Richness of arbuscular mycorrhizal fungi increases with ecosystem degradation of temperate eucalypt woodlands

Purpose Degraded ecosystems can be maintained by abiotic and biotic legacies long after initial disturbances, preventing recovery. These legacies can include changes in arbuscular mycorrhizal fungi (AMF). To inform potential restoration pathways, we aimed to elucidate differences in AMF between intact and degraded ecosystems, their responses to modified soils, and interactions with invasive plants. Methods We used a state-and-transition framework to characterise AMF communities, native and exotic plant cover, and soil physicochemical properties across little-modified reference states and degraded states, which were carbon (C) and nitrogen (N) -depleted, intermediate, and CN-enriched, in temperate eucalypt woodlands of south-eastern Australia.


Introduction
Human livelihoods and agricultural productivity depend on global biodiversity conservation (Miller et al. 2017), with more than US$2 trillion per annum spent on the ecological restoration of degraded landscapes (Cunningham 2008). To inform practices aiming to restore biodiversity, we must first understand the abiotic and biotic limitations to ecosystem recovery towards reference states. The persistence of invasive species, altered propagule banks, and altered physicochemical soil properties are commonly identified as barriers to ecosystem recovery (Cuddington 2011;Corbin and D'Antonio 2012). Albeit understudied, altered soil microbial communities are increasingly recognised as potential contributors to the persistence of ecosystems modified from their reference state (hereafter 'degraded' in the sense of altered biodiversity and ecosystem processes) (Hawkes et al. 2005;Ba et al. 2018) and equally, to ecosystem restoration (Neuenkamp et al. 2019). This is true of arbuscular mycorrhizal fungi (Vogelsang and Bever 2009;Asmelash et al. 2016;Grove et al. 2017;Stover et al. 2018).
Arbuscular mycorrhizal fungi (AMF) contribute to ecosystem functioning above and belowground. They are best known for their capacity to enhance host uptake of phosphorus (P) and they can also improve water uptake (Smith and Read 2010), offer pathogen defence (Veresoglou and Rillig 2012), improve soil physicochemical properties (Willis et al. 2013), and contribute to nutrient cycling (Powell and Rillig 2018). Altered AMF communities caused by disturbances can also promote plant invaders, leading to further ecosystem degradation (Hilbig and Allen 2015;Ba et al. 2018). Restoration studies show that re-introduction of native AMF can promote the reestablishment of native flora and ecosystem recovery (Asmelash et al. 2016;Koziol et al. 2018). These findings suggest that key AMF taxa are missing from degraded sites, but this assumption is rarely tested despite its prevalence in restoration thinking (Maltz and Treseder 2015;Neuenkamp et al. 2019).
Ecosystem degradation can impose abiotic or biotic limitations on recovery (Suding et al. 2004). For example, land management practices like overgrazing and land clearing, can lead to ecosystem states depleted in soil physicochemical properties, that directly impair native vegetation productivity. It has been shown that these states are mainly depleted in carbon (C) and total nitrogen (N), as well as being compromised in other properties such as water holding capacity (Prober et al. 2002(Prober et al. , 2014McIntyre and Lavorel 2007). In contrast, atmospheric N deposition, disturbance of native vegetation, and land uses such as industrial agriculture can lead to states enriched in the same soil properties (Prober et al. 2002(Prober et al. , 2014. These enriched states promote invasive over native plants, while invasive plants help to maintain the altered soil physicochemical properties (Hawkes et al. 2005;Ba et al. 2018), often causing the competitive exclusion of native plants by invaders (Cuddington 2011). Given the relationship between soils, plants, and AMF, it is plausible that this turnover from native to exotic vegetation during ecosystem degradation could partially alter AMF communities (i.e., AMF community composition, abundance, and richness).
The role of and consequences for AMF of these depletion and enrichment dynamics are poorly characterised, but emerging evidence shows AMF communities respond to changes in soil nutrient availability and aboveground vegetation during ecosystem degradation (e.g., Phillips et al. 2019;Han et al. 2020;Albornoz et al. 2022). As noted above, the most common assumption for restoration is that AMF taxa have been lost during degradation, and hence re-introduction of those taxa can facilitate restoration. This could also imply reduced AMF richness with degradation. On the other hand, theory suggests a range of potential responses, depending on environmental factors, including changes in host nutrient limitations, and AMF life history traits (Chagnon et al. 2013;van der Heyde et al. 2017;Lilleskov et al. 2019).
