Specialist species of wood-inhabiting fungi struggle while generalists thrive in fragmented boreal forests


Correspondence author. E-mail: jenni.norden@nhm.uio.no


  1. The loss of suitable habitats is one of the main causes behind the loss of species and communities. Habitat fragmentation, that is, the division of the remaining habitat into small and isolated fragments, often co-occurs with the process of habitat loss. The spatial division of habitats decreases connectivity among local populations and generally has a negative effect on population viability, but it can also have a positive effect for some species, for example, due to released competition pressure.
  2. In both animals and plants, certain characteristics such as low dispersal ability and narrow ecological niche are known to be associated with fragmentation vulnerability, but in fungi, systematic analyses have so far been lacking. With their small and highly dispersive spores, fungi could be mainly resource-limited, not dispersal-limited.
  3. In this study, we analysed spatial occurrence data on 119 species of wood-inhabiting fungi to identify the species characteristics that are associated with high extinction risk and fragmentation vulnerability in particular. We modelled resource use and connectivity dependence separately for each species using the presence–absence data on 98 318 dead trees in 496 sites located on a gradient in the duration and intensity of land use in eastern Fennoscandia. We then related species' responses to connectivity to their resource-use patterns, life-history characteristics and red-list status.
  4. Our results show that red-listed species are highly specialized in their resource use and suffer from loss of connectivity at three spatial scales: along the large-scale gradient, at the landscape scale and at the scale of a forest stand. In contrast, many of the non-red-listed generalist species are actually more likely to occur (per resource unit) in fragmented managed forests than well-connected natural forests.
  5. Synthesis. We show that the expected number of red-listed species per a fixed amount of similar resources (dead trees) can be even more than 10 times higher in well-connected than in fragmented surroundings, and thus, protecting high-quality areas that are well connected is conservationally more effective than protecting small fragments distributed across the landscape.


Habitat loss is globally one of the greatest threats to biodiversity (World Resources Institute 2005). However, it is often difficult to assess which component of habitat loss eventually drives populations to extinction – loss of habitat area, quality, connectivity or continuity (Hanski 2005) – and at which spatial and temporal scales. In particular, habitat loss and fragmentation typically appear simultaneously and it can be difficult to disentangle whether the isolation of the habitat fragments has any effect on top of the negative effect due to loss of habitats (St-Laurent et al. 2009). Loss of connectivity has often had a negative effect (Löbel, Snäll & Rydin 2009), but it may have no effect (Trzcinski, Fahrig & Merriam 1999) or even be beneficial for part of the species community (Spanhove et al. 2009).

In Fennoscandia and elsewhere in the boreal zone, intensive forest management with short rotation times has led to the loss and fragmentation of natural forests and profoundly changed forest structure and quality both at the stand and landscape levels (Östlund, Zackrisson & Axelsson 1997; Siitonen 2001). Habitat fragmentation has direct consequences to spatial population dynamics, mainly through dispersal limitation, but also through abiotic, structural and successional changes in habitat quality that result, for example, from increased edge effects (Lindenmayer & Fischer 2006). Indirect impacts of fragmentation are often mediated through interspecific interactions. For example, a competitively inferior species may become abundant in small and isolated fragments if competitively superior species are lost through dispersal limitation (see Debinski & Holt 2000).

All species are not equally vulnerable to the effects of habitat fragmentation or more generally, to extinction, but extinction risk can be associated with particular species characteristics (including both biological traits and other characteristics). Identifying such characteristics is essential, for example, for designing cost-effective conservation measures targeted for the most vulnerable species. Extinction vulnerability is generally associated with high trophic level, low population density, slow life history and small geographic range (Purvis et al. 2000; Thomas, Lanctot & Szekely 2006; Waldron et al. 2006; Cardillo et al. 2008; Fritz, Bininda-Emonds & Purvis 2009). In a comprehensive review of empirical and theoretical fragmentation studies, Henle et al. (2004) identified six characteristics as predictors of increased sensitivity to fragmentation in particular: specialized habitat requirements, low natural abundance, high population fluctuations, low ability to persist over unfavourable environmental conditions, intermediate or low dispersal power and low reproductive potential. Good dispersal ability is essential for tolerating habitat fragmentation for those species that have a high risk of local extinction. In contrast, less dispersive species may be insensitive to fragmentation if they have sufficiently stable local dynamics. Henle et al. (2004) noted that a number of other characteristics (e.g. sensitivity to disturbance and competition, matrix use, rarity, annual survival, sociality, body size and trophic position) have been suggested as predictors of fragmentation sensitivity, but the results concerning these characteristics are controversial or taxon-specific.

While the number of studies that try to establish the link between species characteristics and fragmentation sensitivity is high (see e.g. Tscharntke 2002; Aguilar et al. 2006; Barbaro & van Halder 2009), few studies have resulted in conclusive evidence. This may not be surprising, given the data requirements for such a study. First, to obtain the statistical power required to measure the effect of fragmentation for a given species, the sampling design should include a relatively large number of habitat fragments, with sufficient variation in habitat area and isolation. Second, to assess how species characteristics link to the effect of fragmentation, a sufficiently large number of species needs to be included. Due to logistical constraints, most studies that include a sufficiently large number (say, > 50) of species and sites have been based on presence–absence data at the fragment level. However, abundance data at the fragment level, or presence–absence data at the resource-unit level, are needed for deriving density-area relationships (DARs; Hambäck et al. 2007) or more generally for assessing how the number of individuals per one unit of resource depends on the spatial context.

