## Main

Lignin is an irregular phenylpropanoid biopolymer and one of the three major components of lignocellulose, the composite material forming secondary cell walls in higher plants. Lignin can comprise up to one-third of the dry mass of plant cell walls, making it the second most abundant biopolymer in the terrestrial biosphere, after cellulose, and the most abundant aromatic polymer1,2. The aromatic groups in lignin impart essential properties to plant cell walls, including resistance to degradation, structural rigidity and hydrophobicity, properties that facilitate fluid transport, defence against pathogens and biomass accumulation3. Plant cells synthesize lignin from three different primary p-hydroxycinnamyl alcohols: p-coumaryl, coniferyl and sinapyl alcohol. These monomers polymerize through free-radical coupling mechanisms, giving rise to p-hydroxyphenyl (H), guaiacyl (G) and syringyl (S) subunits that are present in varying ratios in different lignins2. Stochastic polymerization processes combined with variations in S:G:H proportions contribute to heterogeneity, diverse bond types and varying degrees of branching in lignin structures2. Lignin’s high recalcitrance defines its biogeochemical role as a carbon sink and presents a significant challenge for biotechnologies seeking to sustainably produce commodity chemicals from lignocellulose4,5,6,7.

Current descriptions of biological depolymerization and modification of lignin focus on aerobic systems, and they are primarily associated with the fungal sub-kingdom of Dikarya6,8. Although some members of Dikarya, such as ascomycete yeasts, are facultative anaerobes, lignin-degrading organisms, such as white rots, thrive in the presence of molecular oxygen6,9. Characterized lignin-modifying enzymes are limited in diversity and reflect the aerobic nature of their hosts; most rely on oxygen-dependent mechanisms that are probably unavailable under anaerobic conditions6,9. Many of these mechanisms indirectly (non-enzymatically) depolymerize lignin through the generation of organic free radicals and are therefore designated as lignin-active enzymes rather than lignin enzymes6,9. The oxygen dependence of characterized lignin-active enzymes has led to the widely accepted view that biological lignin deconstruction cannot occur in anaerobic environments6. The described lignin-active enzymes are categorized as laccases, lignin peroxidases, manganese peroxidases, versatile peroxidases, dye-decolourizing peroxidases, other oxidases and β-etherases6,9. Aerobic bacteria also produce a subset of these enzymes, but no anaerobic organism, bacterial or otherwise, possesses any known lignin-active enzyme9,10.

Anaerobic microbial communities, such as herbivore gut microbiomes, rapidly process lignocellulose under anaerobic conditions, producing greenhouse gases on scales that directly impact the global climate11,12. However, the fate of lignin in anaerobic environments remains largely unknown1,13. At a minimum, herbivore gut communities, and other microbes decomposing lignocellulose anaerobically, must displace lignin from plant cell walls to access cellulose and hemicelluloses, the primary carbon sources for microbiome and animal metabolism. Some previous studies have collected indirect evidence of lignin modification by anaerobic bacteria, but these studies are unable to establish the natural occurrence of anaerobic lignin deconstruction because they interrogate changes to Kraft lignin, a lignin-derived extract that has already been modified from naturally occurring lignins14,15,16,17,18,19. Insights into whether and, if so, how anaerobic microbes remove native lignins to acquire carbohydrates will inform geochemical models by helping to define carbon remineralization processes in diverse environments in which lignocellulose is anaerobically deconstructed13.

Organisms from Neocallimastigomycetes are early-branching anaerobic fungi broadly distributed in herbivore digestive tracts that excel at degrading lignocellulose and occupy a keystone role in herbivore gut microbiomes20,21,22. Anaerobic fungi possess unique root-like morphologies and enzyme complexes, termed fungal cellulosomes, that help localize lignocellulose-active enzymes to target polymers23,24. In addition, their genomes encode more carbohydrate-active enzymes (CAZymes) than industrial enzyme-producing fungi such as Aspergillus and Trichoderma25,26. A considerable fraction (~60–75%) of open reading frames in all sequenced Neocallimastigomycetes genomes are not functionally annotated because of the density of repeats, high adenine–thymine content and significant divergence from their closest known relatives27,28. Although challenging in many contexts, the unannotated gene content of Neocallimastigomycetes also provides an excellent opportunity to discover novel enzymes, particularly novel lignocellulose-active enzymes, for advancing sustainable biotechnologies26,29,30.

Undescribed interactions between anaerobic fungi and lignin constitute a major knowledge gap as Neocallimastigomycetes must at least circumvent lignin to access cellulose and hemicelluloses, their primary carbon sources. Previous efforts to interrogate interactions between ruminal microbes and lignin suggested that lignin might be directly or indirectly affected by anaerobic fungi, as implied by the accumulation of lignin degradation products31,32,33,34 or a reduction in lignification for processed plant material35,36,37,38,39,40. However, in the absence of further, direct evidence of lignin deconstruction, the observed release of monoaromatics from lignocellulose was attributed to hemicellulosic sidechain cleavage, a process anaerobic fungi facilitate31,41. Advances in two-dimensional heteronuclear single-quantum coherence nuclear magnetic resonance (2D-HSQC-NMR) spectroscopy allow direct examination of native lignin bonds in whole cell wall materials before and after the action of organisms or enzymes42,43,44,45. Recently, 2D-HSQC-NMR helped describe delignification by white rot46 and brown rot47 fungi, identified a novel ligninase produced by Parascedosporium putredinis NO148 and suggested anti-lignin activity in consortia derived from termite microbiomes49. In this Article, we use a similar approach to determine that members of Neocallimastigomycetes deconstruct lignin during the degradation of plant biomass.

We demonstrate that Neocallimastigomycetes anaerobically deconstruct lignin. Compelling evidence is provided by using 2D-HSQC-NMR supported with gel-permeation chromatography (GPC) analysis for molecular mass, by using liquid chromatography-mass spectrometry (LC-MS) and by elucidating changes in the percentage composition of lignocellulose. Using RNA sequencing (RNA-seq), we further implicate a set of conserved anaerobic fungal genes, the products of which might be responsible for anaerobically deconstructing the constituent polymers of lignocellulose. The identified genes of interest are highly expressed in culture conditions containing complex lignocellulose, and we curate them using bioinformatic methods with predictive models to identify those most likely to encode novel lignocellulose-active enzymes.

## Comparing activity across carbon sources

The anaerobic fungus Neocallimastix californiae was cultivated on grass and wood lignocellulosic substrates (sorghum, switchgrass and poplar) with different lignin compositions. The metabolic activity of the fungus, the extent of lignocellulose solubilization, changes to lignocellulose composition and accumulation of lignocellulose degradation products in fungal growth media were measured (Fig. 1). Cultures of N. californiae produced fermentation products (Fig. 1a,b), depolymerized lignocellulosic biomass (Fig. 1c,d) and freed diverse aromatic monomers from lignocellulose (Fig. 1e). We measured no monoaromatics in the purified cellulose or cellobiose treatments indicating that the fungus did not produce detectable aromatic monomers from simple carbohydrates. The metabolic activity of the anaerobic fungus was comparable when grown on purified carbohydrates (cellulose and cellobiose) and grass (sorghum and switchgrass) lignocelluloses but was lower on poplar (Fig. 1a,b). Cultures grown on cellobiose exhibited an extended lag phase (Fig. 1a).

