Processive Enzymes Kept on a Leash: How Cellulase Activity in Multienzyme Complexes Directs Nanoscale Deconstruction of Cellulose

Biological deconstruction of polymer materials gains efficiency from the spatiotemporally coordinated action of enzymes with synergetic function in polymer chain depolymerization. To perpetuate enzyme synergy on a solid substrate undergoing deconstruction, the overall attack must alternate between focusing the individual enzymes locally and dissipating them again to other surface sites. Natural cellulases working as multienzyme complexes assembled on a scaffold protein (the cellulosome) maximize the effect of local concentration yet restrain the dispersion of individual enzymes. Here, with evidence from real-time atomic force microscopy to track nanoscale deconstruction of single cellulose fibers, we show that the cellulosome forces the fiber degradation into the transversal direction, to produce smaller fragments from multiple local attacks (“cuts”). Noncomplexed enzymes, as in fungal cellulases or obtained by dissociating the cellulosome, release the confining force so that fiber degradation proceeds laterally, observed as directed ablation of surface fibrils and leading to whole fiber “thinning”. Processive cellulases that are enabled to freely disperse evoke the lateral degradation and determine its efficiency. Our results suggest that among natural cellulases, the dispersed enzymes are more generally and globally effective in depolymerization, while the cellulosome represents a specialized, fiber-fragmenting machinery.

The probe head was kept in the middle of the sample to improve liquid circulation. The Dismembrator was used in pulse mode with a pulse time of 2 s, a pause time of 5 s and an intensity of 40 %, while constantly cooling the sample in an ice bath. A maximum sonication time of 20 min was chosen. This ensured that single cellulose fibers were released from the three-dimensional fiber network that the bacterial cellulose film represents (Fig. S2) and at the same time, mechanical disruption of the crystalline nanostructure of the cellulose fiber was minimized.

Cellulose substrate -Characterization
The bacterial cellulose was analyzed for its crystallinity according to Segal et al (1). The content of cellulose allomorph Iα was determined according to Wada et al (2).
X-ray diffraction patterns where measured on the Elettra beamline XRD2 (Trieste, Italy), using an energy of 12.398 keV and a Pilatus 2M detector placed in a distance of 134.2 mm from the sample. An approximately 1 mm 2 small piece was cut from the wet cellulose sheet and mounted on a metal loop aligned perpendicular to the beam. Diffraction patterns were acquired for 90 s at room temperature. Note: additional treatment of the sample with a cryostream resulted in water crystal spots in the diffraction pattern. The procedure used at room temperature resulted in quick drying of the sample, so a large part of the intensity measurement was contributed from dry cellulose sheet. The measurement was done for two different regions of one cellulose sheet. The empty metal loop was also measured and subtracted from the signal. Integrating the Debye-Scherrer rings resulted in 1D intensity signals, which were linearly baseline corrected and normalized with data analysis software (OriginPro 2019b, OriginLab Corporation, Northhampton, MA, US) (Fig. S4A). The peak height calculation after the method of Segal el al (1) resulted in crystallinity indices between 4 85 % and 91 %. This was in accordance with the previously reported high crystallinity of bacterial cellulose (3,4).
The d spacing calculations and Z-plot generation according to Wada et al (2) showed that all regions from the cellulose sheet were in the Iα rich category. Individual d spacings used to calculate the ratio regions (white dotted lines in Fig. S4B) were taken from published data for Cladophora algal cellulose (2). Because bacterial cellulose was expected to behave similarly as algal cellulose regarding its allomorph ratios, an estimation of the Iα:Iβ ratio was therefore possible. The cellulose sheet showed variation of the Iα:Iβ ratio between 10:0 and 9:1, depending on the region analyzed.

