADR1a) exhibiting the bdar morphotype was shown to lack unknown extracellular substance(s) (Römling et al., 2000). However, no conclusion could be drawn from this mutant about the nature of the extracellular substance, as sequence analysis suggested that the adrA gene most probably has a regulatory function. Therefore, a search for additional mutants that show the bdar morphotype was begun by performing random mutagenesis in MAE52, using MudJ as described previously (Hughes and Roth, 1988), and plating the mutants on LB–EGTA plates. Thirteen thousand mutants were screened by patching the colonies onto CR plates. Mutants showing the bdar morphotype were isolated, and the bdar morphotype was confirmed by restreaking the colonies. These were next screened to eliminate mutants that are not directly responsible for the synthesis of the extracellular matrix, but lead to the bdar morphotype. For example, the agfD mutant, which lacks all multicellular behaviour, is a prototroph. Thus, the mutants were tested for prototrophy to exclude mutants in precursor pathways connected with unrelated cell physiology. RpoS mutants, which also display the bdar morphotype (Römling et al., 1998a), were excluded by screening for a lack of catalase activity. To obtain mutants useful for the measurement of transcriptional activity of the affected gene, β-galactosidase activity was tested on Xgal plates. Finally, to confirm that the bdar morphotpye was caused by the introduction of the MudJ element, MudJ insertions were transduced into MAE52. Back-crossing identified over 60% of secondary mutation events, indicating that the pathway to the unknown extracellular substance(s) is unstable under the conditions used for transduction. Sequencing of polymerase chain reaction (PCR) fragments of flanking genome regions generated by inverse PCR and comparison of the S. typhimurium sequence with the annotated E. coli genome revealed that the MudJ element was independently inserted four times in the yhjL and yhjO genes. YhjL and yhjO are part of the yhjO, yhjN, yhjM and yhjL gene cluster encoding a putative cellulose synthase catalytic subunit, cellulose synthase regulatory subunit, endo-1,4-β-glucanase (cellulase family D) and oxidoreductase. The putative functions were deduced from the homology of the yhj genes to genes involved in cellulose biogenesis in A. xylinus, A. tumefaciens and R. leguminosarum bv. trifolii (Kawagoe and Delmer, 1997; Ausmees et al., 1999). Polar effects of the MudJ element in yhjO on downstream genes were excluded by constructing an in frame deletion in yhjO. The resulting strain, MAE171, showed a bdar morphotype. YhjO shows the highest sequence similarity to other bacterial catalytic subunits of cellulose synthase (up to 64% amino acid similarity). The D,D,D35QXXRW motif characteristic of repetitive transferases is distributed over the two domains A and B of the cellulose synthases (Stasinopoulos et al., 1999). Domain A harbours the conserved UDP-glucose binding motif. The conserved KAG motif (domain A) and QTP motif (between domains A and B) have no known function. The conserved aspartates and the QXXRW signature present in all cellulose synthase sequences are represented by domains U1 to U4 (Blanton et al., 2000). Based on these homologies and the analysis described below, yhjO, yhjN, yhjM and yhjL were renamed bcsA (bacterial cellulose synthesis), bcsB, bcsZ and bcsC. The organization of Salmonella and E. coli bcs genes is similar to some of the cellulose biosynthesis operons in A. xylinus (Fig. 1). Consequently, most of the bcs genes have the highest homology to genes found in A. xylinus (data not shown).

Figure 1.

Operon structure of bcs genes and protein domain structure of BcsA.

A. Comparison of the operon structure of bcs genes with the highest homology to the E. coli sequence. S. typhimurium, S. typhi and P. putida KT2442 have the same operon structure as E. coli (data not shown). Operon structure from A. xylinus JCM7664 (AB015803) and A. xylinus ATCC53582 (X54676). Homologous genes are indicated by arrows with the same pattern. Open arrows show genes with no homology within the operons.

B. Domain structure of BcsA. Coloured boxes indicate homologies to BcsA from A. xylinus ATCC53582 (X54676); lightest colour: 39% amino acid similarity; middle colour: 47% amino acid similarity; deepest colour: 72% amino acid similarity. Domains A and B according to Saxena et al. (1995); domains U1 to U4 from Blanton et al. (2000); UDP-glucose (UDP-glc) binding, KAG, QTP, D,D,D,35QXXRW motifs from Stasinopoulos et al. (1999).

