Resurrection of efficient Precambrian endoglucanases for lignocellulosic biomass hydrolysis

Cellulases catalyze the hydrolysis of cellulose. Improving their catalytic efficiency is a long-standing goal in biotechnology given the interest in lignocellulosic biomass decomposition. Although methods based on sequence alteration exist, improving cellulases is still a challenge. Here we show that Ancestral Sequence Reconstruction can “resurrect” efficient cellulases. This technique reconstructs enzymes from extinct organisms that lived in the harsh environments of ancient Earth. We obtain ancestral bacterial endoglucanases from the late Archean eon that efficiently work in a broad range of temperatures (30–90 °C), pH values (4–10). The oldest enzyme (~2800 million years) processes different lignocellulosic substrates, showing processive activity and doubling the activity of modern enzymes in some conditions. We solve its crystal structure to 1.45 Å which, together with molecular dynamics simulations, uncovers key features underlying its activity. This ancestral endoglucanase shows good synergy in combination with other lignocellulosic enzymes as well as when integrated into a bacterial cellulosome. The efficient biocatalytic processing of cellulose is a longstanding goal. Here, an ancient endogluconase is recreated using ancestral sequence reconstruction and the basis for its thermal stability and favourable catalytic properties are probed computationally.

C ellulose is one of the major components in plant cell walls and is the most abundant organic polymer on the planet 1 . This widespread substrate offers a great opportunity to generate bioproducts, such as biofuels and nanocellulose. There is an enormous variety of raw materials rich in cellulose, such as agricultural, industrial, and urban wastes that can be used as sources for cellulose 2,3 . However, generating bioproducts from cellulose is still complex and expensive. Cellulose is a highly recalcitrant substrate difficult to obtain from plant cell walls, because it is protected by hemicellulose and lignin. In many processes, cellulases must withstand the harsh conditions of the industrial bioconversion process, such as high temperature, generally above 50°C, and low or high pH 4,5 . The lower efficiency of the enzymes under these conditions makes the saccharification process a critical bottleneck in the bioconversion of cellulose. Increasing the thermal operability and activity of cellulases is perhaps the most investigated aspect for their industrial implementation 6,7 .
In order to improve cellulases, several strategies ranging from rational and computational design to de novo enzyme design and directed evolution have been implemented, aimed at obtaining biocatalysts with improved performance 5,8,9 . Despite these advances, the limitations of engineered cellulases under the highly demanding industrial conditions (in terms of pH, temperatures, the presence of nonconventional media, and more) are still a barrier that must be overcome. The natural trade-off between activity and stability of proteins makes extremely complicated to enhance, for instance, the temperature and pH operability, the expression level or the activity of enzymes, all or at least some of them at once. The development of a strategy capable of finding more suitable blueprints, whereby improving the catalytic properties of enzymes in a cost-efficient manner, may revolutionize the biotechnology and chemical industries.
In the past decade or so, the so-called ancestral sequence resurrection technique (ASR) has been used to study the evolution of genes and proteins [10][11][12] . ASR utilizes sequences of proteins or genes from different species to create phylogenetic relationships, from which the sequences of their ancestors can be predicted and reconstructed in the laboratory 13 . Using a diverse combination of sequences, it is even possible to reconstruct Precambrian proteins belonging to organisms that lived shortly after the origin of life 10,11 . Reconstructed ancestral proteins have displayed enhanced thermal or mechanical stability, better pH response, improved activity and expression level, chemical promiscuity, and in some cases, all of these at once [10][11][12]14 . These traits are thought to reflect the conditions in which these ancestral proteins lived. Nevertheless, the molecular bases behind the high efficiency of ancestral enzymes are not fully understood. In addition, ancestral enzymes have been suggested to work like a "Swiss army knife" due to their versatility, provided that primitive cells likely relied on a limited but efficient set of enzymes that worked as generalists rather than as specialists 15 . Precambrian enzymes from the Hadean and Archean eons (older than 2500 million years) were adapted to work under temperature, pH, and environmental conditions that often resemble those of industrial settings 11 . Following this assumption, we propose ASR as a paleoenzymology method to generate efficient enzymes beyond the evolutionary implications.
In this work, we test the ability of ASR to generate efficient enzymes by reconstructing ancestral endoglucanases (EG) from 1.7-to 2.8-billion-year-old bacterial species. The ancestral EGs showed higher activity than those of contemporary EGs under a broad range of temperatures and pH. The oldest enzyme works well with various substrates even displaying processive endoglucanase and exoglucanase activity. The ancestral EG enzyme also displays higher efficiency when integrated into a bacterial cellulosome, a macromolecular machine for cellulose degradation 16 , which has been also proposed for industrial implementation 8,17 . To investigate the determinants of its activity, we solve the crystal structure to 1.45 Å, demonstrating that the fold is highly conserved. Interestingly, using the solved structure, we perform atomistic molecular dynamics simulations (MD) in the presence of substrate, which suggest that the balance between accessibility and dynamics of the substrate on the enzyme active site seems to play an important role on the high efficiency of the ancestral endoglucanase. Surprisingly, we determine that an efficient bioconversion can be potentially achieved by reconstructing very few enzymes as compared with other methodologies, where hundreds of variants need to be tested. This work represents a proof of concept, which may open new avenues toward efficient enzyme improvement in a single step.

Results
Ancestral sequence reconstruction of bacterial EG. To generate ancestral sequences of bacterial EG enzymes, we use 32 EG Cel5A sequences from extant bacteria (see the ID list of extant sequences in Supplementary Note 1), which are obtained from the UniProt database (www.uniprot.org), using the sequence form Bacillus subtilis (Bs_EG) as query. We target EG from family Cel5A because of their interest in biotechnology industry. EG enzymes are present in different bacterial phyla, such as Firmicutes, Actinobacteria, and Proteobacteria, which diverged more than 3 billion years ago (~3000 million years ago), indicating that these enzymes are ancient and were present in organisms that lived in the Archean eon. We select sequences from these three phyla. A sequence alignment is generated, and the catalytic domains of all the sequences are well resolved, forming a block with no major gaps or unstructured portions. In contrast, the carbohydratebinding module (CBM), a smaller subunit responsible for cellulose binding, does not align well, as some sequences had the CBM at the C terminus, while others have it at the N terminus, and there are numerous gaps. In addition, not all the sequences contain a CBM. As the CBM is poorly aligned, we conclude that this module is heterogeneous and poorly conserved within the family 5 of cellulases. Therefore, we focus on the catalytic region.
