Two Subgroups within the GH43_36 α-l-Arabinofuranosidase Subfamily Hydrolyze Arabinosyl from Either Mono-or Disubstituted Xylosyl Units in Wheat Arabinoxylan

Fungal arabinofuranosidases (ABFs) catalyze the hydrolysis of arabinosyl substituents (Ara) and are key in the interplay with other glycosyl hydrolases to saccharify arabinoxylans (AXs). Most characterized ABFs belong to GH51 and GH62 and are known to hydrolyze the linkage of α-(1→2)-Ara and α-(1→3)-Ara in monosubstituted xylosyl residues (Xyl) (ABF-m2,3). Nevertheless, in AX a substantial number of Xyls have two Aras (i.e., disubstituted), which are unaffected by ABFs from GH51 and GH62. To date, only two fungal enzymes have been identified (in GH43_36) that specifically release the α-(1→3)-Ara from disubstituted Xyls (ABF-d3). In our research, phylogenetic analysis of available GH43_36 sequences revealed two major clades (GH43_36a and GH43_36b) with an expected substrate specificity difference. The characterized fungal ABF-d3 enzymes aligned with GH43_36a, including the GH43_36 from Humicola insolens (HiABF43_36a). Hereto, the first fungal GH43_36b (from Talaromyces pinophilus) was cloned, purified, and characterized (TpABF43_36b). Surprisingly, TpABF43_36b was found to be active as ABF-m2,3, albeit with a relatively low rate compared to other ABFs tested, and showed minor xylanase activity. Novel specificities were also discovered for the HiABF43_36a, as it also released α-(1→2)-Ara from a disubstitution on the non-reducing end of an arabinoxylooligosaccharide (AXOS), and it was active to a lesser extent as an ABF-m2,3 towards AXOS when the Ara was on the second xylosyl from the non-reducing end. In essence, this work adds new insights into the biorefinery of agricultural residues.


Introduction
Arabinoxylan (AX) is one of the major plant cell wall polymers present in monocotyledon biomass and can be utilized via enzymatic degradation for the biofuel or prebiotic industries [1,2]. AX composition and content is highly variable among various botanical species and tissue types. For instance, corn bran constitutes up to 50% hemicellulose consisting of mainly AX [3], whereas wheat bran constitutes up to 23 to 32% AX, and wheat endosperm only constitutes up to 2 to 4% AX [4]. Furthermore, different AX populations have varying substitution patterns and, based on water solubility, AX is generally separated into water-extractable and water-unextractable fractions [4,5].

Class
Enzyme Organism Selectivity Source
With only two characterized fungal ABF-d3 enzymes (in GH43_36) so far, we hypothesized that a subgroup of GH43_36 aligns with ABF-d3 activity, and this study sought to employ phylogenetic analysis to identify a new candidate with a different function to this subfamily and to biochemically characterize this candidate. Based on the phylogenetic analysis, a novel GH43_36 candidate from Talaromyces pinophilus was selected and heterologously produced and characterized by examination of its product profiles from AX and specific AXOS by 1 H-NMR, HPAEC-PAD, and MALDI-TOF MS. Furthermore, its catalytic rate and specificity were compared to previously characterized fungal GH51, GH62, and GH43_36 ABFs.

Structure-Based Phylogenetic Analysis of GH43_36 Reveals Further Subdivision with Distinct GH43_36a and GH43_36b Subgroups
For this study, about 810 publicly available GH43_36 sequences of the phylum Ascomycota were used for the phylogenetic analysis. This phylogeny showed that these GH43_36s were subdivided into two main clusters, cluster 1 contained the major clade A and minor clades C, D, and E, whereas cluster 2 was defined by the other major clade B (Figure 1). Nevertheless, multiple sequence alignment of a subset of representative GH43_36 sequences from clades A and B exhibited high overall similarity between the two clades with general conservation of the residues involved in substrate interaction (SI 1, Figure S1). A noticeable exception was the amino acid positioned at D291 of H. insolens ABF-d3 (SI 1, Figure S1). Interestingly, in clade A (in cluster 1), harboring H. insolens ABF-d3, for nearly all 500 sequences studied, exclusively aspartate (Asp) was found in position 291 (D291). In contrast, in clade B (in cluster 2) the nearly 250 sequences studied showed a tryptophan (Trp) at the equivalent position (W291) (Figures 1 and 2). In the remaining 7% of the sequences making up several smaller clades, either Asp, glutamic acid (Glu), aspargine (Asn), or serine (Ser) were found at position 291 ( Figure 1).