A further difficulty in linking species characteristics to fragmentation vulnerability is that often the response to fragmentation is not analysed individually for each species, but by examining how species richness in particular species groups (e.g. species with high dispersal ability vs. species with low dispersal ability; Herault & Honnay 2005) correlates with the level of fragmentation. While such analyses are helpful in recognizing general patterns, some of the results may be masked by the dissimilar habitat requirements of the species within each group. Finally, a vast majority of large-scale fragmentation studies are based on comparative mensurative rather than manipulative approaches (McGarigal & Cushman 2002). As the results of non-manipulative studies can be affected by a number of confounding factors, such as correlations among habitat area, isolation and quality, these data pose challenges for the analysis phase (Ewers & Didham 2006).

Henle et al. (2004) found that the characteristics predisposing species to fragmentation are similar in plants and animals, but regarding fungi, we are not aware of any systematic analyses. One might assume that with their small and highly dispersive spores, fungi could be mainly resource-limited, not dispersal-limited. Wood-inhabiting fungi are known to respond to fragmentation, as species richness and the occurrence of particular species have been shown to vary with forest patch isolation and size (Berglund & Jonsson 2001; Penttilä et al. 2006; Laaksonen et al. 2008; Berglund, O'Hara & Jonsson 2009), the latter at least partly due to edge effects (Snäll & Jonsson 2001; Siitonen, Lehtinen & Siitonen 2005). Further, the occurrence of wood-inhabiting fungi is known to depend on habitat availability at the landscape scale (Siitonen, Penttilä & Kotiranta 2001; Sverdrup-Thygeson & Lindenmayer 2003; Edman et al. 2004a; Penttilä et al. 2006; Hottola & Siitonen 2008). A review of the conservation ecology of polypores in boreal forests (Junninen & Komonen 2011) concluded that a species-rich polypore assemblage is likely only if the area of the forest stand exceeds 20 ha, the volume of dead wood is at least 20–40 m2 ha−1 and most of the dead wood consists of logs with a diameter of at least 20–30 cm – conditions that are not met in the majority of the current Fennoscandian boreal forests. Polypores form a well-known group of wood-inhabiting basidiomycetes. A substantial fraction of polypores is classified as red-listed species in Fennoscandia, for example 42% in Finland (Kotiranta et al. 2010).

In this study, we use a large-scale data set (Fig. 1) to examine how species characteristics relate to response to habitat fragmentation in wood-inhabiting basidiomycete fungi. We use a modelling approach to examine how the probability that a given species occurs on a given resource unit (a dead tree) depends on the density of other resource units at three spatial scales. At the smallest scale of the local forest stand (scale c. 100 m), we measure the density of resources directly by the amount of dead wood suitable for the focal species. At the landscape level (c. 5 km), we do not have direct data on the availability of dead wood, and thus, we use the age of the surrounding forests as a proxy. At the largest scale, we use as the explanatory variable the position of the study site along an overall gradient in forest use history (c. 600 km). The modelling approach accounts for the effect of resource quality (e.g. host-tree species, diameter and decay class) and allows us to systematically classify the species in terms of their specialization level. We use the parameterized model to ask how fragmentation vulnerability is linked to specific characteristics. Based on the results of Henle et al. (2004) on other species groups, we hypothesize that also in fungi sensitivity to fragmentation increases with decreasing niche width, with decreasing natural abundance and with decreasing dispersal ability (for which characteristic we use spore size and shape and the thickness of the spore cell wall as proxies). In the analysis, we included fruit-body size, trophic position (successor and non-successor species) and a number of other characteristics that might affect how sensitive to fragmentation a species is.

Figure 1.

Locations of the 496 study sites in the southern half of Finland and north-west Russia. The sizes of the circles are proportional to landscape-level connectivity (S2; see Material and methods) and the grey colour depicts forest. Study regions: 1 = south-western and southernmost Finland; 2 = central Finland; 3 = eastern Finland and Russian Karelia.

Materials and methods

Study area and study sites

We surveyed dead trees and wood-inhabiting fungi on a set of forest sites in southern and middle boreal vegetation zones (Ahti, Hämet-Ahti & Jalas 1968) in eastern Fennoscandia. The data were acquired in autumns of 2000–2005 in 496 study sites, covering managed, semi-natural and natural forests of different successional stages, ranging from young to old-growth forest. The study area extended from south-western Finland to eastern Finland and north-western Russia (Fig. 1), along a gradient in the history of land use and consequent differences in the amount and quality of forests (see below). Most of the forests were admixed, dominated by Norway spruce (Picea abies), some by Scots pine (Pinus sylvestris) and a few by different deciduous tree species.

Inventory methods

The inventories were conducted using sample plots (size 0.2–4.0 ha). All dead trees, called resource units below, with a minimum diameter of 5 cm were measured within the sample plots. In most study sites, also the remaining stand area outside the sample plot was inventoried using a minimum diameter of 15 cm. For each dead tree, we recorded the tree species, type of resource unit (nine classes, e.g. uprooted or cut logs), diameter and decay stage (seven classes, 0 representing recently died and six almost completely decayed; Hottola & Siitonen 2008).

On each resource unit, we recorded the presence–absence of all polypores (poroid Aphyllophorales) and, in addition, 13 other wood-inhabiting fungal species (corticioid or hydnoid Aphyllophorales) that are easy to identify in the field. One reason for selecting this species group is that our focal species have macroscopic fruit bodies that can be reliably detected and distinguished in the field, thus enabling a large-scale survey. The fruit bodies that could not be identified with certainty to the species level in the field and were collected for later microscopical identification. Species were surveyed once in each site. The inventory method, resource-type classification, key characteristics of the study sites and the life-history characteristics of the inventoried species are described in detail in Appendix S1 in Supporting Information.