Changes in AMF communities have been proposed to depend on soil P and N availability, with higher AMF abundance and biomass expected when P is more limiting than N (Johnson et al. 2013;Lilleskov et al. 2019). In such cases, greater benefit to the plant is expected from allocation of C to AMF for P-uptake, than from C allocation for N-uptake (e.g., through more extensive root systems). Plant C allocation can mediate both abundance and diversity of AMF, and thus it is possible that this hypothesised dependence on N vs P limitation could also apply to diversity. For example, if plants reduce their C allocation to AMF, competitive exclusion could eliminate rare species, decreasing AMF diversity (Liu et al. 2015;Knegt et al. 2016). If so, we would expect higher AMF diversity in N-enriched states, if N-enrichment leads to P becoming more limiting. Conversely, where N depletion results in N becoming the most limiting nutrient, we might predict plants will allocate C to N-uptake rather than to AMF, potentially reducing their diversity. However, the effects of N-enrichment and depletion on P-limited ecosystems is unresolved.
It has also been proposed that changes in AMF communities might reflect the life-history strategies of both fungus and host (competitor, stress-tolerator or ruderal, C-S-R; Grime 1979, Chagnon et al. 2013. Chagnon et al. (2013) proposed the classification of Gigasporaceae, Acaulosporaceae, and Glomeraceae, into competitor, stress-tolerator, and ruderal, based on functional traits, respectively. Competitor AMF have high extraradical hyphal production, high P transfer to host, and late production of spores. Stress-tolerator AMF have low growth rate, long-lived mycelium, and resistance to abiotic stressors. Ruderal AMF have high growth rates, early production of spores, and high hyphal turnover rates (Chagnon et al. 2013). Under this C-S-R framework, Chagnon et al. (2013) proposes a match between the life-history of host and their symbiotic AMF (e.g., ruderal plants associate with ruderal AMF). Invasive plant species are typically ruderal (Alexander et al. 2016), and can be found in great abundance in enriched states (Prober et al. 2002), while depleted states might favour stresstolerator plants due to extreme nutrient limitations. Hence, we expect a dominance of stress-tolerator AMF (i.e., Acaulosporaceae) in depleted states, compared with dominance by ruderal AMF (i.e., Glomeraceae) in weedy, enriched states. Whilst having plausible theoretical or practical foundations, these predictions lack empirical evidence. A better understanding of the effects of different types of degradation on AMF, linked to potential causal factors, is important if AMF are to be incorporated into restoration practices (e.g., Neuenkamp et al. 2019).
The temperate eucalypt woodlands of Australia are critically endangered due to broad scale clearing for agriculture and degradation of soils and remnant vegetation (Prober et al. 2002(Prober et al. , 2014. Recovery of these woodlands is limited by abiotic (e.g., nutrient depletion and enrichment) and biotic (e.g., plant invasions) factors depending on their degradation pathway, thus providing an opportunity to evaluate the effects of different types of degradation on AMF communities. Depleted states of these eucalypt woodlands are associated with lower C and total N (hereafter 'CN-depleted' ecosystem states), while the enriched states are associated with an increase in the same elements (hereafter 'CN-enriched' ecosystem states; Prober et al. 2002Prober et al. , 2014. Nutrient-depletion and associated disturbance often result in decreased native plant productivity and moderate plant invasions; while nutrient-enrichment promotes the persistence of exotic plants that prevent ecosystem recovery (Prober et al. 2002, 2014, Farrell and Prober 2021Fig. 1).
Here, we characterised AMF communities across ecosystem degradation states characteristic of temperate eucalypt woodlands, as a first step towards understanding potential feedbacks and opportunities for incorporating AMF into ecological restoration practices. To do so, we used a state-and-transition framework ( Fig. 1, already employed to compare soil chemical and biophysical properties; Prober et al. 2002, 2014, Farrell and Prober 2021, to compare AMF communities characteristic of reference sites with four states of ground-layer degradation (Fig. 1). We hypothesised that AMF richness would be lowest in CN-depleted states due to N-limitation (which is proposed to direct C to N uptake rather than to AMF), and highest in CN-enriched states due to N-enrichment (which strengthens P-limitation and hence benefit AMF) (H1). Using Chagnon et al. (2013) CSR classification, we also hypothesised that there would be a shift in AMF composition from dominance of stress-tolerator AMF (i.e., Acaulosporaceae) in reference and CN-depleted states, to dominance of ruderal AMF (i.e., Glomeraceae) in invaded, CN-enriched states (H2). Finally, we hypothesised that changes in AMF community composition will reflect the changes in N, P, and the shift from native-dominated to exoticdominated vegetation across ground-layer (H3).