Using the block of catalytic domains, we construct a phylogenetic chronogram using Bayesian inference 18 , in which the three bacterial phyla are well resolved (Fig. 1). We date the phylogenetic tree using data from the Time Tree of Life (TTOL) 19 . Using the alignment and tree, we reconstruct the most likely ancestral sequence for each node. We select three nodes, the oldest node belonging to the last Firmicutes common ancestor (LFCA) that lived~2.8 billion years ago. We speculate that this may have been one of the earliest cellulase enzymes. This is consistent with the idea that the earliest cellulose producers were likely bacteria before the endosymbiotic transfer of cellulose synthase to eukaryotic plant cells 20 . The second node belongs to the last Clostridia common ancestor (LCCA), which is~2.1 billion years old; the third one from the last Actinobacteria common ancestor (LACA),~1.7 billion years old. The ancestral reconstruction utilizes a maximum-likelihood assignment at each site for the residue, with the highest posterior probability. The posterior probabilities of all 297 sites are presented in Supplementary Fig. 1. The average posterior probability values are 0.91-0.99, which ensures reliability of the reconstruction. Overall, the ancestral sequences display between 50% and 73% identity with respect to the modern Bs-EG. The mutations in the ancestral sequences with respect to the modern Bs_EG are distributed all over the sequence ( Supplementary Fig. 2). To reconstruct the ancestral EGs, the gene sequences of the domains are synthesized and cloned into an expression vector and expressed in the E. coli strain BL21 (DE3). The ancestral EG enzymes demonstrate a high level of expression, as shown by SDS/PAGE for LFCA_EG ( Supplementary Fig. 3). The reason behind this high expression level is unknown, but it seems to be a common feature among ancestral proteins.
Endoglucanase activity measurements. To test the performance of the ancestral enzymes, we carry out activity assays under different conditions. We first test enzyme activity in the temperature range 30-90°C and compare the activity of the reconstructed enzymes with those of contemporary enzymes from Thermotoga maritima (Tm_EG) and Bs_EG, at the same temperatures. Both enzymes belong to the Glycoside Hydrolase Family 5 (GH5).
Tm_EG is interesting because T. maritima is a hyperthermophilic organism that lives at temperatures up to 90°C; T. maritima is one of the most extremophile bacteria known today. The thermal range tested is broader that the typical testing range, which normally covers from 40 to 70°C. We first use a standard soluble substrate such as carboxymethylcellulose (CMC), using the dinitrosalicylic acid (DNS) assay to assess the release of reducing sugars by EGs 21 . The oldest ancestral LFCA_EG and LCCA_EG shows higher activity than the modern enzymes, with soluble CMC at all temperatures until 70°C. At 80 and 90°C, the activity of these enzymes is similar to that of the hyperthermophile Tm_EG (Fig. 2a). Interestingly, such high operational temperature has only been achieved by contemporary archaeal cellulases 7 .
Another important factor in cellulose hydrolysis is the pH at which the reaction is carried out. The pretreatment of lignocellulosic material can be performed at low or high pH values 22 . Therefore, improving cellulase activity in a broad range of pH values is of interest from an industrial point of view, as it would minimize the need for neutralization and the associated cost. We determine the activity of the reconstructed cellulases from pH 4 to 10 at 50°C using CMC. LFCA_EG and LCCA_EG show the highest activities (Fig. 2b). We find that the younger LACA_EG performs like Tm_EG.
From the temperature profiles in Fig. 2a, we wonder whether any evolutionary trend can be devised. We plot the relative activity of the ancestral enzymes plus Bs_Eg against the evolutionary time and a clear decreasing trend can be observed (Fig. 2c). We do this at 50°C, but the same is true for most temperatures. Surprisingly, this trend runs parallel to the cooling trend of seawater temperature over the past 3.5 By, as determined from δ 18 O in marine cherts 23 , which suggests that enzyme stability, activity, and environmental temperature are all linked. This trend seems to be general in ancestral protein stability 24,25 , but we prove it here for activity.
From the three ancestral enzymes, we take the most efficient one, i.e., LFCA_EG, for further experimental testing. We study the kinetics of the enzymatic reaction for the studied cellulases by applying the Michaelis-Menten model. From the experimental data in Fig. 2c, we determine that the K M , a measure of affinity, is quite similar for the three enzymes although slightly lower for LFCA_EG (1.25 mg mL −1 ). The highest turnover rate, k cat , corresponds to the ancestral LFCA_EG (0.04 s −1 ). Similarly, the highest catalytic efficiency, k cat /K M , is also achieved by LFCA_EG (0.032 mL mg −1 s −1 ), doubling the value of the modern enzymes. The kinetic parameters determined from the plot are shown in Table 1. Overall, these parameters indicate that LFCA_EG shows a higher substrate affinity, is faster, and more efficient that the modern enzymes.
We also evaluate how stable to temperature incubation is the ancestral EG, compared with Tm_EG and Bs_EG. We determine the T 50 value (defined as the temperature at which the enzyme loses half of its activity after 30 min of incubation). T 50 values of 85, 79, and 68°C are obtained for Tm_EG, LFCA_EG, and Bs_EG, respectively. The activity is determined at 60°C after the incubation (Fig. 2d). The ancestral EG was performed short behind the extremophile Tm_EG, highlighting its thermophilic nature.
Apart from their resistance to temperature and pH, ancestral enzymes have been suggested to show chemical promiscuity, which might be reflected in the ability to operate over more than one substrate or in the ability to display more than one mechanism of action 14 . An interesting promiscuous behavior in EG is to display processive activity, that is, to show both endoglucanase and exoglucanase activity. This is typical for EG from family GH9. We decide to test our LFCA_EG for such processive activity by measuring the relation between soluble and   19 . Geological scale and times are indicated in the upper bar. The internal node corresponding to the last Firmicutes common ancestor (LFCA), last Clostridia common ancestor (LCCA), and the last Actinobacteria common ancestor (LACA) was selected for reconstruction and laboratory testing insoluble sugars in the reaction. Surprisingly, LFCA_EG shows a higher ratio of soluble to insoluble sugars after 30 min of incubation, as compared with Tm_EG and Bs_EG that remain nearly constant at all times (Fig. 2e). The measured ratio for LFCA_EG is similar to that of other natural or designed EG with processive activity 26 . Although some hydrolases from family GH5 have been shown to be processive 27 , they have not been included in our phylogeny, which makes the processivity of the ancestral LFCA_EG a surprising feature. Besides bacterial cellulase, we also test our LFCA_EG against a fungal EG, given that fungal EG is much widely used than bacterial ones in biotechnological applications. We compare LFCA_EG with endoglucanase from Trichoderma reesei from family Cel5A (Tr_EG). From the activity experiments, we still observe that the ancestral LFCA_EG shows better performance in most conditions tested (Supplementary Fig. 4 and Supplementary Table 1). This result shows that designer bacterial EGs might be a good alternative to fungal ones, due to their diversity, complexity, thermal and pH operability, and even higher activity, as well as the high growth rate of bacteria 8 .
