Caldicellulosiruptor strains C. saccharolyticus DSM 8903, C. hydrothermalis 108 (DSMZ 18901), and C. kronotskyensis 2002 (DSMZ 18902) were grown anaerobically, and genomic DNA was isolated in order to amplify the sequences of the genes with locus tags Csac_0437 (ABP6075.1), Calhy_0274 (ADQ06027.1) and Calkro_0290 (ADQ45201), respectively.

After cloning in expression vector pET24c(+), the expected constructs were verified by sequencing. Expression in E. coli BL21 Star and the subsequent purification resulted in sufficient amounts of the proteins for further study. Separation by SDS-PAGE showed bands corresponding to the expected molecular weights including the His₆-tags (Cs_Gaf159A: 37.6 kDa; Ch_Gaf159A: 37.6 kDa; Ck_Gaf159A: 37.7 kDa) for all enzyme preparations used in this study (Figure 1). A high degree of purity as judged from SDS-PAGE was achieved by means of immobilized metal affinity chromatography. Heat treatment of crude extracts from the recombinant E. coli strains at 55°C for depletion of host proteins did not result in a much higher purity, although only small amounts of the respective enzymes precipitated (Supplementary Figure S1).

Activity assays with para-nitrophenyl-β-d-galactofuranoside (pNPG), pNP-α-l-arabinofuranoside (pNPA), pNP-α-d-galactopyranoside, pNP-α-l-fucopyranoside, pNP-β-l-fucopyranoside, pNP-β-d-glucopyranoside, pNP-β-d-xylopyranoside, pNP-β-d-mannopyranoside, pNP-α-l-rhamnopyranoside, and pNP-α-d-glucopyranosiduronic acid as substrates showed that the enzymes were active only on pNPG and pNPA. The activity with pNPG was on average 3.6-fold higher at 65°C and 2.6-fold higher at 80°C than the activity with pNPA (Figure 2). K_m values with pNPG ranged from 4.2 to 8.5 mM, while the maximum velocity (V_max) was determined to be 2.7–2.9 U mg⁻¹ (Supplementary Figure S2). Therefore, the enzymes can be classified as galactofuranosidases (Gafs).

Since arabinofuranosidase activity was not previously reported for GH159 enzymes, we mainly focused on this newly detected activity in the present work. The mean specific activities on pNPA for Cs_Gaf159A, Ch_Gaf159A, and Ck_Gaf159A were 0.04, 0.04, and 0.03 U mg⁻¹, which corresponds to 1.5, 1.5 and 1.3 U µmol⁻¹, all respectively. To uncover the influence of different metal ions on the activity, standard assay reactions were supplemented with various salt solutions. The addition of 1 mM of MnCl_2, CaCl₂, MgCl₂, and CoCl₂ increased the activity (see Supplementary Table S1). NaCl and KCl had nearly no or a slightly negative effect, whereas addition of 1 mM ZnCl₂ or CuCl₂ reduced the activity by 80% (Figure 3A). At 10 mM, the effects of the salts on enzymes activity were similar, the difference being that the higher concentrations of NiSO₄ and CoCl₂ had an inhibitory effect (Figure 3B). The data for MnCl₂, CoCl₂, NiCl₂, and CuCl₂ that is shown in Figure 3 should not be overinterpreted due to color and precipitate formation under the assay conditions and hence interference with the assay used. Based on its activity-enhancing effect, 10 mM CaCl₂ was included in the standard assay buffer. This led to at least a two-fold increase in specific activities, i.e., about 0.1 U mg⁻¹ (3.76 U µmol⁻¹), 0.09 U mg⁻¹ (3.39 U µmol⁻¹), and 0.08 U mg⁻¹ (3.02 U µmol⁻¹) with pNPA as substrate, of Cs-, Ch- and Ck_Gaf159A, respectively. Addition of 10 mM of the chelating agent EDTA resulted in an almost complete inactivation of all three enzymes, thus confirming their dependency on divalent metal ions (Figure 3B).

The temperature ranges in which the enzymes displayed at least 60% activity (pH 6.5) were 78.4–99.0°C for Cs_Gaf159A, 71.7–89.6°C for Ch_Gaf159A and 78.4–96.3°C for Ck_Gaf159A, whereas the temperatures for maximal activity (20 min assay with pNPA as substrate) were 90°C, 82°C, and 93°C, respectively (see Figure 4).