Structure-Based Phylogenetic Analysis of GH43_36 Reveals Further Subdivision with Distinct GH43_36a and GH43_36b Subgroups
For this study, about 810 publicly available GH43_36 sequences of the phylum A mycota were used for the phylogenetic analysis. This phylogeny showed that t GH43_36s were subdivided into two main clusters, cluster 1 contained the major clad and minor clades C, D, and E, whereas cluster 2 was defined by the other major cla (Figure 1). Nevertheless, multiple sequence alignment of a subset of representa GH43_36 sequences from clades A and B exhibited high overall similarity between two clades with general conservation of the residues involved in substrate interactio 1, Figure S1). A noticeable exception was the amino acid positioned at D291 of H. ins ABF-d3 (SI 1, Figure S1). Interestingly, in clade A (in cluster 1), harboring H. insolens A d3, for nearly all 500 sequences studied, exclusively aspartate (Asp) was found in pos 291 (D291). In contrast, in clade B (in cluster 2) the nearly 250 sequences studied sho a tryptophan (Trp) at the equivalent position (W291) (Figures 1 and 2). In the remai 7% of the sequences making up several smaller clades, either Asp, glutamic acid (G aspargine (Asn), or serine (Ser) were found at position 291 ( Figure 1).  tains two pockets within its active site, a catalytic main-pocket and a shallower sidepocket in which D291 is positioned [12] (Figure 2A). Surprisingly, a D291A mutation has previously been found to render the H. insolens ABF-d3 fully inactive [12], which points to the importance of D291 for the ABF-d3 function. It was speculated that the Asp/Trp partition between GH43_36 from clades A and B might reflect functional differentiation, hypothesized to result in the loss of activity towards dXyls for the GH43_36b enzymes. Modeled structures of two GH43_36 enzymes from clade A (HiABF43_36a) and clade B (TpABF43_36b) used in this study are displayed in Figure 2. Obviously, a single amino acid position in a protein never solely results in clustering. Indeed, several positions showed a preferential amino acid correlating with either the A or the B clade. Close to the pKa modulator E216 [12], position 218 of the A clade contained either alanine (Ala) or Ser, while in the B clade an Asn was almost exclusively found (SI 1, Figure S1). Although other similarities were found, none of the other positions with a preferential pattern for either the A or the B clade were as strictly conserved as the abovementioned amino acids at positions 291 and 218. All these other positions appeared to define structural elements that were located away from the active site.
Phylogenetic clustering within an enzyme family of broad taxonomic origin often indicates a functional differentiation, as seen in the Ascomycetes phylogenetic tree ( Figure  1). Nevertheless, a phylogenetic tree of about 52 credible public Basidiomycete GH43_36 sequences produced an imperfect clustering, yet still shows a major A clade having Asp at position 291 and a minor B clade with Trp or Ser at an equivalent position (SI 2, Figure  S2). Furthermore, the nearly 393 publicly available bacterial GH43_36 sequences all carry an Asp at the 291 position (SI 3, Figure S3), as found for the fungal clade A sequences.

Production, Purification, and General Characteristics of TpABF43_36b
One ABF43_36b from clade B (see Section 2.1) was mainly selected for its availability. It originates from an Ascomycete well known for producing carbohydrases with a significant impact in the hydrolysis of plant cell walls, Talaromyces pinophilus [23] (TpABF43_36b). The gene for this TpABF43_36b was heterologously expressed in a fungal production host. As a fungal host, an Aspergillus oryzae strain was used which had been optimized for monocomponent enzyme production [24]. After induced expression, the cultivation broth was sterile-filtered and subsequently purified by column chromatography. Subsequently, the SDS-PAGE of the TpABF43_36b (for sequence see SI 1, Figure  S1) showed a thick band at 55 kDa (SI 4, Figure S4), which was comparable to the theoretical Mw of 57178.11 Da. Based on the crystal structure of HiABF43_36a (PDB:3ZXJ_B), the fungal ABF-d3 contains two pockets within its active site, a catalytic main-pocket and a shallower sidepocket in which D291 is positioned [12] (Figure 2A). Surprisingly, a D291A mutation has previously been found to render the H. insolens ABF-d3 fully inactive [12], which points to the importance of D291 for the ABF-d3 function. It was speculated that the Asp/Trp partition between GH43_36 from clades A and B might reflect functional differentiation, hypothesized to result in the loss of activity towards dXyls for the GH43_36b enzymes. Modeled structures of two GH43_36 enzymes from clade A (HiABF43_36a) and clade B (TpABF43_36b) used in this study are displayed in Figure 2.
Obviously, a single amino acid position in a protein never solely results in clustering. Indeed, several positions showed a preferential amino acid correlating with either the A or the B clade. Close to the pKa modulator E216 [12], position 218 of the A clade contained either alanine (Ala) or Ser, while in the B clade an Asn was almost exclusively found (SI 1, Figure S1). Although other similarities were found, none of the other positions with a preferential pattern for either the A or the B clade were as strictly conserved as the above-mentioned amino acids at positions 291 and 218. All these other positions appeared to define structural elements that were located away from the active site.
Phylogenetic clustering within an enzyme family of broad taxonomic origin often indicates a functional differentiation, as seen in the Ascomycetes phylogenetic tree ( Figure 1). Nevertheless, a phylogenetic tree of about 52 credible public Basidiomycete GH43_36 sequences produced an imperfect clustering, yet still shows a major A clade having Asp at position 291 and a minor B clade with Trp or Ser at an equivalent position (SI 2, Figure S2). Furthermore, the nearly 393 publicly available bacterial GH43_36 sequences all carry an Asp at the 291 position (SI 3, Figure S3), as found for the fungal clade A sequences.

Production, Purification, and General Characteristics of TpABF43_36b
One ABF43_36b from clade B (see Section 2.1) was mainly selected for its availability. It originates from an Ascomycete well known for producing carbohydrases with a significant impact in the hydrolysis of plant cell walls, Talaromyces pinophilus [23] (TpABF43_36b). The gene for this TpABF43_36b was heterologously expressed in a fungal production host. As a fungal host, an Aspergillus oryzae strain was used which had been optimized for monocomponent enzyme production [24]. After induced expression, the cultivation broth was sterile-filtered and subsequently purified by column chromatography. Subsequently, the SDS-PAGE of the TpABF43_36b (for sequence see SI 1, Figure S1) showed a thick band at 55 kDa (SI 4, Figure S4), which was comparable to the theoretical Mw of 57178.11 Da.

