We used published data of understory vegetation of the hop-hornbeam forest in the central Apennines (central Italy), in the hilly sectors of the Umbria-Marche Apennines (Marche Region) (Tardella et al. 2019). The bedrock is calcareous and climatically the area belongs to the transition zone between the Mediterranean and Temperate climate zones, defined as the sub-Mediterranean climate (Pesaresi et al. 2017). The mean annual rainfall ranges between 900 and 1,100 mm and the mean annual temperature is between 12 and 13 °C (Tardella et al. 2019). The landscape is dominated by hop-hornbeam (Ostrya carpinifolia) forests coexisting mainly with Fraxinus ornus subps. ornus, Acer opalus subsp. obtusatum, and Quercus cerris. These forests are one of the most widespread forest types in the Central Apennines (Blasi 2010; Casavecchia et al. 2021) and they have been managed mainly as coppice-with-standards. This management practice consists of the logging of young shoots on short rotation favoring vegetative re-sprouting of new shoots from dormant buds on the cut stumps, while a variable number of trees is left uncut (hereafter “standards”), to ensure seed production and prevent soil erosion. However, after the rural exodus started at the end of World War II, this forest management was progressively abandoned (Ferrara et al. 2021). Thus, while the younger forest stands were still in their rotation period during the vegetation survey, the older forest stands are over their turn (>40 years for the regional law of the Marche Region).

We extracted vegetation data on species distribution from Tardella et al. (2019). Specifically, the sampling design was based on a random stratified approach to select plots in similar environmental conditions in terms of bedrock composition (limestone), elevation (between 600–950 m a.s.l.), slope (20°–40°), and aspect (from north-west to north-east) (Table 1). In total, 38 vegetation plots (20 m × 20 m) were selected, partitioned into 19 plots in younger forest stands and 19 plots in older forest stands. Vegetation data consists of visually estimated percent cover values of forest-floor species inside each plot. The largest proportion of plots was located in private areas. More detailed information on sampling design and data collection is present in Tardella et al. (2019).

We selected a set of plant traits capturing a wide spectrum of plant functional variation of forest understory species (Burton et al. 2020; Chelli et al. 2024b; Table 2), specifically, the specific leaf area (SLA) which captures the leaf economic spectrum, the plant height (H) for the plant size spectrum and the seed mass (SM) for sexual reproduction and dispersal ability. These traits made up the LHS scheme of Westoby (1998). In addition, we selected lateral spread (LS), the number of clonal offspring (CO), and the persistence of clonal growth organs (PCGO). These three clonal traits capture different functional dimensions that have received less attention, such as space occupancy, resource foraging and sharing, and ability to recover after physical damage, all factors that affect plant persistence (Klimešová et al. 2016).

We retrieved data on LHS from the LEDA database (Kleyer et al. 2008) and Campetella et al. (2020) publication, whereas clonal traits were retrieved only from the CLO-PLA3 database (Klimešová et al. 2017) (Table 2). We focused the analysis only on herbaceous understory and we removed shrubs, tree seedlings, and saplings since trait values of mature individuals obtained from databases and publications assigned to young shrubs or tree seedlings and saplings would overestimate their functional role. Therefore, in this study, the understory layer consisted only of herbaceous forest species. Values of plant traits were available for all those species whose relative cumulative cover reached at least 95% of the total cover of all species for at least one trait (Suppl. material 1: table S1) (Májeková et al. 2016). No significant relationships (p < 0.05) were detected between traits (Suppl. material 1: fig. S1).

We generated a phylogenetic tree using the most inclusive and updated phylogeny for vascular plants (Smith and Brown 2018). We adopted “Scenario 1” because it is the most cautious and avoids random solutions by adding genera or species as basal polytomies within families or genera (Jin and Qian 2019). Species nomenclature was standardized according to The Plant List (http:/­/­www.­theplantlist.­org/­) before building the phylogenetic tree (Smith and Brown 2018). The phylogenetic tree was created with the V.PhyloMaker function in the V.PhyloMaker package (Jin and Qian 2019) in the R environment.