AFM -Real time experiments
Atomic force microscopy (AFM) measurements were performed with a commercially available device (Dimension Fast Scan Bio™, Bruker, Billerica, MA, USA). The controller (Nanoscope V, Bruker) was operated with the included software (NanoScope 9.2, Bruker).
The experiments were done in tapping mode in liquid environment.
Unless stated otherwise, Fast Scan DSS (Bruker AFM Probes, Camarillo, CA, USA) probes were used with nominal frequency, spring constant and tip radius of 110 kHz, 0.25 N/m and 1 nm, respectively. During laser alignment in liquid environment care was taken that the incoming signal at the detector was at least 1 V.
As substrate, a highly ordered pyrolytic graphite (HOPG) crystal with grade I (SPI supplies, West Chester, PA, USA) was used. Cleavage with adhesive tape was immediately followed by immobilization of bacterial cellulose fibers. This was done by incubating the surface with 300 µL heavily diluted (<0.01 g/L) bacterial cellulose suspension for 15 min. Subsequently, the HOPG surface was rinsed with deionized water. Residue water droplets were removed by spraying carbon dioxide on the surface for 2-3 s.
Mounting the substrate on the AFM stage was immediately followed by generating the liquid environment by pipetting 250 µL of the relevant buffer onto the surface and carefully driving 5 down the scan head until it was completely immersed in the droplet. During long measurements (≥ 3 h), it was necessary to add further liquid due to evaporation. Unless stated, the measurements were performed at 35 °C. The stage was heated with a temperature controller (Bruker) and the added buffers and enzymes were preheated to this temperature prior to injection.
The equilibration times varied between 10 and 60 min. The system was determined as equilibrated once the signal at the photodiode was stable. The surface was scanned for single, isolated and firmly attached cellulose fibers. Once a suitable fiber was found, the tip was stopped (but not withdrawn) followed by careful injection of 10-100 µL enzyme (15 mg/L) solution into the droplet. This was performed with a 100 µL pipette in 10 µL portions.

AFM -Parameter selection
AFM is a very adaptive technique, requiring iterative adjustment of imaging settings based on the feedback from the apparatus, therefore we report our general approach and systematic regarding imaging settings.
For every image the topography, phase and amplitude information were acquired. Two distinct measurement types were performed, dependent on the focus of the measurement being on the cellulose fiber or the enzymes. Scanning with regard to the structural changes of the cellulose fibers was performed by slow scan rates (> 1 frames/min). For imaging the enzymes, time resolution was improved by choosing small imaging regions (< 0.05 µm 2 ) and increasing the scan rate until the adjustment of feedback parameters no longer resulted in improved image quality. In more detail, tracking errors or other artefacts became predominant, resulting in the enzymes not being recognizable anymore. With this procedure, it was possible to achieve a maximum measurement speed of 1.7 frames/s. In any case, the spatial resolution was adjusted to 2 nm/pixel or better.
Settings regarding the force load on the sample were chosen individually for each measurement. The Drive Amplitude was set to the smallest possible voltage while still being 6 in a stable measurement region, which was monitored by the phase image. Sudden phase jumps between scan lines or after objects, regardless of Amplitude Setpoint adjustments, were taken as indications for a too small Drive Amplitude. Therefore, the Drive Amplitude was increased in 5 mV steps until the signal stabilized. The Amplitude Setpoint was adjusted to just below contact so as to prevent any disruption of the enzymes or a change/destruction of the cellulose. Amplitude Setpoints of 70-90 % resulted in phase shifts under 20 °, enabling minimally invasive visualization. Generally, a constant Setpoint was used during a measurement sequence, however, in case of any z-drift the Setpoint was adjusted accordingly.
The Integral Gain was set just below resonance (and adaptively adjusted to stay there) and the Proportional Gain was kept at a low value (about 3 -5 times the Integral Gain value). To increase resolution further, the z-range of the scanner was reduced to half of its maximum range.

AFM -Nanomechanical characterization of cellulose fibers
Peakforce QMN was done in liquid at room temperature using a Dimension Fast Scan Bio™.
Preparation of bacterial cellulose and immobilization on HOPG was done as described above.
Prior to each measurement the probes deflection sensitivity was measured on HOPG and k was calibrated by thermal tuning method. The effective tip radius was estimated by using the nominal tip radius and opening angle.
For imaging, the peak force set point was set to 15 nN and the cantilever was oscillated at 2 kHz. The Scan rate was adapted based on the imaged area to avoid artifact formation but was typically ≤ 1 Hz.
Data evaluation was done using Gwyddion (Version 2.55.