Site selection
We sampled five ground-layer states ('Groundlayer state' factor) following the state-and-transition framework proposed by Prober et al. (Prober et al. 2002(Prober et al. , 2014 Fig. 1). Ground-layer states reflected dominance by invasive or different native grasses, and correlated with patterns of soil modification (in particular, depleted or enriched in C and N compared to reference plots). We first identified eleven little-modified remnants of grassy Eucalyptus albens Benth, Eucalyptus melliodora Cunn. ex Schauer, and Eucalyptus blakelyi Maiden (boxgum) woodlands in south-eastern Australia to represent the 'reference state' (Fig. 1 in Prober and Thiele 1995). These remnants have been excluded from substantive livestock grazing, cropping, and fertilising since very early in the history of European colonisation of the region (>100-150 years ago), typically due to fencing for use as cemeteries (although they were never fully utilised as cemeteries) (Prober and Thiele 1995). These plots provide unique value to our study, in that such areas with close to intact examples of grassy box-gum woodland understorey and soils are typically not available in similar modified, multi-use landscapes. The reference state was dominated by tall native perennial tussock grasses (Themeda triandra Forssk. and Poa sieberiana Spreng), with typically high native forb diversity and low exotic plant abundance (Prober et al. 2002). The reference state was low in soil available N (particularly nitrate-N) and had moderate soil fertility in terms of total N, potassium, available P, and exchangeable cations as well as amounts of total C for microbial activity (Prober et al. 2014(Prober et al. , 2002; Appendix S1: Table S1). Mean annual rainfall ranges from 600 to 700 mm, Fig. 1 State-and-transition model of ground-layer states in grassy eucalypt woodland according to dominant grass species and soil chemistry associated with changes in soil conditions (Prober et al. 2014; Appendix S1: Table S1). The reference state is dominated by the tall native tussock grasses Themeda triandra and Poa sieberiana and is low in soil nitrogen (N) (axis 1) and has moderate soil fertility described by total carbon (C), total N, potassium, phosphorus (P), exchangeable cations. Carbon and nitrogen (CN)-Depleted states are dominated by short to moderate native tussock (Rytidosperma spp. -Austrostipa scabra; Aristida ramosa) with low to moderate frequency of annual exotics and native forbs, and soils contain lower C and N levels and higher bulk density than the reference state. The intermediate state is dominated by Bothriochloa macra and a similar composition of native and exotic grasses and forbs as the depleted state, but soils contain relatively higher values of N, P, and bulk density than the reference state. The CN-enriched state is dominated by robust annual exotics with low perennial cover and native richness, and soils with higher values of nutrients and C than the reference state. Black solid arrows represent a transition from reference state (1) to depleted and enriched states (2-5), i.e., degradation pathways, while grey dashed arrows represent a transition, also on a degradation pathway, from depleted (2-3) or intermediate (4) states to a highly degraded (enriched) state dominated by exotic annuals (5). Reverse (restoration) pathways are uncertain and not shown mean maximum temperature from 27.5-29.5 °C and mean minimum temperature from 6 to 7 °C (Su et al. 2019).
We sampled one plot of each of the four degraded ground-layer states close to and on the same lithological type and topographic position as each of the eleven reference remnants (i.e. fence-line comparison or within 1 km apart) (Table 1). Most plots of degraded states had been grazed by livestock, varying from occasional grazing for short periods (such as passing through travelling livestock reserves, roadsides, and other public lands by travelling stock), to grazing for extended periods for farm production in productive paddocks. Only two of the paddocks (and not public lands) had undergone other disturbances, e.g. cropping (>20 years ago) and/or superphosphate addition (>10 years ago).
As characterized in Prober et al. (Prober et al. 2002(Prober et al. , 2014, the two 'CN-depleted states' were dominated by short-to-moderate native tussock grasses (Rytidosperma spp. -Austrostipa scabra, hereafter 'Rytidosperma state'; Aristida ramosa, hereafter 'Aristida state') with low-to-moderate abundance of annual exotics and native forbs. These states were associated with lower soil total N and C and higher bulk density relative to the reference state (Appendix S1: Table S1). The 'intermediate state' was dominated by Bothriochloa macra with typically higher exotic annual abundance than the CN-depleted states, and soils with relatively higher total N, available P, and bulk density than the reference state (Appendix S1: Table S1). The 'CN-enriched state' was dominated by robust annual exotics (e.g. Avena spp.) with low perennial cover and native plant richness. Soils from the CN-enriched state typically had higher nitrate-N and P availability, total N and C, and similar bulk density to the reference state (Appendix S1: Table S1).