## Measuring lignocellulose deconstruction

Compositional analyses determined the relative cellulose, hemicellulose, lignin and ash fractions of lignocellulose before and after fungal growth, providing a glimpse into the extent of deconstruction for the various polymers (Fig. 1d). The amount of cellulose and hemicellulose decreased for all lignocellulose types examined, and the proportion of acid-soluble lignin (ASL) also fell in grass lignocelluloses (Fig. 1d). Conversely, acid-insoluble lignin (AIL) made up a greater proportion of the total lignocellulose mass after growth. We further observed similar patterns of cellulose, hemicellulose and ASL loss for a second anaerobic fungal isolate, Anaeromyces robustus, confirming that observations of ASL reduction were not limited to a single isolate or genus (Extended Data Fig. 1). Remarkably, A. robustus reduced both ASL and AIL when grown on poplar, and this action contrasts with all other lignocellulose–fungus pairings (Extended Data Fig. 1).

The liberation of monoaromatics from lignocellulose was also observed in cultures of four additional isolates within Neocallimastigomycetes (Fig. 2). Cultures fed with grass lignocelluloses accumulated more than 450 µM of total measured monoaromatics, and the dominant chemical species was p-coumaric acid (pCA). Fungi liberated more variable concentrations of monoaromatics from poplar (200–750 µM), and the predominant chemical species freed was p-hydroxybenzoic acid (pB). All isolates accumulated greater than 200 µM pCA from grass lignocellulose or pB from poplar, and all other concentrations of monoaromatics liberated were less than 10 µM (Fig. 2b). Diverse fungi freed similar rank-ordered profiles of monoaromatics from sorghum and switchgrass, and the top five compounds released were consistent. In order of descending concentration, these compounds were pCA, ferulic acid, vanillic acid (VA), protocatechuic acid (PA, 3,4-dihydroxybenzoic acid) and syringic acid. Poplar–fungus pairings resulted in a different profile of monoaromatics, namely pB, pCA, VA, PA and salicylic acid, in order of decreasing concentration. Some isolates, specifically A. robustus and Piromyces sp. E1M (Supplementary Text and Extended Data Fig. 2), liberated ten times more pB from poplar than Neocallimastix isolates and Caecomyces churrovis (Extended Data Table 1).

We next used 2D-HSQC-NMR to determine any modifications to lignin by characterizing lignocellulose spectroscopically before and after the activity of Neocallimastigomycetes. To ensure that any observed lignin-modifying activity was not restricted to a single isolate, strain or genus, we chose to collect data from two different fungal isolates: N. californiae and A. robustus. We found that both anaerobic fungal isolates deconstruct parts of lignin in diverse types of lignocellulose (Fig. 3, Extended Data Fig. 3 and Extended Data Table 2). Expected signatures such as hemicellulose re-modelling and the removal of pCA, ferulic acid and pB pendent units were apparent in the spectra of all feedstocks treated with anaerobic fungi. The lignin of uninoculated controls for sorghum, switchgrass and poplar exhibited varying S and G subunit content that shifted after incubation with the fungal isolates. The S:G ratios of the sorghum samples fell from 0.76 in the control sample to 0.68 when treated with N. californiae and 0.58 when treated with A. robustus (Fig. 3a–c). Similarly, both fungi reduced switchgrass S:G ratios from 0.64 in the uninoculated control to 0.46 after incubation with N. californiae and 0.34 after incubation with A. robustus (Fig. 3d–f). The selective removal of S subunits from sorghum, but not necessarily switchgrass and poplar lignins, coincides with a loss of β-aryl-ether (β–O–4-linked) units and phenylcoumarans (β–5). In sorghum, the almost complete depletion of these β–O–4s and β–5s provides strong evidence of extensive modifications to the lignin polymer. Selective removal of β-aryl-ether and phenylcoumaran units by fungi grown on switchgrass and poplar lignins is more limited than in sorghum, although altered distributions of these linkages are apparent. Changes to poplar contrasted with changes to grass lignocelluloses by resulting in higher S:G ratios in samples treated with N. californiae (1.27) and marginally lower with A. robustus (0.93) than in the uninoculated control (0.99) (Fig. 3g–i). It proved instructive to normalize the integration data on a methoxyl integral basis in addition to the S and G subunit (ΣSG) basis, even though this cannot clarify all features (Fig. 3 and Extended Data Table 2). Although methoxyl levels change with the S:G shifts, some alterations to S and G subunits appear to retain the methoxyl groups such that the S:G reductions and the losses of some lignin linkages (β–O–4 and β–5) in the grasses become more apparent when using this normalization.

Complementary GPC of lignin oligomers derived from the same lignocellulose fractions used to generate the NMR samples showed a loss of lignin oligomers under 3,500 Da after fungal modification but a relative increase of >3,500 Da lignin oligomers (Extended Data Fig. 4a–f). A subsequent GPC experiment, incubating N. californiae with alkaline lignin extract, also resulted in a reduction of molecular weight for lignin-derived oligomers, highlighted by the appearance of a new low-molecular-weight peak in the treated samples (Extended Data Fig. 4g).

## Identifying novel lignocellulose-active enzymes

To identify what proteins might facilitate unknown aspects of anaerobic fungal lignocellulose deconstruction, we matched the deconstruction measurements to anaerobic fungal genes, upregulated in the presence of lignocellulose versus a purified cellulose control, using RNA-seq. The differential expression experiment examined N. californiae gene regulation on sorghum, switchgrass and poplar, comparing expression to control cultures growing on purified cellulose (Figs. 4 and 5, and Extended Data Fig. 5). Genes were considered of interest for this study if they were more expressed (q < 0.05)50 on sorghum, switchgrass and poplar. The most differentially regulated Kyoto Encyclopedia of Genes and Genomes (KEGG) categories51 meeting the specified conditions for functionally classified genes were metabolism, cellular processes and organismal systems (Fig. 4). Within the metabolism KEGG category, carbohydrate metabolism, energy metabolism and amino acid (AA) metabolism were the most impacted subcategories (Fig. 4).

The 25 most expressed and differentially upregulated CAZymes on each lignocellulose type were glycoside hydrolases (GHs) from various families. Six of the 25 most expressed, differentially upregulated CAZymes were shared across all three lignocellulose conditions, and all are annotated as diverse GHs; specifically, one representative was present from each of the GH9, GH1, GH3, GH27, GH32 and GH5 (subfamily one) classifications. None of the annotated carbohydrate esterases, enzymes associated with the cleavage of aromatic-hemicellulose ester linkages (CE1s), were differentially expressed in any condition. Of the 18 putative CE1s in the N. californiae genome, 11 had marginally increased transcript levels in all 3 lignocellulose-containing conditions, but none met the cut-off for statistically differential regulation.

