INTRODUCTION
Humans rely on cellulose for building material, clothing, and fuel (
1–3). Recently, the polymer has sparked interest in the biotechnology field as a potential source of biofuel feedstock (
4) and in the biomedical industry as a biologically neutral scaffold to promote tissue regeneration (
5,
6). Cellulose is a linear polymer of glucose molecules connected with β-1,4 linkages by a glucosyltransferase. Individual linear glucan chains can pack via hydrogen bonding and van der Waals interactions in various ways to form different types of celluloses, with different properties (
3,
7,
8). The most common way glucan chains organize in nature is to form hydrogen-bonded planes stacked into parallel layers via van der Waals interactions (
9,
10). These stacked layers give rise to cellulose I microfibrils, or native cellulose, that can then coalesce to form larger arrays. Because glucan chains pack in a regular lattice but cannot sustain this regular pattern over their entire length, cellulose I is considered paracrystalline. Depending on how the lattice is organized, cellulose I can be of the α form, bearing a triclinic unit cell, or β form, bearing a monoclinic unit cell (
11,
12). Cellulose Iβ is mainly found in plants, where it is a major structural element of the cell wall (
13).
In the prokaryotic world, cellulose is an important component of bacterial biofilms (
14,
15), which increase cells’ tolerance for a range of biotic and abiotic stresses and enhance surface adhesion, cell cooperation, and resource capture (
14). Cellulose-containing biofilms have also been shown to be involved in pathogenicity, enabling bacteria to resist antibiotics and disinfection (
16,
17). Most cellulose-synthesizing bacteria produce amorphous (noncrystalline) cellulose, but a few genera, including
Gluconacetobacter, can produce cellulose Iα microfibrils. In
Gluconacetobacter, these paracrystalline cellulose microfibrils can further aggregate into wide ribbon structures and larger arrays (
18), giving rise to thick biofilms that are predominantly pure cellulose I.
Bacterial cellulose is synthesized by an envelope-spanning machinery called the bacterial cellulose synthase (BCS) complex, encoded by the BCS gene cluster and first identified in
Gluconacetobacter (
15). While the components vary, most of the species encode BcsA, a component in the inner membrane that, with BcsB, catalyzes transfer of UDP-glucose to the nascent glucan chain (
15,
19,
20). BcsD forms a periplasmic ring thought to gather glucan chains from several BcsA/B units (
21,
22). BcsA and BcsB are essential for cellulose synthesis
in vivo, and BcsD is essential for the crystallization of nascent glucan chains (
23). BcsC forms a pore in the outer membrane (OM), and very recent work has solved its crystallographic structure (
24). Consistent with previous data relying on sequence homology with the exopolysaccharide secretin components AlgE and AlgK from
Pseudomonas aeruginosa, BcsC is found to form an outer-membrane β-barrel pore at its C-terminal end, secreting the nascent elementary cellulose fibrils into the environment (
23–27). It is hypothesized that the elementary cellulose fibrils can aggregate with neighboring elementary fibrils upon secretion to form microfibrils (
28,
29). Genes flanking the operon,
cmcAx (endo-β-1,4-glucanase),
ccpAx (unknown function), and
bglxA (β-glucosidase), are essential for cellulose crystallization, and despite knowledge of their enzymatic functions, how they take part in this process is unclear (
29–32).
In this report, the terms used to describe the cellulose assembly process are adapted from the ones defined in reference
29, elaborating on the cell-directed hierarchical model for cellulose crystallization (
7,
10). Glucan chains are linear polymers of β-1,4-linked glucose residues synthesized by a single catalytic site of a cellulose synthase. An elementary fibril (also termed a minicrystal in previous work [
10,
33,
34]) is the product of the periplasmic aggregation of multiple glucan chains which is then extruded through a single BcsC subunit into the environment. Microfibrils result from the aggregation of several elementary fibrils, at least three according to earlier work (
34), outside the cell. These microfibrils can then crystallize into sheets that stack on each other to form ribbons. The latter terminology differs somewhat from previous usage in that our definition of a sheet is equivalent to the “bundles of microfibrils,” the polymerization step prior to the ribbon, described in reference
29.