Because trees also influence soil properties (Prober et al. 2002(Prober et al. , 2014Eldridge and Freudenberger 2005), we stratified our sampling 'under trees' and in 'canopy gaps' ('Canopy state' factor). This was done only for the reference, one CNdepleted (i.e., Rytidosperma), and the CN-enriched states. For other ground-layer states we sampled only in canopy gaps (beyond the dripline of trees) because these ground-layer states tended not to occur under trees. We successfully located 6-10 replicate plots of the eight Ground-layer state × Canopy state combinations, for a total of 64 samples (Table 1). A detailed description of plot selection, sampling, and processing can be found in (Prober et al. 2014) and is summarised here.  Sodosol  TSR  TSR  TSR  TSR  TSR  TSR  TSR  TSR  Bookham  Sample collection and processing Each plot was 100 m 2 and vegetation and topsoils were sampled in April 2012. Vegetation and soil physicochemical properties (Prober et al. 2014;Farrell and Prober 2021) were used here as explanatory variables for AMF community composition and richness. Cover of native grass species and annual exotics, litter, and bare ground in each plot was estimated using a point-intercept technique (Prober et al. 2005). Soil samples for each plot comprised a bulk of 10, 2 cm diameter × 10 cm depth cores collected randomly across each plot. After thoroughly mixing the ten cores per sample, a subsample was transferred to microcentrifuge tubes for DNA analysis. All samples were kept in a refrigerator for up to 2 days until further processing; samples for DNA analysis were then transferred to a freezer at −80 °C. Samples for chemical analyses were air dried at up to 40 °C (in a glasshouse), then sieved through a 2 mm sieve.

Soil physicochemical analyses
Soil bulk density was estimated by weighing soil (dried at 105 °C) from each of three soil cores of known volume (50 mm diameter and depth) per plot. Processed soil samples were sent to CSBP Laboratories (Bibra Lake, Western Australia) for nutrient analyses. Unless specified otherwise, the methods for soil analyses followed those of Rayment and Lyons (2012). Plant-available P was measured using the Colwell test (Colwell 1963), ammonium-and nitrate-N were measured as per Searle (1984). Soil pH was measured in CaCl 2 in a solution ratio of 1:5. We also estimated total C in 0-10 cm using a modified Dumas method with a Leco CNS 2000 analyser (Leco Corporation, St. Joseph, MI, USA).
DNA extraction and sequencing DNA was extracted from 0.25 g of soil using the Power Soil DNA Isolation kit (MO-BIO) and quantified spectrophotometrically (Nanodrop ND-100, Thermoscientific). Arbuscular mycorrhizal fungi were targeted using the nested PCR approach of Krüger et al. (2009) to initially generate an approximately 1500-bp fragment specific to AMF, from which the ITS regions (i.e., ITS1, 5.8S, and ITS2) were then sequenced using primers ITS1F (Gardes and Bruns 1993) and ITS4 (White et al. 1990 (Wang et al. 2007) with the 'classify.seqs, method = wang' function in Mothur against the Mothur-formatted UNITE6_SH database at 60% probability cut-off.

Statistical analyses
Two samples failed to amplify DNA and were removed from further analyses (final N = 62). We rarefied the raw dataset (i.e. the initial denoised zOTU abundance table) to the smallest sequencing depth (12,088 sequences) to avoid sequencing depth bias (Dickie 2010; Appendix S1: Fig. S1). All non-AMF sequences (i.e. non-Glomeromycota sequences) were removed from further analyses. All data were analysed, and figures were created, in R v4.2.1 (R Core Team 2016). We evaluated differences in rarefied zOTU richness (hereafter 'richness') of AMF among groups (Ground-layer state × Canopy state). Richness of AMF was analysed using generalised linear models with location as random effect using a Poisson distribution. Location was the general area where replicate plots from multiple ground-layer states were found within 1 km of their respective reference site and on the same lithological type (equivalent to a statistical 'block', Table 1).
To test how environmental attributes associated with transition between ground-layer states related to AMF richness, we built a piecewise structural equation model (SEM). A piecewise approach is a flexible variant of SEMs that allows for the fitting of various model forms and specifications, like variance structures and distribution families (Lefcheck et al. 2016). First, we calculated the 'native:exotic ratio' (i.e., native to exotic plant cover ratio) to represent the relative cover of native and exotic plants independent of ecosystem productivity. We used this ratio instead of absolute values of individual groups of plants to simplify models. Also, due to high covariation among most soil variables (Appendix S1: Fig. S2), we simplified models to only include variables from which we had a priori knowledge of being involved in state transitions in eucalypt woodlands. These variables were organic matter (OM), nitrate-N, and Colwell P, which are hypothesised to promote exotic annuals (Prober and Wiehl 2012;Prober et al. 2014). Individual models testing relationships between single response and all explanatory variables were fitted with location as random effect, and residuals were inspected for assumption violations. In these individual models, each degraded state was included as a binomial variable, meaning the reference state was implied if all degraded states had values = 0. Individual models were fitted using the nlme package (Pinheiro et al. 2017). Then, path models were fitted with the piece-wiseSEM package (Lefcheck et al. 2016). During this step, proposed pathways were removed if they were non-significant, and their removal did not compromise model fit. To include a proxy of community composition into the SEMs, we used the first axis of a Principal coordinates analysis using Bray Curtis dissimilarity of the log-transformed AMF community matrix.