Commonalities between the 25 most expressed and differentially regulated CAZymes in sorghum and switchgrass treatments distinguished these grass-induced CAZyme profiles from those of cultures grown on poplar. The transcriptomes from cultures grown on the two grass lignocelluloses shared most (19 out of 25) of their top 25 expressed CAZymes, but 13 of these 25 were not in the top 25 expressed CAZymes for the poplar treatment. All 13 genes prioritized in grass-induced transcriptomes but not poplar transcriptomes were various GHs (3 GH1s, 3 GH5s, 2 GH9s, 2 GH3s, 2 GH2s and 1 GH6). The occurrence of GH2s and GH6s and the presence of some carbohydrate-binding modules (CBMs), specifically fungal dockerin domains (CBM10s), were features of these grass-specific genes. The most upregulated and expressed genes in poplar transcriptomes were different, less diverse GHs. Ten of the top 25 expressed CAZymes specific to poplar were GH5s: 5 from subfamily 7, and 5 from subfamily 4. The other six uniquely poplar-associated genes were two glycoside transferases (GT8s), one polysaccharide lyase (PL1) and two GHs (GH26 and GH13) with predicted xylose binding modules (CBM35s).

We found 1,578 genes without functional classifications that had higher expression levels in the presence of lignocellulose versus the cellulose control and further curated this list with bioinformatic models (Fig. 5). Of the 1,578 identified genes without a KEGG classification, there were only 338 unannotated genes with both predicted signal peptides and increased expression levels across all three lignocellulose types (Fig. 5a). Of these, 29 had more than one transmembrane helix, 93 had exactly one transmembrane helix, whereas 216 had only a signal peptide and were hypothesized to represent extracellular and soluble proteins. Once filtered for a transcripts per million (TPM) cut-off of 10, only 179 out of 338 were identified as high-priority targets for characterization. Multiple sequences alignments completed on each subset of the 179 high-priority targets reveal some clusters of homologues within the gene sets (Extended Data Fig. 5)52. In addition, almost all predicted protein sequences for high-priority targets had partial homologues of >50% AA identity in one or more of the other sequenced anaerobic fungal genomes (Supplementary Table 1). In most cases, homologues were present across multiple genera of anaerobic fungi. Many of the predicted homologues from other genomes shared more than 90% AA identity, and most of these highly conserved peptide sequences were present across the entire diversity of available Neocallimastigomycetes genomes.

Domain predictions helped curate high-priority genes of interest by implicating some unclassified, upregulated gene products in lignocellulose degradation. Of the 179 high-priority targets identified in this experiment, 29 of them contained CAZyme domains predicted by the database for automated carbohydrate-active enzyme annotation (dbCAN; Supplementary Table 2)53. Among the CAZyme domains most identified by dbCAN were those of putative acetyl xylan esterases, ferulic acid esterases, pectin acetyl esterases and diverse xylanases. Predictions from a three-track neural network that accounts for primary sequences and also considers atomic coordinates in folded proteins54 complemented and enhanced dbCAN predictions by finding the same CAZyme domains in all test cases in which dbCAN provided predictions (Supplementary Table 2 and Extended Data Fig. 5). The neural network analysis additionally identified several putative expansin-like proteins that may be involved in swelling plant cell walls, as well as a xylanase that was not predicted by dbCAN. Activity assays using lignin dimeric β-aryl-ether model compounds48 suggested that the anti-β-ether unit activity we observe in NMR spectra might be sensitive to environmental redox potential and non-enzymatic (Supplementary Text and Extended Data Fig. 6).

## Discussion

The 2D-HSQC-NMR data presented here provide evidence for the deconstruction of unpretreated naturally occurring lignin by cultures of anaerobic fungi. In all lignocellulose–fungus pairings, the S:G ratio of lignin after fungal growth was different than before, but the nature of that difference seemed to depend mainly on the lignin type. Sorghum and switchgrass lignins predictably had lower initial S:G ratios than the S-subunit-enriched poplar lignin. For grass lignins, fungal activity preferentially removed S subunits in sorghum, reducing S:G ratios by between 0.08 and 0.18, but in the case of poplar, S:G ratios increased by 0.28 in the N. californiae treatment but stayed about the same upon treatment with A. robustus. The magnitude of observed changes in S:G ratios is comparable to those reported for consortia enriched from termite microbiomes when incubated for 3 weeks with wheat straw49 and a brown rot fungus grown on aspen wood for 16 weeks47. Changes in S:G ratios are associated with the re-modelling of β-aryl-ether units (β–O–4 bonds) and phenylcoumarans (β–5 bonds) in the spectra we acquired. These shifts in S:G ratios are often also associated with disruption of resinols (β–β) and dibenzodioxocin (5–5/4–O–β bonds), but these units were not readily observed here. This activity against diverse bond types contrasts with recent findings in which activities were restricted to β-aryl-ether units, including those to tricin pendant units48,49. Non-specific activity against varied bond types suggests that indirect, small-molecule-mediated reactions could be responsible for some of the observed lignin bond cleavage. We hypothesize this action to be analogous to the way oxidative depolymerization processes in higher fungi operate non-enzymatically through the formation of aromatic acid radicals6.

The observations reported here can most accurately be summarized as lignin re-modelling accompanied by some lignin depolymerization, especially in the sorghum spectra. This extent of lignin deconstruction seems consistent with lignin removal from lignocellulose by anaerobic fungi to gain access to carbohydrate polymers. The most striking changes to lignin–lignin linkages we observed are in sorghum. In samples of sorghum, we observed complete elimination of phenylcoumaran signals and the drastic reductions of resonances corresponding to β-aryl-ether units after activity by both N. californiae and A. robustus. We also observed decreases of phenylcoumarans in all treated poplar and switchgrass samples, and the reduction of these signals in poplar samples was especially conspicuous. In both switchgrass samples and one poplar sample, β-aryl-ether units decreased in relative abundance after treatment. The observed changes extend to the core lignin polymer in the case of sorghum, but lignin deconstruction might be more dominated by phenolate (hydroxycinnamate and hydroxybenzoate) pendent groups in the cases of switchgrass and poplar.

The observed loss of β–O–4 and β–5 units in NMR spectra might be responsible for the lignin fragmentation we observed in GPC traces. The GPC data derived from the same lignocellulose samples as NMR spectra indicated a loss of lignin oligomers in the small to medium size range but a marginal accumulation of high-molecular-weight lignin oligomers. Incubations of N. californiae with alkaline lignin extract also resulted in fragmentation of lignin-derived oligomers as measured by GPC. These GPC observations, combined with the 2D-HSQC-NMR data, are consistent with the fragmentation of lignin into smaller oligomers but could indicate subsequent recondensation of lignin fragments. The NMR data, supported by fragmentation observed in GPC traces, show that Neocallimastigomycetes act on lignin itself through unknown means, which contrasts with the previously accepted hypothesis that changes to lignin caused by anaerobic fungi were incidentally induced by the cleavage of lignin–polysaccharide linkages.