The pH optimum, as determined at the respective apparent 20 min temperature optimum of each enzyme, differed less between the enzymes than the temperature optima. Using assays carried out at 80°C all three Gafs displayed the highest activity at around pH 5.6–6 (Figure 5). At room temperature, the optimal pH levels were found to be at pH 5.5–6 in McIlvaine+ buffer and pH 6.4–6.8 in MOPS buffer (Supplementary Table S2). The specific activities in McIlvaine+ buffer were from two to more than four times lower than in MOPS, with 0.02 U mg⁻¹ (0.8 U µmol⁻¹) for Cs_Gaf159A and 0.03 U mg⁻¹ (1.0 U µmol⁻¹) for Ch_Gaf159A and Ck_Gaf159A, possibly due to complex formation between the divalent cations and the phosphate present in the McIlvaine+ buffer system.

Regarding resistance against thermo-inactivation, incubation of the enzymes at their apparent “optimal” temperatures (temperature of maximum activity in a 20 min assay with substrate pNPA) indicated a high level of thermostability, especially for Cs_Gaf159A. Incubation of up to 24 h at 90°C did not seem to affect the activity of the latter. Ch_Gaf159A lost 50% of its activity after 5 h at 82°C, and Ck_Gaf159A lost nearly 50% of its activity after 2.5 h at 93 °C (Supplementary Figure S3). Although Cs_Gaf159A revealed an apparently extraordinary heat stability at 90 °C, variation of the assay time between 10 and 30 min indicated that heat inactivation at this temperature over a short time span did indeed occur (see Supplementary Figure S4). The explanation for the extreme long-term stability at 90°C may lie in the refolding of previously (partially) inactivated enzyme. This refolding may occur during cooling to room temperature after sample withdrawal and before addition of the pNP-arabinoside substrate and the measurement of residual activity. To assess the enzymes’ structural thermostability more directly, the melting temperatures (T_m) of the enzymes were determined using differential scanning fluorimetry (DSF). The apparent T_m values for Cs_Gaf159A, Ch_Gaf159A, and Ck_Gaf159A were 95.5°C, 92.5°C, and 99.0°C respectively (Figure 6), which was consistent with a high intrinsic stability of all three orthologous enzymes.

As for kinetic parameters and product inhibition, the latter was low under presumed physiological conditions, as the enzymes retained more than 50% of their activity in the presence of 500 mM arabinose in the reactions (Supplementary Figure S3). Due to the instability of the substrate pNPA at the high assay temperatures (90, 82, and 93°C) used, only short reaction times were applied.

Kinetic parameters with the chromogenic arabinoside substrate pNPA were able to be determined only up to a concentration of 10 mM. Results obtained at the standard enzyme concentration of 0.1 µg μl⁻¹ (2.65 µM) after 10, 20 and 30 min at the enzymes’ optimal temperature are shown by way of example for Cs_Gaf159A at 90°C in Supplementary Figure S4. Substrate saturation was not reached with substrate concentrations up to 10 mM using a 10 min assay. On the other hand, the curves generated after incubation for 20 or 30 min showed some flattening at higher substrate concentrations, which is presumed to be the result of a loss of enzyme activity over time at 90°C (see above). Therefore, the reactions were stopped after 10 min in the following experiments. The curves of substrate concentration vs. reaction rates were recorded for Cs_Gaf159A and Ch_Gaf159A at various enzyme concentrations. Although the conditions for near saturation of the substrate were not met, the curves suggest a minimum Km for pNPA of at least 10 mM.

Overnight incubation of 5 mg ml⁻¹ arabinan, arabinoxylan, arabinogalactan, and gum arabic with the three enzymes at different temperatures (65, 70, 75, and 80°C) were analyzed via the DNSA assay. Increased reducing sugar content was measureable only with arabinogalactan (containing arabinose residues linked either α-1,6 to the galactan backbone, or α-1,4 to galactose side chains) when compared to control reactions without enzyme addition. Product analysis by HPAEC-PAD revealed that only arabinose (about 5 mg L⁻¹ corrected by enzyme and substrate controls) was liberated after an incubation period of 23 h (Supplementary Figure S5). This only accounts for about 0.6% of the possible release of arabinose, which was calculated to be 0.76 mg ml⁻¹. This was calculated assuming 95% purity and 14% arabinose content in larch arabinogalactan (Megazyme data sheet), a substrate concentration of 5 g L⁻¹ and the ratio in molecular weight of hydrolyzed to non-hydrolyzed arabinose of 150.13 g mol⁻¹: 132.11 g mol⁻¹. The calculated specific activity was 0.024 nmol min⁻¹ mg⁻¹.