Hierarchical model for species occurrence

The full data consist of the presence–absence of 174 focal species on 98 318 resource units in 496 sites. We included in the analyses those 119 species (including 12 corticioid or hydnoid species, half of which are red-listed species) with at least 10 occurrences, with a total of 42 893 occurrences out of which 1019 represent 23 red-listed species. Given the hierarchical nature of the data (resource units in sites), we applied a two-level hierarchical model. We next define the model, then describe how the model was parameterized using Bayesian inference and finally describe how the parameterized model was applied to address the aims of this study.

We modelled the data independently for each species and denote by oij the indicator variable with the value oij = 1 if the focal species was present on the resource unit j of site i and with the value oij = 0 if it was absent. We modelled the presence–absence of the species at the resource unit by probit regression,

display math(eqn 1)

Where si is the effect of site i, math formula is the resource-unit level covariate k for resource unit j of site i and math formula is the resource-unit level regression coefficient for covariate k. To study the effects of resource-unit variables on the occurrence probability of species, we included four resource-unit level covariates: tree species, resource-unit type, log-transformed diameter and decay class (see Appendix S1 for details). The probit link function is given by the cumulative distribution of the standard normal distribution, and it is an alternative to the more commonly applied logit link function for a binary response. We selected to use probit regression rather than logistic regression because neither the logit nor the probit link function can be considered more natural from biological grounds, but the probit link function is computationally more convenient.

The effect of the site was modelled as si = ci + εi, where math formula, with math formula being the site-level covariate k of site i and math formula the corresponding regression coefficient.

Computation of site-level covariates

To study the effects of site-level variables on the occurrence probability of species, we included six site-level covariates S1–S6 (Table 1). Three of these (S1–S3) quantify the connectivity of the resource unit to other resource units at different spatial scales, whereas two (S4–S5) describe stand quality. Small-scale connectivity (S1), that is, the local density of resources available for the focal species in site i was measured as

Table 1. Site-level variables describing connectivity or site quality (S1–S6), and variables describing the responses of the species to resource-use level variables (R1–R7)
  1. a

    C refers to continuous variables and Dx to discrete variables with x classes.

Site level
S1Small-scale connectivity: local density of resourcesC
S2Intermediate-scale connectivity: the age of forests in the surroundingsC
S3Large-scale connectivity: location along the gradient from SW to NEC
S4Naturalness: negative of the log-transformed basal area (m2 ha−1) of cut stumpsC
S5Canopy closure: basal area (m2 ha−1) of the living standC
S6Unexplained variation among sitesC
Resource-unit level
R1Rarity on preferred resource unitsC
R2Tree-species preferenceD8
R3Resource-type preferenceD3
R4Mean decay classC
R5Specificity in decay-class useC
R6Size dependencyC
R7Overall specialization levelC
display math(eqn 2)

Where Ai is the area of the site inventoried. S1 thus utilizes information on the number and quality of all resource units (math formula) in the focal site i. By eqn 1, S1 measures the expected number of occurrences per unit area, the expectation being computed using information about the quantity and quality of resource units at the focal site. We computed S1 accounting only for trees at least 15 cm in diameter because trees of this size were surveyed over the whole site.

As we did not have data on the amount and quality of dead wood at larger scales, we approximated the connectivity of the focal resource unit to other resource units at the landscape (S2) and regional (S3) scales using the best available proxies. For the intermediate-scale connectivity (S2), we utilized data from the Finnish multisource National Forest Inventory (ms-NFI; Tomppo et al. 2008) which cover entire Finland with a spatial resolution of 25 × 25 m. The predictions employed here are based on the field data from NFI10 from years 2004–2005 and the computationally updated field data from 1996 to 2003 (NFI9), satellite images from 2004 to 2005 and digital-based map data. To obtain a measure of forest age around each study site on a coarse-grained 100 × 100 m grid, we averaged forest ages over the sixteen 25 × 25 m cells over the combined land and water area. The age zero was attached to the cells outside of forestry land, that is, outside of the combined forest land, poorly productive forest land and unproductive land. Twelve of the sites are located in the territory of Russian Federation, outside the coverage of the ms-NFI data. For these sites, we predicted the landscape data using the Finnish NFI field plots and satellite images covering both the Finnish and Russian sites, more precisely, the field plots from year 2000 and the Landsat ETM+ images 186-1617 (10.06.2000) and 186-15 (28.07.2000). We computed the age of forests around each inventory site as a sum of the grid-cell-specific values weighted by exp(−αd), where d is the distance to the focal site and the scale parameter was set to 1/α = 5 km. The reasoning here is that the lower the average forest age is, the more intensively the surrounding forests have been managed, and the more fragmented the landscape generally is.

The duration and intensity of forestry and other types of land use have been substantially greater in south-western Finland than in eastern Finland (Kalliola 1966; Rouvinen, Kuuluvainen & Karjalainen 2002; Lilja & Kuuluvainen 2005 and references therein). This is reflected in the proportion of natural-like, old-growth forests, as well as in the average amount of dead wood in the landscape. The proportion of semi-natural forest is 0.3% in south-western Finland and 2.9% in the middle boreal parts of eastern Finland (Virkkala et al. 2000). The average volume of dead wood is 1.8 m3 ha−1 in south-western Finland (Korhonen et al. 2000) and 4.5 m3 ha−1 in eastern Finland (Korhonen et al. 2001). We defined S3 as the sum of the latitude and longitude, and it thus refers to the location of the site along the gradient from SW to NE, measuring connectivity at spatial scales larger than the landscape scale.