The LC-MS and compositional analyses we conducted relate our novel lignin deconstruction observations to published results, and our measurements of monoaromatic accumulation and ASL loss agree well with previous observations of lignocellulose dissolution by anaerobic fungi. We hypothesize that the monoaromatics we observe in solution are the combined result of hemicellulose and lignin pendent ester hydrolysis and, in addition, lignin degradation, given the extent of deconstruction observed in NMR spectra. Many previous publications, using LC-MS or biomass compositional metrics to interrogate activity against lignin, fell short of being able to convincingly claim observation of anaerobic lignin modification by Neocallimastigomycetes31,32,35,36,37,38,39,40 because these methods cannot directly show reorganization of lignin–lignin linkages. The monoaromatics we observe in solution after fungal growth are of similar concentrations to those in previous observations and indicate that the isolates used in this study probably participate in lignocellulose deconstruction processes comparable to those of unidentified strains from previous reports31,35. Likewise, compositional analyses of the proportions of various polymers in lignocellulose, performed before and after fungal growth, indicated that the fungi in this study delignify plant cell walls in a manner consistent with literature reports32,36,37,39,40.

Anaerobic fungi have a predictable preference for depolymerizing cellulose, followed by a secondary preference for hemicelluloses, but they also reduce the percentage of ASL. The extent to which anaerobic fungi impact these polymers is consistent with their established preference for hexose sugars over pentose sugars and the removal of lignin to access carbohydrates55. In contrast to ASL, AIL increased in all pairings except one, a possible indication of lack of access to AIL moieties, the relative increase in AIL due to removal of cellulose and hemicellulose, or even recondensation, a well-documented phenomenon in lignin degradation processes56. Observed increases in the relative abundance of AIL in digested cell walls, combined with the selective elimination of β–O–4 signatures from sorghum NMR spectra, offer indications that the changes to the lignin we observe cannot result from hemicellulose modification and subsequent solubilization of peripheral, pendant lignin fragments alone. The almost complete removal of β–O–4 bonds implies that this bond type is specifically affected by the processes we observed as the lignin left behind in digested cell walls is more abundant but observably deficient in this specific feature. The primary conclusions we draw from the LC-MS and lignocellulose compositional change data are that our observations are consistent with literature reports, and the lignocellulose deconstruction we observe is probably conserved across the Neocallimastigomycetes class. These data speak to phenomena (dissolution and depolymerization) that could follow the lignin deconstruction we observe, but they do not establish anaerobic lignin deconstruction independently from the NMR observations.

The observed anaerobic lignin disruption by Neocallimastigomycetes contrasts with described aerobic processes in both completeness of lignin deconstruction and rate, particularly in sorghum. The lignin deconstruction processes we attribute to anaerobic fungi occur more quickly than those of other biological lignin deconstruction processes but do not reach the extent of deconstruction achieved by white rot fungi6. Characterization of anti-lignin processes by aerobic fungi typically occurs on the order of months47, whereas consortia derived from termite microbiomes49 took 33% longer to achieve less extensive lignin modification than that achieved by N. californiae and A. robustus. The accelerated nature of anaerobic fungal lignin deconstruction is logical in the context of herbivore digestion. This is the first report of a rumen microbe deconstructing lignin, but ruminant microbiomes extract nutritional value from lignocellulose on short timescales with limited residence times, and so anaerobic fungi might reorganize lignin rapidly to gain access to carbohydrates32,38.

Some of the upregulated, unannotated gene products we identified might be responsible for the observed anti-lignin chemistry attributed to anaerobic fungi, although many biological lignin deconstruction processes are non-enzymatic. An analysis of functionally classified N. californiae genes revealed expected regulation patterns, such as differential regulation of genes associated with carbohydrate metabolism in response to lignocellulose availability20. Furthermore, observations of the extended lag phase in the cellobiose condition generally agree with previous descriptions of catabolite repression in anaerobic fungi57,58. As there are no detectable homologues of known lignin-active enzymes in anaerobic fungal genomes, we hypothesize that observed lignin bond cleavage could be facilitated by the products of some of the currently unannotated genes that make up 72% of detectable open reading frames in the N. californiae genome20,26. The differential expression experiments and subsequent bioinformatic filtering identified 216 soluble, secreted proteins, as well as the 93 single transmembrane helix proteins that constitute gene targets for heterologous expression and characterization. We included proteins with one predicted transmembrane helix in this set of genes as some Neocallimastigomycetes lignocellulose-active enzymes, especially cellulosome complexes, are thought to be anchored to cell membranes via a single, terminal transmembrane helix23.

Several features of genes identified in the differential expression analysis reinforce the notion of characterizing them as encoding novel lignocellulose-active enzymes. Most of the genes of interest we identified are conserved across all sequenced Neocallimastigomycetes genomes. The ubiquity of these genes within a class of organisms that specializes in lignocellulose deconstruction indirectly supports the hypothesis that these unannotated genes might be involved in lignocellulose depolymerization. Furthermore, some AA sequences derived from these genes contain predicted lignocellulose-active enzyme domains identified using hidden Markov models53 or neural networks54. Among the enzyme domains most frequently identified are several classes of enzymes involved in hemicellulosic and hemicellulosic-substituent modification, including ferulic acid esterases, acetyl xylan esterases, pectin acetyl esterases and various xylanases. The bias of predictions towards domains that act on non-cellulose moieties is logical as control conditions in the differential expression experiment contained purified cellulose but no xylan or lignin and might allude to the importance of non-carbohydrate bond modifications in lignocellulose depolymerization by anaerobic fungi. The prevalence of uncharacterized CAZymes in this gene set indicates that our experimental design was effective at inducing the expression of uncharacterized lignocellulose-active enzymes and that novel lignin-active enzymes might eventually be identified within the genes of interest.

The biochemical mechanism by which Neocallimastigomycetes achieve anaerobic lignin deconstruction might be challenging to characterize, as was the case for aerobic lignin-active enzymes. Several aerobic lignin-active enzymes have only been described after decades of concerted efforts revealed their complex redox chemistry and non-enzymatic nature6,9,59. The re-modelling of multiple bond types within lignin polymers, as presented in our 2D-HSQC-NMR dataset, could be an indication that free radicals or other small molecules generated by anaerobic fungal enzyme systems non-specifically cause bond scission in lignin. Furthermore, the only measurable activity against lignin-mimicking model compounds we observed occurred in the small-molecule fraction of fungal supernatants, possibly implicating non-protein mediators in lignin deconstruction (Supplementary Text). Currently, only tentative hypotheses exist to explain how anaerobic microbes might be able to generate lignin-degrading radicals in the absence of oxygen14,15,16,17,18,19, and testing these hypotheses in Neocallimastigomycetes cultures and with heterologously expressed Neocallimastigomycetes proteins will be essential for unravelling the biochemical mechanism of anaerobic lignin deconstruction.

We expect this discovery that certain anaerobic organisms have the ability to deconstruct unpretreated, naturally occurring lignin will change perceptions of how lignocellulose can be processed, not only in ruminant guts but also in anaerobic environments throughout the biosphere4,7,13. Although the changes reported here are somewhat limited in comparison to those induced by previously described lignin depolymerizing organisms, such as white rot fungi, they are of broad significance as they alter understandings of lignin–anaerobe interactions that are commonplace in diverse environments. Removal of lignin allows anaerobic microbes access to labile carbon in carbohydrate polymers that can be directly remineralized to greenhouse gases or diverted into other processes such as animal metabolism. Furthermore, decarbonization biotechnologies, such as lignocellulosic biorefineries, could benefit from the increased availability of diverse, biological lignin deconstruction mechanisms that operate under anaerobic conditions5,8.