Main and Side Activities of TpABF43_36b and HiABF43_36a on AX and AXOS
The reaction product profiles from AX generated by TpABF43_36b, HiABF43_36a, MgABF51, and PoABF62 were analyzed by HPAEC and MALDI-TOF MS (Figures 4 and 5, respectively). MgABF51, PoABF62, and HiABF43_36a mainly released arabinose from AX ( Figure 4), showing that these enzymes possessed, predominantly, ABF activity. Po-ABF62 also released xylose, albeit in minor amounts ( Figure 4). Likely, it could act on the xylan-chain in an exo-acting matter after arabinosyl substituents were removed, yet only at a low rate. The co-occurrence of these two functionalities is not uncommon and has been previously reported among ABFs, nevertheless not yet among GH62 ABFs [10,26,27]. MgABF51 seemed to release slightly more arabinose compared to PoABF62 after 24 h incubation (153 and 127 μg mL −1 , respectively), whereas after 1 h incubation, arabinose release was nearly identical for the two GH51 and GH62 enzymes used (127 and 122 μg mL −1 , respectively). In comparison, HiABF43_36a released 117 μg mL −1 arabinose after 24 h ( Figure 4) from dXyls. As the combination of HiABF43_36a and MgABF51 should, in theory, remove all arabinosyl substituents from the AX, a theoretical concentration of 380 μg mL −1 arabinose released (~38% total arabinose in AX; supplier information) was expected from their combined activity. This was indeed close to the theoretical combined activity from the sum of arabinose released by HiABF43_36a and MgABF51 when acting alone on AX after 24 h incubation (384 μg mL −1 ). Nevertheless, the combined arabinose release of a mixture of HiABF43_36a and MgABF51 after 1 h incubation was found to be incomplete, as only ~320 μg mL −1 arabinose was released (see Section 2.6.3).
As a further comparison, TpABF43_36b released 119 μg mL −1 arabinose after 24 h, and, thus, released slightly lower amounts of arabinose compared to MgABF51 and Po-ABF62 after 24 h incubation ( Figure 4).
More surprisingly, the new TpABF43_36b substantially released xylo-oligosaccharides (XOS) and arabinoxylo-oligosaccharides (AXOS), besides arabinose (Figures 4 and 5), while the other ABFs did not. All enzymes were produced in the same production host, and purified, hence, it was not expected that this oligomer release was due to impurities (i.e., other enzymes co-produced and ending up in the TpABF43_36b fraction used). To confirm this, proteomic analysis was carried out and no xylanase was identified (SI 5, Table S1). Nevertheless, in comparison to a well-characterized GH10 endoxylanase from Aspergillus aculeatus (AaXyn10), the xylanase-like activity of TpABF43_36b was only 0.3% compared to AaXyn10 (based on insoluble AZCL-xylan; results not further shown).
In summary, these findings demonstrated that the novel TpABF43_36b was mainly active as ABF-m2,3 with minor endoxylanase activity, albeit with a relatively low rate compared to the other ABFs tested and the AaXyn10 endoxylanase.

Main and Side Activities of TpABF43_36b and HiABF43_36a on AX and AXOS
The reaction product profiles from AX generated by TpABF43_36b, HiABF43_36a, MgABF51, and PoABF62 were analyzed by HPAEC and MALDI-TOF MS (Figures 4 and 5, respectively). MgABF51, PoABF62, and HiABF43_36a mainly released arabinose from AX ( Figure 4), showing that these enzymes possessed, predominantly, ABF activity. PoABF62 also released xylose, albeit in minor amounts ( Figure 4). Likely, it could act on the xylanchain in an exo-acting matter after arabinosyl substituents were removed, yet only at a low rate. The co-occurrence of these two functionalities is not uncommon and has been previously reported among ABFs, nevertheless not yet among GH62 ABFs [10,26,27]. MgABF51 seemed to release slightly more arabinose compared to PoABF62 after 24 h incubation (153 and 127 µg mL −1 , respectively), whereas after 1 h incubation, arabinose release was nearly identical for the two GH51 and GH62 enzymes used (127 and 122 µg mL −1 , respectively). In comparison, HiABF43_36a released 117 µg mL −1 arabinose after 24 h ( Figure 4) from dXyls. As the combination of HiABF43_36a and MgABF51 should, in theory, remove all arabinosyl substituents from the AX, a theoretical concentration of 380 µg mL −1 arabinose released (~38% total arabinose in AX; supplier information) was expected from their combined activity. This was indeed close to the theoretical combined activity from the sum of arabinose released by HiABF43_36a and MgABF51 when acting alone on AX after 24 h incubation (384 µg mL −1 ). Nevertheless, the combined arabinose release of a mixture of HiABF43_36a and MgABF51 after 1 h incubation was found to be incomplete, as only~320 µg mL −1 arabinose was released (see Section 2.6.3).
As a further comparison, TpABF43_36b released 119 µg mL −1 arabinose after 24 h, and, thus, released slightly lower amounts of arabinose compared to MgABF51 and PoABF62 after 24 h incubation ( Figure 4).
More surprisingly, the new TpABF43_36b substantially released xylo-oligosaccharides (XOS) and arabinoxylo-oligosaccharides (AXOS), besides arabinose (Figures 4 and 5), while the other ABFs did not. All enzymes were produced in the same production host, and purified, hence, it was not expected that this oligomer release was due to impurities (i.e., other enzymes co-produced and ending up in the TpABF43_36b fraction used). To confirm this, proteomic analysis was carried out and no xylanase was identified (SI 5, Table S1). Nevertheless, in comparison to a well-characterized GH10 endoxylanase from Aspergillus aculeatus (AaXyn10), the xylanase-like activity of TpABF43_36b was only 0.3% compared to AaXyn10 (based on insoluble AZCL-xylan; results not further shown).
In summary, these findings demonstrated that the novel TpABF43_36b was mainly active as ABF-m2,3 with minor endoxylanase activity, albeit with a relatively low rate compared to the other ABFs tested and the AaXyn10 endoxylanase.