AFM -Image processing and video construction
Processing of the images shown in the time-lapse movies was done by an automated MATLAB routine (developed in R2017b, Version 9.3.0.713579). Unless stated otherwise, data correction was only performed for topography channel data. The routine was inspired by generally used consecutive processing steps in single image processing with established software, like Gwyddion. It consists of three main steps: object masking, data manipulation and scaling.
Masking resulted in a classification of every data point in either object or background. This was done by, on the one hand, finding the edges of the objects and, on the other hand, by finding the surfaces of the objects. The combination of both produced the mask. The edges of the objects were identified by calculating the numerical gradient for every pixel in x-and ydirection. If the root sum square of both gradients exceeded a user defined "gradient parameter" (GP), the pixel was identified as object and masked. The surfaces of the objects were identified by exploiting that the absolute numerical values of the surfaces and background lied in other ranges. The median was calculated for every row and the overall average of all medians was defined as background. If a pixel exceeded the sum of the average median and a user defined "median parameter" (MP), the pixel was masked. Prior to the median process, a rough baseline correction was performed by fitting a plane through all datapoints using least square method. This was merely done to avoid also masking the background in cases of pronounced tilts and should not be confounded with the actual tilt correction performed in the data manipulation part.
GP as well as MP were held constant within the entire measurement sequence. Appropriate choice of the parameters was critical for the quality of the masking process (Fig. S5).
The obtained mask was used in all actual data manipulation steps that followed. The tilt of the data was corrected by fitting a plane through all data points marked as background with least square method. The plane was then subtracted from every pixel in the raw image (Fig. S6).
Mismatching profiles of pixel rows were corrected by row alignment. By user choice, either a "Degree 0" or "Degree 1" correction was performed for the entire measurement sequence.
Degree 0 determined the mean of the background pixels in every row and subtracted this value from all the pixels in this row (Fig. S7). This routine was mainly used for data sets with scan areas below 500 nm -100 nm. Bigger areas often resulted in additional ascending or descending rows. Therefore, "Degree 1" correction was done, by least square fitting a polynomial of degree one through the background data points in every row and successively subtracting these lines from the whole, unmasked row. If rows consisted of 80% or more masked pixels, no correction was performed in either case for these rows. Unless stated otherwise, the phase and amplitude channel data were only processed regarding the color range. The lowest value was defined as zero and a maximum value was set by the user. Additionally, it was necessary to counteract two often-occurring problems in the amplitude channel data. Due to feedback errors some pixels can contain outlier values and the histogram of the whole image can be asymmetric. Both cause a shift in the false color scale in the upper color region, resulting in unclear visibility of the region of interest. This was handled by eliciting the number of bins between the peak of the histogram and it's maximum.
This number was used to define a new minimum for the histogram (Fig. S8). This mirroring of bin numbers resulted in cutting off distorting pixel points and increased visibility of the presented features.
A drift correction routine was adapted from Sugar et al (5). Briefly, a reference image was chosen and compared to the remaining images in the sequence. To determine the number of pixels an image must be shifted in x-, and y-direction the cross-correlation between both images was calculated. In more detail, utilizing the convolution theorem, the two-dimensional