## Methods

### N. californiae cultivation conditions

To explore the possibility that anaerobic fungi deconstruct lignin, we completed primary experiments with the robust anaerobic fungus N. californiae that we previously isolated20. We measured the activity of N. californiae cultures on three types of lignocellulose and two purified carbohydrates. Switchgrass (Panicum virgatum) was obtained from the laboratory of D. Putnam at the University of California, Davis, whereas sorghum (Sorghum bicolor) and poplar (Populus) were obtained from the laboratory of G. L. Gresham of Idaho National Laboratory. All three lignocelluloses were ground using a Thomas-Wiley Mini Mill (Model 3383-L10, A. H. Thomas), and fragments were size-selected through a 2 mm screen by a vibratory sieve system (Endecotts). We chose the specified lignocellulose types because of their diverse lignin contents and their importance as feedstocks for biorefineries and livestock. Controls included purified cellulose (catalogue number 09-805, Fisher Scientific) and cellobiose (catalogue number A1455322, Fisher Scientific). Purified cellulose was cut into small strips (~3 mm) in lieu of milling.

N. californiae cultures for all primary experiments were grown in a minimalist version of Medium C described previously22. We supplied growth substrates (various lignocelluloses, cellulose or cellobiose) at 10 g l−1 in 80 ml serum bottles closed with butyl stoppers and aluminium crimp seals. Culture vessels had 100% CO2 headspaces, and cultures were 40 ml in liquid volume. Haem (protoporphyrin IX) and a vitamin solution were 0.22 µm filtered and added to cultures before inoculation as described and at the concentrations described previously22,60. Uninoculated controls for each feedstock were included alongside cultures of N. californiae to ensure that growth media were aseptic. Experimental cultures were grown in six replicates with three replicates earmarked for RNA collection and three replicates earmarked for endpoint analysis of N. californiae liquid metabolites, continued monitoring of fungal growth over time, and endpoint analysis of changes to lignocellulose. Samples for RNA-seq were collected 93 h after inoculation by combining the entire culture volume with RNA later in a 1:1 ratio (v/v).

At the final time point of this experiment (336 h after inoculation), liquid samples for high-performance liquid chromatography (HPLC) plus LC-MS were separated from solids (feedstock and fungal biomass) by centrifuging cultures at 16,000 × g for 20 min. Solids were resuspended in 10 ml of autoclaved MilliQ water to wash off salts and fungi and then re-pelleted at 16,000 × g for 20 min. Nine millilitres of water was removed, and solids plus the remaining 1 ml MilliQ were lyophilized for 24 h in a FreeZone 4.5 Liter Benchtop Freeze Dry System (part number 77500200, Labconco). These washed and dried lignocellulose samples were retained to measure changes to lignin and percentage lignocellulose composition.

### Measurement of N. californiae metabolic activity with HPLC and total pressure accumulation

As a proxy for fungal growth, total fermentation gas accumulation was monitored daily using the pressure transducer technique (PTT) as described previously61,62. After measuring daily pressure accumulation, headspaces were vented to 6.9 kPa.

We further quantified fungal metabolites as a secondary metric of fungal activity using HPLC as previously described21,22. Samples were run on a 1260 Infinity HPLC (Agilent Technologies) equipped with an Aminex HPX-87H analytical column (part number 1250140, Bio-Rad). Separation conditions were 0.6 ml min−1, at 50 °C and used a 5 mM sulfuric acid (H2SO4) mobile phase. The refractive index detector was set to 35 °C, and the variable wavelength detector was set to 210 nm. Guard columns were a 0.22 µm physical filter, followed by a Coregel USP L-17 guard cartridge (Concise Separations). Compounds monitored were acetate, formate, succinate, ethanol, lactate, cellobiose and glucose. All HPLC standards of 0.1%, 0.05% and 0.01% (w/v) were prepared in minimalist Medium C to account for media background as this medium contains sterilized and clarified rumen fluid22. During analysis, blank media chromatograms were subtracted from all standard and experimental chromatograms using OpenLab CDS analysis software (version 2.6, Agilent Technologies).

Samples and standards for HPLC were prepared by acidification to a concentration of 5 mM H2SO4, incubated for 5 min at room temperature and spun at maximum speed in a tabletop centrifuge for 5 min to pellet fungal cells, proteins and other debris. The acidified samples were removed from above the pellet and 0.22 µm filtered through a polyethersulfone (PES) membrane into HPLC vials with 300 µl polypropylene inserts.

### Lignocellulose compositional analysis before and after N. californiae growth

We assayed relative lignocellulose composition before and after growth to determine whether N. californiae preferentially removed certain polymers from lignocellulose using the National Renewable Energy Laboratory’s two-step acid hydrolysis analytical protocols63. All compositional analyses were conducted in technical duplicates. Briefly, 200 mg of biomass and 2 ml of 72% H2SO4 were incubated at 30 °C while shaking at 200 rpm for 1 h. The resultant slurry was diluted to 4% H2SO4 with 56 ml of deionized water for secondary hydrolysis at 121 °C for 1 h. The reaction was quenched by cooling the flasks to room temperature before removing the solids by filtration. The filtrate was used to determine glucan, xylan and ASL composition, while the AIL and ash amounts were ascertained from the solid residue.

Glucose and xylose concentrations were determined from the filtrate using an Agilent HPLC 1260 Infinity system equipped with a Bio-Rad Aminex HPX-87H column and a refractive index detector at 35 °C. A solution of sulfuric acid (4 mM) was used as the mobile phase with a flow rate of 0.6 ml min−1 and a column temperature of 60 °C. The amount of cellulose (glucan) and hemicellulose (xylan) was calculated from the glucose and xylose content multiplied by the anhydro correction factors of 162/180 and 132/150, respectively.

ASL was estimated by measuring the ultraviolet absorption of the acid hydrolysis supernatant at 240 nm using a Nanodrop 2000 ultraviolet–visible spectrophotometer (Thermo Fisher Scientific). AIL was quantified gravimetrically from the remaining solids after heating samples overnight at 105 °C to obtain the weight of AIL and ash and subsequently repeating this procedure at 575 °C for at least 6 h to obtain the weight of ash alone.

### LC-MS sample preparation and run conditions for quantification of monoaromatics

Samples for LC-MS were prepared by filtering N. californiae supernatants through 0.22 µm PES membranes, then through 3,000 Da molecular weight cut-off, PES, centrifugal filter units (12,000 × g, 30 min, 20 °C). The filtrate from the 3,000 Da centrifugal filter was diluted with one-part HPLC grade methanol (v/v). Analytical standards of aromatic compounds, samples and uninoculated controls were analysed using an Agilent Technologies HPLC-ESI-TOF-MS64. Calibration curves were used for the absolute quantification of each analyte of interest. The theoretical mass-to-charge ratios of the deprotonated analytes were used to identify the aromatics of interest. The aromatics of interest were pCA (163.040068), ferulic acid (193.050632), pB (137.024418), vanillin (151.040068), VA (167.034982), syringic acid (197.045547), caffeic acid (179.034982), PA (153.019332), catechol (109.029503) and salicylic acid (137.024418).