However, the high performance of LFCA_EG is not only limited to a soluble laboratory substrate such as CMC. In industry, the actual interest resides on the hydrolysis of crystalline cellulose 28 . For this reason, we compare the activity of the EG  Table 1. The ancestral EG displays the highest relation k cat /K m indicating higher efficiency. e Pre-incubation experiments at different temperatures conducted for 30 min. Residual activity was determined on CMC for 30 min at 60°C using DNS. Relative activity is determined for each individual enzyme. Each enzyme was preincubated at its best performing pH. f Ratios of soluble to insoluble reducing sugars generated using PASC as a substrate by LFCA_EG, Tm_EG, and Bs_EG, respectively. For each data point, three to five replicates were collected. The average ± S.D. values are shown for each measurement Table 1 Kinetic parameters of the cellulases determined from the Michaelis-Menten plot in Fig. 2c Kinetic parameters LFCA_EG Tm_EG Bs_EG enzymes using a microcrystalline substrate such as Avicel. Avicel requires long digestion times. We perform the assay at different times ranging from 4 to 72 h. For the assay, we use LFCA_EG in two forms, only the catalytic domain and the catalytic domain incorporating a CBM from Clostridium thermocellum, since an ancestral CBM cannot be reconstructed, as we have discussed. From Fig. 3a, we can observe that the maximum conversion percentage corresponds to the LFCA_EG form incorporating the CBM, but it is also surprising that the LFCA_EG catalytic domain by itself also displays remarkable activity against Avicel. The conversion at 70 h of hydrolysis reaches 60% for LFCA_EG with CBM, 45% for LFCA_EG, and around 25% for Tm_EG and Bs_EG, at equal enzyme load. Importantly, for industrial applications, cellulases must be able to hydrolyze cellulose in lignocellulosic materials, such as agricultural, industrial, or the organic fraction of city waste, in which cellulose in crystalline and amorphous forms together with lignin and hemicellulose is present. The digestion occurs in synergy with other enzymes, such as laccase and hemicellulases that given the recalcitrant nature of the biomass, helps by breaking down lignin and hemicellulose, making cellulase accessible for hydrolysis. This is important, for instance, for the Time pretreatment of lignocellulosic biomass, using enzymes for biofuel production. To test this aspect, we use cardboard, newspaper, and softwood from pine tree as a source of cellulose. These three materials have different contents of cellulose, lignin, and hemicellulose. While cardboard contains around 60% cellulose and around 15% of lignin and hemicellulose, newspaper and pine softwood contain less cellulose, 50% or less 29,30 , and more lignin, 22% and~30%, respectively, and~18% and~25% of hemicellulose, respectively. We perform activity assays using isolated LFCA_EG and in combination with an evolved laccase mutant from Myceliophthora thermophila 31 and xylanase from Trichoderma viride (endo-1,4-β-xylanase M1), enzymes that can help to break down lignin and hemicellulose, respectively. We determine the percentage of cellulose hydrolyzed in a 50 mg sample of lignocellulosic material 29,30 , within 1 h at 50°C and pH 4.8. In the case of cardboard, the three EG enzymes degrade very small amounts of cellulose on their own, no more than~19% (Fig. 3b), suggesting that cellulose is not easily accessible. LFCA_EG worked best when used synergistically with laccase and xylanase, hydrolyzing close to 40% of the cellulose present in the sample, as compared with Bs_EG, which degraded~27% and Tm_EG, which degraded~14%. In the case of newspaper and softwood, similar efficiency of cellulose degradation than cardboard is obtained, although still LFCA_EG shows higher conversion (Fig. 3c, d). These results highlight not only the potential of LFCA_EG to work with lignocellulosic substrates, but also the advantage of using multienzyme cocktails containing cellulases, laccases, xylanases, and other enzymes for efficient enzymatic pretreatment of raw materials and subsequent hydrolysis of cellulose. Similar measurements are carried out comparing LFCA_EG with Tr_EG, in which LFCA_EG also demonstrates better performance ( Supplementary Fig. 5).
Introduction of the LFCA_EG in a cellulosome. Another attempt to increase the activity of LFCA_EG is to incorporate it into scaffoldin, a non-catalytic scaffolding protein from a cellulosome, which is a macromolecular complex containing several lignocellulose-degrading enzymes anchored via dockerin protein domains. Anaerobic cellulolytic bacteria such as C. thermocellum utilize the cellulosome to degrade cellulose very efficiently, and its use has been suggested for industrial applications, due to the increased cellulolytic activity observed when compared with the free enzymes 32 . We make different constructs fusing EG enzymes to dockerin domains present in this cellulosome to convert LFCA_EG into the cellulosomal mode. We fuse dockerin at the C terminus of the ancestral EG (LFCA-Dock) to allow its incorporation into a mini-scaffoldin containing a single (Scaf1) or two tandem (Scaf2) cohesin modules (Fig. 4a). As controls, we use LFCA EG (LFCA-Dock) fused to a cellulose-binding module (LFCA-CBM) and C. thermocellum Cel8A EG (CtCel8A), a major EG in its cellulosome 33 . LFCA-Dock incorporation into two mini-scaffoldins occurs at molar ratios of 1.1:1 (LFCA-Dock: Scaf1) and 2:1 (LFCA-Dock:Scaf2), which is close to the expected ratio since cohesin-dockerin-binding occurs in a 1:1 ratio 32 , indicating precise complex formation (Fig. 4b). Furthermore, LFCA-Dock incorporated into the cellulosome and LFCA_EG-CBM is capable of binding microcrystalline cellulose Avicel ( Supplementary Fig. 6), while the other proteins fail. This indicates that, as expected, only when a CBM is present, specific microcrystalline cellulose binding can occur.