We used non-metric multidimensional scaling, with Bray Curtis dissimilarity of the log-transformed matrix, to visualise variation in AMF community composition among groups (Ground-layer state × Canopy state). To indicate species contributions to the ordination we constructed two-way tables using the rarefied sequence abundance in the log-scale with the 'inkspot' function from the rioja package (Juggins and Juggins 2019). Plots were ordered on the x-axis by ground-layer state. Two-way tables summarise and present the raw community data as a powerful way of visualising the distribution of species along environmental, spatial, or temporal gradients.
To support data visualisation with statistical tests, we used permutational multivariate analysis of variance 'adonis2' within the vegan package (Oksanen et al. 2007) to test for differences in community composition among groups (Ground-layer state × Canopy state). Then, to quantify the extent of dissimilarity in community composition among groups, we used 'anosim' (9999 permutations) with the vegan package (Oksanen et al. 2007). To test which vegetation and soil chemistry variables correlated with changes in AMF community composition among groups, we calculated vectors of maximum correlation with the vector-fitting procedure using 'envfit' (9999 permutations) with the vegan package (Oksanen et al. 2007). A biplot was drawn on the ordination to display the relationships between the explanatory variables and the ordination axes. To identify zOTUs with a disproportionate influence on community patterns, we used indicator species analysis (9999 permutations) using the indicspecies package (De Caceres et al. 2016). Finally, we assessed the number and identity of zOTUs that were unique to each group using the basic functions in R (R Core Team 2016).

Results
We obtained 14,538,162 sequences and 2454 zOTUs across all samples, from which 11,914,494 sequences and 861 zOTUs belonged to Glomeromycota. These zOTUs belonged to eight AMF families and a large proportion of unclassified AMF (14% of zOTUs). The most diverse families were Glomeraceae (40% of zOTUs) and Diversisporaceae (15% of zOTUs), while the least diverse was Gigasporaceae (2% of zOTUs). The other families ranged between 7 and 4% of zOTUs.

AMF richness
There was a significant interaction between Groundlayer state × Canopy state on AMF richness (F = 0.41; d.f. = 2; P < 0.005). For canopy gaps, AMF richness was highest in the intermediate and CN-enriched states (Bothriochloa, 'Annual Exotics', respectively), intermediate in one CN-depleted state (Rytidosperma), and lowest in the reference state (Fig. 2a). Under trees, AMF richness was highest in the CN-enriched state and lowest in both the CNdepleted and reference states (Fig. 2b).
Within ground-layer states, AMF richness differed between canopy gaps and under trees only in the CN-depleted state (Rytidosperma), being higher in canopy gaps than under trees (Fig. 2). In canopy gaps, higher AMF richness in intermediate and CNenriched states was driven by higher richness of Archaeosporaceae, Diversisporaceae, Glomeraceae, Paraglomeraceae, and unclassified AMF (Appendix S1: Fig. S3a). Under trees, higher AMF richness in the CN-enriched state was driven by higher richness of Ambisporaceae, Archaeosporaceae, Glomeracaeae, Paragomeraceae, and unclassified AMF (Appendix S1: Fig. S3b).
The multivariate piecewise SEM linking ground-layer state, soil chemistry, and native:exotic ratio with AMF communities in canopy gaps was well supported by the data (χ 2 = 26.3; d.f. = 28; P = 0.24; Fig. 3a). The best model showed that ecosystem degradation was indirectly linked with AMF richness and community composition through changes in native:exotic ratio, mediated by soil OM, nitrate-N and P availability (Fig. 3a). Indeed, the higher OM, soil nitrate-N and P availability found in CN-enriched and intermediate states (relative to the mean across other states) was associated with lower native:exotic ratio (i.e. more exotic and less native cover). In turn, higher native:exotic ratio was negatively linked with AMF richness and community composition (Fig. 3a). By contrast, lower OM, soil nitrate-N and P availability found in CN-depleted states was associated with a higher native:exotic ratio and lower AMF richness relative to the mean across other states (although AMF  Table 1 for sample size. Symbols and error bars represent estimate and 95% confidence intervals from generalised mixed models using location as random effect. Different letters represent statistical differences among Ground-layer state× Canopy state using post hoc Tukey test (alpha = 0.05)  richness was still higher in the Rytidosperma state than the reference state, Fig. 2). The multivariate piecewise SEM linking groundlayer state, soil chemistry, and native:exotic ratio with AMF communities under trees was also well supported by the data (χ 2 = 23.6; d.f. = 0.6; P = 26; Fig. 3b). The best model showed similar results to canopy gaps, with the CN-enriched state indirectly linked with AMF richness through soil OM and the native:exotic ratio (Fig.3b). However, the CNdepleted states and soil P availability did not have a significant effect in either the native:exotic ratio or AMF richness under trees, and P was the only variable that influenced AMF community composition (Fig. 3b). It should be noted that the best model for canopy gaps explained almost twice as much variation than the best model for under trees (Fig. 3).