### Cultivation of A. robustus for generation of additional lignocellulose samples

To collect corroborating lignocellulose samples representing the activity of a second fungal isolate, we repeated a limited version of the primary experiment previously described for N. californiae with A. robustus20. The cultivation conditions matched those described for N. californiae in the ‘N. californiae cultivation conditions’ section. Cultures were vented to 6.9 kPa every other day to avoid inhibition of fungal growth. After 14 days, solid samples of processed lignocellulose were collected and washed as described in the ‘N. californiae cultivation conditions’ section. These washed and dried lignocellulose samples were retained to measure changes to lignin. They were also used to measure changes to lignocellulose caused by fungal activity as described in the ‘Lignocellulose compositional analysis before and after N. californiae growth’ section.

### Extending LC-MS observations to four additional Neocallimastigomycetes isolates

To extend observations of monoaromatics released from lignocellulose to other anaerobic fungal isolates, we grew five different strains of anaerobic fungi, representing four genera, on sorghum, switchgrass and poplar. We grew N. californiae20, A. robustus20, Piromyces sp. E1M (Supplementary Text and Extended Data Fig. 2), Neocallimastix lanatii28 and C. churrovis65,66 with each of the lignocellulosic feedstocks (sorghum, switchgrass and poplar), in triplicate and using Medium C20,61. These growth conditions were required instead of those described in the ‘N. californiae cultivation conditions’ section as not all isolates grow well in the minimal version of Medium C that was used for the primary experiment and subsequent generation of A. robustus lignocellulose samples. Cultures were otherwise as described previously. Lignocelluloses were loaded at 1% (w/v), cultivation media was amended with haem and vitamins as previously described21,22,60, and culture growth was monitored with PTT followed by venting to 6.9 kPa61. Samples from this experiment were prepared and run as described in the ‘LC-MS sample preparation and run conditions for quantification of monoaromatics’ section.

### Isolation and classification methods for the Piromyces sp. E1M strain

Piromyces sp. E1M was isolated from the faeces of an Asian elephant at the Santa Barbara Zoo as previously described66,67. Briefly, faeces were collected and diluted in Medium C under anaerobic conditions in the presence of chloramphenicol, and lignocellulose was supplied as a carbon source. After dilution and subsequent growth, observed as pressure production accompanied by a clumped matt of lignocellulose, fungi were diluted in an anaerobic roll tube and allowed to grow for 4 days. A single colony representing a clonal fungus was picked from the wall of the roll tube in an anaerobic chamber and reinoculated into Medium C with chloramphenicol66. The roll tube process was repeated three times to obtain pure cultures, alternating between rolling out single colonies for picking and growing liquid cultures in the presence of chloramphenicol.

The novel isolate was classified as belonging to the genus Piromyces on the basis of its internal transcribed spacer region (ITS1) and large ribosomal subunit (LSU) gene sequences. The ITS1 region of the genomes and the ribosomal RNA gene of the fungus were amplified with primers JB206/JB205, and a region of the large subunit 28 S rDNA was amplified using primers NL1/NL468. Tentative taxonomy was assigned to the novel isolate using multiple sequence alignments conducted by the Ribosomal Database Project (RDP) Classifier69,70. The UNITE Fungal ITS database was used to classify the ITS1 sequence71, and the RDP Classifier Fungal 28 S database was used to classify the LSU sequence. The ITS and LSU sequences can be found in GenBank under BioProject accession number PRJNA800048.

### Whole plant cell wall 2D-HSQC-NMR sample preparation and data acquisition

Dried, washed lignocellulose samples for 2D-HSQC-NMR were collected from both N. californiae and A. robustus cultures. These lignocellulose samples were acted on by fungi, separated from liquid media and washed as described in the ‘N. californiae cultivation conditions’ and ‘Cultivation of A. robustus for generation of additional lignocellulose samples’ sections. Lignocellulose (~200 mg) was ground with an MM300 Mixer Mill (Qiagen) using 2-mm-diameter stainless steel balls and a mixing frequency of 20 s−1. Samples were ground for 45 min in the cases of sorghum and switchgrass samples or 80 min for poplar samples. These conditions are not the fine ball milling specified in the original methods, but grinding times were selected to generate material with particle sizes that are optimal for gelling, according to ref. 42. The ground samples were prepared for 2D-NMR experiments as described previously42,43. In a 5 mm NMR tube, ~120 mg of ground plant material was added along with 1 ml of pre-mixed DMSO-d6/pyridine-d5 solvent (4:1, v/v) to swell the lignocellulose and form a gel. The NMR tubes were sealed and sonicated for 4 h with a 30 min interval every hour until the gel became apparently homogeneous.

The 2D-HSQC-NMR spectra from lignocellulose processed by N. californiae or A. robustus were collected on a Bruker Avance I 800 MHz spectrometer equipped with a Bruker Triple Resonance Probe (TXI) at 310 K. A standard Bruker pulse sequence (hsqcetgpsisp2.2) was used with parameters that are typical for lignocellulose samples. Data were acquired using Bruker’s TopSpin software (version 4.1.0). The HSQC spectra were collected from 11 ppm to −1 ppm in the F2 (1H) dimension with 1,024 data points and 53 ms acquisition time with an interscan pulse delay of 1 s, and from 165 ppm to −10 ppm in the F1 (13C) dimension with 256 data points and 3.5 ms acquisition time. For each evolution period (t1) increment, 256 scans were recorded. The central DMSO solvent peak was used as a reference for the chemical shift calibration for all samples (δC 39.5 ppm, δH 2.5 ppm). All HSQC spectra were processed using typical cosine-squared apodization in both F2 and F1 dimensions, and the contours were integrated using MestreNOVA (version 14, Mestrelab Research). Peaks were assigned according to published data42,43,72,73.

Two sets of controls were included in this NMR experiment to account for the unlikely possibility that lignin was inadvertently deconstructed by autoclaving lignocellulose in fungal growth media. These unautoclaved controls were treated identically to the experimental samples by washing them with MilliQ water and then freeze-drying them. As the spectra of unautoclaved controls were similar to the no-inoculum controls, we include the no-inoculum controls in the main text comparisons to treated samples as this comparison is more direct. Extended Data Fig. 3 shows comparisons of the no-autoclave controls and no-inoculum controls.

### Cultivation of N. californiae with added alkaline lignin extract

To investigate whether anaerobic fungi affect the size of lignin-derived extractives, we grew the anaerobic fungus N. californiae in M2 media with both cellulose and alkaline lignin. All cultures received 0.22 µm, PES-filtered amendments of vitamins and haem, as previously described21,22. Alkaline lignin (catalogue number L0082, TCI America) was dissolved in MilliQ water and added to a final concentration of 2.5 g l−1. Cultures were 40 ml in liquid volume and were supplied 10 g l−1 cellulose (catalogue number 09-805, Fisher Scientific) as a primary carbon source. Experimental cultures were grown in triplicate, and an uninoculated control, with the same 0.22 µm filtered amendments, was included. At the cessation of growth, as measured by PTT61, bottles were collected, and solids were separated from supernatants by centrifugation (5,000 × g, 5 min). Supernatants were then freeze dried for 48 h in a FreeZone 4.5 Liter Benchtop Freeze Dry System (part number 77500200, Labconco). Alkaline lignin retrieved from these freeze-dried samples was then processed for GPC.