HiABF43_36a Cleaves α-(1→2 or 3)-Ara from Xyls Disubstituted on the Non-Reducing End and α-(1→2 and 3)-Ara Monosubstituted on the Second Xyl from the Non-Reducing End
The specificities of the GH43_36 ABFs were further studied with selected AXOS. The well-characterized HiABF43_36a (ABF-d3) revealed a so far unknown specificity in this study, highlighted by the release of both α-(1→2)-Ara and α-(1→3)-Ara from A2,3XX with 100% conversion (Figure 6). In this oligomer, the disubstitution is located on the non-reducing terminal end, which might have resulted in the release of either arabinoses.   The specificities of the GH43_36 ABFs were further studied with selected AXOS. The well-characterized HiABF43_36a (ABF-d3) revealed a so far unknown specificity in this study, highlighted by the release of both α-(1→2)-Ara and α-(1→3)-Ara from A 2,3 XX with 100% conversion ( Figure 6). In this oligomer, the disubstitution is located on the non-reducing terminal end, which might have resulted in the release of either arabinoses.

HiABF43_36a Cleaves α-(1→2 or 3)-Ara from Xyls Disubstituted on the Non-Reducing End and α-(1→2 and 3)-Ara Monosubstituted on the Second Xyl from the Non-Reducing End
The specificities of the GH43_36 ABFs were further studied with selected AXOS. The well-characterized HiABF43_36a (ABF-d3) revealed a so far unknown specificity in this study, highlighted by the release of both α-(1→2)-Ara and α-(1→3)-Ara from A2,3XX with 100% conversion (Figure 6). In this oligomer, the disubstitution is located on the non-reducing terminal end, which might have resulted in the release of either arabinoses.  Incubation of HiABF43_36a with a mixture of XA 2 XX and XA 3 XX revealed another novel specificity, namely the cleavage of arabinose from monosubstituted AXOS (Figure 7). The ABF-m2,3 activity was unexpected and not in line with activity on AX (Section 2.3, Figure 3) and previous results from GH43_36a ABFs [8,12]. Based on our results, we propose that HiABF43_36a can cleave arabinose from mXyls specifically when monosubstituted on the second xylosyl from the non-reducing end. This ability might be the result of a structural similarity between AXOS monosubstituted on the second Xyl from the nonreducing end and AXOS disubstituted on the non-reducing end, as Xyl and Ara are spatially similar [27,28]. The same specific arabinose release was observed for HiABF43_36a on an endoxylanase digest from AX (produced by GH11 from Thermomyces lanuginosus; TlXyn11) that was pretreated with an excess of HiABF43_36a (ABF-d3) to remove disubstitutions. GH11 enzymes are endoxylanases that require three consecutive unsubstituted backbone units for activity and can only cleave the chain between two unsubstituted Xyls [25,29]. Hence, the resulting AXOS mainly have arabinosyl substituents on the second xylosyl from the non-reducing end (i.e., XAXX) [25,29]. Subsequently, the resulting AXOS were subjected to a second incubation with excess HiABF43_36a. This second incubation even released~40% arabinose compared to the total arabinose still present in these AXOS, which were basically devoid of dXyls, pointing at the relevance of this, so far, unnoticed side activity. Total arabinose release was achieved from these AXOS by MgABF51 ABF-m2,3 (SI 6, Figure S5). Incubation of HiABF43_36a with a mixture of XA2XX and XA3XX revealed another novel specificity, namely the cleavage of arabinose from monosubstituted AXOS ( Figure  7). The ABF-m2,3 activity was unexpected and not in line with activity on AX (Section 2.3, Figure 3) and previous results from GH43_36a ABFs [8,12]. Based on our results, we propose that HiABF43_36a can cleave arabinose from mXyls specifically when monosubstituted on the second xylosyl from the non-reducing end. This ability might be the result of a structural similarity between AXOS monosubstituted on the second Xyl from the nonreducing end and AXOS disubstituted on the non-reducing end, as Xyl and Ara are spatially similar [27,28]. The same specific arabinose release was observed for HiABF43_36a on an endoxylanase digest from AX (produced by GH11 from Thermomyces lanuginosus; TlXyn11) that was pretreated with an excess of HiABF43_36a (ABF-d3) to remove disubstitutions. GH11 enzymes are endoxylanases that require three consecutive unsubstituted backbone units for activity and can only cleave the chain between two unsubstituted Xyls [25,29]. Hence, the resulting AXOS mainly have arabinosyl substituents on the second xylosyl from the non-reducing end (i.e., XAXX) [25,29]. Subsequently, the resulting AXOS were subjected to a second incubation with excess HiABF43_36a. This second incubation even released ~40% arabinose compared to the total arabinose still present in these AXOS, which were basically devoid of dXyls, pointing at the relevance of this, so far, unnoticed side activity. Total arabinose release was achieved from these AXOS by MgABF51 ABF-m2,3 (SI 6, Figure S5).