Naturalness (S4) of each stand was measured as the negative of the log-transformed basal area (m2 ha−1; set to 0.1 if smaller than that) of cut stumps and canopy closure (S5) as the basal area (m2 ha−1) of the living stand. The random effect εi (S6), assumed to be normally distributed with zero mean and variance σ2, models among-site variation not captured by the measured covariates.

Given the rather complex hierarchical structure of the model (note the link between the two levels, see eqn 2), we chose to estimate the model parameters using Bayesian inference, as this framework enables the use of flexible Markov Chain Monte Carlo (MCMC) methods for model fitting. The prior distributions and the MCMC algorithm used for sampling the posterior distribution are described in Appendix S1.

The variables S2, S3, S4 and S5 were normalized to zero mean and unit variance across the sites. We note that when measuring connectivity at a spatial scale that is similar or larger than the distance between study sites, it is unavoidable that the connectivity values show spatial autocorrelation. This is the case here for both S2 and S3 (Fig. 1), in the former one due to smoothing by the exponential kernel and the fact that the amount of old forests varies in an autocorrelated manner. The variables S2 and S3 correlate positively with each other, but there is a lot of variation around this trend (R2 = 0.10).

Resource-use classification

While the estimated resource-unit level regression coefficients math formula as such include information on the resource use of each species, they are not straightforward to be interpreted biologically. To aid biological interpretation, we converted these parameter estimates into a quantification or classification of resource use of each species with respect to seven characteristics R1–R7 (Table 1). To do so, we predicted the occurrences of the species in hypothetical resource units, which we varied in characteristics such as tree species, and compared the model predictions to examine, for example how specialized the focal species is to a given resource type.

To compute the variables R1–R5, we used the model to predict the occurrence oH of the species on a hypothetical resource unit H as

display math(eqn 3)

Where math formula represents the characteristics of the hypothetical resource unit. In this formula, we have taken the mean over the site level effects to locate the hypothetical resource unit to a site with average properties in terms of connectivity, naturalness and canopy closure. We integrated eqn 3 over the posterior distribution of parameter estimates.

To estimate the rarity of each species on its preferred resource units (R1), we assumed that the hypothetical resource unit consists of the most suitable resource quality (for the particular species in question) concerning the tree species, resource type and decay class, and that it is 25 cm in diameter. We then defined R1 as the negative of the probit-transformed occurrence probability (eqn 3). Note that R1 differs from the raw number of recorded occurrences in the sense that it is independent of the availability of the preferred resource qualities.

To compute tree-species preference (R2), we constructed one hypothetical resource unit for each tree species. We set the other properties of these resource units (resource type, decay class and diameter), except the tree species, to the same values as for the preferred resource unit constructed for R1. We then predicted the occurrence probability of the species for all of these resource units using eqn 3. We first compared the expected number of occurrences on coniferous trees (occurrence probability summed over classes 1–4; Table S1–S2 in Appendix S1) and on deciduous trees (classes 5–11). If either class contained at least 90% of the predicted occurrences, we classified the species as a coniferous-dweller or a deciduous-dweller, otherwise as a generalist. If the species was classified as a coniferous-dweller, we investigated whether one of the conifer tree species (classes 1–3) contained at least 90% of the predicted occurrences within this group. Similarly, for the deciduous-dwellers, we examined if one of the tree species (5–10) contained at least 90% of the predicted occurrences.

We classified the resource-type preference (R3) following a similar logic as in the case of the tree species, again setting the other properties (tree species, decay class and diameter) to the most preferred values and allowing the resource type vary. We first examined if at least 90% of the predicted occurrences belonged to resource units that had died naturally (classes 1–6; Table S1–S3 in Appendix S1) or that were man-made (7–9). If neither was the case, we classified the species as a generalist with respect to resource-type use. If the species mainly occurred on resource types that had died naturally, we examined if at least 90% of the predicted occurrences within this class belonged to downed logs (classes 1–3), standing dead trees (4–5) or natural stumps (6). If the species was associated with man-made resource types, we used the same criteria to examine whether the species was specialized in cut stumps (7) or cut downed logs (8–9).

To compute the decay-class related classifications R4 and R5, we used the median (over the posterior density) values for the predicted occurrence probabilities for hypothetical resource units belonging to each of the seven decay classes. We measured the mean decay class (R4) by weighting each decay class with its occurrence probability and defined the decay-class specificity (R5) as the reciprocal of the standard deviation of occurrence probabilities over the decay classes.

As a measure for the resource-size dependency (R6), we used the median estimate (over the posterior distribution) for the regression coefficient for tree diameter.

To compute the overall specialization level (R7), we did not consider the hypothetical resource units, but examined what fraction of all of the 98 318 real resource units would be suitable for the focal species. To do so, we used eqn 1 in the main text to predict the occurrence probability of the species on all resource units using both the resource-unit level and the site-level characteristics. We then ordered the resource units in terms of the predicted occurrence probability and computed the fraction f of resource units that were needed to cover 50% of the predicted occurrences, if selecting the resource units with the highest occurrence probabilities. We defined the specialization level as −log f, so that a high value indicates that only a small fraction of the resource units represent suitable resources for the focal species. Note that unlike R1–R6, the variable R7 does not depend solely on the resource-unit level attributes but also on the site-level variables.

As some of the tree species and resource-unit types were very rare in the data, the data were not sufficient to confirm with high confidence that the species could not occur on these resource classes. We thus included in the above classification procedures only those tree species and those resource-unit types for which at least one occurrence of the focal species was found.