### GPC for lignin oligomer size

GPC was used to determine the relative molecular weight distribution of lignin in lignocellulose before and after treatment with N. californiae or A. robustus74. These lignocellulose samples were the same as those processed for NMR in the ‘Whole plant cell wall 2D-HSQC-NMR sample preparation and data acquisition’ section. Following standard protocols, 10 mg of lignocellulose was incubated in 2.5 ml of acetic acid and acetyl bromide (92:8) and stirred at 50 °C for 2 h to dissolve the lignin. Excess acetyl bromide and acetic acid were removed with a rotary evaporator connected to a high-vacuum pump and a cold trap. Acetylated lignin was immediately dissolved in tetrahydrofuran and filtered through 0.2 µm polytetrafluoroethylene filters. Samples of alkaline lignin from the experiments detailed in the ‘Cultivation of N. californiae with added alkaline lignin extract’ section were prepared for GPC analysis in the same way, but 10 mg of alkaline lignin was added to the acylation reaction instead of 10 mg of lignocellulose.

GPC analysis of both lignocellulose-derived lignin fragments and alkaline lignin fragments was conducted using an Ecosec HLC-8320GPC (Tosoh) equipped with Agilent Technologies PLgel 5 μm Mixed-D column and a Diode Array Detector. Tetrahydrofuran, spiked with 250 ppm of butylated hydroxytoluene, was used as the mobile phase with a flow rate of 1 ml min−1 and a column temperature of 40 °C. The GPC standards for oligomer size were polystyrene oligomers ranging from 162 g mol−1 to 29,150 g mol−1 (part number PL2013, Agilent Technologies).

### RNA extraction from N. californiae samples for differential expression analysis

To conduct differential expression analysis, samples of N. californiae for RNA, collected as described in ‘N. californiae cultivation conditions’ section, were lysed, and RNA was extracted. Samples were pelleted at 16,000 × g for 20 min to separate RNAlater and cultivation medium from solids, then solids were flash frozen in liquid nitrogen and stored at −80 °C until extraction. Fungal pellets were ground to a fine powder in liquid nitrogen using a mortar and pestle, and purification of RNA followed the Qiagen RNAeasy kit (part number 74004, Qiagen) protocol, including homogenization with a QIAshredder column (part number 79656, Qiagen) and an on-column DNAse digestion (part number 79254, Qiagen). The quality of extracted RNA was assessed using a TapeStation microfluidic electrophoresis device equipped with RNA tape (part number 5067–5576 and 5067–5577, Agilent Technologies), whereas RNA quantity was assessed using a Qubit fluorometer (Thermo Fisher Scientific) and the RNA broad range assay (part number Q10211, Thermo Fisher Scientific).

### RNA library construction, sequencing, read quality control and alignment of reads to the genome

Sequencing libraries were created and quantified at the Joint Genome Institute (JGI). Stranded complementary DNA libraries were generated using the Truseq Stranded mRNA Library Prep (catalogue number 20020595, Illumina). Messenger RNA was purified from 200 ng of total RNA using magnetic beads containing poly-T oligos (catalogue number 20020595, Illumina). Purified mRNA was then fragmented using divalent cations (catalogue number 20020595, Illumina) and incubation at 94 °C for 2 min. The fragmented RNA was reverse transcribed using random hexamers and Super Script II enzyme (catalogue number 18064-022, Thermo Fisher Scientific), followed by second-strand synthesis. The fragmented cDNA was treated with end-pair, A-tailing, adapter ligation and 10 cycles of PCR. The prepared libraries were then quantified using KAPA Illumina Library Quantification Kit (Roche) and run on a LightCycler 480 real-time PCR instrument (Roche). The quantified libraries were multiplexed, and the pool of libraries was prepared for sequencing on the Illumina NextSeq 500 sequencer using NextSeq 500 High-Output Sequencing kit v2 (Illumina) and following a 2 × 150 indexed run protocol. This sequencing run resulted in 15 sequence libraries totalling 396 million 2 × 150 reads before quality control (QC). Raw reads of differential expression data are deposited in NCBI’s Sequence Read Archive and can be found with NCBI Sequence Read Archive accession numbers SRP288871SRP288885.

We filtered and trimmed raw reads using a QC pipeline developed by JGI. Raw reads were assessed for artefact sequences with kmer matching, and detected artefacts were trimmed from the 3′ ends of the reads (kmer = 25, 1 mismatch allowed) using BBDuk (version 38.90)75. Reads matching RNA spike-ins, PhiX reads and reads containing any ambiguous nucleotides were removed. We performed quality trimming using the phred trimming method set to Q6. Finally, reads under a length threshold of 25 bases or that ended QC as less than one-third of the original read length were removed. This QC pipeline resulted in 235 million reads that were of high enough quality to align to the N. californiae reference genome.

We used raw gene counts to evaluate the level of correlation between biological replicates. Pearson’s correlations were calculated between replicates to determine which replicates could be used in the differential expression analysis. For all five treatments, all three biological triplicates were included in differential expression analyses as these replicates showed high correlation coefficients ranging from 0.89 to 1.00 (Extended Data Fig. 5).

Filtered reads from each library were aligned to the reference genome using HISAT2 version 2.1.076,77. Strand-specific coverage bigWig files were generated using deepTools version 3.178. FeatureCounts (version 2.0.0)79 was used to generate the raw gene counts file using an appropriate genome feature file (Neosp1_GeneCatalog_genes_20170918.gff). On average, 92.6% of quality-controlled reads in each sequence library were mapped to the reference genome. We only included primary hits assigned to the reverse strand in the raw gene counts.

### Differential expression analysis of N. californiae transcriptomes

We determined which genes were differentially expressed between the purified cellulose control condition and other culture conditions using DESeq2 (version 1.28.1)50. We used an adjusted P value of q < 0.05 as the statistical cut-off for labelling a gene as differentially upregulated or downregulated. Genes were included as genes of particular interest if they were upregulated on all three lignocellulosic substrates compared to purified cellulose, and they were not assigned a KEGG functional class.

Characterized and annotated genes were labelled with KEGG categories on the basis of assigned KEGG Orthology (KO) numbers retrieved from JGI’s Mycocosm. The total relevant transcripts for each KEGG category were determined by summing the total transcripts aligned to all differentially expressed genes within a given KEGG category for both the control condition and each treatment condition. We then calculated a log2 fold change for each KEGG category by log2 transforming the ratio obtained from dividing experimental transcripts by purified cellulose control transcripts. Gene counts for genes with multiple KO numbers, or KO numbers that reference multiple KEGG categories, were included in calculations for all categories they were classified in, but gene transcript counts were only included once per KEGG category even if they were mapped to a KEGG category more than once by multiple KO numbers. All parsing of differential expression data was conducted in Python 3.9.13, using tools from Biopython (version 1.79) and NumPy (version 1.23.1)80,81,82.