To study the effect of the incorporation of LFCA_EG into the cellulosome, we first perform activity assays with Avicel, which is targeted by the CBM used (Fig. 4c). According to the thermal stability measurements, we perform these assays at 70°C, a temperature at which no major loss of activity is expected to occur during the long incubation time needed. Free LFCA_EG shows higher activity with this substrate than native CtCel8A (4.3 ± 0.2 vs. 2.9 ± 0.13 mmol sugars mmol −1 enzyme min −1 , respectively). Dockerin incorporation into LFCA_EG results in a lower activity (3.41 ± 0.03 mmol sugars mmol −1 enzyme min −1 ) than that of the original LFCA_EG, which is still slightly higher than that of CtCel8A (Fig. 4c). Importantly, when LFCA_EG-Dock is incorporated into Scaf1, the resulting activity is remarkably enhanced, 6.2 ± 0.7 mmol sugars mmol −1 enzyme min −1 , with a high degree of synergy of 1.8 ± 0.2 (defined as the ratio of the activity of the bound enzyme over that of the free one). In the case of CtCel8A, the activity measured is 4.2 ± 0.7 mmol sugars mmol −1 enzyme min −1 in the presence of Scaf1 and a degree of synergy of 1.4 ± 0.3 was found.
The complex LFCA_EG-CBM shows a similar activity than that of LFCA-Dock, 7.0 ± 0.3 mmol sugars mmol −1 enzyme min −1 , supporting the idea that this enhancement is due to a substratetargeting effect. Incorporation into Scaf2, whereby two tandem identical cohesins allow for the formation of a cellulosome with two enzymes, does not provide further activity enhancement in either case, 4.4 ± 1.8 for LFCA_EG and 4.3 ± 0.6 mmol sugars mmol −1 enzyme min −1 for CtCel8A. Nevertheless, this result does not preclude the possibility of further synergy, if different enzymes are used in the future, together with LFCA_EG. Similar results are observed at all of the tested pH values (Supplementary Fig. 7) and at lower temperatures ( Supplementary Fig. 8). However, at temperatures above 80°C, the situation is reversed and CtCel8A shows higher activities ( Supplementary Fig. 8), perhaps due to the long reaction times.
The activity of the different proteins and complexes is then tested for different substrates to investigate the origin of the enhanced activity upon cellulosomal incorporation. First, we use PASC (Fig. 4d), an amorphous cellulose substrate. The results obtained are similar to those presented for Avicel, where LFCA_EG shows higher activity than CtCel8A when studied free in solution. Incorporation of both enzymes into Scaf1 resulted in an increased activity, which is higher for LFCAcontaining mini-cellulosomes than in CtCel8A ones. However, when tested on CMC, to which the CBM used in this study does not bind, neither the fusion with dockerin or the CBM fusion, nor the integration into a mini-cellulosome, significantly alters the activity of LFCA_EG or CtCel8A (Fig. 4e). Since the CBM used in this study is expected to bind PASC but not CMC, the results obtained also support the idea that scaffoldin CBM is capable of further enhancing the activity of LFCA_EG on certain substrates. Importantly, the activity of LFCA_EG is found to be greater than that of CtCel8A in all substrates, although these results seem to depend on the particular conditions of the assay, especially above 70°C (Supplementary Fig. 8). Taken together, these results indicate that the incorporation of LFCA_EG into a minicellulosome enhances its activity, especially for substrates that are difficult to degrade, which are the most interesting ones for biotechnological applications.
Comparison of the crystal structures of the EG. The experiments described above show that the ancestral EG is more active than the modern enzymes in almost any condition. To shed light into the structural basis of this high efficiency, we solve the crystal structure of LFCA_EG to 1.45 Å resolution (PDB ID: 6GJF) from data collected at a synchrotron source. The crystal belongs to the P2 1 space group (Supplementary Table 2) with six polypeptide chains in the asymmetric unit and a water content of 45.5%. All chains present the conserved EG canonical fold typical for enzymes from the GH5 family, composed of an internal β-barrel surrounded by an array of α-helices, (β/α) 8 -barrel (Fig. 5a, b). The maximum root mean square deviation (RMSD) is of 0.34 and 0.28 Å between chains A and C for the Cα and all atoms, respectively.
To investigate any substantial structural change, we compare the structure of LFCA_EG with that of Bs_EG (PDB ID: 3PZT), sharing 73% of their sequence, also used as a query model for the molecular replacement phasing procedure. From the superposition of both structures (Fig. 5c), we see that all major structural elements are equivalent with an all-atom RMSD of 0.5 Å. We do not observe any major difference other than a small displacement, lower than 2.5 Å, in several loops. A structural alignment of the two enzymes also reveals the location of the conserved and mutated residues (Supplementary Fig. 9). Mutations mainly occur in α-helices and loops.
We also compare the structure of LFCA_EG with that of Tm_EG (PDB ID: 3MMU). The all-atom RMSD is of 2.6 Å, in the best of the cases, which is not surprising since sequence identity between Tm_EG and our ancestral enzyme is only~20%. Although all three structural models have similar fold, relevant differences are observed between the ancestral and Tm_EG structures with significant movement of some secondary elements (Fig. 5d). The internal β-barrel is quite conserved, whereas the outer α-helices show structural changes. There is also relevant displacement in several loops accumulating the higher sequence discrepancy between both enzymes, with important deletions in the ancestral reconstructed one. Nevertheless, the residues E136 and E224 of LFCA_EG, essential for the catalytic reaction, are in a similar position than those in Bs_EG and Tm_EG. Also, W174 and W258, that serve for substrate recognition and stacking, are conserved and equivalent to those of Bs_EG and Tm_EG (Fig. 5e,  f). Whether the small structural differences observed can explain the difference in activity is hard to tell. bind to its glucosidic substrate, without mutating the active site to freeze the substrate, is virtually impossible due to the hydrolyzing activity of the enzyme. Hence, MD simulations are a good alternative to study the positioning and dynamics of the substrate in the active site prior to the hydrolysis reaction. We use the experimental structures for the cellulases from Tm_EG, Bs_EG, and LFCA_EG in the presence of a tetrasaccharide (Fig. 6a). Since the experimental structures lack the ligand, we insert it by fitting each experimental structure onto that of a cellotetraose-bound mutant of Cel5 from Thermotoga maritima (PDB ID: 3AZT) and carefully replacing the substrate for a cellotetraose (see further details in the "Methods" section). Two independent equilibrium MD simulations are prepared for each complex, with the total simulation data for each enzyme adding up to 1 ms (the results of one set of simulations are reported in Fig. 6b, while those for the other set are shown in Supplementary Fig. 10). This simulation timescale allows for probing the dynamics of the ligand within the active site cavity. Although we cannot recover exhaustive sampling with only two runs, our results show important qualitative differences in terms of protein-substrate interactions for the three enzymes. On the one end, we find that the Tm_EG keeps the substrate closest to the catalysis-competent position, where the nucleophile (E253) and proton donor (E136) are closest to the glycosidic oxygen, d nuc and d AB in Fig. 6b plots, respectively (see also Supplementary  Fig. 10). This is facilitated by a long loop that forms a clamp for the substrate (via a tryptophane residue, W210), which is possibly required for efficient binding at the high temperatures where this thermophile grows. In the two simulation runs of the Bs_EG enzyme with the sugar, we find that the substrate escapes from the binding site, as monitored by the distances of the glycosidic oxygen to the donor and nucleophile glutamic residues (E169 and E257) (Fig. 6b and Supplementary Fig. 10), suggesting a lower affinity in good accordance with experiment. In what appears an intermediate situation between Tm_EG and Bs_EG, the ancestral enzyme LFCA_EG is able to retain the substrate close to the position compatible with catalysis during the full duration of our simulations, albeit with stronger fluctuations than in the case of Tm_EG.