In contrast to the lack of activity against polymeric arabinoxylan, all three wildtype enzymes were able to efficiently cleave arabinoxylooligosaccharides (AXOS). HPAEC-PAD analysis (Figure 7; Supplementary Figure S6) revealed that, in 24 h reactions containing 0.1 g L⁻¹ substrate, the arabinofuranosidase activity of Cs_Gaf159A, Ch_Gaf159A, and Ck_Gaf159A was able to fully cleave the α-l-1,3-O linkage in A^2,3XX, yielding A²XX and arabinose as the only products. In contrast, in A³X-containing reactions under identical conditions only small amounts of arabinose and xylobiose (X2) were released, indicating incomplete hydrolysis of A³X.

Ch_Gaf159A was crystallized and its structure determined through molecular replacement to a resolution of 1.7 Å (R_free = 15.3%; PDB ID: 7ZEI, see Supplementary Table S3). The stable monomeric state of Ch_Gaf159A was confirmed by means of polymerization-induced self-assembly (PISA) (Krissinel and Henrick 2007) calculations and size-exclusion chromatography. The PDB Pisa analysis predicted a solvent-accessible surface area of 13519.7 Å² and −324.2 kcal/mol free energy for folding. The fold of Ch_Gaf159A displays striking similarities to that of glycoside hydrolase family 43. A search for homologous topologies using the DALI-server (Holm and Sander 1995) revealed structures with high Z scores but low sequence identities of <20% (structural superposition of representatives with Ch_Gaf159A are shown in Supplementary Figure S7). The best hits turned out to be an exo-α-1,5-l-arabinofuranosidase from GH43_26 (root-mean-square deviation (rmsd): 1.5 Å for 173 C_α atoms, sequence identity (SI): 17%, Z = 27.0, PDB ID: 3AKI) (Fujimoto et al., 2010) and a mutant derivative (E186Q) of Bacillus pumilus β-xylosidase XynB in complex with xylobiose (rmsd 1.7 Å for 181 C_α atoms, sequence identity 15%, Z = 26.2, PDB ID: 5ZQX) classified in GH43_11 (Hong et al., 2018). However, Ch_Gaf159A is substantially smaller than its homologues, consisting of only a five-bladed β-propeller domain (Figure 8A). The metal ion present in the central cavity was modelled as Ca²⁺ due to the absence of any anomalous signal and the experimental findings for metal ion dependence described above (Figure 8A). Each blade is composed of four mainly antiparallel beta-strands. The overlay with XynB in complex with xylobiose (X2) suggests the substrate binding on the upper part of the barrels center (Figure 8B). The positions of E190, D142, and D19 in Ch_Gaf159A exactly matched E186, proposed to be the nucleophilic catalytic base, D127, and D14 in XynB (Figure 8C). The latter two possible catalytic aspartic acid residues of Ch_Gaf159A surround the non-reducing end of the substrate in subsite -1 of the binding site. The specific interaction with hydroxides of C2, C3, and C4 in xylobiose determines the orientation of the glycosidic linkage and therefore substrate specificity. E190 in Ch_Gaf159A is suggested as being catalytically important given its proximity to the glycosidic O and its description as a catalytic base in XynB. H244 located between D19 and D142, also adopts a putatively important position in coordinating the metal ion (Figure 8C). This matches the findings for CoXyl43 as previously described by Matsuzawa et al. (2017) (Supplementary Figure S8).

Analyses for pronounced specificity pockets at the active site identified a solvent exposed negatively charged ligand binding site with approximately 8 Å in length and 6 Å in diameter (Berka et al., 2012). The surface representation showed two electro negative pockets, one of them capable of incorporating the X2 ligand (Figure 8D). This substrate pocket structurally explains why Ch_Gaf159A acts as an exo-arabinofuranosidase, or -galactofuranosidase. The substrate can only enter the pocket at one end and is specifically bound in the -1 subsite (Figure 8D).