In canopy gaps, trends in AMF community composition correlated with changes in several soil and vegetation properties across ground-layer states ( Fig. 4a; Appendix S1: Table S2). The CN-enriched state was mainly distinguished from the reference state along the second axis of the NMDS, which correlated with changes in cover of native and exotic plant species, soil available P, and nitrate-N (Fig. 4a). In contrast, the CN-depleted states were  Table 1 for sample size. Vectors show significant variables selected from maximum correlation analyses (Permutations = 9999). correlated with ordination. Vector length is proportional to the relative importance (Appendix S1: Table S4). BG = bare ground; other variables listed in Appendix S1: Table S1 mainly distinguished from the reference state along the first axis of the NMDS, which correlated with changes in OM, soil pH, and other soil parameters (Fig. 4a). Under trees, both degraded states were distinguished from the reference state along both axes of the NMDS (Fig. 4b). These changes in AMF communities correlated with only two variables, cover of Poa spp. and soil pH; cover of exotic plants was marginally non-significant ( Fig. 4b; Appendix S1: Table S2).
Changes in AMF community composition among ground-layer states were associated with distinct patterns in the relative abundance of AMF at the family level. Distinctions among ground-layer states were clearest in canopy gaps ( Fig. 5; Appendix S1: Fig. S5).
In canopy gaps, Acaulosporaceae was most abundant (i.e., highest sequence abundance) in the reference state, driven by a single zOTU in two plots, while rarest (i.e., lowest sequence abundance) in the CN-enriched state (Fig. 5b Appendix S1: Fig. S6). Ambisporaceae were absent or rare in the reference and CN-enriched states, except for one reference plot, while Glomeraceae were most abundant in these ground-layer states (Fig. 5c, h; Appendix S1: Fig. S6). Archaeosporaceae were rarest in the reference state than in all other ground-layer states, and Diversisporaceae were most abundant in the two most CN-enriched states (Fig. 5d, f; Appendix S1: Fig. S6). Beyond these distinctions, Claroideoglomeraceae, Paraglomeracaeae, and Gigasporaceae were rare in canopy gaps.  Table 1 for sample size. First panel (a) shows all AMF, while all other panels (b-j) show individual taxonomic families of AMF. Y-axis shows each zOTU but was compressed to show full dataset on each panel. Bubble size is proportional to a rank scale of sequence number within, but not among, panels within the figure Under trees, patterns were less clear, with Acaulosporaceae being most abundant in the CN-depleted state, and Diversisporaceae being most and least abundant in the CN-enriched and CN-depleted states, respectively (Appendix S1: Fig. S5, Fig. S6). Claroideoglomeraceae was most abundant in the CNdepleted state, while both Glomeraceae and uncultured AMF were most and least abundant in reference and CN-enriched states, respectively (Appendix S1: Fig. S5, Fig. S6). All other families were rare or absent in all three ground-layer states under trees (Appendix S1: Fig. S5, Fig. S6).
No families were unique to any given ground-layer state, although each state did contain unique zOTUs. For canopy gaps, all ground-layer states showed relatively similar numbers of unique zOTUs, albeit the reference and CN-enriched states showed slightly higher values (Appendix S1: Fig. S7; Table S3). Indicator species analysis showed that the two most CNenriched states (Bothriocholoa and Annual Exotics) contained four-to-five times more indicator zOTUs than the reference state (Appendix S1: Table S4). From these, the Diversisporaceae were the most represented in the intermediate state, and Glomeraceae in the CN-enriched state (Appendix S1: Table S4). However, this does not suggest there were no zOTUs from these families in the other ground-layer states (Fig. 5). The two CN-depleted states showed few-tono indicator OTUs (Appendix S1: Table S4). Under trees, the reference, CN-enriched, and CN-depleted states, showed the highest, intermediate, and lowest number of unique zOTUs, respectively (Appendix S1: Fig. S7; Table S3). However, indicator species analysis showed that the CN-enriched state showed six times more indicator zOTUs than the other two ground-layer states, mainly from Diversisporaceae and Glomeraceae (Appendix S1: Table S4).

Discussion
We used an existing state-and-transition framework for box-gum woodlands (Prober et al. (2002(Prober et al. ( , 2014 to structure our sampling and characterise changes in AMF with degradation. As described in this framework, transitions from reference to degraded states are linked to either depletion or enrichment of soil properties (particularly C and total N), which in turn promotes exotic over native plant productivity. Here, we show that differences among degraded states are also associated with differences in AMF communities.