Rate and Synergy of the GH43_36, GH51, and GH62 Enzymes used in this Study
2.6.1. Rate of TpABF43_36b, HiABF43_36, MgABF51, and PoABF62 Enzymes on AX Commonly, ABF (GH51) rates are determined with p-nitrophenyl α-L-arabinofuranoside, but GH62 and GH43_36 ABFs, generally, do not show activity on such artificial substrates [8,10,30]. Therefore, reaction rates of TpABF43_36b, HiABF43_36a, MgABF51, and PoABF62 acting on AX were compared by reducing end groups released with increasing enzyme concentration (Figure 8). MgABF51 and PoABF62 showed the fastest rates of all ABFs tested, followed by HiABF43_36a.
The novel TpABF43_36b showed an even lower reaction rate compared to the Hi-ABF43_36a and, apparently, did not reach maximum activity after 1 h incubation ( Figure  8). In hopes of finding a more suitable substrate for TpABF43_36b, its activity on arabinan, 2.6. Rate and Synergy of the GH43_36, GH51, and GH62 Enzymes used in this Study 2.6.1. Rate of TpABF43_36b, HiABF43_36, MgABF51, and PoABF62 Enzymes on AX Commonly, ABF (GH51) rates are determined with p-nitrophenyl α-L-arabinofuranoside, but GH62 and GH43_36 ABFs, generally, do not show activity on such artificial substrates [8,10,30]. Therefore, reaction rates of TpABF43_36b, HiABF43_36a, MgABF51, and PoABF62 acting on AX were compared by reducing end groups released with increasing enzyme concentration (Figure 8). MgABF51 and PoABF62 showed the fastest rates of all ABFs tested, followed by HiABF43_36a. rhamnogalacturonan, and insoluble corn fiber was investigated, however, no arabinose was released.

Rate of TpABF43_36b, HiABF43_36, MgABF51, and PoABF62 on Specific AXOS
The rates and specificities of TpABF43_36b, HiABF43_36a, PoABF62, and MgABF51 enzymes on specific AXOS were examined by incubation with A2,3XX, A2XX, and a mixture of XA2XX and XA3XX (Figures 9-11, respectively).  The novel TpABF43_36b showed an even lower reaction rate compared to the HiABF43_36a and, apparently, did not reach maximum activity after 1 h incubation (Figure 8). In hopes of finding a more suitable substrate for TpABF43_36b, its activity on arabinan, rhamnogalacturonan, and insoluble corn fiber was investigated, however, no arabinose was released.

Rate of TpABF43_36b, HiABF43_36, MgABF51, and PoABF62 on Specific AXOS
The rates and specificities of TpABF43_36b, HiABF43_36a, PoABF62, and MgABF51 enzymes on specific AXOS were examined by incubation with A 2,3 XX, A 2 XX, and a mixture of XA 2 XX and XA 3 XX (Figures 9-11, respectively). The rates and specificities of TpABF43_36b, HiABF43_36a, PoABF62, and MgAB enzymes on specific AXOS were examined by incubation with A2,3XX, A2XX, and a m ture of XA2XX and XA3XX (Figures 9-11, respectively).    The PoABF62 and MgABF51 were highly active on all monosubstituted AXOS (Figures 10 and 11), which was expected as they act on mXyls from AX and AXOS as their preferred substrate [8,11,31]. Nevertheless, MgABF51 showed significant activity with the full conversion of A2,3XX to arabinose and xylotriose at 1 μg mL −1 (Figure 9). The latter is   The PoABF62 and MgABF51 were highly active on all monosubstituted AXOS (Figures 10 and 11), which was expected as they act on mXyls from AX and AXOS as their preferred substrate [8,11,31]. Nevertheless, MgABF51 showed significant activity with the full conversion of A2,3XX to arabinose and xylotriose at 1 μg mL −1 (Figure 9). The latter is in line with previous research, as for some GH51 ABFs it has been reported that they cleave arabinose from dXyls exclusively when substituted on the non-reducing end [11]. The PoABF62 and MgABF51 were highly active on all monosubstituted AXOS (Figures 10 and 11), which was expected as they act on mXyls from AX and AXOS as their preferred substrate [8,11,31]. Nevertheless, MgABF51 showed significant activity with the full conversion of A 2,3 XX to arabinose and xylotriose at 1 µg mL −1 (Figure 9). The latter is in line with previous research, as for some GH51 ABFs it has been reported that they cleave arabinose from dXyls exclusively when substituted on the non-reducing end [11]. The MgABF51 required at least 25 times more protein loading to achieve 100% conversion of A 2,3 XX compared to A 2 XX and XA 2 XX/XA 3 XX, indicating that its action towards dXyls on the non-reducing end can be considered as a side activity. In addition, no remaining A 2 XX or A 3 XX were detected from the product profile of the A 2,3 XX digest (Figure 9), as these compounds were immediately hydrolyzed to arabinose and xylotriose due to high ABF-m2,3 preference. Unexpectedly, PoABF62 converted a small amount of disubstituted compound to arabinose and xylotriose (Figure 9), but only at high enzyme concentration.
As shown above (Section 2.5), HiABF43_36a had mainly ABF-d3 activity, but could also cleave to either α-(1→2)-Ara or α-(1→3)-Ara from AXOS disubstituted on the nonreducing end ( Figure 9) and could cleave arabinose when monosubstituted on the second xylosyl from the non-reducing end ( Figure 11). Furthermore, HiABF43_36a showed 100% arabinose release from A 2,3 XX at the lowest enzyme concentration tested (Figure 9). This high reaction rate of HiABF43_36a towards A 2,3 XX suggested that AXOS disubstituted on the non-reducing end could be its preferred substrate over AX. Calculations were made to compare the reaction rate of HiABF43_36a on A 2,3 XX and AX. The HiABF43_36a was almost twice as fast when acting on dXyls from A 2,3 XX compared to AX (SI 7). The calculation was performed by extrapolation of the reaction conditions, and, thus, is a rough estimate. Nevertheless, this calculation suggested that HiABF43_36a prefers either AXOS in general or specifically AXOS disubstituted on the non-reducing end over AX.
In line with our previous results, TpABF43_36b was active towards A 2 XX, XA 2 XX, and XA 3 XX, but only with low rates compared to MgABF51 and PoABF62 enzymes, yet with a higher rate compared to HiABF43_36a (Figures 10 and 11). In addition, TpABF43_36b was more active on XA 2 XX compared to XA 3 XX (Figure 11), suggesting that its preferred substrate might contain more O-2 substituted Xyls. Nevertheless, after cleaving arabinosyl from the monosubstituted AXOS, the formed xylotetraose could be further hydrolyzed to mainly xylobiose, which confirms the endoxylanase activity already observed towards AX by TpABF43_36b (Figure 4, Section 2.4). Furthermore, minor xylosidase activity was also observed by the release of minor amounts of xylose from either xylotriose or xylotetraose, which is also in line with previous activity by TpABF43_36b on AX (Figure 4, Section 2.4). It seems that TpABF43_36b prefers arabinofuranosidase activity over xylanase activity, or xylanase activity is hindered by substituents since no substituted arabinoxylobiose was formed from XA 2 XX and XA 3 XX (Figure 11). The TpABF43_36b was not active on A 2,3 XX ( Figure 9).