The sequence of Ch_Gaf159A also structurally aligned (Supplementary Figure S9) with the catalytic residues of GH43_33 HoAraf (PDB 4QQS), an α-l-arabinofuranosidase from Halothermothrix orenii (Hassan et al., 2015). The known conserved catalytic nucleophile of HoAraf (E195) aligned with E190 in Ch_Gaf159A, the catalytic base D17 (HoAraf) with D19, and D126 in HoAraf corresponded to D142 in Ch_Gaf159A. This indicated that all three residues are catalytically important.

Therefore, three variants of Ch_Gaf159A were generated at the possible catalytic amino acid position E190, D19 and D142, each containing an alanine residue instead of the proposed catalytic residue. All three mutants lost the ability to cleave pNPA, as well as A^2,3XX and A³X (Figure 7), nor did they show any activity against the other tested AXOS (Supplementary Figure S10). The behavior during expression, purification, or protein analysis exhibited change in none of the mutants (Mut I D19A, Mut II D142A, or Mut III E190A) as compared to the wildtype Ch_Gaf159A. DSF analyses revealed that the melting temperature T_m of Mut II (D142A) was unchanged. In contrast, the T_m of Mut I (D19A) was reduced by 7.5°C, and that of Mut III (E190A) increased by 3°C (Figure 6). Since differences in melting temperatures could indicate differences in folding properties, Circular Dichroism (CD) measurements were additionally performed. The far UV CD-spectra of the mutants compared to wild type indicated no changes in secondary structure and probably none in the overall folding properties of the backbone either (Rodger 2018), at 20°C (Figure 9A). In addition, the curves were identical for WT and mutants at 60°C (Figure 9B). Since the wild type still showed about 50% of its activity at 60°C, it can be assumed that the wild type was correctly folded at this temperature, which also applied to the mutants. According to the secondary structure composition deduced from the CD spectra by computational analysis, all Ch_Gaf159A variants are composed of 31–32% antiparallel beta-strands, 21% beta turns, and 8–9% helical structures. The proportion of random coils was calculated to be 35–36% in WT and mutants (Figure 9C).

The Carbohydrate Active Enzymes (CAZy) database (http://www.cazy.org/Glycoside-Hydrolases.html; Lombard et al., 2014) was scoured, and homology analysis was performed using the NCBI tool BLASTp (https://blast.ncbi.nlm.nih.gov/Blast.cgi). Signal peptides were predicted using the SignalP 4.1 server (http://www.cbs.dtu.dk/services/SignalP-4.1/; (Petersen et al., 2011). Structural information was extracted from the RCSB PDB Protein Data Bank (http://www.rcsb.org/). Structural alignments were created using HHpred, a MPI Bioinformatics Toolkit (https://toolkit.tuebingen.mpg.de/tools/hhpred; Zimmermann et al., 2018) to compare conserved catalytic amino acids.

The bacterial strains C. saccharolyticus DSM 8903, C. hydrothermalis 108 (DSMZ 18901), C. kronotskyensis 2002 (DSMZ No. 18902) were obtained from the DSMZ (Braunschweig, Germany). Dried cells were rehydrated with 500 µl GS2 medium (Johnson et al., 1981; Koeck et al., 2015) for 30 min under anaerobic atmosphere in an anaerobic chamber (98% N₂, 2% H₂; Coy Laboratory Products). An aliquot of 250 µl was used to inoculate 50 ml GS2 medium in butyl rubber-stoppered serum bottles with 0.1% cellobiose. Another 100 µl were plated in the anaerobic chamber onto a GS2 agar plate (1.8% agar). Cells were cultivated anaerobically at 65°C for 5 days (bottles without shaking). Genomic DNA of a clonal culture inoculated from a single colony was isolated using the Gentra® Puregene® kit for Yeast and Bacteria Kit (Qiagen). Briefly, 5 ml of the bacterial culture was centrifuged for 1 min at 13,000 rpm. After discarding the supernatant, the protocol for DNA isolation was followed in accordance with the manufacturer’s description for Gram-positive bacteria. However, 1.5 μl of lysozyme (730 U ml⁻¹) was used instead of 1.5 μl of the Lytic Enzyme Solution. PCR-amplified DNA fragments were purified using NucleoSpin® Gel and PCR Clean-up kits (Macherey-Nagel).