The rdar morphotype of S. typhimurium produces cellulose

As the bcs genes were highly homologous to cellulose biosynthesis genes, we reasoned that the unknown extracellular substance might be cellulose. We used strain MAE97 (pink colony morphology; pdar morphotype), which produces the unknown extracellular substance(s) but not thin aggregative fimbriae, the other component of the extracellular matrix, to elucidate this question. Calcofluor, a dye detecting polysaccharides with (1 → 3)-β- and (1 → 4)-β-d-glucopyranosyl units, was shown by examination with fluorescent microscopy to bind to filament-like material in the cell clumps of MAE97 (Fig. 2A). In contrast, matrix-deficient MAE51 showed neither cell clumps nor Calcofluor binding. Further experiments showed that the huge cell clumps of MAE97 could be dissolved into single cells by digestion with cellulase for 2 h. Incubation with proteinase K for up to 72 h had no effect on the cell clumps that still bound Calcofluor. Next, cells of MAE97 and MAE51 were incubated in acetic–nitric reagent, which destroys all polysaccharide material with the exception of crystalline cellulose (Updegraff, 1969). The carbohydrate content of both strains was then analysed by carbohydrate compositional analysis using gas chromatography-mass spectrometry (GC-MS). In MAE97, glucose was exclusively detected in amounts two orders of magnitude higher than in MAE51. In order to exclude the presence of glucose polymers with linkages different from cellulose, a methylation analysis of the glucose polymers was performed. Exclusively 1,4,5-Tri-O-acetyl-2,3,6-tri-O-methyl-glucitol, the derivative characteristic of a 1 → 4-linked polymer, was detected in MAE97, whereas in MAE51, only traces of this compound were identified, confirming the results described above.

Figure 2.

Molecular architecture of the matrix and cell arrangement in the rdar morphotype and its mutants.

A. Arrangement of Calcofluor-stained cellulose fibrils in MAE97 (cellulose production) and MAE52 (cellulose and AGF fimbriae production) clumps. Top, fluorescence microscopy; bottom, phase contrast. Magnification ×400.B. Arrangement of cells in agar-grown colonies as seen by confocal scanning laser microscopy of the x–y plane. Cells without matrix (MAE51) are widely spaced and exhibit no structural arrangement within the colony. Cells with a full matrix are closer together and show parallel alignment perpendicular to the agar surface. A similar arrangements of cells is also seen when only one matrix component is expressed (data not shown). Bar represents 10 µm.

Production of cellulose was also shown for MAE52 (rdar morphotype at 28°C and 37°C) and UMR1 grown at 28°C (rdar morphotype), but not at 37°C (white colony morphology; Tables 2 and 3). As expected, cellulose was not produced by MAE150, MAE171 and MAE155, the bcsA and bcsC mutant strains of MAE52 (Table 1). In conclusion, the pdar morphotype in S. typhimurium results from cellulose production alone, whereas the rdar morphotype also produces thin aggregative fimbriae (Table 1).