Much work has already been done to understand the synthesis of paracrystalline cellulose (
18,
20,
21,
23,
30–33,
35–41). In particular, freeze fracture/freeze-etching electron microscopy (EM) studies have found that the
Gluconacetobacter hansenii BCS complexes are arrayed linearly along the side of the cell (
18,
33,
38,
39), and this arrangement seems to determine the extracellular organization of cellulose I into ribbons (
18,
33,
39). How this linear arrangement is achieved is not known.
Here, we used cryo-electron tomography (cryo-ET) of isolated cells and cryo-focused-ion-beam (cryo-FIB) milling of biofilms to visualize native cellulose production in G. hansenii and Gluconacetobacter xylinus, allowing the morphological characterization of the cellulose ribbons in a near-native state. We identified a novel cytoplasmic structure, which we call the cortical belt. We found that this cortical belt is absent from Escherichia coli 1094, which produces amorphous cellulose, and Agrobacterium tumefaciens, which produces crystalline microfibrils but not higher-order sheets, suggesting that the cortical belt functions to align BCS complexes to produce cellulose sheets.
MATERIALS AND METHODS
Cell culture.
Gluconacetobacter hansenii (ATCC 23769) was cultured as previously described (
37) in SH medium: 2% glucose, 0.5% Bacto peptone, 0.5% yeast extract (pH 6). For solid medium, 2.5% Bacto agar was added. Cells were separated from the cellulose biofilm by mechanical disruption as previously described (
38). Briefly, the bacterial cellulose biofilm developing at the air-medium interface was picked up with a single-use sterile inoculating loop and transferred to fresh medium, where it was vigorously shaken and then removed. In preparation for freezing, cells were pelleted by centrifugation for 10 min at a relative centrifugal force (RCF) of 2,500 at 20°C and resuspended in 0.5 ml of SH medium. The culture was incubated for the desired length of time at 30°C without shaking before plunge freezing. For cellulose digestion, 0.2 g/liter cellulase (purified exo- and endoglucanases; number LS002598; Worthington) was added.
Gluconacetobacter xylinus (ATCC 700178/BPR2001) was cultured as described above in fructose-peptone-yeast extract (FPY) medium: 2% fructose, 1% Bacto peptone, 0.5% yeast extract, and 0.25% K2HPO4.
Escherichia coli 1094 was cultured in lysogeny broth (LB) and induced for cellulose production in minimal medium: 0.2% (NH4)2SO4, 1.4% KH2PO4, 0.1% MgSO4, 0.5% FeSO4·7H2O, 0.4% glucose, 0.01% thiamine (pH 7). A saturated overnight LB culture was diluted 1:50 in 3 ml of minimal medium with or without 0.2 g/liter cellulase (purified exo- and endo-glucanases; number LS002598; Worthington). Cultures were incubated at 37°C with shaking at 220 rpm. When the medium transitioned from turbid to clear and white flakes appeared (cellulose and bacteria), the induction of cellulose synthesis was considered successful.
Agrobacterium tumefaciens was cultured as described in previous work (
84). Briefly,
A. tumefaciens C58 was cultivated in liquid AB medium (0.2% glucose, 18.7 mM NH
4Cl, 2.5 μM MgSO
4, 2 mM KCl, 0.07 mM CaCl
2, 0.01 mM FeSO
4, 8.4 mM K
2HPO
4, 4.16 mM NaH
2PO
4·7H
2O [pH 7]) at 30°C overnight. Induction was done by pipetting 100 μl of overnight culture and spreading it onto AB induction plates (0.2% glucose, 18.7 mM NH
4Cl, 2.5 μM MgSO
4, 2 mM KCl, 0.07 mM CaCl
2, 0.01 mM FeSO
4, 8.4 mM K
2HPO
4, 4.16 mM NaH
2PO
4.7H
2O, 1.7% Bacto agar, 100 μM acetosyringone [pH 5.8]). Plates were then incubated for 3 days at 20°C. Cells were resolubilized by scraping a small amount from the plate with an inoculation loop and resuspending it in 100 μl of liquid induction AB medium.