Recombinant Escherichia coli strains DH10B (Thermo Fisher Scientific, Waltham, MA, United States) and BL21 Star (Invitrogen, Carlsbad, CA, United States) were aerobically grown in LB-Lennox medium with 50 µg ml⁻¹ kanamycin at 180 rpm and 37°C. Plasmids were isolated from overnight cultures using the NucleoSpin Plasmid kit from Macherey-Nagel (Düren, Germany). Amplified DNA was purified using the NucleoSpin® Gel and PCR Clean-up (Macherey-Nagel, Düren, Germany). Plasmid pET24c (+) (Merck, Darmstadt, Germany) was amplified in E. coli DH10B, cut with NdeI and XhoI (New England Biolabs GmbH, Frankfurt, Germany) and purified from an 1% agarose gel. The genes coding for the GH159 enzymes with arabinofuranosidase activity from C. saccharolyticus (Cs_Gaf159A; ABP66075.1), C. hydrothermalis (Ch_Gaf159A; ADQ06027.1) and C. kronotskyensis (Ck_Gaf159A; ADQ45201.1) were amplified from genomic DNA by PCR using Phusion High-Fidelity DNA Polymerase (Thermo Fisher Scientific, A, United States), genomic DNA of the respective strain and primer pairs as listed in Table 1. Cloning of pET24c(+) constructs was accomplished by Gibson Assembly (Gibson et al., 2009) using purified PCR product and restricted vector as described above, followed by transformation of strain E. coli DH10B. Sanger sequencing at Eurofins Genomics (Ebersberg, Germany) verified sequence integrity before enzyme production in E. coli BL21 Star™, isolation and purification by immobilized metal affinity chromatography (IMAC) was carried out as described previously (Baudrexl et al., 2019). The eluates from IMAC were incubated at 50°C for 15 min to denature remaining E. coli proteins followed by precipitate removal by centrifugation. Size and purity of the proteins were confirmed by SDS-PAGE (Laemmli 1970). Protein concentration was determined spectrophotometrically at 280 nm under denaturing conditions (5 M urea) using the molar extinction coefficients, calculated by the ExPaSy ProtParam tool (https://web.expasy.org/protparam/) from the Swiss Institute of Bioinformatics (Artimo et al., 2012).

Crystals of Ch_Gaf159A (Genbank: ADQ06027.1, Uniprot: E4QB49) from C. hydrothermalis were grown at 20°C within 14 days to a final size of 300 μm × 100 μm × 100 μm by using the sitting drop vapour diffusion method. The drops contained 0.3 µl protein (30 mg ml⁻¹) and 0.1 µl reservoir (100 mM 2-(N-morpholino)-ethanesulfonic acid, pH 6.5, 30% (w/v) polyethylene glycol 8000, and 5% (v/v) glycerol). Before exposure to X-rays, crystals were soaked in a mixture of mother liquor and glycerol (5:1, v/v) for 30 s and vitrified in liquid nitrogen. A native diffraction data set was collected using synchrotron radiation at the X06SA-beamline, SLS, Switzerland. Reflection intensities were evaluated with the program package XDS and data reduction was carried out with XSCALE (Kabsch 2010) (Supplementary Table S3). Initial phases were obtained by Patterson search calculations with PHASER (McCoy et al., 2007) using a starting model predicted by alphafold2 (Jumper et al., 2021). Six monomers could be positioned in the asymmetric unit. Phases were improved by cyclic non-crystallographic averaging methods, resulting in an electron density map that allowed to correct misaligned secondary structure elements and loop connections. The structure was completed in successive rounds of model building with COOT (Emsley and Cowtan 2004) and refined with REFMAC5 (Vagin et al., 2004). Water molecules were placed with ARP/wARP solvent (Perrakis et al., 1997). The determined crystal structure has superb crystallographic values (Rcrys = 0.128 and Rfree = 0.153), was proven to fulfil the Ramachandran plot (Supplementary Table S3) and evaluated by MolProbity (Williams et al., 2018). Interface areas were calculated with PISA (‘Protein interfaces, surfaces and assemblies’ service at the European Bioinformatics Institute) (Krissinel and Henrick 2005). Coordinates and structure factor amplitudes have been deposited in the RCSB Protein Data Bank under the accession code 7ZEI.