Synergy of TpABF43_36b, HiABF43_36, MgABF51, and PoABF62 on AX
The synergy between TpABF43_36b, HiABF43_36, MgABF51, and PoABF62 on AX was analyzed by reducing end groups released from the hydrolysis of AX. The synergy between HiABF43_36a with MgABF51 or PoABF62 enzymes was expected as the action of HiABF43_36a generates more mXyls for MgABF51 and PoABF62 to act on [8,11,12,[31][32][33]. Indeed,~54% more arabinose was released by their combination ( Figure 12A). Synergy experiments were also performed for TpABF43_36b with HiABF43_36a and MgABF51 ( Figure 12B). TpABF43_36b performed as expected based on our previous results on AX, and acted in synergy with HiABF43_36a, nevertheless with a relatively low reaction rate compared to HiABF43_36a and MgABF51. The synergy of TpABF43_36b with HiABF43_36a resulted in~56% more arabinose released (SI 8, Figure S6), whereas no synergy was observed between MgABF51 and TpABF43_36b.

Discussion
Previously, only two fungal GH43_36 arabinofuranosidases (ABFs) have been characterized that are exclusively active towards α-(1→3)-arabinose from disubstituted xylosyl (dXyl), one from Humicola insolens (HiABF43_36a) [8,12] and the other from Chrysosporium lucknowense (Abn7) [15]. In the current study, phylogenetic analysis of available sequences within the GH43_36 ABF subfamily of Ascomycetes showed a further subdivision into two main clusters, where cluster 1 consisted of major clade A and minor clades C, D, and E, and cluster 2 which consisted of the second major clade B. A functional activity difference was expected between the two clusters based on the topology of their active site. Most notably, GH43_36 members from clade A (GH43_36a) contain an aspartate (D291) in the side-pocket of their active site, which is essential for the aforementioned specificity towards dXyls [12], whereas GH43_36 members from clade B (GH43_36b) contain a tryptophan at equivalent position (W291). Nevertheless, a less conserved active site topology was found for the two corresponding clades in the Basidiomycetes GH43_36 phylogeny.
In this research, the product profiles from AX and specific AXOS of a novel fungal GH43_36 enzyme that aligns with clade B (TpABF43_36b) were examined and compared to previously well-characterized GH51 (MgABF51), GH62 (PoABF62), and clade A GH43_36 (HiABF43_36a) ABFs. An overview of specificities and reaction rates is shown in Table 2. Table 2. Comparison of substrate selectivity and rate of arabinofuranosidase (ABF) and xylanase activity on AX and AXOS by HiABF43_36a, TpABF43_36b, MgABF51, and PoABF62. +++ Excellent rate (used as reference), ++ High rate, + Intermediate rate, +/− Low rate,-trace activity, --no activity.