The picture that we recover from the MD simulations is that of a greater retention of the substrate for the Tm_EG enzyme and lower affinity for Bs_EG, with the ancestral enzyme (LFCA_EG) being somewhere in-between. The greater retention of the substrate in Tm_EG is consistent with its higher degree of active site burial. We show a representation of the active site cavities for each protein, derived from the CASTp3.0 server 34 , in Supplementary Fig. 11. Clearly, the Tm_EG has a greater surface area than that of Bs_EG and LFCA_EG, due to the longer loop encompassing tryptophan W210. This loop may be involved in large-amplitude motions that modulate the access of the substrate and release of the product. Unfortunately, the MD simulations we perform are too short relative to the timescales in which these movements may occur. For this reason, we derive an elastic network model (ENM), which can efficiently predict conformational fluctuations related to ligand binding 35 (see Supplementary  Information). The slow modes from the ENM for Tm_EG predict opening and closing motions that are highly localized in the loop region ( Supplementary Fig. 12). These slow dynamics may hinder ligand binding compared with the very easy access of the substrate to the active site in Bs_EG and LFCA_EG. We speculate that, for LFCA_EG, the greater ability to retain the substrate relative to Bs_EG together with the easy access to the active site in the opened cavity may contribute to the increased activity observed in the experiments, providing a "best of both worlds" situation relative to its extant counterparts. In addition, these results provide a clue for the structural origin of substrate promiscuity.

Discussion
Numerous ancestral proteins and enzymes have been reconstructed in the past few years, but most of them mainly aim to prove evolutionary hypotheses. Although the possibility of using ancestral enzymes in biotechnology has been pointed out c Structural overlay of LFCA_EG (dark gray) with Bs_EG (PDB: 3PZT, blue) and d overlay with Tm_EG (PDB: 3MMU, red). A zoom perspective of the active site highlighting the conserved residues E136, E224, W210, and W258 is shown in e for LFCA_EG and Bs_EG and f for LFCA_EG and Tm_EG. The residues refer to positions in LFCA_EG that overlap with the corresponding residues in Bs_EG (E169, E257) and Tm_EG (E136, E253) before 14,15,24 , such a goal still remained unexplored. Here, we use ASR to improve an example of an enzyme relevant in biotechnology, and we focus on most of the aspects that are of interest in a possible industrial setting, i.e., thermostability, pH tolerance, broad substrate usage, chemical promiscuity, and synergy with other enzymes.
In this work, we show that an ancient reconstructed endo-βglucanase displays high activity over a broad range of temperatures, pH values, and substrates, both as a free enzyme, and in combination with other lignocellulosic enzymes, as well as part of a cellulosome complex. This enzyme also shows processive endoglucanase activity, which is remarkable given that its modern counterparts from family GH5 do not display exoglucanase activity. Overall, the ancestral enzyme displays chemical properties that make it an interesting catalyst for possible biotechnological and protein engineering applications. From the crystal structure, we can infer that the ancestral EG maintains the same fold as modern cellulases. Simulating a complete enzymatic reaction would require complex quantum mechanical calculations that are beyond the scope of this work. However, using the crystal structure, it is possible to run classical MD that can shed light into the structural rearrangement between the enzyme and the substrate prior to the reaction itself. These simulations show that the ancestral enzyme seems to share features with the other two enzymes studied, on the one hand accommodating the substrate for the whole duration of the simulations (like in the simulations of Tm_EG), and on the other hand, allowing for greater dynamics in the more opened active site (as is the case of Bs_EG). We speculate that the greater dynamics and the opened cavity in the active site may contribute to the promiscuity of substrates that is a characteristic feature of ancestral enzymes.
A relevant aspect of this new ancestral enzyme is its elevated activity even at high temperature. Typically, ancestral enzymes are not necessarily more stable than those from modern extremophiles 11 . Enzymes present in the Hadean and Archean eons, when the temperature of the oceans was estimated to be 60-70°C 10 , were thermophiles 11,25,36 . This thermophilic phenotype is captured by ASR and exhibited by our ancestral EG. However, our LFCA_EG goes beyond the thermophilic range, working at 30-90°C. This range covers a good portion of temperatures from mesothermophiles to hyperthermophiles. However, ancestral enzymes display other properties, such as broad  pH usability, higher expression yields, or substrate and catalytic promiscuity 11,14 , which makes them stand up vs. extant enzymes, including extremophiles. These features make ancestral enzymes an interesting alternative for industry. In general, ancestral enzymes are considered to be generalists having a broader range of applicability than contemporary enzymes, which are considered specialists 37 , including extremophiles, for which the evolution to substrate and organismal specificity may limit their efficiency outside their natural environment. Thus, ASR emerges as a potential methodology for protein engineering with multiple applications in biotechnology 38,39 , beyond its possible evolutionary implications. Also, it is remarkable that our resurrected EG enzyme works well at 30°C. This temperature is interesting for future applications in processes, such as consolidated bioprocessing of biomass, which is carried out to obtain bioethanol in a single step combining saccharification and fermentation 40 .