Mutations were introduced into the gene sequences via site directed mutagenesis. For this, linear fragments with overlapping ends were amplified using PCR reactions with reverse complementary primer pairs with single nucleotide changes as indicated in Table 1 and the constructed vector pET24c-ChGaf159A as template. Reactions were subjected to complete digestion with DpnI to remove non-mutated methylated template DNA isolated from the cloning strain E. coli DH10b (Marinus and Løbner-Olesen 2014). Transformation of this strain using the resulting linear DNA with overlapping ends lead to self-circularization upon replication in the transformed cells. The expected mutations were verified by sequencing.

Standard reactions with pNP-glycosides, purchased from Megazyme (Wicklow, Ireland), were performed in a total volume of 50 µl with 1 mM substrate and 2.65 µM enzyme (1 µg μl⁻¹) in standard reaction buffer (100 mM MOPS pH 6.5, 50 mM NaCl, 10 mM CaCl₂) for 20 min at the indicated temperature. For a first screening, pNP-glycoside reactions were incubated at 65°C, the optimal growth temperature of Caldicellulosiruptor. For further biochemical characterization, the respective apparent “temperature optima” (temperatures at which maximum activity was measured under standard assay conditions in a 20 min assay) were used, i.e., 90°C for Cs_Gaf159A, 82°C for Ch_Gaf159A, and 93°C for Ck_Gaf159A. Reactions were performed in triplicates and stopped by adding 100 µl 1 M Na₂CO₃. Color formation from released pNP was quantified by spectrophotometric measurement of 100 μl at 405 nm and calculated using a standard curve created with pNP as reference.

Activity on various pNP glycosides was assessed by prolonged overnight incubation to detect even very low activity. Dependency on and response to ions was examined by using 100 mM MOPS buffer pH 6.5 in standard reactions with and without addition of salt solutions (1–10 mM of CaCl₂, NaCl, CuCl₂, NiSO₄, KCl, MnCl₂, MgCl₂, FeCl₃, ZnCl₂, CoCl₂, NH₄Cl, K₂HPO₄, KSO₄, KNO₃, or K₂CO₃). For this purpose, the storage buffer for the enzymes was changed to 0.1 M MOPS pH 6.5 without additional ions, either by diafiltration using Vivaspin columns, cutoff 10 kDa (Sartorius AG, Goettingen, Germany), or by using PD-10 desalting columns (Cytiva Europe GmbH, Freiburg, Germany). Temperature optima were determined by measuring activities in a gradient thermocycler (TAdvanced 96S, Biometra, Göttingen, Germany), with a span of 40°C over 12 reactions.

Because the pH tolerance strongly depends on the temperature, an established high throughput method was used to analyze the pH optima as described by Herlet et al. (2017). Citric acid—phosphate buffer was prepared in the style of (McIlvaine 1921), but stock solutions were supplemented with 0.1 M NaCl (McIlvaine+) as described by Herlet et al. (2017). This buffer system (pH 4.5–8) was used in a ratio of 1:2, whereas MOPS buffer (pH 6–7.4) was used in final concentrations as in the standard reactions (0.1 M). Buffers were prepared at room temperature. To take the temperature dependency into account, pH values were additionally measured at 80 °C to use actual pH values under reaction conditions to visualize the pH optima. Values for both buffers at RT and 80°C can be found in Supplementary Table S2.

Kinetic parameters v_max and K_m as known from the Michaelis Menten equation ( v 0 = v max × [ S ] K m [ S ] ) were calculated from substrate concentration vs. reaction velocity graphs. Standard reactions were performed with final substrate concentration up to 20 mM and enzyme concentrations were reduced to 0.001 µg μl⁻¹ (equal to 0.027 µM Cs_Gaf159A). The reaction time was reduced to 10 min to minimize the effect of decrease in specific activity with time.