AFM -Image analysis
Qualitative analysis was done visually by comparing consecutive pre-processed images of measurement sequences. To identify cellulose fibers, additionally to the topography, also the phase channel was considered. The high phase contrast, due to different material properties of the fiber and the HOPG background (6) The sum of all pixel entries for each frame was calculated, whereby the sum of the first frame was defined as 100 %. Shortly increasing heights, caused by temporally appearing enzymes where excluded. The temporal evolvement of this percentage is representative of the global degradation behaviour of the enzymes (Fig. 2E and 3D).
Quantiative analysis regarding the local mode of degradation was also done in MATLAB.
The temporal change of the height profile along the vertical axis of the fiberbundle (Fig. S9A) was tracked. To minimize noise, three neighbouring pixel rows were used for profile generation (Fig. S9B). The time dependent height decrease can be represented by plots, where the abscissa represents time, the ordinate the fiberbundle cross-section at a certain position and the color map the height. (Fig. S9C), using 15 colors from the color scale. The top view of this plot enables close tracking of the surface and dimensional change of the fiber over time (Fig. S9D). The temporal change of the maximum value of every pixel column was tracked and the first derivative was used to calculate the degradation speed. To minimize noise and to disregard suddenly increasing heights, caused by temporally occurring enzymes, the lines where smoothed as follows: Outliers, values that are more than three scaled median absolute deviations from the median, were replaced by the previous non-outlier value. Thereafter, the data was flattened by calculating the moving average of the values, using Gaussian weighing function and a fixed window length. The mean height difference per time step for the moving average was calculated for each individual column (Fig. S9E), resulting in the global estimates of degradation speed for the whole fiber (Fig. 6J). Height increasing steps were excluded to avoid distortion of the mean degradation steps by occasionally appearing enzymes.
Quantitative analysis regarding properties such as fiber length were performed using common tools in Gwyddion (Version 2.55) (7).
Fractions containing the cellulosomes were pooled, concentrated (~ 200 µg/mL) and stored at 4 °C. Analysis by SDS PAGE (Fig. S12A; see (10)) confirmed that the cellulosome preparation used had the protein composition as expected from the literature (8,11).

Enzyme preparation -Non-complexed (dispersed) cellulosome
The dispersed cellulosome was prepared according to an earlier method (12), with minor modifications. The method involves disassembly of the enzymatic subunits while the cellulosome is adsorbed on cellulose (12). Therefore, for cellulose binding, 750 µg/mL purified cellulosome was incubated with 150 mg/mL Avicel at 22 °C, 800 rpm for 1 h in a heating block (Thermomixer comfort, Eppendorf AG, Hamburg, Germany). Thereafter, the reaction was centrifuged and the cellulose pellet was washed thrice using 30 mM MOPS, pH 7.0. For disassembling, the cellulose pellet was resuspended in 15 mM EDTA, 30 mM MOPS, pH 7.0, and incubated at 70 °C for 10 min without agitation. The supernatant from centrifugation was collected. The disassembly step was repeated twice and all the supernatant fractions were pooled, concentrated and buffer exchanged using 10 kDa Vivaspin. This preparation was stored in 30 mM MOPS, pH 7.0, and is referred to as dispersed cellulosome.
Analysis by SDS PAGE revealed the presence of the major enzymes of the cellulosome while the scaffoldin CipA was characteristically lacking (Fig. S12A) (11).

Enzyme preparation -Dispersed cellulases
Complete (free) cellulase was obtained from the supernatant of the fungus Trichoderma reesei (Strain SVG17, Institute of Biotechnology and Biochemical Engineering, University of Technology, Graz, Austria). The SVG17 strain was reported earlier (13) and is an efficient cellulase-producing "wild-type" strain of T. reesei. For cellulase production, the strain was grown on wheat straw as described recently (14). TrCel7A (the major component of T. reesei SVG17 cellulases) was purified from the preparation of the complete cellulase as described recently (15). In brief, cellulase was buffer exchanged to 20 mM triethanolamine, pH 7.0, using disposable centrifugal concentrators (Vivaspin Turbo 15; Sartorius) with a 10 kDa molecular mass cut-off and loaded onto a 6 mL pre-packed column (Resource Q, GE Healthcare) equilibrated with the same buffer. Elution was performed at room temperature with a linear gradient of 0-300 mM sodium chloride over 10 column volumes. TrCel7A elutes in a discrete protein peak at about 180 mM sodium chloride, clearly separated from other major cellulases such as TrCel7B and TrCel6A that elute both at about 0 mM sodium chloride. The purified TrCel7A was buffer exchanged to 50 mM sodium acetate buffer, pH 5.0, and stored at 4 °C with a final concentration of about 5 mg/mL. Analysis by SDS PAGE (data not shown) revealed a single protein band migrating to a mass of 65 kDa, consistent with the molecular size of the full-length enzyme.