Table 2. Correlation between cellulose production and phenotypic features in bacterial strains.
StrainCellulose productionMorphotypeCalcofluor binding
E. coli ECOR10+rdar+
E. coli ECOR12+rdar+
E. coli MC4100bdar-like
E. coli TOB1+rdar+
K. pneumoniae DSM12082+pdar+
S. enteritidis 27655-3b+rdar+
S. enteritidis 728+rdar+
S. typhimurium UMR1 (28°C)+rdar+
S. typhimurium UMR1 (37°C)saw
S. typhimurium SR-11b+rdar+
Table 3. Bacterial strains and plasmids.
Strain or plasmidGenotype or relevant phenotypeSource
S. typhimurium ATCC14028
 ADR1aMAE52 adrA101::MudJ; bdar28/37 Römling et al. (2000)
 MAE51MAE32 ΔagfD101; saw Römling et al. (2000)
 MAE52UMR1 PagfD1; rdar28/37 Römling et al. (1998a)
 MAE97MAE52 ΔagfBA102; pdar28/37 Römling et al. (2000)
 MAE110MAE52 zxx::gfp; rdar28/37To be published
 MAE119MAE51 zxx::gfp; sawTo be published
 MAE150MAE52 bcsA101::MudJ; bdar28/37This study
 MAE155MAE52 bcsC101::MudJ; bdar28/37This study
 MAE160MAE52 bcsC101::MudJ rpoS::pRR10(ΔtrfA); bdar28/37This study
 MAE161MAE32 bcsC101::MudJ ΔagfD101; sawThis study
 MAE164MAE52 bcsA101::MudJ rpoS::pRR10(ΔtrfA); bdar28/37This study
 MAE165MAE32 bcsA101::MudJ ΔagfD101; sawThis study
 MAE171MAE52 ΔbcsA102; bdar28/37This study
 MAE174MAE52 ΔbcsA102 adrA101::MudJ; bdar28/37This study
 MAE190MAE52 bcsA::MudJ ΔagfBA102; sawThis study
 MAE192MAE52 bcsC::MudJ ΔagfBA102; sawThis study
 UMR1ATCC14028–1s Nalr; rdar28 saw37 Römling et al. (1998b)
S. typhimurium LT2
 DA1705 HisD9953::MudJ his-9949::Mud1D. Andersson
 LB5010 metA22 metE551 ilv-452 leu-3121 trpC2
xyl-404 galE856 hsdL6 hsdSA29 hsdSB121
H1-b H2-e,n,x fla-66 nml(–) Fel-2(–)
Bullas and Ryu (1983)
E. coli K-12
 DH5α EndA1 hdR17 SupE44 thi-1 recA1 gyrA relA1
(lacZYA-argF)U169 (m80lacZΔM15)
Laboratory collection
 pBAD30Arabinose-regulated expression vector, Ampr Guzman et al. (1995)
 pMAK700Cmr, temperature-sensitive replicon derived from pSU101 Hamilton et al. (1989)
 pXZO1pMAK700ΔbcsA::102This study
 pWJB30pBAD30::adrATo be published
Table 1. Phenotypic characterization of strains.
CharacteristicsaStrain (genotype)
MAE52 (agfD+)MAE97 (adrBA)Adr1a (adrA)MAE150 (bcsA); MAE155 (bcsC)bMAE51 (agfD)MAE190 (bcsA agfBA); MAE192 (bcsC agfBA)b
  • a

    . Phenotypes were investigated at 37°C unless otherwise stated.

  • b

    . Both genotypes were investigated, but have been found indistinguishable by the phenotypic assays performed.

  • c

    . Formic acid-sensitive AgfA in Western blots was assessed from strains grown at 37°C for 24 h.

  • d

    . Assessed by scanning electron microscopy of strains grown at 37°C for 48 h.

  • e

    . Cells were grown in LB without salt medium and minimal medium, respectively, at 28°C for 24 h, and the adherence to the glass wall was measured.

  • f

    . Cells were grown in LB without salt medium at 28°C for 24 h.

CR binding+ (red)+ (pink)+ (brown)+ (brown)
Calcofluor binding+++
Cellulose production++
Polymerized AgfAc+++
Extracellular matrixd++++
Biofilm formatione+++/++++/+++/++++/++
Autoaggregation in liquid culturef++++++++++++

Characterization of strains lacking cellulose production

The phenotypic characteristics of bcsA and bcsC mutant strains are summarized in Table 1. The phenotype of the cellulose minus mutants (bdar morphotype, production of polymerized thin aggregative fimbriae on the cell surface, abundant extracellular matrix observed by electron microscopy, adherence to glass) is indistinguishable from the ADR1a strain lacking the adrA gene encoding a putative regulatory protein of the unknown extracellular substance(s). A comparison of the phenotypes of the bcs and adrA mutants would suggest that adrA only regulates cellulose synthesis. However, in order to prove the presence of only two components in the extracellular matrix, the cellulose biosynthesis genes and the genes responsible for the production of thin aggregative fimbriae would have to be deleted in the same strain. To this end, the double mutants bcsA agfBA and bcsC agfBA were constructed. Both were of the saw morphotype and completely lacked multicellular behaviour, similar to adrA agfBA double mutants and agfD mutants (Table 1; Römling et al., 2000). No extracellular matrix was observed by scanning electron microscopy (Table 1).