For activity based stability tests standard reactions were preincubated at different temperatures and then briefly cooled on ice before the substrate was added. To assess intrinsic protein thermal stability, DSF was performed in a C1000 Touch Thermal Cycler equipped with the CFX96 Real-Time System (Bio-Rad). Reactions contained 4 µM protein in reaction buffer (100 mM MOPS pH 6.5, 10 mM CaCl₂, 50 mM NaCl) and 20x SYPRO® Orange Protein Gel Stain (Merck KGaA, Darmstadt, Germany) diluted in the respective buffer in a total volume of 50 µl. Triplicates were pipetted on ice in white RT-PCR 96-well plates, sealed with Microseal®B Adhesive Sealer (Bio-Rad, MSB-1001). Samples were incubated at the initial temperature (25°C, or 50°C) for 10 min before the temperature was increased every 15 s by 1°C each. Fluorescence was measured each time before the temperature changed. The measured values and respective temperatures were recorded at the respective time points. Interpretation of the saturation curves was performed using the xy analysis method “Smooth differentiate curve” in Graphpad Prism 7.00.

CD spectroscopy was conducted with 0.1 µg μl⁻¹ enzyme in 10 mM Tris pH 6.5, using a Jasco–J710 CD spectrometer (Oklahoma City, OK, United States), a 1 mm cuvette, a range from 195 to 260 nm, and the signal average of 10 scans at 20 and 60°C as described by Stelzer et al. (2008). Secondary structure compositions were obtained after calculation of molar elipticity from signals in mdeg on the basis of reference spectra (Poschner et al., 2007).

The reactions used to test the ability of the enzymes to cleave arabinose from polysaccharides contained 0.5% wheat medium viscosity arabinoxylan (P-WAXYM, Megazyme, Wicklow, Ireland), sugar beet arabinan (P-ARAB, Megazyme), larch arabinogalactan (P-ARGAL, Megazyme) or acacia gum arabic (Merck, Darmstadt, Germany), 1x RP (100 mM MOPS pH 6.5, 50 mM NaCl, 10 mM CaCl₂), and 2.65 µM of the respective enzyme. Reactions were incubated overnight at 65, 70, 75, or 80°C. Preferences for specific arabinosyl linkages in arabinoxylan-derived oligosaccharides (AXOS) purchased from Megazyme were tested using an 24 h incubation at 75°C and just 0.01% (100 mg L⁻¹) of the respective substrate as follows: Either 3²-α-l-arabinofuranosyl-xylobiose (A³X), 2³-α-l-arabinofuranosyl-xylotriose (A²XX), 2³,3³-di-α-l-arabinofuranosyl-xylotriose (A^2,3XX), 2³,3³-di-α-l-arabinofuranosyl-xylotetraose (XA^2,3XX), 3³-α-l-arabinofuranosyl-xylotetraose (XA³XX), or a mixture of XA³XX and 2³-α-l-arabinofuranosyl-xylotetraose (XA²XX) which is referred to as XAXX-Mix. The number in the exponent indicates which xylose counting from the reducing end, and the integer of the base shows which oxygen atom of this xylose is substituted with arabinose (Supplementary Figure S12; abbreviations adopted from Fauré et al., 2009).

Hydrolysates were analyzed for the liberated reducing ends using the dinitrosalicylic acid method (DNSA), as described elsewhere (Wood and Bhat 1988). To summarize, 50 µl of reactions, or arabinose standard solutions (0.2–2 mg ml⁻¹) were mixed with 75 µl reagent (10 g L⁻¹ dinitrosalicylic acid, 200 g L⁻¹ potassium sodium tartrate, 10 g L⁻¹ NaOH, 0.5 g L⁻¹ Na₂SO₄, 2 g L⁻¹ phenol) and heated at 95°C for 10 min. After measuring 100 μl at 540 nm, values were interpolated from arabinose standard curves using non-linear regression in GraphPad Prism 7.

High performance anion exchange chromatography with pulsed amperometric detection (HPAEC-PAD) of polysaccharide and oligosaccharide hydrolysates was conducted as described (Mechelke et al., 2017; Baudrexl et al., 2019) at an adjusted gradient: During the first 10 min, a gradient from 0 to 100 mM sodium acetate was used; in the following 20 min a second linear gradient to a final concentration of 800 mM NaOAc was suitable for separating the arabinoxylan-derived oligosaccharides in a shorter time. In addition, the arabinose was quantified using arabinose solutions (12.5–200 mg L⁻¹) as external standards.