A single EG enzyme is needed to achieve high activity under different conditions and substrates, which is difficult for any other protein engineering technique currently available. Certainly, our result could be a valuable departure point for further EG engineering through directed evolution, which typically starts from modern enzymes of limited evolvability, given that they have been already specialized by natural selection toward a given function. Indeed, the first successful example of laboratory evolution of a Precambrian enzyme has been reported, opening an unexplored path for more challenging objectives 41 . Among them, improving the thermal stability of an enzyme while maintaining its catalytic activity unchanged is a milestone for protein engineers, given the complex interrelation between structure and function in proteins 42 . Furthermore, while our ancestral EG is both thermoactive and thermostable, it also shows a noticeable catalytic efficiency. Finally, the ancestral EG also shows very good synergy with other lignocellulosic enzymes, such as laccase and xylanase, and its activity can be further enhanced by incorporation into the cellulosomal mode. We anticipate that other lignocellulosic enzymes, including fungal cellulases and ligninases, could benefit from ancestral reconstruction, which could also help to generate very efficient cocktails for the saccharification step of cellulosic substrates. This would provide a long-awaited improvement that could be used in numerous industrial applications.

Methods
Phylogenetic analysis and ancestral sequence reconstruction. We downloaded 32 endoglucanase sequences from different species divided in three bacterial phyla (Proteobacteria, Actinobacteria, and Firmicutes) from UniProt database. All sequences belong to the family Cel5A and are classified as 1,4-(1,3:1,4)-β-D-glucan-4-glucano-hydrolases (EC 3.2.1.4). All sequence ID numbers are listed in the Supplementary Information. The sequences were aligned using MUSCLE 43 software and further edited manually. The alignment was tested for best model of protein evolution using ProTest 44 , resulting in the Jones-Taylor-Thornton (JTT), with gamma distribution model as the best evolution model. The phylogeny was performed using Bayesian inference using Markov chain Monte Carlo (MCMC). We used BEAST v1.8.4 package software 18 incorporating the BEAGLE library for parallel processing. We set monophyletic groups for Proteobacteria, Actinobacteria, and Firmicutes. We set the JTT model with eight gamma categories and invariant distribution, Yule model for speciation, and length chain of 25 million generations, sampling every 1000 generations. We estimated divergence times using the uncorrelated log-normal clock model (UCLN), using molecular information from the TTOL 19 . Birth and death rates were set to default. Calculations were run for 2 days in a 12-core iMac computer. We discarded the initial 25% of trees as burn-in using the LogCombiner utility from BEAST. The MCMC log file was verified using Tracer, with all parameters showing effective sample size (ESS)>100. Tree Annotator was used to estimate maximum clade credibility. All nodes were supported by posterior probabilities above 0.69, with most of them nearly 1. FigTree v1.4.2 was used for tree representation and editing. Ancestral sequence reconstruction was performed by maximum likelihood using PAML 4.8 45 , incorporating a gamma distribution for variable replacement rates across sites and the JTT model. Posterior probabilities were calculated for all 20 amino acids. In each site, the residue with the highest posterior probability was selected. Three internal nodes LFCA, LCCA, and LACA were selected for laboratory resurrection.
Protein expression and purification. Ancestral LFCA_EG, LCCA_EG, LACA_EG, extant Tm_EG (ID: Q9X273), and Bs_EG (ID: P23549) proteins encoding genes were synthesized and codon-optimized for expression in E. coli cells. They were cloned into pQE80L vector (Qiagen) and transformed onto E. coli BL21 (DE3) (Life Technologies). Cells were incubated in LB medium at 37°C, and after reaching OD 600 of 0.6, IPTG was added to a final solution of 1 mM to induce protein expression overnight. Cells were harvested by centrifugation at 4000 rpm. Cell pellets were resuspended in extraction buffer containing 50 mM sodium phosphate, pH 7, 300 mM NaCl, and lysed using a French press. Cell debris was removed by centrifugation at 40,000 rpm for 40 min. For purification, His 6 -tagged proteins were loaded onto His GraviTrap affinity column (GE Healthcare) and eluted in 50 mM sodium-phosphate buffer, pH 7, 300 mM NaCl, and 150 mM imidazole. Finally, proteins were further purified by size-exclusion chromatography using a Superdex 200HR column (GE Healthcare) and eluted in 50 mM citrate buffer at pH 4.8. For the verification of purified proteins, sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) was used on 12% gels. The protein concentration was estimated by measuring absorbance at 280 nm using a Nanodrop 2000C. Tr_EG is a commercial preparation and two different batches were used for the experiments: one as lyophilized powder (Sigma reference C8546 from T. reesei ATCC 26921) and a second one as enzyme solution from Sigma-Aldrich (C2730), both sold as 1,4-(1,3:1,4)-β-D-glucan-4-glucano-hydrolase (EC 3. 2.1.4). The determination of the protein concentration in the solution (C2730) was first made by the dry-weight method 46 for protein content determination. Sizeexclusion chromatography was used with a Superdex 200HR column and eluted in water. Then the sample was frozen, dried, and weighted. Protein concentration was also determined by the BCA assay (Pierce TM , Thermo Fisher 23227) using a BSA standard supplied with the kit and also our ancestral LFCA_EG. A total protein concentration of about 125 mg mL −1 was determined with both methodologies. Using Tr_ EG in the powder or solution form provided nearly identical results at the same concentration. An evolved laccase (KyLO mutant) from M. thermophila was heterologously produced in Saccharomyces cerevisiae, as reported elsewhere 31 .