Effect of site-level variables on species richness

We quantified the overall effect sizes of the site-level variables (S1–S5) by examining how much their variation influences predicted species richness, separately for the red-listed (IUCN statuses VU, vulnerable and NT, near threatened) and non-red-listed (LC, least concern) species. We constructed a representative forest site by selecting randomly 200 resource units from the entire data set of all surveyed dead trees. We then used eqn 1 to predict the probability pHj that a given species is present on resource unit j in a hypothetical forest site H. Assuming that these probabilities are independent, the probability that the species is present in none of the resource units is math formula, and thus the probability of occurrence at the site level is

display math

The expected number of species present in the hypothetical site H was obtained by summing over the species-specific probabilities. Baseline species diversity in the hypothetical forest site H was estimated by setting the site-level variables (S1–S5) at their mean values. The effect of each site-level variable (e.g. S1) was measured by comparing the predicted species richness for the cases in which the focal site-level variable was set to its maximum or minimum value (among the 496 sites), while keeping the other site-level variables (e.g. S2–S5) at their mean values.

Relating species characteristics to resource use and response to fragmentation

To make the link from species characteristics to fragmentation vulnerability, we acquired information on various potentially relevant species attributes (T1–T15 in Table 2) and examined how they were associated with the resource use (R1–R7) and site-level responses (S1–S5) of the species. We first summarized the relationships between the patterns of species occurrences (R1–R7 and S1–S5) by a two-dimensional non-metric multidimensional scaling (NMDS; Venables & Ripley 2002; Oksanen et al. 2012) in which we located the species according to the pairwise dissimilarities (Gower 1971) of the variables R1–R7 and S1–S5. We calculated the locations (vector endpoints for continuous variables and factor-level centroids for categorical variables) of the variables R1–R7 and S1–S5 in the ordination space as averages of species' locations (Venables & Ripley 2002; Oksanen et al. 2012). We then fitted the species attributes T1–T15 in the ordination space so that maximal correlation between the variable and the species values was achieved. We tested for the significance of all the R-, S- and T-variables by a permutation test (Oksanen et al. 2012). Some of the patterns arising from the ordination analysis may be masked by a dominating effect of the host-tree species. We thus conducted the ordination analyses also separately for the species classified as coniferous-dwellers, deciduous-dwellers and generalist species.

Table 2. Variables relating to biological traits and other life-history characteristics (T1–T11), taxonomy, phylogeny, red-list status and the prevalence of the species. Parameter values were obtained from the literature (Appendix S1)
  1. LC, least concern; NT, near threatened; VU, vulnerable.

  2. a

    C refers to continuous variables and Dx to discrete variables with x classes.

  3. b

    Information was missing for 42 out of the 119 species.

  4. c

    Spore volume was approximated from width and length by assuming a cylindrical shape.

T1Decay type: brown (b) or white (w) rotD2
T2Sexualityb: homothallic (h), or bi- (b) or tetrapolar (t) heterothallicD3
T3Successor species: yes (y) or no (n)D2
T4Fruit-body size: small (s), intermediate (i) or large (l)D3
T5Fruit-body life span: annual (a) or perennial (p)D2
T6Fruit-body type: resupinate (r), half-resupinate (h) or pileate (p)D3
T7Hyphal system: mono- (m), or di- or trimitic (d)D2
T8Thickness of the spore cell wall: thin (n) or thick (y)D2
T9Log-transformed spore volumecC
T10Spore shape measured by the length/width ratioC
T11Chlamydospores: yes (y) or no (n)D2
T12Genus of the speciesD20
T13Phylogenetic cladeD14
T14Red-list status (LC, NT, VU) in FinlandD3
T15Log-transformed total abundance in the dataC

Excluding confounding effects for the effect of the fragmentation gradient

As our study design includes only one large-scale fragmentation gradient, it lacks statistical power to conclusively measure the effect of fragmentation at this scale. While the climatic conditions measured, for example, by mean temperature and precipitation do not vary markedly along the gradient (Hottola & Siitonen 2008), it is nevertheless possible that some confounding factors may cause an apparent positive or negative response to the fragmentation gradient. To at least partially exclude the possibility of such alternative explanations, we conducted the following analysis (see Appendix S2 for more details). For species that seemed to suffer from large-scale fragmentation (posterior probability for occurrence probability increasing towards north-east at least 0.95), we examined whether the species is known to occur also south-west from our study area: southern Sweden, Baltic countries or Central Europe. If this was the case, we classified the negative response to fragmentation as reliable, whereas in the opposite case, we classified it suspect to confounding factors. Similarly, for the species that seemed to benefit from large-scale fragmentation (posterior probability for occurrence probability decreasing towards north-east at least 0.95), we classified the positive response as reliable if the species occurs also north-east from the study area: northern Finland or north-western Russia.


Overall patterns of resource use and response to fragmentation

Based on our classification, 42% of the species grew mainly on deciduous tree species, but only 4 species were specialized in a single deciduous tree species (Table 3). A roughly similar proportion (36%) of the species grew on conifers, but more than half of these species were specialized in one host-tree species (spruce or pine). Relatively few species (22%) were classified as true generalists using both deciduous and coniferous wood. Regarding different resource types, none of the species showed a preference for man-made dead wood, but 20 species were specialized in natural dead wood and 12 of them were further specialized in natural logs (Table 3). The occurrence probability of almost all of the species increased with increasing diameter of the resource unit (Table 4). For species-specific results, see Appendix S2.