Discussion
Previously, only two fungal GH43_36 arabinofuranosidases (ABFs) have been characterized that are exclusively active towards α-(1→3)-arabinose from disubstituted xylosyl (dXyl), one from Humicola insolens (HiABF43_36a) [8,12] and the other from Chrysosporium lucknowense (Abn7) [15]. In the current study, phylogenetic analysis of available sequences within the GH43_36 ABF subfamily of Ascomycetes showed a further subdivision into two main clusters, where cluster 1 consisted of major clade A and minor clades C, D, and E, and cluster 2 which consisted of the second major clade B. A functional activity difference was expected between the two clusters based on the topology of their active site. Most notably, GH43_36 members from clade A (GH43_36a) contain an aspartate (D291) in the side-pocket of their active site, which is essential for the aforementioned specificity towards dXyls [12], whereas GH43_36 members from clade B (GH43_36b) contain a tryptophan at equivalent position (W291). Nevertheless, a less conserved active site topology was found for the two corresponding clades in the Basidiomycetes GH43_36 phylogeny.
In this research, the product profiles from AX and specific AXOS of a novel fungal GH43_36 enzyme that aligns with clade B (TpABF43_36b) were examined and compared to previously well-characterized GH51 (MgABF51), GH62 (PoABF62), and clade A GH43_36 (HiABF43_36a) ABFs. An overview of specificities and reaction rates is shown in Table 2. The HiABF43_36a was active towards dXyls as its preferred substrate (Table 2). Activity towards disubstituted xylosyl was expected as GH43_36a enzymes contain two pockets in the active site, in which the secondary pocket interacts with the α-(1→2)-Ara to direct the α-(1→3)-Ara into the catalytic main-pocket [8,11,12]. This is different compared to so far characterized GH51 or GH62 ABFs, which only have a single pocket allowing for the entry of one arabinosyl substituent into the active site, thus hydrolyzing monosubstituted residues [8,[31][32][33]. As expected, based on the above-described mechanism, synergy for HiABF43_36a together with MgABF51 or PoABF62 resulted in 54% more arabinose released from AX. Accordingly, HiABF43_36a was highly active on A 2,3 XX and showed no activity on A 2 XX (Table 2). Surprisingly, HiABF43_36a generated both A 2 XX and A 3 XX from A 2,3 XX ( Figure 6), even though GH43_36a ABFs were thought to be exclusively active on α-(1→3)-Ara from disubstituted residues [8,12]. Most likely, selectivity towards α-(1→3)-Ara is lost for Xyl units disubstituted on the non-reducing terminal end. This selectivity is likely exclusive to fungal GH43_36a enzymes, as the bacterial GH43_10 from Bifidobacterium adolescentis only generated A 2 XX from A 2,3 XX [11]. Furthermore, the reaction rate of HiABF43_36a towards A 2,3 XX was almost double compared to the reaction rate on AX, suggesting that HiABF43_36a might act in synergy with xylanases. Another unexpected selectivity was demonstrated as HiABF43_36a hydrolyzed monosubstituted arabinose from XA 2 XX and XA 3 XX (Table 2). Based on our results, it was speculated that HiABF43_36a can hydrolyze AXOS monosubstituted on the second xylosyl from the non-reducing end as they are structurally similar to AXOS disubstituted on the non-reducing terminal end (i.e., when the non-reducing end Xyl is seen as one of the Aras on A 2,3 XX). Nevertheless, reaction rates on XA 2 XX and XA 3 XX were lower compared to A 2,3 XX and AX, indicating that activity towards these monosubstituted AXOS is a side activity. Furthermore, the bacterial GH43_10 from Bifidobacterium adolescentis was not able to hydrolyze arabinose from XA 2 XX [11] and the GH43_36a from Chrysosporium lucknowense showed no activity on GH10 endoxylanase digest of AX pretreated with ABF-d3 [15]. Therefore, the ability to hydrolyze AXOS monosubstituted on the second xylosyl from the non-reducing end is likely not a common functionality among ABF-d3 enzymes. Surprisingly, the newly discovered specificities of HiABF43_36a towards A 2,3 XX, XA 2 XX, and XA 3 XX were different compared to the observed specificities of HiABF43_36a by Sørensen et al. [8], where HiABF43_36a's action on AXOS produced by GH10 or GH11 endoxylanases only released α-(1→3)-Ara.
Unlike previously characterized ABF-d3 GH43_36a enzymes [8,12,15], TpABF43_36b cleaved arabinose exclusively from monosubstituted residues and showed minor endoxylanase side activity ( Table 2). Loss of ABF-d3 activity was expected for GH43_36b enzymes compared to GH43_36a, based on the topology of their active site. GH43_36b enzymes are structurally similar to GH43_36a, yet they have a tryptophan, whereas GH43_36a have an aspartate at an equivalent position (D291) which is essential for activity towards dXyls [12]. In fact, synergy was observed between TpABF43_36b and HiABF43_36a which resulted in 56% more arabinose released from AX. This, along with 1 H-NMR results from AX, confirmed that TpABF43_36b was active as ABF-m2,3. Besides ABF activity, TpABF43_36b also possessed minor endoxylanase activity as a side activity. Endoxylanase activity was predicted to be a property of GH43_36 enzymes in previous research [12], where mutation of a tyrosine residue on the rim of the active site of HiABF43_36a opened its pocket and introduced endoxylanase activity while keeping its ability to act on disubstituted residues [12]. This suggested that GH43_36 enzymes already possess the catalytic apparatus for endoxylanase activity, which was confirmed here. Nevertheless, the presence of the tryptophan in the active site does not open the catalytic pocket of TpABF43_36b. Instead, it is expected that the xylanase activity of TpABF43_36b could be induced via stacking interactions with the xylan backbone, as TpABF43_36b contains two tryptophan residues in close proximity (Figure 2, Section 2.1), which is a feature also used by xylanbinding family 15 carbohydrate-binding modules [34]. Moreover, despite the relatively low reaction rate of TpABF43_36b on AX compared to other ABFs tested in this study, its inactivity on arabinan, rhamnogalacturonan, and insoluble corn fiber suggested that the β-(1→4)-linked xylan backbone with simple arabinosyl substituents (i.e., AX from wheat) is its preferred substrate.
The MgABF51 and PoABF62 showed the fastest rates out of all ABFs tested and released arabinosyl from both α-(1→2) and α-(1→3) monosubstituted Xyls from all substrates ( Table 2). The MgABF51 also hydrolyzed arabinosyl from the disubstituted A 2,3 XX, yet at a lower rate, indicating that this is a side activity. In literature, some GH51 ABFs have been determined to also hydrolyze disubstituted residues exclusively when substituted on the terminal non-reducing end [11]. Overall, MgABF51 showed slightly higher rates on AXOS compared to PoABF62, yet they have similar activity on AX (Table 2).