13
The major endoglucanase Cel7B was obtained commercially (Megazyme, Dublin, Ireland). It is a preparation from Trichoderma longibrachiatum, an anamorph of T. reesei. Note that the enzyme is nearly identical to its counterpart in T. reesei on an amino acid level (identity: 95 %, similarity: > 95 %). The enzyme was stored as provided by the manufacturer at 4 °C.
Prior to experiments the Cel7B was buffer exchanged to 50 mM sodium acetate buffer,

Enzyme preparation -TrCel7A core
TrCel7A core was prepared using limited papain treatment as described previously (16).
Papain was activated in 50 mM ammonium acetate buffer, pH 6.0, at 30 °C and 250 rpm for 30 min. For limited digestion, 200 µg activated papain was added to 1.0 mg purified TrCel7A and incubated at 30 °C and 250 rpm for 2 h. Thereafter, the digested reaction mixture was immediately buffer exchanged to 10 mM TEA-HCl buffer, pH 7.6, using disposable centrifugal concentrators (Vivaspin Turbo 10; Sartorius) and TrCel7A core was purified using anion exchange chromatography (Resource Q GE Healthcare). TrCel7A core purification protocol was similar to TrCel7A purification as described above. Purified TrCel7A core was also analyzed on 12 % SDS PAGE (Fig. S12B). After purification CBH I core was buffer exchanged to 50 mM sodium acetate buffer, pH 5.0 and stored at 4 °C till further use.

Enzyme characterization -Hydrolysis
All reactions were conducted in triplicates in 1 mL total reaction volume in micro tubes (1.5 mL, Eppendorf AG, Hamburg, Germany). Reactions employing free cellulases were done in 50 mM sodium acetate buffer, pH 5.0, and reactions employing cellulosomes were done in freshly prepared 30 mM sodium acetate buffer, pH 5.5, supplemented with 100 mM NaCl,  (19)).

Enzyme characterization -Total bacterial cellulose concentration
Bacterial cellulose concentration was determined using the total cellulose solubilization method (20) as reported previously. Briefly, 10 mg bacterial cellulose was hydrolyzed using 72% sulfuric acid for 1 h at room temperature in tightly capped glass tubes. The slurry was mixed at 15 min interval. The hydrolyzed cellulose mixture was diluted to 4 % sulfuric acid and autoclaved at 121 °C for 1 h. Glucose standard was also prepared the same way. Samples 15 were cooled (overnight at room temperature) and the pH adjusted to 5.0-5.5. Samples were assayed for glucose using a commercially available kit as stated above. The total mass of bacterial cellulose was calculated as anhydroglucose.

Enzyme characterization -Supplementation of the cellulosome in hydrolysis
Hydrolysis experiments were done as described above for reactions of the cellulosome. A standard cellulosome loading of 1.0 µg/mL was used. The cellulosome was supplemented with Cel7A (0.5 µg/mL), Cel7A core (1.0 µg/mL), Cel6A (0.5 µg/mL), Cel7B (0.5 µg/mL) and the disassembled cellulosome (0.4 µg/mL). The Cel7A core loading was selected based on carful comparison with native Cel7A (Fig. S13) to guarantee comparable conversions in the supplemented reactions. The disassembled cellulosome was also used as the main enzyme in a loading of 0.8 µg/mL. This was supplemented with Cel7A at a loading of 0.5 µg/mL. The degree of synergy (DS) was calculated for every combination at 24 h by the ratio of glucose produced by the mixture of main and supplemented enzymes ( + ) to glucose produced by the sum of the individual enzymes ( + ) according to equation S1.        Image acquisition rate and resolution were 0.9 frames/min and 2 nm/pixel, respectively. Scale Image acquisition rate and resolution were 3 frames/min and 2 nm/pixel, respectively. Scale bar, time stamps and false color scale are included in the video.