Table 3. Community-level patterns of resource use with respect to tree species (R2) and resource-unit types (R3). The values in the table give the number of species (proportion of all 119 species) classified to belong to each resource-use class
VariablesClassSpecies (%)
Tree-species preference (R2)Generalist26 (22)
Coniferous wood43 (36)
Any coniferous wood20 (17)
Spruce12 (10)
Pine11 (9)
Deciduous wood50 (42)
Any deciduous wood46 (39)
Aspen2 (2)
Birch1 (1)
Goat willow1 (1)
Resource-type preference (R3)Generalist99 (83)
Man-made dead wood0 (0)
Natural dead wood20 (17)
Any natural dead wood8 (7)
Natural logs12 (10)
Table 4. Responses of species to site-level and resource-unit level variables (Table 1)
  1. a

    Responses in brackets refer to cases with low statistical support (posterior probability < 95%).

S1Connectivity (small scale)2386118
S2Connectivity (intermediate scale)2743409
S3Connectivity (large scale)20383526
S5Canopy closure37313417
R6Size dependency142094

Regarding the site-level variables, the local amount of suitable resources (S1) had a positive effect on the occurrence per resource unit of 18 species (15%), while two species (1.7%) showed the opposite response (Table 4). Somewhat unexpectedly, only nine species (8%) benefited from high age of the surrounding forests (S2), while the occurrence probability decreased with this measure of connectivity for 27 species (23%). The data showed a high level of statistical support for 26 species (22%) being more likely to occur (per resource unit) in the less fragmented (north-eastern) side of the geographical gradient (S3). As the distributional ranges of all of these species extend to south-west from our study area (Appendix S2), we judged these species to be negatively affected by large-scale fragmentation. Twenty species (17%) were more likely to occur (per resource unit) in the more fragmented end of the gradient, but for three of these species, the apparent positive response to fragmentation is likely affected by climatic conditions (Appendix S2). A larger number of species preferred natural (20%) than managed (5%) forests, and somewhat surprisingly a clearly larger number of species preferred open (31%) than closed (14%) forests. For species-specific results, see Appendix S2.

Species characteristics and sensitivity to fragmentation

Species that suffered from fragmentation at the largest spatial scale (S3) occurred generally more often in sites which were well connected also at the landscape scale (S2) and especially in sites which had an abundance of resources at the local scale (S1) (Fig. 2). These fragmentation-sensitive species had several characteristics in common. They had a high level of overall ecological specialization (R7), being specialized in natural logs (R3l), large-diameter dead trees (R6), one host-tree species (pine or spruce) and preferred a particular stage of decay (R5), often an advanced one (R4). These species were rare also on their preferred resources (R1) and mainly occurred in closed (S5) forests.

Figure 2.

A two-dimensional non-metric multidimensional scaling ordination locating the 119 study species based on their resource use and responses to the site-level variables. The shape of the symbol indicates tree-species preference (R2), classified as generalist (square), deciduous-dweller (circle), conifer-dweller (diamond), spruce-specialist (triangle up) or pine-specialist (triangle down). The colour of the symbol indicates IUCN classification, classified as VU or NT (red) or LC (green). Contour lines show how the probability of a species being red-listed changes in the ordination space. The variables R1–R7 and S1–S5 that significantly affected the location of the species in the ordination space are shown with black and blue circles, respectively, whereas species characteristics (T1–T11 and T15; Table 2) with a significant relationship with the ordination configuration are shown with orange circles. Continuous variables are shown by vectors and factors by their centroids. For statistical tests, see Appendix S2. For R-variables, g = generalist, c = coniferous, d = deciduous, s = spruce, p = pine, n = any natural dead wood and l = natural log.

The location of the species in the ordination space (Fig. 2) was based solely on the field data and thus independent of the species attributes (T1–T15), including their red-listed classification. The VU and NT species were distinctively concentrated in the connectivity-dependent part of the ordination space and conspicuously separated from the LC species (Fig. 2, Appendix S2). Successor species (T3y) showed fragmentation-sensitive characteristics, while non-successor species (T3n) did not.

Host tree had a strong structuring effect on the ordination space (Fig. 2), deciduous-dwellers forming a distinct group and coniferous-dwellers (including specialists on spruce or pine) forming another group. In a separate analysis for coniferous-dwellers (n = 43), the species again formed two distinctive groups, one by the highly specialized, fragmentation-sensitive VU and NT species and the other by the generalist, non-fragmentation-sensitive LC species (Fig. 3). Neither deciduous-dwellers (n = 50) nor generalists (n = 26) responded strongly to connectivity, but the occurrence of the rare and specialized red-listed species depended on the presence of natural logs in advanced stages of decay (for deciduous-dwellers) or in a specific stage of decay (for generalists) (Fig. 3b,c, Appendix S2). The pattern of the successor species being more fragmentation-sensitive than non-successor species (Fig. 2) was present both for conifer-dwellers (Fig. 3a) and deciduous-dwellers (Fig. 3b).

Figure 3.

Patterns of resource use and site-level responses within each tree-species preference group. Panels a–c show the two-dimensional non-metric multidimensional scaling of Fig. 2 conducted separately for conifer-dweller (a), deciduous-dweller (b) and generalist species (c). Panel d shows the dependency of specificity in decay-class use (R5) on spore volume (T9). The regression lines correspond to tree-species preference groups. Main effects of species group (< 0.001) and slope (= 0.013) were significant. Non-significant interaction was excluded from the model.

Only few of the life-history characteristics (T1–T11) correlated strongly with the occurrence patterns in the field. One exception was spore size and shape, either of which was significant for the separate analysis conducted for deciduous-dwellers (Fig. 3b) and generalist species (Fig. 3c). Spore volume correlated negatively especially with the specificity in decay-class use so that within each tree-species group, species with small spores were more specific to decay class than species with large spores (Fig. 3d).