The following strains were included in the tomogram analysis. NT1 is a C58 strain without plasmid pTiC58 (tumor inducing). A139 is NT1REB(pJK270) + pJZ041. NT1REB is a “bald” strain, i.e., a no-flagellin mutant, derived from NT1. The plasmid pJK270 is pTiC58 with the transposed NPTII gene for kanamycin resistance. Plasmid pJZ041 carries a green fluorescent protein (GFP)-tagged VirB8 gene, encoding a component of the type IV secretion system (T4SS) (
85). Strain JX148 is a C58-derived mutant with a mutation of the
rem gene. The strain is nonmotile. AD348 is a GV3101(pMP90) strain with its whole VirB system deleted. GV3101 is a pTiC58-free, rifampin-resistant C58 strain, and pMP90 is a helper pTiC58 without the T-DNA. AD1484 is an AD348 variant, transformed with pAD2079 containing the whole VirB system.
Confocal microscopy.
Cellulose was stained with calcofluor white (number 18909; Sigma-Aldrich) at a concentration of 0.001%, and cell membranes were stained with MitoTracker Deep Red FM (number M22426; Thermo Fisher) at a concentration of 0.5 μg/μl. Stack acquisition was done on a Zeiss LSM880 Airyscan microscope. Airyscan acquisitions were performed in superresolution mode with the Z-step set at the optimal optical sectioning. The Mito-Tracker Deep Red FM channel was set as follows: excitation at 633 nm, use of the 488/561/633 main beam splitter, and a band-pass 570–620 + long-pass 645 filter. The calcofluor white channel was set as follows: excitation at 405 nm, use of the 405 main beam splitter, and a band-pass 420–480 + band-pass 495–550 filter. Airyscan processing was performed on the fly by the in-built algorithm of Zeiss Black.
Sample preparation for cryo-EM.
For isolated cells, Quantifoil Cu R2/2 Finder grids (Quantifoil Micro Tools GmbH) were glow discharged at 15 mA for 1 min. The grids were preincubated with fiducial marker solution prepared as follows: 50 μl of 10 nm colloidal gold (Ted Pella, Inc.) was mixed with 50 μl of 5% bovine serum albumin (BSA), vortexed for 1 min, and centrifuged at an RCF of 15,000 for 15 min; the supernatant was discarded, and the pellet was resuspended in 40 μl of phosphate-buffered saline (PBS) buffer. A 3-μl sample was deposited on each grid, left for 1 min, and then back-blotted with Whatman paper. Cells were plunge frozen with a Vitrobot Mark IV (Thermo Fisher Scientific) with 100% humidity at 30°C and back-blotted for 3 to 5 s.
For native biofilms, Quantifoil gold R2/2 Finder grids were placed in 35-mm glass-bottom petri dishes (number P35G-1.0-2.0C; MatTek Corporation) containing 1 ml of SH medium inoculated with a 2-day-old biofilm. The dishes were sealed with Micropore tape (3M) and incubated without shaking at 30°C for 3 to 6 h. Plunge freezing was done at 22C and 50% humidity, either with manual blotting on both sides of the grids (first back-blotted and then front-blotted) or by using the automatic blotting function of the Vitrobot with a blotting time of 5 to 6 s, a blotting force of 15, and a drain time of 2 s.
For E. coli 1094, after 4 h of incubation in minimal medium, the medium should turn from turbid to clear with white flakes. The optical density at 600 nm (OD600) of the cultures was monitored using the culture (always turbid) where cellulose induction was performed in the presence of cellulase to keep the cells from aggregating. It was then used as a reference to concentrate the cells to a high OD600 (10 to 20), in order to form bacterial mats on the EM grids, for control and cellulase conditions. Plunge freezing was done at 20°C and 100%, either with manual back-blotting for 5 to 7 s and a drain time of 1 s or by using the automatic blotting function of the Vitrobot with a wait time of 10 s, a blotting time of 5 to 6 s, a blotting force of 3, and a drain time of 1 s.