Enzymatic activity assays. Cellulolytic activity of ancestral EG was tested in 50 mM citrate buffer, pH 4.8, containing 2% CMC (Sigma-Aldrich reference 21902) for 30 min at various incubation temperatures and a final volume of 500 µL. Cellulases from Tm_EG and Tr_EG (from Sigma-Aldrich, reference C8546 for lyophilized powder and C2730 for enzyme solution) were used as controls. Enzyme dosage 5 mg per gram of glucan. Enzymatic reactions were terminated by placing the tubes into an ice-water bath. Enzymatic activity was determined quantitatively by measuring soluble reducing sugars released from the cellulosic substrate by the DNS method 21 . All assays were performed in 5% glycerol that is used as a stabilizer. A volume of 1.5 mL of the DNS solution was added to each sample, and after boiling the reaction mixture for 5 min, absorbance was measured at 540 nm using a NanoDrop 2000C. A glucose standard curve was used to determine the concentration of the released reducing sugars. All assays were performed in triplicate and the average value with standard deviation was determined. On determination of the pH dependence, purified enzymes were diluted in 50 mM citrate buffer at different pH values between 4 and 10; citrate buffer for pH 4 and 5, phosphate buffer for pH 6, 7, and 8, and carbonate buffer for pH 9 and 10. Activities were measured with 2% CMC at 70°C for 30 min. The amount of reducing sugars was measured and quantified by the DNS and BCA methods. Avicel (Sigma-Aldrich Ref S3504) was used for the determination of the enzymatic activity in crystalline substrates. A volume of 0.4 mL of enzyme solution was placed together with 1.6 mL of 1.25% Avicel solution. Enzyme dosage 15 mg/g of glucan. Substrate and enzymes blanks were also prepared. Enzymatic reactions were stopped by placing the tubes into an ice-water bath, and the tubes were then centrifuged for 2 min at 14,000 rpm at room temperature. Enzymatic activity was determined quantitatively by measuring soluble reducing sugars released from the cellulosic substrate by the DNS. A volume of 1.50 mL of the DNS solution was added to 1 mL of sample (supernatant fluids), and after boiling the reaction mixture for 5 min, absorbance at 540 nm was measured.
Kinetic parameters determination. To determine the kinetics parameters of the cellulases, K M and V max , numerous substrate concentrations were used in the range of 1-20 mg mL −1 of CMC for measurement of endoglucanase activity. The K M and V max were determined directly from the linearized fitting of the Michaelis-Menten model, generated using Phyton in-house written script. k cat was determined from the relation V max /E t , where E t is the total enzyme concentration in μmol mL −147 . The parameters are reported in the main text and Table 1.
Processivity assay. In order to determine the processivity 26 of the cellulases, there was a ratio of soluble to insoluble reducing sugar from PASC. The reaction was carried out at 45°C with 0.5% of PASC, and a sample was removed from the mixture at different time points. After centrifugation, the quantity of the released reducing sugars in the supernatant and in the remaining PASC fraction was determined by the DNS method.
Lignocellulosic substrates hydrolysis. We used 50 mg of milled lignocellulosic material (cardboard, newspaper, and pine softwood) in 50 mM citrate buffer at pH 4.8. Enzyme hydrolysis was performed for 1 h at 50°C. Endoglucanase alone or in combination with laccase and xylanase was used for hydrolysis of the lignocellulosic material. Three different enzyme combinations were used differing in the endoglucanase used: LFCA_EG, Tm_EG, Bs_EG, and Tr_EG. EG enzyme dosage was 14 mg/g of cellulose in the case of cardboard and 15 mg/g of cellulose in the case of newspaper and softwood, 50 µL of a solution of~4 U mL −1 of laccase, and 5 µL of a solution of~1700 U mL −1 of xylanase (endo-1,4-β-xylanase M1 from Trichoderma viride, Megazyme) in a total volume of 500 µL. Released sugars are quantified with the DNS method. Cellulose hydrolysis yield was determined as described elsewhere 48,49 .
Cellulosome constructs. Two mini-scaffoldins were designed in this study consisting of components from C. thermocellum CipA scaffoldin. In particular, the Xmodule and type II dockerin dyad and the CBM were amplified from pET28-XDock and pET28-CBM, respectively. Cohesin 7 was amplified from pAFM-c7A 50 . First, XDock was amplified with primers incorporating NdeI, NheI, KpnI, and SpeI sites at the 5′ end and two STOP codons and a XhoI site at the 3′ end. The resulting fragment was cloned into a pET28 vector using NdeI and XhoI sites. Then the CBM was amplified and cloned into the previous vector using NdeI and NheI sites. Next, cohesin 7 sequence was cloned using KpnI and SpeI sites to generate pET28-Scaf1. A second copy of cohesin 7 was then cloned into this vector in the SpeI site to generate pETScaf2, containing two tandem cohesins. Both mini-scaffoldins carried a His 6 tag at the N terminus.
Integration of the LFCA_EG into the mini-cellulosome was accomplished by cloning the LFCA_EG sequence into a pET28a vector between the NcoI and EcoRI sites. Then, the sequence of C. thermocellum Cel8A dockerin (and N-terminal linker) was PCR amplified and cloned at the C terminus of the LFCA_EG sequence between EcoRI and XhoI sites, thus generating pET28-LFCA_EG-Dockerin that carries a C-terminal hexa-histidine tag. LFCA_EG-CBM was generated by replacing the Cel8A dockerin with a sequence containing the linker between Cel8A catalytic domain and dockerin, followed by the CipA CBM. Both mini-scaffoldins and LFCA_EG fusion proteins were expressed in E. coli BL21 star (DE3). Expression of mini-scaffoldins was carried out at 16°C with 0.1 mM IPTG overnight, while LFCA_EG fusions and Cel8A were expressed at 37°C for 3 h in 1 mM IPTG. Cultures were lysed by enzymatic digestion in 1 mg mL −1 lysozyme, 1% Triton X-100, 5 µg mL −1 DNAseI, and 5 µg mL −1 RNAse A and centrifuged to remove cell debris. Clarified samples were incubated at 55°C for 20 min, cooled in ice, and centrifuged to eliminate aggregated proteins. Affinity purification was then carried out using HisTrap columns in an ÅKTA Purifier FPLC (GE Healthcare). Sample purity was evaluated by SDS-PAGE and proteins were concentrated in Tris 50 mM, NaCl 300 mM, and CaCl 2 1 mM, pH 7, quantified by absorbance at 280 nm with a NanoDrop (ThermoScientific) and stored in 50% glycerol. Mini-cellulosome assembly assays were performed by native PAGE. Different relations of proteins were incubated in 50 mM Tris, 300 mM NaCl, and 1 mM CaCl 2 , pH 7 at 37°C for 1 h before running the gel. SdbA cohesin was also added to block XDock in the scaffoldin. The true enzyme-scaffoldin ratio was determined from this analysis and according to that ratio, no free protein was found in excess. This ratio was used in the following experiments.
Microcrystalline cellulose binding was assayed as described previously 32 . Briefly, 10 µg of protein was incubated with 10 mg of Avicel (Sigma-Aldrich) at 4°C for 1 h with gentle agitation. Samples were centrifuged and the supernatant was stored as the unbound fraction. The pellet was washed three times and used as the bound fraction. Both samples were then analyzed by SDS-PAGE and BSA was used as a control.