Contrary to expectations, both CN-depleted and CN-enriched states of box-gum woodlands harboured higher AMF richness than the reference state. This result partially supports our first hypothesis (H1), where we predicted that AMF richness would be highest in CN-enriched states and lowest in CNdepleted states. We found little support for our second hypothesis (H2), that there would be a shift from dominance (i.e, higher relative abundance) of stresstolerator and competitor AMF in reference and CNdepleted states, to dominance of ruderal AMF in CN-enriched states. Indeed, only Acaulosporaceae followed the expected trajectory (higher relative abundance in the refence state), but this was driven by a single zOTU found in two reference plots. We found support for our third hypothesis (H3), that transition between ground-layer states would indirectly affect AMF communities via changes in soil properties and vegetation. The differences in AMF community composition and richness among ground-layer states were associated with differences in OM, nitrate-N, P, and native:exotic ratio, all variables previously linked to box-gum woodland degradation (Prober et al. 2014;Farrell and Prober 2021).

Effect of ecosystem degradation on AMF communities
The reasons for higher AMF richness in degraded states are unclear but potentially related to changes in plant composition and soil nutrients with ecosystem degradation. Replacement of native plant species that do not rely on AMF for P-uptake with plant species that do, could promote AMF. Equally, if N-enrichment exacerbates P-limitation, it could directly promote AMF. We develop these ideas below.
With respect to altered plant composition as a driver of change in AMF richness, P-limitation is the norm in many Australian ecosystems and many plant species have evolved diverse P-acquisition strategies, resulting in more non-AMF plants than in many other ecosystems globally (Lambers et al. 2011). Subsequent arrival of exotic plants with reliance on AMF in these ecosystems would increase host availability for AMF, likely promoting their diversity (Lekberg et al. 2013;Kim et al. 2015;Gomes et al. 2018;Wang et al. 2020). Hence, changes in AMF communities during ecosystem degradation might be responding to the arrival of AMF-dependent exotic plants, contributing to the maintenance of the degraded states by providing competitive advantage to exotic over native plants. Consistent with this idea, both the changes in AMF community composition and increase in richness from reference to CN-enriched states were related to the increase in exotic relative to native plant cover. Box-gum woodlands may thus harbour relatively low AMF diversity, albeit involving distinct AMF communities, probably due to there being few plant hosts. Indeed, up to 45% of the native plant species found in the box-gum woodlands are non-AMF, while only 22% of the invasive ones are non-AMF (Prober et al. 2002;Brundrett 2009).
With respect to changes in soil nutrients as a potential driver of change in AMF richness, it has been hypothesised that the effects of nutrient enrichment on AMF depend on the limiting nutrients of the reference ecosystem. If the ecosystem is P-limited, as in our study, an increase in N-availability is expected to intensify P-limitation, which in turn is expected to promote AMF (Johnson et al. 2013;Lilleskov et al. 2019;Han et al. 2020). This is indeed consistent with the higher AMF richness we observed in CNenriched states, and also with the relationships we observed between AMF community composition and soil N and P. We note that our study excluded P-rich sites affected by intensive agriculture (i.e., high P inputs), such as fertilised pastures. To further explore the nutrient hypothesis, it would be valuable to test whether increases in N-availability in such P-enriched states would lead to lower AMF richness, through amelioration of nutrient limitations and plants' reliance on AMF.
Even though we found support for both the nutrient enrichment and plant invasion hypotheses, nutrient depletion and enrichment are intrinsically linked to exotic plant cover in our study, making it very difficult to tease these factors apart. It is likely that soil physicochemistry and exotic plant cover persist via positive feedback (Prober et al. 2002;Van der Putten et al. 2013), which we acknowledge was difficult to model statistically (i.e., in our SEM or in other casual models). We conclude that from an ecological perspective, it is likely that both exacerbation of P-limitation and exotic plant invasion co-occur and promote AMF diversity in degraded box-gum woodlands.
Response of AMF families to ecosystem degradation Similar to aboveground plant communities (Prober et al. 2002), AMF community composition was a good descriptor of ground-layer state, however, these differences in AMF communities provided little general support for the Chagnon et al. (2013) CSR framework. Rather, contrasting differences in relative abundance of individual zOTUs within each AMF family were observed. This suggests that the CSR strategies of AMF might not necessarily be generalisable at the family level, rather, species within each family could potentially have different CSR strategies (Klironomos 2000). Recently, another grouping of AMF families into functional groups have been proposed: Weber et al. (2019) classified AMF into functional groups based on biomass allocation (extra-vs intra-radical hyphae). For example, they classified both Gigasporaceae and Diversisporaceae as 'edaphophilic' due to greater allocation to extra-rather than intra-radical hyphae. However, our results do not support this classification of broad taxonomic groups. Like Phillips et al. (2019), we found that Gigasporaceae and Diversisporaceae behave differently.