Regulation of cellulose production

The regulatory cascade elucidated so far suggests that agfD is the positive (putative transcriptional) regulator of multicellular behaviour as a whole. As adrA is regulated by AgfD at the transcriptional level (Römling et al., 2000), we tested whether bcsA and bcsC are also transcriptionally regulated by agfD (and consequently adrA). Upon transduction of the bcsA101::MudJ and bcsC101::MudJ mutations into MAE51 (ΔagfD), no change in transcriptional activity could be measured (Fig. 3A). Furthermore, the bcs genes were also not dependent on the alternative sigma factor rpoS, which has been shown to regulate the pdar morphotype and, partially, adrA transcription (Fig. 3A, Römling et al., 1998a; 2000).

Figure 3.

Dependency of bcsA and bcsC transcription on agfD, rpoS and environmental conditions.

A. Expression of bcsA and bcsC is not dependent on agfD and rpoS. Strains used are MAE150 (bcsA::MudJ), MAE165 (ΔagfD bcsA::MudJ), MAE164 (rpoS bcsA::MudJ), MAE155 (bcsC::MudJ), MAE161 (ΔagfD bcsC::MudJ) and MAE160 (bcsC::MudJ rpoS).

B. The bcs genes are preferentially expressed in the stationary phase of growth. Growth (open symbols) and β-galactosidase activities (closed symbols) of strain MAE150 (bcsA::MudJ; circles) and MAE155 (bcsC::MudJ; triangles), which were grown under aerobic conditions at 28°C in LB medium without salt, are shown.

C. Dependency of bcsA transcription on environmental conditions. MAE150 (bcsA::MudJ) was grown for 24 h at 37°C on plates, under microaerophilic conditions in liquid medium (m.a.) and on plates under: anaerobic conditions (ana.), iron depletion (Fe), in M9 minimal medium (m.m.), 0.5 M NaCl and 0.4% glucose.

For all experiments, average values are shown with standard errors from three independent experiments with duplicate measurements.

As the transcription of the bcs genes was not dependent on agfD, we next asked whether other unknown agfD regulated genes might participate in the biogenesis of cellulose or whether the expression of adrA alone is sufficient to switch on cellulose production. MAE51 carrying an in frame deletion in agfD was transformed with pWJB30 carrying the adrA gene under the control of the arabinose-inducible araBAD promoter in pBAD30 (Table 3). Whereas MAE51 harbouring the vector control remained white, MAE51 with adrA on the plasmid produced cellulose, as indicated by the development of the pdar morphotype (Fig. 4).

Figure 4.

Cellulose synthesis in the absence of agfD needs only AdrA expression. MAE51 (ΔagfD) harbouring pWJB30 (pBAD::adrA) expresses the pdar morphotype (cellulose synthesis; right) in contrast to MAE51 with the vector control (left).

AgfD is homologous to members of the FixJ family of transcriptional response regulators. Proteins of the response regulator family consist of an N-terminal receiver domain and a C-terminal DNA-binding domain. The absence of most of the highly conserved amino acids involved in the phosphorylation of a conserved aspartate in the N-terminal receiver domain in AgfD prompted us to speculate whether the activation of AgfD might occur by the binding of a putative precursor molecule produced constitutively by the bcs genes and accumulating in the stationary phase of growth in the cell. To test the hypothesis described, the transcriptional activity of an adrA::MudJ fusion was compared in MAE174 harbouring ΔbcsA102 and ADR1a. No difference was found in the transcriptional activities of the two strains (data not shown). Hence, there is no feedback control by precursor gene products accumulating in the stationary phase of growth.

Environmental regulation of cellulose biosynthesis genes

As the transcriptional pattern of the bcs genes is independent of the expression of agfD, we evaluated the correlation between the transcription of bcsA and bcsC and the rdar morphotype under various environmental conditions. The rdar morphotype is not expressed in the exponential phase of growth in liquid culture, at high osmolarity or when supplemented with 0.4% glucose (Römling et al., 1998a; U. Gerstel et al., in preparation). However, the bcs genes were already transcribed during the exponential phase of growth, although transcription was enhanced in stationary phase (Fig. 3B). On LB agar medium containing 0.5 M NaCl or 0.4% glucose, the cells expressed the bcs genes but not agfD, adrA and the rdar morphotype (Fig. 3C, Römling et al., 1998a; 2000). In addition, the bcs genes were co-expressed with the rdar morphotype under all environmental conditions tested, such as growth at 28°C, at 37°C, on plates, under anaerobic and microaerophilic conditions, growth on M9 minimal agar or on iron depletion agar medium (Fig. 3C). In conclusion, the bcs genes appear to be transcribed constitutively and independently of the rdar morphotype.