FIB milling.
Grids were clipped in Autogrid holders (Thermo Fisher) machined with a notch to allow FIB milling closer to the edge of the grid. Autogrids were placed in a custom-built shuttle and inserted into a Versa 3D dual-beam FIB/SEM microscope with a field emission gun (FEG) (FEI) equipped with a PP3000T cryo-transfer apparatus (Quorum Technologies). They were maintained at −175°C at all times by a custom-built cryo-stage (
86). To reduce sample charging and protect the sample from curtaining during milling, the grids were sputter coated with platinum at 15 mA for 60 s. Thin lamellae were generated with the Ga
+ ion beam at 30 kV at angles ranging from 10 to 17°. Rough milling was done at high currents, ranging from 0.3 nA to 100 pA, until the lamellae measured 1 μm in thickness under the FIB view. Current was then progressively brought down to 10 pA for the final milling steps until the measured thickness was between 100 and 200 nm. Final polishing of the back end of the lamella was also done at 10 pA, where the sample was tilted +0.5 to 1° to homogenize the lamella thickness. During the whole procedure, imaging with the SEM beam was done at 5 kV and 13 pA.
Electron cryotomography.
Tomography of whole cells and FIB-milled lamellae was performed on either a Titan Krios or Tecnai G2 Polara transmission electron microscope (Thermo Fisher) equipped with a 300-kV field emission gun, energy filter (Gatan), and K2 or K3 Summit direct electron detector (Gatan). The Krios microscope is equipped with a Volta phase plate (Thermo Fisher) (
87). Tilt-series acquisition was done with SerialEM (
88) with a 2 to 3° tilt increment for a total range of ±60° or ±50°, a defocus of −4, −6, or −8 μm, and a total dose up to 180 e
−/Å
2. Volta phase plate images are shown in
Fig. 1,
2,
5, and
7A and
B with a defocus of −2 μm and a measured phase shift of 0.5 π/rad before tilt series acquisitions.
Low-magnification tomography on the biofilm lamellae was performed at a magnification of ×6,500 (14-Å2 pixel size) with a −10 or −15 μm defocus and a total dose between 5 and 10 e−/Å2.
Tomography of FIB-milled lamellae was done exclusively on the Titan Krios instrument. Because samples were thinner, the total dose was limited to ∼80 e−/Å2.
Data processing.
Tomograms were reconstructed using the IMOD software (
http://bio3d.colorado.edu/imod/) (
89). Alignment was done on 1,000-by-1,000 binned tilt series with fiducial-marker-based alignment. Aligned stacks were low-pass filtered (0.35, σ = 0.05) to eliminate high-frequency noise. Weighted back projection reconstruction was performed, and the SIRT-like filter was used with 20 iterations.
Segmentation was also done using IMOD and drawing tools developed by Andrew Noske (
http://www.andrewnoske.com/student/imod.php). To better distinguish features during the segmentation steps, tomograms were filtered with the three-dimensional (3D) nonlinear anisotropic diffusion filter in IMOD. The cell contours and cortical belt were segmented manually on a Cintiq 21uX tablet (Wacom), and cellulose was segmented using a semiautomated thresholded method. (i) A denoising nonlinear anisotropic diffusion filter was applied (included in the etomo package [
http://bio3d.colorado.edu/imod/]) on the tomogram; (ii) precise boundary models were drawn around the structures to be thresholded; (iii) thresholding segmentation was performed with 3Dmod using the isosurface function, and the previously drawn contours were used as a mask. When the contours are precisely following the features of interest, this technique makes it possible to raise the isosurface threshold without picking up background noise.
Measurements for all distances between elements (cellulose sheet to outer membrane, width of the cellulose ribbon, cortical belt to inner membrane) were taken by generating normalized density profile plots and measuring the distances between the density peaks of the corresponding subcellular features (
Fig. 3). This was automated with a custom script, sideview-profile-average, written by Davi Ortega (
https://www.npmjs.com/package/sideview-profile-average).