Phylogenetic and functional diversity are not necessarily independent from each other since a community characterized by similar functional species could be the consequence of phylogenetic clustering. In such cases, a strong association between traits and phylogeny due to underlying trait evolution (i.e., phylogenetic signals or trait conservatism), can lead to misleading interpretation since phylogenetically clustered species may still exhibit substantial functional diversity that is not captured by phylogenetic structure alone (de Bello et al. 2017). The phylogenetic signal was assessed by performing a Mantel correlation test between the phylogenetic dissimilarity matrix and the functional dissimilarity matrix calculated with multiple traits (Jucker et al. 2013). The phylogenetic dissimilarity matrix was obtained from the phylogenetic tree using the cophenetic distance. Values of cophenetic distance were then square-rooted and scaled in a range of 0–1, with 0 indicating the closest related species and 1 the furthest species in the phylogenetic tree (de Bello et al. 2017). The functional dissimilarity matrix was obtained using the Gower distance on species trait values. Gower distance standardizes the functional species distance values in a range of 0–1 (with 0 if two species have the same trait values, and 1 if two species have completely different trait values). Moreover, Gower distance handles missing values and therefore it is suitable for calculating a functional dissimilarity based on multiple traits (Pavoine et al. 2009). Since some species had missing values for certain plant traits, and Gower distance requires at least one shared trait without missing values between species, we excluded 9 out of 103 species that lacked information on plant height (the trait with the fewest missing values across species). Then, the significance of the Mantel correlation test was assessed by comparing observed values of the Mantel statistic to a random distribution generated through 999 permutations of the rows and columns of the functional dissimilarity matrix (Legendre and Legendre 2012). A more positive correlation than expected by chance indicates trait divergence, conversely a more negative correlation than expected by chance indicates trait conservatism (Jucker et al. 2013). Traits did not result phylogenetically conserved (r = 0.01; p-value = 0.35), therefore, we did not need to decouple trait information from phylogeny (de Bello et al. 2017).

To calculate cophenetic distance we used the cophenetic function in the picante package. We used the rescale function in the scales package to rescale cophenetic distance values in a range of 0–1. Gower distance was calculated with the gowdis function in the FD package. A Mantel correlation test was performed with the mantel function in the vegan package.

All the analyses were done in the R environment (R Foundation for Statistical Computing, Vienna, Austria; http://www.R-project.org).

We analyzed the species composition change over time by running a Non-metric Multidimensional Scaling (NMDS) for the two groups of stands (20–25 years and 40–45 years since the last coppices). Before running NMDS, we log(x+1) transformed cover data, and we calculated a distance matrix using the Bray-Curtis distance. Then, we square-rooted the Bray-Curtis distance matrix to have a distance with Euclidean properties and finally, we ran the NMDS (with 3 dimensions). With the same sqrt-Bray-Curtis dissimilarity matrix we tested whether the two groups of stands have i) different extents in beta diversity by performing multivariate homogeneity of groups dispersion (variances; Anderson et al. 2006); and ii) the amount of distinctiveness running analysis of similarity (ANOSIM). For ANOSIM, R-values close to 1 indicate highly dissimilar groups, while R-values close to 0 identify highly similar groups (Clarke 1993). Both analyses were run with 999 permutations to assess their significance. NMDS, multivariate homogeneity of groups dispersion, and ANOSIM were run with NMDS, betadisper, and ANOSIM functions in the vegan package.

Then, we investigated how the two understory plant communities differ in terms of social behavior type (SBT). Specifically, for each species we assigned an SBT from the European forest vascular plant species list (Heinken et al. 2022): i) species of forest interiors (SBT 1.1) – hereafter “forest specialist species”; ii) species of forest edges and forest openings (SBT 1.2) – hereafter “gap species”; iii) species that can be found in the forest as well as open vegetation (SBT 2.1) – hereafter “forest generalist species”; iv) species that can be found partly in the forest, but mainly in open vegetation (SBT 2.2) – hereafter “marginal species”; and v) species typical for non-forest vegetation (SBT 0) – hereafter “non-forest species”. Since the list for Italy has not been published yet, we have used the list for the French mountains. This region is close to Italy in terms of biogeography. For species missing from the lists, we assigned the SBT by consulting the national flora (Pignatti et al. 2017–2019). Then, we quantified the relative frequencies (expressed as CWM) of each SBT class (Ricotta and Moretti 2011), and we compared them with time since the last coppicing event using a t-test.