Co-expression of cellulose and thin aggregative fimbriae dramatically alters colony features

The rdar morphotype that co-expresses thin aggregative fimbriae and cellulose was recently reported to show differences in multicellular behaviour when compared with the morphotypes expressing only the individual components of the extracellular matrix. For example, the former form a tight, rigid cell network on plates and in liquid culture in contrast to the elastic or fragile network that is formed when cellulose or thin aggregative fimbriae are expressed alone (Römling et al., 2000). As a test for susceptibility, cell clumps of MAE52 (rdar), MAE97 (pdar; cellulose production) and MAE150 (bdar; thin aggregative fimbriae) were digested with cellulase and proteinase K as described in Experimental procedures. Whereas MAE150 clumps were so fragile that they fell apart even in the buffer control without enzymatic treatment, MAE97 clumps resisted proteinase K treatment (see above). Cell clumps of MAE52 conserved their consistency even after 72 h of digestion with cellulase or proteinase K, which is indicative of a tightly packed, inaccessible matrix. The architectural and physical parameters of bacterial colonies from the different morphotypes were investigated to explain the inert behaviour. When only cellulose was expressed, staining cellulose fibres with Calcofluor showed freely floating bundles protruding from cell clumps taken from agar plates. Co-expression of thin aggregative fimbriae captured the cellulose bundles and wrapped them tightly around the cells (Fig. 2A). Additionally, because of the expression of the extracellular matrix, the cells themselves altered their orientation to each other. Using green fluorescent protein (GFP)-labelled cells, observations with a confocal microscope revealed that, within a colony, the cells without matrix were randomly distributed, whereas matrix-enclosed cells were tightly packed and aligned in parallel perpendicular to the agar surface (Fig. 2B; further material under The consequences of matrix production on the relative hydrophobicity of the colony surface were measured by estimation of the contact angle of pure water (Mills and Powelson, 1996). Although MAE51 (no matrix) and MAE97 (cellulose production) had a low contact angle (21° and 27° respectively), indicating a relatively hydrophilic surface, an angle of 51° was obtained for ADR1a (Agf-fimbriae biosynthesis). Co-expression of the two extracellular substances in MAE52 raised the hydrophobicity significantly, as indicated by a contact angle of 77°.

Wild-type E. coli and other Salmonella spp. express the rdar morphotype

The bcs genes, the regulatory gene adrA and the two agf operons have been identified in E. coli and in all the Salmonella spp. genomes currently in the process of being sequenced (data not shown). The correlation between the expression of the rdar morphotype and cellulose was investigated in four strains of E. coli, two strains of Salmonella enteritidis and in S. typhimurium strain SR-11b. The laboratory strain E. coli MC4100, a K-12 derivative, has been used to study the expression of curli, the E. coli orthologue of thin aggregative fimbriae (Hammar et al., 1995). However, MC4100 showed a bdar-like morphotype and, accordingly, did not produce cellulose (Table 2). All other E. coli and Salmonella strains tested, which were natural isolates, showed the rdar morphotype and expressed cellulose (Fig. 5A; Table 2, data not shown). The rdar morphotype of E. coli TOB1, a faecal isolate, was investigated further. TOB1 expressed thin aggregative fimbriae, produced biofilms on glass surfaces and displayed other forms of multicellular behaviour such as clumping in liquid culture and pellicle formation that were indistinguishable from the rdar morphotype of S. typhimurium MAE52 (Fig. 5B; data not shown).

Figure 5.

Rdar morphotypes of Salmonella and E. coli strains.

A. Differential strength of expression of the rdar morphotypes of Salmonella and E. coli strains on CR plates at 28°C grown for 72 h. E. coli MC4100 shows a bdar-like morphotype. Upper row: S. enteritidis 728-b, S. enteritidis 27655-3b, S. typhimurium SR11-b; middle row: E. coli ECOR10, E. coli ECOR12, E. coli MC4100, E. coli TOB1; lower row: S. typhimurium MAE52.

B. Biofilm formation of E. coli TOB1 (right) with the positive control MAE52 (left) and the negative control MAE51 (middle) grown for 48 h in LB without salt medium with shaking.