Cellulosome activity assays. Proteins were incubated in acetate buffer, pH 5.5 containing 100 mM NaCl, 12 mM CaCl 2 , and 2 mM EDTA for 1 h at 37°C to allow complex formation. Enzymes were used at 0.5 µM for Avicel and PASC analyses, and at 0.35 µM for CMC assays. Scaffoldins were added at equimolar concentration according to native-PAGE analysis. BSA was added in all samples to minimize unspecific enzyme-substrate interactions. Avicel assays were conducted for 24 h in an orbital shaker in 2-mL tubes containing a wing magnet to improve stirring, so that this insoluble substrate did not precipitate. PASC was prepared as described elsewhere 51 . Assays with this substrate were conducted in similar tubes but in a heating block for 30 min. After incubation time, samples were centrifuged and the soluble sugars present in solution in the supernatant were determined by the DNS assay. Absorbance was measured in a 96-well plate using a FLUOstar fluorimeter (BMG Labtech, Germany) in the absorbance mode. CMC assays were conducted in a heating block using azo-CMC (Megazyme) as a substrate. The activity was determined according to the manufacturer's indications.
Crystallization data collection and structure determination. Protein, at an initial concentration of 8.0 mg mL −1 in 50 mM sodium citrate buffer, pH 4.8, was incubated with CMC at double concentration for 30 min. The excess of ligand was removed by washing with the same buffer with centrifugation, using 0.5 ml concentration units (Amicon). The final protein concentration was 17.0 mg mL −1 , as determined spectrophotometrically. An initial crystallization screening was done using the vapor-diffusion technique in its hanging-drop configuration. Crystallization experiments were set up in 24-well crystallization plates VDX (Hampton Research), using the 100-conditions kit from the Hampton Research Screen I&II. Hanging droplets were prepared by mixing protein solution (1 µL) with reservoir solution (1 µL) on a 22 mm siliconized round coverslip inverted over a 500 µL reservoir. Crystals were obtained in conditions C22 (0.2 M sodium acetate trihydrate, 0.1 M Tris hydrochloride, pH 8.5, and 30% w/v PEG 4 K) and C32 (0.1 M MES monohydrate, pH 6.5, 12% w/v PEG 20 K) of rod and hexagonal plate shapes, respectively. For data collection, crystals were cryoprotected in the mother liquid containing 15% (v/v) glycerol, cryocooled in liquid nitrogen, and stored until data collection. Crystals were tested at the European Synchrotron Radiation Facility (beam line ID30B). Data were indexed and integrated with XDS 52 , and scaled and reduced with AIMLESS 53 of the CCP4 program suite 54 . The structure was determined by molecular replacement, using the coordinates of the endoglucanase from B. subtilis (PDB:3PZV) as the search model as suggested from Phyre2 55 . The molecular replacement solution was found using Phaser 56 locating the six monomers in the asymmetric unit. Several cycles of manual building steps, Coot 57 , and structure refinement, phenix.refine 58 , were done followed by continuous model check with MolProbity 59 , as implemented within the Phenix suite 58  Computational structural characterization. We have run atomistic MD of Bs_EG, Tm_EG, and LFCA_EG, in the presence of a cellulosic substrate starting from the experimental structures (PDB IDs: 3AMC, 3PTZ, and 6GJF, respectively). None of these structures were resolved in the presence of an oligosaccharide, and for this reason, we had to introduce it manually using the following procedure. First, we fitted the structure of a tetrasaccharide formed by four units of D-glucose linked by β (1 → 4)-glycosidic bonds on the corresponding atoms of the cellotetraose-bound Tm_EG E253A mutant (pdb 3PZT), so that the four glucose monomers corresponded to subsites −3, −2, −1, and +1, resulting in a conformation that is compatible with catalysis. Having the tetrasaccharide well positioned, then we used the MultiSeq plugin 60 available in the VMD software to fit each of the three enzyme structures of interest on that of the Tm_EG mutant based on their structural alignment. The coordinates of sugar and enzyme were then combined, solvated, and energy minimized. Simulations were run using an identical protocol for all three enzymes, involving a short NVT run with position restraints on the enzyme and the sugar, followed by removal of restrains on the enzyme, a 20 ps NPT run to equilibrate the box volume and a production run at 300 K in the NVT ensemble, using a stochastic dynamics integrator with 2 fs time steps. The particle mesh Ewald method 61 was used for the electrostatics and the distances for all the hydrogen-heavy atom bonds were constrained using LINCS. All the simulations were run using the Gromacs 2018 software package 62 . We used the optimized Amber03* force field 63 for the protein with the TIP3P water model 64 . The doglycans tool 65 was used to generate parameters for the oligosaccharide, so that they can be read by the Gromacs software package. We chose the GLYCAM parameter set that is compatible with the Amber force field family 66 . Specifically, we used the prepreader.py script to prepare the parameters for the carbohydrate chain.
Elastic network models. To gain further insight on the slow conformational dynamics of the proteins of interest, we resort to ENMs. ENMs are based on the assumption that the dynamic properties of proteins are dictated by the topology of native contacts 67 . This type of model, combined with normal mode analysis, has been very useful for a variety of applications related to the study of protein dynamics, including the identification of functional conformational changes in enzymes and the comparison of ensembles of experimental structures 35,67 . Here, we limit our study to the simplest and most broadly used type of ENM, the anisotropic network model (ANM).
We have used the Python package ProDy 68 (http://prody.csb.pitt.edu/) to generate ANMs of Tm_EG, Bs_EG, and LFCA_AG, using the same PDB files as for the MD simulations. The ANM is built using the C α trace of the protein, whose atoms are connected by harmonic springs, resulting in an energy function where the sum runs over pairs ij of residues under a cutoff distance (r cut ), the terms r ij and r ij 0 correspond to the distances between pairs of C α atoms in instantaneous and reference configurations, respectively, and γ i are the force constants 67 . Here, we use the default parameters in ProDy for both cutoff distances (r c = 15 Å) and force constants (γ = 1). The analysis of the Hessian of the potential returns the normal modes of the system. The lowest frequency normal modes are of greatest interest because they contain information about the large-amplitude motions in the biomolecule.

Data availability
Data supporting the findings of this study are available from the corresponding author upon reasonable request. Coordinates and structure factors have been deposited in the Protein Data Bank repository with accession code 6GJF.