Gigasporaceae and Diversisporaceae tended to follow opposite trends among ground-layer states in our study. Gigasporaceae was most abundant in reference and one CN-depleted state, consistent with its classification as edaphophilic (i.e., better at scavenging limiting resources and soil exploration). Diversisporaceae, on the other hand, was most abundant in the intermediate states, where resources are more available, despite its classification as edaphophilic. Our finding for Diversisporaceae is more consistent with Chagnon et al. (2022), who found their indicator species analysis selected Diversisporaceae for an intermediate land-use management, and Wang et al. (2020) found that relative abundance of Diversispora (genus within Diversisporaceae) was highest under N-only addition, and lowest under N and P addition. Hence, Diversisporaceae may prefer intermediately enriched soils.
The other AMF family that warrants discussion is Archaeosporaceae, which was found in much higher relative abundance in all degraded states than in the reference state. Archaeosporaceae has been considered 'ancestral' due to lack of clear preference for intra-or extra-radical hyphae biomass . In our study, Archaeosporaceae was the only family that appeared to benefit from degradation, irrespective whether it was CN-depletion or enrichment. Zheng et al. (2016) found that richness of this family can be strongly influenced by plant identity. We surmise that species of this family might prefer exotic over native plant hosts in the box-gum woodlands, but this needs further investigation.

Effect of trees on AMF communities
Trees showed equalising effects on AMF by homogenising their communities: AMF communities were clearly different among ground-layer states in canopy gaps, but less so under trees. Further, the changes in zOTU relative abundances of each AMF family were also less clear under trees than in canopy gaps. This may be due to effects of understorey state being overridden by the strong effects of trees on soil properties: most soil nutrients, including soil OM, were found in higher levels under trees than in gaps (Appendix S1: Table S1; Farrell and Prober 2021;Prober et al. 2014). Given this homogenising effect of trees, we recommend future studies stratify as we did, as trees can potentially obscure the effects of ecosystem degradation.

Implications for restoration
Our application of a state-and-transition approach allowed us to elucidate changes in AMF communities with ecological degradation, in turn providing a potential framework for testing interventions for ecological restoration (Sinclair et al. 2019). It stands to reason that restoration interventions will be more successful if they account for the belowground component of ecosystems. For example, guided by the stateand-transition model, Prober and Lunt 2009 showed that N-depletion of CN-enriched states reduced exotic grasses and promoted a transition towards reference states (Prober and Lunt 2009). However, the same practice may not be effective in states that are already CN-depleted, highlighting the value of these models.
Targeting AMF could similarly trigger transitions between ground-layer states (Coban et al. 2022). However, whether altered AMF communities are contributing or responding to the maintenance of degraded ecosystems (i.e., driver vs passenger theory; Hart et al. 2001, Shah et al. 2009), remains unknown. If AMF are drivers of ecosystem degradation, state transitions could be achieved by manipulating AMF to shift plant communities and nutrient levels towards a reference state. This intervention has been explored with promising results in the context of re-introducing missing AMF taxa (Neuenkamp et al. 2019 and references therein). If AMF are passengers, on the other hand, state transitions could be achieved by using plant hosts that do not associate with weedy AMF to shift AMF communities towards a reference state. This intervention has been tested before, whereby re-introduction of native grasses induced state transitions (Prober et al. 2005;Prober and Lunt 2009), but the role of AMF in these dynamics needs testing in field conditions.
Our study highlights the importance of further research into the mechanisms by which AM is involved in ecosystem degradation. We found that, albeit overall higher AMF richness in degraded states, there was both a loss of AMF taxa from the reference state as well as introduction of new AMF taxa in degraded states. This is important because current restoration attempts only involve the use of native AMF as inoculum to promote native vegetation (Neuenkamp et al. 2019). To our knowledge, no study has yet attempted to manage 'weedy' AMF to facilitate ecological restoration, although there is increasing research on the global distribution of particular weedy AMF species (Thomsen and Hart 2018).

Conclusion
This is the first time that AMF have been linked with data from vegetation and soil using a state-and-transition framework to understand ecosystem change. Our results suggest that transitions among ground-layer states are related to changes in soil physicochemistry, vegetation, AMF, and plant-soil feedbacks. This approach improved our ecological understanding of the different pathways of degradation, potentially elucidating constraints to ecosystem recovery. Understanding the type of degradation and its effects on AMF, particularly the arrival of 'weedy' AMF, is paramount if soil microbes are to be incorporated into ecosystem restoration. Indeed, our study indicates that even though loss of key AMF taxa from the reference state occurs, 'weedy' AMF are also involved, either as passengers or drivers (sensu Hart et al. 2001), in the maintenance of degraded states of box-gum woodlands.
Restoration interventions to manipulate AMF are thus not necessarily restricted to re-introducing AMF that have declined or been lost, as is often assumed.