Similarity between species in their responses to resource-unit and site-level variables could be partly due to their phylogenetic affinity. In the case of deciduous-dwellers and generalists, the species belonging to the same genus (T12) or clade (T13) were located non-randomly in the ordination space (Appendix S2), and thus, these results should be interpreted with caution. Genus or clade was not significant in any other of the analyses presented here.

Effect size of fragmentation in terms of numbers of species

The ordination analyses strongly suggest that the LC species and the red-listed species differ from each other with respect to their responses to the site-level variables. This view is supported by Fig. 4 that shows the effect of the site-level variables in terms of predicted species richness in a hypothetical forest site with 200 dead trees. While the number of LC species is independent of the location along the large-scale gradient (S3), the predicted number of red-listed species is more than ten times higher in the less fragmented as compared with the highly fragmented end of the gradient. The predicted number of red-listed species is higher in sites that are well connected at the landscape level (S2), whereas the opposite is true for LC species. In both groups, but more so in red-listed species, the occurrence probability per resource unit increases with increasing local amount of resources (S1). Red-listed species but not LC species show a preference for natural forests (S4). Red-listed species preferred closed-canopy forests (S5), whereas LC species are more numerous in open forests.

Figure 4.

Predicted numbers of least concern (LC) species (a) and red-listed species (b) on 200 randomly selected dead trees. The dotted line corresponds to the predicted number of species (16.8 for LC species and 0.17 for red-listed species) when all the site-level variables are at their mean values. The black (grey) circles show the predicted value when the focal site-level variable (S1, S2, S3, S4 or S5) was set to its maximum (minimum) value among the 496 study sites, while the remaining site-level variables were kept at their mean values.


In this study, we have built hierarchical Bayesian models to relate the occurrence of wood-inhabiting fungi to resource-level and site-level covariates. Our main result is that red-listed species are generally highly specialized in their resource use and require well-connected natural forests, whereas many of the generalist species thrive especially well in fragmented surroundings. This suggests that highly specialized species are gradually lost from isolated fragments of natural forest (see also Penttilä et al. 2006; Berglund & Jonsson 2008; Laaksonen et al. 2008), which may leave more resources available for the competitively inferior generalist species (Marvier, Kareiva & Neubert 2004; see also Holmer, Renvall & Stenlid 1997). However, note that we measured the species performance per resource unit and thus even for the generalist species the overall effect of forest management and consequent fragmentation, including reduction in the total amount of dead wood, is likely to be negative. The influence of connectivity was more strongly seen at the largest (S3) and smallest (S1) scales than at the intermediate scale (S2), most likely due to the fact that the measure S2 is only a rough proxy for the landscape-level availability of dead wood. Further, low value of forest age (the measure S2) may actually mean high amount of resources for species that use deciduous wood or logging residues.

Out of the six characteristics associated with vulnerability to fragmentation in plants and animals (Henle et al. 2004; see Introduction), we found specialized habitat requirements and low natural abundance to be important predictors also for wood-inhabiting fungi. Both characteristics imply a low overall prevalence, which amplifies the effects of environmental and demographic stochasticities. We did not have data to quantify reproductive potential, the level of population fluctuations or the ability to persist over unfavourable environmental conditions, thus examining the relationship between these characteristics and fragmentation vulnerability remains a challenge for future work. To understand how fragmentation affects species through demographic processes, it would be highly important to bring information about the mycelial stage on fragmentation studies, but the necessary quantitative data are currently not available.

Dispersal ability is by no doubt important for determining the responses of a species to fragmentation. We used spore size and shape and the thickness of the spore cell wall as proxies for dispersal ability. Before discussing our results on the role of spore size, we note that in the size range of the spores of our focal species (the mean of length is 5.5 μm, range 3–18 μm), the terminal velocity is so low that it makes almost no difference to the aerodynamic properties of the spores (Kuparinen et al. 2007), and thus, it is not evident that species with small spores would be able to cover greater dispersal distances than species with large spores. Spore size and shape and the thickness of the spore cell wall may, however, affect dispersal ability through various processes, such as the rate at which the spores are deposited to the vegetation or ground (Petroff et al. 2008), the tolerance of the spores to UV-light (Mitakakis, O'Meara & Tovey 2003) or their likelihood of being carried by insect vectors (Tuno 1999). Further, it seems clear that there must be a trade-off between allocating the available resources to many small or few large spores. The other side of the trade-off is that spore size is proportional to the nutrient supply for the establishment phase.

We did not detect a strong relationship between connectivity dependence and spore size, but we found that specialized species are characterized by small spores (Fig. 3d). We envisage that as the resources for the highly specialized species are scarce in space and time, it is especially important for these species to be in the right place at the right time and thus maximize the spore production rate. High inoculum potential and thus a large spore size is likely to be critical especially for the generalist species that strive to establish on resource units of varying quality in competition with other species which may be more specialized in that particular resource. Spore shape is correlated with spore size, larger spores being more spherical (Kauserud, Colman & Ryvarden 2008), and hence, it may be that the significant effect of spore shape in Fig. 3c has no causal reason. For a given spore size, both spruce- and pine-specialists were more specific in their decay-class use than other conifer-dwellers, deciduous-dwellers or the generalist species. Thus, the fact that spore size was not significant in the ordination analysis of the coniferous-dwellers (Fig. 3a) is partly explained by systematic differences between the coniferous host-tree groups. We note that the result of specialized species being connectivity-dependent does not necessarily mean that these species would have particularly low dispersal ability. This is because the suitable resources are most sparsely distributed for the specialist species and thus – even with equal dispersal ability – the specialists are expected to be the first ones to fall below their extinction threshold in the course of habitat fragmentation (Hanski & Ovaskainen 2002).