Phylogenetic Analysis of GH43_36 from Ascomycetes
Fungal Ascomycete GH43_36 sequences were obtained from the publicly available genome and protein sources: EMBL, ENA, NCBI, JGI, and UniProt. Genome sequences for which proteomes were not available were subjected to assembly by SPAdes [35]. Gene calling was performed using GeneMark ES [36] or Augustus [37]. Complete and credible GH43_36 gene models were aligned using Clustal Omega multiple sequence alignment service version 1.2.4 [38]. Visualization of the phylogenetic tree was created with iTOL as a rooted tree without outgroup and leaf sorting [39]. Table 3 shows the incubation conditions of the different experiments performed by number, denoted as # in the figure captions. Almost all enzyme incubations with arabinoxylan (AX) and arabinoxylo-oligosaccharides (AXOS) were performed in single determination, except the incubation with only HiABF43_36a in exp #3, which was performed in duplicates. Experiment #1 was performed in 2 mL Eppendorf tubes, whereas all other experiments were performed in 96 well plates (Thermo Fisher Scientific, Waltham, MA, USA). All experiments were performed while shaking (600 rpm) in 10 mM NH 4 Ac buffer pH 5.0 at 40 • C. Incubations varied in enzyme type and concentration, substrate type and concentration, incubation time, and analysis method (Table 3). AXOS (1 mg mL −1 ) and AX, AB, and RG stocks (2 mg mL −1 ) were prepared in 10 mM NH 4 Ac buffer pH 5.0, in which AX and RG were dissolved by heating to~75 • C. Insoluble corn fiber was added to the incubation as suspension (30 mg mL −1 ) while stirring. All incubations included a blank of pure MilliQ (not incubated), enzyme blanks (only of the highest protein concentration), and substrate blanks. Samples from experiment #1 were heat inactivated at 99 • C for 15 min after incubation before being subjected to lyophilization and subsequent 1 H-NMR (Section 4.2.3). Samples subjected to HPLC (Section 4.2.4) and MALDI-TOF MS (Section 4.2.5) from experiment #1 were filtered in 0.5 mL 10 kDa centrifugal filter units (Merck, Darmstadt, Germany) by centrifugation (14,500× g rpm, 20 min), whereas incubations from experiment #2 to 6 were filtered in 10 kDa 96 well filter plates (Pall corporation, New York, USA) by centrifugation (1500× g, 45 min). Analysis with PAHBAH from experiments #1 to 4 followed immediately after incubation and were not inactivated or filtered, whereas samples from experiments #5 and 6 were filtered in 10 kDa 96 well filter plates (Pall corporation, New York, NY, USA) by centrifugation (1500× g, 45 min). Samples were stored and frozen before analysis if not analyzed immediately.

Production of AXOS Monosubstituted on the Second Xylosyl from the Non-Reducing End
AXOS monosubstituted on the second xylosyl from the non-reducing end were prepared in three steps. All incubations were performed in 10 mM NH 4 Ac buffer pH 5.0. First, AX (1 mg mL −1 ) was incubated with excess HiABF43_36a (110.5 µg mL −1 ) for 22 h at 40 • C to remove disubstituted arabinosyl substituents. This was followed by enzyme inactivation at 99 • C for 15 min. Secondly, the monosubstituted AX was dialyzed in 3.5 MWCO dialysis discs (Thermo Fisher Scientific, Waltham, MA, USA) against MilliQ water for 24 h and 2 h while stirring to remove arabinose. Thereafter, two 75 µL samples were taken before and after dialysis as a control and analyzed with acid hydrolysis followed by HPLC. Furthermore, a 100 µL sample was taken after dialysis and incubated with excess HiABF43_36a (100 µg mL −1 ) for 24 h at 40 • C as control. Lastly, the dialyzed monosubstituted AX was incubated with excess TlXyn11 (50 µg mL −1 ) for 22 h at 40 • C followed by enzyme inactivation at 99 • C for 15 min. Subsequently, the generated mix of monosubstituted AXOS was incubated with HiABF43_36a (0 to 50 µg mL −1 ) and MgABF51 (0 to 10 µg mL −1 ) in single determination in 96 well plates (Thermo Fisher Scientific, Waltham, MA, USA) while shaking (600 rpm) with a total volume of 250 µL at 40 • C for 24 h. Incubated samples were analyzed with PAHBAH (Section 4.2.6) and HPAEC-PAD (Section 4.2.4).

Proteomics: Protein ID
Tryptic digests were prepared by a filter-aided sample preparation (FASP) method. Following digestion, the extracted peptides were analyzed on a nano LC-MS/MS system: Evosep One (Evosep)/timsTOF Pro (Bruker Daltonics GmbH, Bremen, Germany). For protein identification, the data were searched against available internal and public databases using the Mascot search engine (Matrix science, Boston, MA, USA) using Genedata Expressionist software with a 1% False Discovery Rate cutoff. Relative protein concentrations were calculated by label-free quantification from peptide volumes using a Hi3 standard method in Genedata Expressionist.

Conclusions
Extensive phylogenetic analysis of the GH43_36 subfamily within the Ascomycota phylum revealed the presence of two main clusters. Cluster 1 comprised of the major clade A (GH43_36a) containing the, so far, only two characterized GH43_36 arabinofuranosidases (ABFs) and three minor clades. These two GH43_36a ABFs both cleave α-(1→3)-arabinosyl from disubstituted xylosyl residues (Xyl) (ABF-d3). Cluster 2 was defined by major clade B (GH43_36b), and, at present, only contains uncharacterized sequences. Amino acid sequence alignment from a subset of GH43_36 sequences showed a highly conserved aspartate/tryptophan partition between GH43_36a (D291) and GH43_36b (W291) at amino acid position 291, where D291 is essential for the ABF-d3 activity of the previously characterized clade A HiGH43_36 from H. insolens (HiABF43_36a). Therefore, the D/W291 partition was hypothesized to result in a functional activity difference, most notably in the loss of ABF-d3 activity for GH43_36b enzymes. A novel GH43_36b from Talaromyces pinophilus that clustered in clade B (TpABF43_36b) was heterologously produced and examined for its reaction products on AX and specific AXOS. Indeed, the previously characterized HiABF43_36a was active as ABF-d3, whereas the novel TpABF43_36b was exclusively active towards monosubstituted Xyls (ABF-m2,3), and, surprisingly, also showed minor endoxylanase side activity. Moreover, new activities were identified for the HiABF43_36a, as this ABF was also able to cleave α-(1→2 or 3)-Ara from AXOS disubstituted on the non-reducing end and showed ABF-m2,3 activity exclusively towards AXOS substituted on the second xylosyl from the non-reducing end.