Occurrence and expression of cellulose biosynthesis genes in other enterobacterial genomes

Databank searches revealed that highly homologous bcs genes with the identical operon structure and the adrA gene are also present in the genome of K. pneumoniae MGH78578 (data not shown). Subsequently, screening of K. pneumoniae isolates detected a cellulose-producing strain (Table 2).


In this communication, we have shown that the multicellular morphotypes of S. typhimurium, S. enteritidis and E. coli strains produce cellulose. The genes involved in cellulose biosynthesis, yhjO, yhjN, yhjM and yhjL, have been known to exist in E. coli since 1994 (Sofia et al., 1994). Their existence, however, has remained puzzling because, as a result of the fact that cellulose production is abolished in E. coli K-12 laboratory strains, their function and conditions of expression were never clarified. Detection of cellulose biosynthesis in the well-characterized species Salmonella and E. coli should open the way to elucidation of the regulatory pathways leading to cellulose production by genetic analysis.

The functions of yhjO, yhjN, yhjM and yhjL were deduced by comparison with cellulose biosynthesis genes in A. xylinus, the model organism for cellulose biosynthesis. Three of the four genes in several A. xylinus cellulose biosynthesis operons are highly homologous to yhjO, yhjN and yhjL. The fourth gene, bcsD, is not absolutely necessary, but is required to achieve maximum cellulose synthesis (Ross et al., 1991). In A. xylinus, cellulose biosynthesis proceeds at a linear rate with respect to cell growth, with regulation on the molecular level occurring via the concentration of a low-molecular-weight activator, c-di-GMP (Ross et al., 1987). In S. typhimurium, however, cellulose biosynthesis takes place in the stationary phase of growth only when the adrA gene, encoding a putative membrane protein, is activated, suggesting a novel regulatory pathway. In future, it will be interesting to elucidate common features and differences in cellulose biosynthesis between these two species.

Cellulose biosynthesis in S. typhimurium requires adrA expression, although the bcs genes are constitutively transcribed (Fig. 6). On the other hand, adrA expression is independent of the bcs genes. We propose that AdrA interacts with one or more Bcs proteins causing either stabilization of the Bcs protein(s) or activation of cellulose production. However, collection of additional experimental data is necessary to favour one hypothesis.

Figure 6.

The cellulose synthesis module. AgfD regulates cellulose production by turning on transcription of adrA. The adrA and bcs gene products putatively interact on the protein level to synthesize cellulose.

The four cellulose biosynthesis genes together with the regulatory gene adrA would appear to constitute a cellulose module (Fig. 6) that can be transferred between bacterial species, as the same genes occur in the unrelated bacterium P. putida KT2440. Surprisingly, not only the protein sequences, but also the nucleotide sequences, are highly homologous (data not shown), indicating lateral gene transfer of the cellulose operon. The cellulose biosynthesis genes have most probably been transferred to an enterobacterial ancestor chromosome from a species with higher GC content, as their GC content is higher than the 50.8% average (56% and 58% in E. coli and S. typhi respectively). Regulation of the cellulose module takes place at the level of adrA transcription (Fig. 6). However, the activators of the adrA promoter can obviously be exchanged, as the agfD induction of adrA seen in S. typhimurium does not exist in P. putida and K. pneumoniae (our unpublished data). As a working hypothesis, we propose that agfD acquired the capacity to regulate cellulose biosynthesis when the genes were established in the genome.

Cellulose and thin aggregative fimbriae are two components co-produced by both Salmonella spp. and E. coli. Thin aggregative fimbriae are depolymerized by treatment with 99% formic acid (Collinson et al., 1991), whereas cellulose withstands treatment in strong acidic and alkaline solutions. Produced together, cellulose and thin aggregative fimbriae form a highly inert, hydrophobic extracellular matrix, best described as ‘bacterial wood’, in which the cells are embedded. Clearly, cellulose is the structural component lending stability to the cell–cell interconnections, most probably by binding non-covalently to thin aggregative fimbriae. Consequently, the number of interactions along the polymers could determine the strength of the network structure.