Our results are consistent with several earlier studies that have shown that the number of spruce-associated red-listed fungal species increases with increasing local amount of dead wood and increasing proportion of old forest at the landscape and regional scales (Siitonen, Penttilä & Kotiranta 2001; Edman et al. 2004a,b; Penttilä, Siitonen & Kuusinen 2004; Penttilä et al. 2006; Hottola, Ovaskainen & Hanski 2009; Berglund et al. 2011). Also in studies of fungi on beech (Fagus sylvatica), the fraction of threatened species has been found to increase with increasing dead-wood continuity (Heilmann-Clausen & Christensen 2005) and with increasing connectivity of the forest landscape (Ódor et al. 2006). Regarding species richness, Stokland & Larsson (2011) reported lower numbers of species per spruce log (but not per pine log) in managed than in natural forests and this held particularly when examining the number of spruce-specialist species on large logs. While many of these earlier studies have been based on analyses of species richness, we have conducted our analyses at the level of the individual species, bringing information on how the community-level response to habitat fragmentation sums up from the species-specific responses.

Climatic or soil-fertility factors may partly explain the response to the large-scale fragmentation gradient for some of the 26 species that were more likely to occur (per resource unit) in the north-east and the 20 species that were more likely to occur (per resource unit) in the south-west. It has been suggested that the proportion of polypores of all wood-inhabiting basidiomycetes increases along continentality (Høiland & Bendiksen 1996; Heilmann-Clausen & Boddy 2008). Appelqvist (2008), however, argues that it is rather the perenniality of the fruit body than the polyporous life style that seems like an adaptation to continental climate. Perenniality was associated with the large-scale gradient also in our analysis for deciduous-dwellers. Nevertheless, we find it unlikely that the regional patterns in the present data are merely due to climatic factors, partly due to our attempt to control for the potential confounding effects, partly as resource availability and microclimate have been found to override macroclimate as predictors of the diversity of wood-decaying fungi (Bässler et al. 2010).

The results presented here gave independent support for the IUCN classification of wood-inhabiting fungi in Finland, as the classification was found to be very consistent with the patterns of resource use in the field. However, the few exceptions deserve a closer examination, as they may point out species that are currently classified as LC but should actually be red-listed species or vice versa. For example, Meruliopsis taxicola and Rhodonia placenta might deserve more concern than their LC status suggests, as their responses in the ordination space were very similar to those of the red-listed species. In contrast, Ceriporia excelsa (status NT) was surrounded in the ordination space for deciduous-dwellers by LC species. We note, however, that our results were based solely on snapshot data from southern Finland, whereas the criteria for the IUCN classification also include other considerations such as population trends that are not visible in our data.

Forest conservation strategies have been and still are in a phase of continuous change in Finland and other Nordic countries. In particular, the appropriate balance between the establishment of fully protected nature conservation areas and biodiversity-oriented management has been debated. While a priori one could expect that the efficient protection of threatened species could call for a complex combination of species-specific protection measures, our results draw actually a very simplistic picture. The clear majority of red-listed species occurred much more frequently (per resource unit) in well-connected, closed and natural forests with high amounts of local resources (Fig. 4). Thus, conservation of red-listed wood-inhabiting fungi is by far more ecologically effective through well-connected and high-quality conservation areas than if the same conservation effort would be distributed as small fragments across the landscape. However, we note that biodiversity-oriented management such as preservation of woodland key habitats and retention trees can be beneficial for part of the fungal community. For instance, woodland key habitats rich in deciduous wood may host many deciduous-dwellers (Junninen & Kouki 2006; Hottola & Siitonen 2008) and retention trees may host even some red-listed species (Junninen, Penttilä & Martikainen 2007), especially those adapted to dry and sun-exposed microclimate such as that following a stand-replacing disturbance (e.g. forest fire). However, for the vast majority of red-listed wood-inhabiting fungi, the mitigation of the detrimental effects of fragmentation calls for the protection of the remaining semi-natural forests, combined with ecological restoration of managed forests in the vicinity of the existing protected areas.


We thank O. Manninen, I. Eriksson, T. Kosonen, J. Kinnunen, T. Rämä, J. Kytömäki, T. Ala-Risku, M. Wikholm, M. Oksanen, M. Ikonen, S. Velmala, O. Miettinen, H. Murdoch, M. Lindgren, S. Piirainen, K. Savola, K. Silvennoinen, R. Tuominen and other field assistants who took part in the data collection and specimen identification. J. Mäkinen and T. Ylimartimo are thanked for their assistance during the fieldwork, J. Lehtomäki for his assistance with the ms-NFI files, A. Immonen for converting log size measurements into cubic meters, and A. Siika for help with Fig. 1. K. Korhonen and N. Hallenberg helped with classifying the sexuality of species, and I. Hanski, B. Nordén and two anonymous referees provided valuable comments on the manuscript. R.P. thanks his colleagues, R. Virkkala, P. Punttila and H. Kotiranta in the Finnish Environment Institute for collaboration during the project. This study was supported by the Academy of Finland (Grant no. 124242 to O.O. and Grant no. 137135 to J.N.), the European Research Council (ERC Starting Grant no. 205905 to O.O.), the Ella and Georg Ehrnrooth Foundation (grant to J.N.), the Finnish Society of Forest Science (grant to J.N.) and the Research Council of Norway (Grant 203808/E40 to K.-H. Larsson and J.N.). Data collection was funded by the Finnish Ministry of Agriculture and Forestry, the Finnish Ministry of Environment and the EU Forest Focus research programme.