Estimation of the cell depth in the native biofilm lamellae was calculated as follows: (i) using the two parallel walls of the milled trench, a perpendicular line was traced at the leading edge of the lamella (where the platinum meets the frozen material); (ii) lines were drawn from the center of the cells to the leading edge perpendicular line (
Fig. 6H, red line in top view of lamella); (iii) the distance from the cell center to the limit of the platinum on the leading edge, which is the surface of the sample, was measured. The real depth was then calculated using the following equation: opposite side (real depth) = tan(
a) × adjacent side (distance measured [d in
Fig. 6H]), where
a is the angle between the grid surface and the FIB gun during the milling process, which can be accurately measured during reconstruction with 3dmod.
Statistical analysis.
All statistics were performed with GraphPad Prism software (
https://www.graphpad.com/scientific-software/prism/). All data sets were first analyzed for normality using the Shapiro-Wilk test and homoscedasticity (equal standard deviations). If a data set was normal, appropriate parametric tests were performed, and if not, appropriate nonparametric tests were performed. Detailed statistical tests are listed in order of appearance in the paper.
For
Fig. 2E,
n was 3 and 23 for the loose and tight configurations, respectively. Two-tailed
P was 0.0008 (Mann-Whitney test).
For the OM-to-closest-sheet distances in cells at 20 min versus 300 min postseparation, n values were 4, 2, 23, and 3 for 20-min tight, 20-min loose, 300-min tight, and 300-min loose configurations, respectively. A Kruskal-Wallis test followed by Dunn’s multiple-comparison test was performed. The values for 20-min tight versus 20-min loose, 300-min tight, and 300-min loose showed adjusted P values of 0.12, >0.99, and 0.024, respectively. The comparisons of 20-min loose versus 300-min tight and 300-min loose and for 300-min tight versus 300-min loose showed adjusted P values of 0.23, >0.99, and 0.032, respectively.
For
Fig. 4A,
n values were 6, 15, and 33 for 13, 20, and 300 min, respectively. For
Fig. 4B, there were 6 and 21 tomograms for 20 and 300 min postseparation, respectively. The two-tailed
P value was <0.0001 (one-sample Wilcoxon signed rank test against a theoretical value of 1 [the number of sheets observed at 20 min postseparation]). For
Fig. 4H, 12 and 4 microfibril thickness measurements were performed on two separate tomograms (cells 1 and 2, left side of the graph). There were 47 measurements for intersheet distances performed on 23 tomograms. Analysis of variance (ANOVA) followed by Tukey’s multiple-comparison test was performed. Cell 1 versus cell 2, cell 1 versus 300-min intersheet distances, and cell 2 versus 300-min intersheet distances showed adjusted
P values of 0.073, 0.15, and 0.0015, respectively. For
Fig. 4I, 6 and 45 sheets were measured at 20 and 300 min postseparation. Welch’s
t test (parametric
t test without equal standard deviation [SD] assumption) showed a
P value of 0.23.
For
Fig. 6F, 6 and 4 biofilms were grown for 3 h and 6 h, respectively. An unpaired
t test showed a two-tailed
P value of 0.0011. For
Fig. 6G, 6 and 4 biofilms were grown for 3 h and 6 h, respectively. An unpaired
t test showed a two-tailed
P value of 0.2720. For
Fig. 6H,
n values were 49, 46, 4, and 11 for live and dead cells in 3 h and 6 h biofilms, respectively. Mann-Whitney tests were performed on live versus dead cells in 3-h and 6-h biofilms, showing two-tailed
P values of 0.82 and 0.54, respectively.
For G. hansenii cellulose sheet width versus A. tumefaciens cellulose fibril width, 52, 45, and 6 width measurements were taken for A. tumefaciens and for G. hansenii at 20 min and 300 min postseparation, respectively. Kruskal-Wallis one-way analysis of variance followed by Dunn’s multiple-comparison test was performed. A t test for 20 min versus 300 min, 20 min versus A. tumefaciens, and 300 min versus A. tumefaciens showed adjusted P values of 0.25, 0.11, and <0.0001, respectively.