The high energy costs needed for the production of the extracellular matrix are in contradiction to the time the matrix is made, namely in the stationary phase of growth when nutrient starvation occurs. However, as expression of the rdar morphotype is highly regulated in wild-type isolates (Römling et al., 1998b), at any given time, only a fraction of the total cells express the morphotype. Paying the energy costs would also make sense if one considers the rdar morphotype to be the survival mechanism of Salmonella spp. and E. coli, creating an encapsulated state that is the equivalent of a spore in Gram-positive organisms.

The ease with which Salmonella and E. coli strains expressing the rdar morphotype could be isolated indicates a widespread occurrence of strains expressing this morphotype. The E. coli strains examined were commensals from the human gastrointestinal tract, indicating that the rdar morphotype is expressed by non-pathogenic, extracellular organisms in the host in contrast to invasive organisms (Sakellaris et al., 2000). The advantage of expressing the rdar morphotype in the gastrointestinal tract could lie in the production of indigestible and adhesive cell clumps, which might have a longer duration time in the intestine before being shed.

Besides the natural biological functions of the rdar morphotype, its capacity to form biofilms represents a problem in medical settings, for example, where strains adhering to catheters can be a source of infection (Costerton et al., 1987). On the other hand, environmental strains that produce cellulose might be members of the biofilm consortium that contaminates industrial pipes and is responsible for ship fouling. Determination of the substances comprising the extracellular matrix and elucidation of the regulatory pathways may help in developing strategies for preventing the adhesion process.

Experimental procedures

Bacterial strains and growth conditions

E. coli and S. typhimurium strains constructed in this study are listed in Table 3. Other strains used were: E. coli ECOR 10 (faeces); E. coli ECOR 12 (faeces; Ochman and Selander, 1984); E. coli MC4100; E. coli TOB1 (faeces); K. pneumoniae DSM12082 (pond), S. enteritidis 27655-3b (faeces; Collinson et al., 1991); and S. enteritidis 728-b. Maximal expression of the rdar morphotype was achieved by growing strains on Luria–Bertani (LB) agar plates without NaCl. Other growth conditions were as described by Römling et al. (2000). Microaerophilic conditions in liquid culture are described by U. Gerstel et al. (in preparation). Antibiotics were used at the following concentrations: ampicillin (100 µg ml−1); chloramphenicol (20 µg ml−1); kanamycin (30 µg ml−1); and nalidixic acid (50 µg ml−1).

Construction of strains and plasmids

Phage P22 HT105/1 int-201 was used for the transduction procedure. Random MudJ insertions in strain MAE52 were carried out as described previously (Maloy et al., 1996). To identify the sequences of genes containing MudJ fusions, inverse PCR was carried out with primers MudOut (CCGAATAATCCAATGTCCTCCCGGT) and MudTaq (AGTGCGCAATAACTTGCTCTCGTTC) complementary to the Mu left end (Ahmer et al., 1998) after cutting chromosomal DNA with TaqI and intramolecular religation. An in frame deletion removing amino acids 150–800 in bcsA was performed according to described procedures (Römling, 2001). Primers YHJO5 (GGAGGATCCTGCCATATTCGCCTTCATCAG, BamHI site underlined)–YHJO6 (CCTCTGCAGCGCCAGTATCAGTGAAAAGGT, PstI site underlined) and YHJO7 (CCACTGCAGGATATTCTGAAGCTGGGCTTC, PstI site underlined)–YHJO8 (CCAAAGCTTTCGCATTGCGATCCGGCAG, HindIII site underlined) were used to generate PCR fragments that were subsequently cloned in the BamHI–PstI-cut pMAK700. All constructed strains were checked by Southern hybridization and/or PCR. Plasmids used in this study are listed in Table 3.

General molecular biology methods

Isolation of chromosomal DNA, PCR, purification of PCR products, plasmid transformation, restriction digestion, ligation, cycle sequencing and Southern blotting were carried out using standard protocols (Ausubel et al., 1994). Sequences for K. pneumoniae, P. putida, Salmonella paratyphi A, Salmonella typhi and S. typhimurium were obtained from the following websites:,,, and The E. coli and S. typhi sequences were used for all comparisons on the nucleotide and protein level on account of the quality of sequence data. Analyses of sequences were performed with the Genetics Computer Group package, version 9 (GCG, University of Wisconsin).

Protein techniques

For the detection of AgfA in Western blots, bacteria were treated with 99% formic acid before the addition of SDS–PAGE sample buffer (Römling et al., 1998b). Otherwise, standard methods were applied for the protein detection procedure (Ausubel et al., 1994). The anti-AgfA antibody was diluted 1:4000 before use.