To describe the most drastically upregulated CAZymes, identified in each lignocellulose versus cellulose comparison, we selected the top 25 genes, determined by their mean TPM value for the treatment condition. These selected genes were differentially expressed in each condition and had been labelled with the KEGG category ‘Carbohydrate Metabolism’. We then retrieved protein sequences for these 576 genes and subjected them to the dbCAN2 classifier53 to assign CAZyme families and subfamilies to the CAZyme. After applying dbCAN2 annotations, the top 25 genes having the largest TPM allocations were compared to one another to determine similarities and differences in the most expressed CAZymes between conditions. Annotated CAZymes data were parsed in Python 3.0, using tools from Biopython and NumPy80,81,82.

### Using predictive bioinformatic tools to characterize genes of interest

Open reading frames in the N. californiae genome that are not functionally characterized and that were upregulated on all three lignocellulose types were further analysed using several predictive models. We used SignalP5.0 to determine whether genes of interest were likely to be trafficked to the exterior of the cell83. Likewise, we used the TOPCONS web server to predict whether gene products were likely membrane associated or instead were soluble proteins84. We also predicted CAZyme domains in genes of interest using the dbCAN web server53. All output data files were matched to differential expression data and metadata with Python 3.0, using both the Biopython toolkit and NumPy80,81,82. Further analyses were completed on genes that were considered high priority for characterization because they had TPM values greater than 10 TPM and q values less than 0.05, and we refer to these now as sequences of interest.

Sequences of interest were further submitted to the Rosetta server, which used RoseTTaFold to generate protein structure predictions54,85. This analysis helped to ensure confidence in dbCAN53 predictions, and so sequences with predicted CAZyme domains were analysed as positive controls. When RoseTTaFold predictions had confidence greater than 0.4 as computed by DeepAccNet, the model with the lowest average error estimate was searched for local structural similarity against the PDB100 database and AlphaFold Protein Structure Database (https://alphafold.ebi.ac.uk/) using FoldSeek (https://github.com/steineggerlab/foldseek) in 3 Di/AA mode. The lowest E value matches for bacterial and fungal proteins were compiled to suggest putative functions for each sequence of interest.

Multiple sequences alignments were completed to help identify sequences of interest which might have homologous domains and therefore could participate in similar functions. The evolutionary histories for three sets of genes of interest (secreted, soluble and membrane associated) were inferred using the neighbour-joining method86, and this analysis was replicated 100 times87. We calculated evolutionary distances using the Jones, Taylor, Thornton matrix-based method in units of the number of AA substitutions per site88 and removed all ambiguous positions for each sequence pair (pairwise deletion option). Evolutionary analyses were conducted in MEGA X52,89.

### Preparation of protein and small-molecule fractions for β-ether model assays

Fungal cultures of N. californiae and A. robustus were grown on sorghum in a minimalist version of Medium C22. Fungal supernatants were separated from fungal cells by anaerobically centrifuging whole cultures at 16,000 × g for 3 min. Supernatants were then decanted anaerobically and 0.22 µm filtered with a PES filter. Fungal cell pellets were either anaerobically or aerobically shaken on a vortexer for 20 min in sterilized tubes with sterilized 0.5 mm zirconia-silicate beads (part number 11079105Z, Biospec) and 500 µl of sterile phosphate buffer (100 mM, pH 7.0). Both aerobic and anaerobic lysates were then 0.22 µm filtered (PES) to remove any intact cells or large cellular debris. Using 10 kDa molecular weight cut-off centrifugal filtration devices (part number VS0291, Sartorius), fractions for both supernatants and lysates were buffer exchanged twice with anaerobic or aerobic phosphate buffer (100 mM, pH 7.0). Centrifugal filtrations were conducted both aerobically and anaerobically. The filtrates from the initial filtration, as well as the concentrated buffer, exchanged protein fraction from both cell lysate and supernatant, were retained.

To contrast the activity of fungal protein preparations in anaerobic and aerobic incubations, as described above, phosphate buffers (100 mM, pH 7.0) with and without oxygen were prepared. Aerobic phosphate buffer was prepared from MilliQ water and then stirred at room temperature for 1 week with a gas-permeable cover. Anaerobic phosphate buffer was prepared by boiling MilliQ water before passing that boiling water into an anaerobic chamber (atmosphere, 75% N2, 20% CO2, 5% H2) and dissolving phosphate salts after cooling. The anaerobic phosphate buffer was also stirred continuously for 1 week with a gas-permeable cover before being used in subsequent incubations. Before incubations, all phosphate buffers were 0.22 µm filtered. For small-molecule (<10 kDa) incubations, 200 µl of the desired phosphate buffer (aerobic or anaerobic) was added to each incubation.

### Activity assays for β-ether model compounds

We incubated β-ether model compounds with the fractions prepared in ‘Preparation of protein and small-molecule fractions for β–O–4 model assays’ section to reproduce the anti-lignin bond activity in vitro48. Each reaction was sampled five times before and after incubation, and fluorescence was read on a Tecan M200 plate reader (Tecan). For all assays, model compounds were supplied at 20 mM, and reaction buffers were phosphate buffers (100 mM, pH 7.0). Time-zero measurements were conducted immediately after adding model compounds to incubations, and time final measurements were taken after incubation of the model compounds in the reaction mixture for 24 h at 39 °C.

We added 50 µl of glycine buffer (pH 10.1, 100 mM) to 100 µl of reaction mixture to obtain fluorescence measurements. The mixture’s fluorescence was read in a black 96-well plate using an excitation wavelength of 372 nm and an emission wavelength of 445 nm, which correspond to the excitation and emission maxima for 4-methylumbelliferone.

$$\begin{array}{l}{\mathrm{Normalized}}\;{\mathrm{activity}}\\ = \displaystyle\frac{{\begin{array}{l}\left( T_{\mathrm{final}}\;{\mathrm{sample}}\;{\mathrm{fluorescence}}\;-\;T_{\mathrm{0}}\;{\mathrm{sample}}\;{\mathrm{fluorescence}} \right)\\ - \left({T_ {\mathrm{final}}}\; {\mathrm{blank}}\;{\mathrm{fluorescence}}\;-\; T_{0}\; {\mathrm{blank}}\;{\mathrm{fluorescence}} \right)\end{array}}}{{{{\upmu}}\,{\mathrm{g}}\;{\mathrm{of}}\;{\mathrm{protein}}\;{\mathrm{in}}\;{\mathrm{reaction}}}}\end{array}$$
(1)

The anti-probe activity in each incubation, in arbitrary fluorescence units, was calculated by subtracting the time-zero fluorescence measurements (T0 sample and blank fluorescence) from the time final measurements (Tfinal sample and blank fluorescence) and then dividing the change in fluorescence by the protein content of the incubation (μg of protein in reaction) (equation (1)). The changes in fluorescence for blank incubations, consisting of model compounds in phosphate buffer with no protein, were subtracted from all experimental assay values (equation (1)). Positive controls were tyrosinase from button mushroom (part number T3824, Sigma Aldrich), supplied at a concentration of ~35 µg purified protein per 1 ml reaction. The amount of protein in experimental treatments was determined using the Pierce Coomassie Plus (Bradford) Assay Reagent (part number 23238, Thermo Fisher Scientific). The uncertainties associated with time-zero means and the blank reaction means are propagated to the errors associated with the final normalized activity values, while protein concentrations are treated as constants.

### Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.