β-Galactosidase levels were measured as described by Miller (1972). β-Galactosidase activity is calculated using the formula: units = 1000 × {[OD420–(1.75 × OD550)]/(t × V × OD660)} with t = reaction time in min, V = volume of cell suspension in assay in ml, OD420 and OD550 of reaction solution, OD660 of original cell suspension.

To digest the extracellular matrix with enzymes, whole bacterial clumps were taken from agar plates and put into the reaction mixtures without resuspension. Digestion with 0.1 mg ml−1 cellulase (Trichoderma viride; Sigma) was carried out in 0.05 M citrate buffer, pH 4.6, 5 µg ml−1 chloramphenicol at 45°C for up to 72 h. Digestion with 1.8 mg ml−1 proteinase K (Boehringer) was carried out in 50 mM Tris, pH 7.5, 5 mM CaCl2, 5 µg ml−1 chloramphenicol at 37°C for up to 72 h.

Evaluation assays for multicellular behaviour

The rdar morphotype was judged visually on CR plates. Adherence assays to glass were carried out as described previously (Römling and Rohde, 1999).

Cellulose detection assays

Calcofluor (fluorescent brightener 28) binding of agar-grown colonies was detected by observing the cells under a 366 nm UV light source. For fluorescence microscopy, cells grown on plates were suspended in 0.025% Calcofluor.

To isolate cellulose, bacteria were treated with acetic nitric reagent (58% acetic acid, 19% nitric acid) at 95°C for 30 min (Updegraff, 1969). The pellet was washed twice with water, hydrolysed (4 N TFA, 72 h, 100°C) and then subjected to methanolysis (0.625 M HCl in MeOH, 16 h, 70°C) followed by pertrimethylsilylation. The resulting pertrimethysilylated methylglycosides were then analysed on a ThermoQuest GCQ ion trap GC-MS system (MS: EI mode; GC: 30 m DB 5 column) and identified by their characteristic fragmentation pattern. The sugar identified by GC-MS was almost exclusively glucose. For quantification, the relevant glucose peaks were compared with an internal standard of 1 µg of myoinositol. For linkage analysis (methylation analysis), the sample was permethylated as described previously (Anumula and Taylor, 1992). After hydrolysis, reduction and acetylation, the resulting partially methylated alditol acetates were analysed by GC-MS. 1,4,5-Tri-O-acetyl-2,3,6-tri-O-methyl-glucitol was exclusively identified by its retention time and its characteristic fragmentation pattern, indicating the presence of a 1 → 4-linked glucose polymer.

Hydrophobicity measurement

The relative hydrophobicity of the bacterial colony was measured by calculating the contact angle between a drop of water and the bacterial colony on filter paper (Mills and Powelson, 1996). A filter paper was placed on an agar plate; a bacterial film was overlaid and incubated for 24 h. The filter paper with the bacterial film was dried on a glass slide for at least 30 min, and the contact angle of a drop of water was measured 30 times under a stereomicroscope with an ocular micrometer (gonimometer).

Confocal microscopy

Confocal analysis was carried out with a Zeiss inverted microscope and a dual laser-scanning confocal imaging system (MRC-1024 UV; Bio-Rad) equipped with a 100 mW argon ion UV laser and a 5 mW krypton–argon laser. For detection of the GFP-labelled bacteria, the 488 nm line and filter sets T1/E2 were used. Images were collected with a Plan-Neofluar 63× (oil)-NA 1.25 objective, preprocessed with lasersharp 3.1T software and converted to Tiff files.


We appreciate the generous support and interest of G. Maaß and J. Wehland. We are indebted to H. Biebl, C. Guzman, H. Ochman and M. Strätz for providing bacterial strains. We thank J. Müller for permission to use the gonimometer. Thanks to P. Hagendorf and R. Munder for sequencing. The excellent technical assistance of A. Tiepold and E. Müller is acknowledged. X.Z. appreciates the assistance of A. Secchi and M. Geese in using the light microscopy unit. U.R. is a recipient of a fellowship from the program ‘Infektionsbiologie’ from the Bundesministerium für Forschung und Technologie (BMFT). This work was supported in part by a DFG grant to U.R. (RO 2023/3-1).