We calculated taxonomic diversity (TD), functional diversity (FD), and phylogenetic diversity (PD) using Rao’s Quadratic Entropy (Q). We selected Rao’s Q because it provides a common methodological framework that efficiently synthetizes the different facets of diversity (de Bello et al. 2010). Rao’s Quadratic Entropy expresses the expected dissimilarity between two individuals of a given assemblage randomly selected with replacement:

Q=∑i,jsdijpipj (1)

where S is the number of species, d_ij is the distance or dissimilarity between the i-th and j-th species, and p_i and p_j are the relative covers of i-th or j-th species in the sampling unit. For FD and PD, we used the functional and phylogenetic dissimilarity matrices used for the Mantel test (see above). For TD, the species distance can assume only two values: 1 for all i ≠ j and 0 for all i = j. In this context, TD consists of the well-known Simpson index of dominance (D = ∑^S_{i= 1}p²_i) and it represents the upper limit that FD and PD may achieve. However, to remove the influence of species composition on FD and PD indices, and to shed light on assembly rules, we used the null-model approach in which observed functional and phylogenetic diversity values were compared with a random distribution of expected values (Götzenberger et al. 2016; de Bello et al. 2017). Expected values were generated by shuffling all species traits together for FD and by shuffling species’ distance in the phylogenetic distance matrix. We calculated the standardized effect size (SES) for both FD and PD as follows:

SES = (I_obs − I_sim)/σ_sim (2)

where I_obs is the observed value of the index, I_sim is the mean of the expected index, and σ_sim is the standard deviation of the expected index. Then, we assessed whether the distribution of SES values for both FD and PD was significantly different from zero using a two-tailed t-test. Significant distribution of positive SES values (>0) indicates higher observed values than expected (i.e., “trait or phylogenetic divergence”), while significant distribution of negative values (<0) indicates lower observed values than expected (i.e., “trait or phylogenetic convergence”). Values close to zero indicate a random assembly pattern (de Bello et al. 2017). Thus, variation of TD, SES-FD, and SES-PD about time since the last coppicing event was quantified using a t-test. Time since the last coppicing event was treated in the model as a categorical variable with two levels: younger and older forest stands. We are aware that pooling together all traits in a multi-FD index can mask finer functional patterns, thus, we also calculated the SES-FD for each single trait, i.e., plant height (SES-FD_H), specific leaf area (SES-FD_SLA), seed mass (SES-FD_SM), clonal offspring (SES-FD_CO), lateral spread (SES-FD_LS) and persistence of clonal growth organs (SES-FD_PCGO) and we tested if their values changed across the two forest systems. To calculate them, we shuffled species traits independently.

Taxonomic diversity (TD), functional diversity (FD), and phylogenetic diversity (PD) at the plot level were calculated with the RaoRel function in the cati package. A two-tailed t-test was performed using the t-test function in the stat base package. A list of references for each package is reported in Suppl. material 1: table S2.

Theoretically, under closed forest stands TD should be greater because of the presence of different micro-habitats preventing the establishment of few dominant species (Chelli et al. 2023). However, we did not find any significant variation. This can be attributed to the different trajectory of TD since the last disturbance event (Scolastri et al. 2017), or to the length of our regeneration gradient that might not be broad enough to capture it. According to the literature, TD should grow after logging, but during the succession (especially in the first 25–30 years after logging), it should decline due to the progressive tree canopy closure (Roberts and Gilliam 1995; Howard and Lee 2003; Catorci et al. 2011; Chelli et al. 2024a). Eventually, TD rebounds when the forest reaches an old-growth stage, following a U-shaped trajectory (Hilmers et al. 2018). In our study, the older forest stands cannot yet be classified as old-growth, as they are only a few decades old. Consequently, our sampling likely captured two points along the plateau of the U-shaped curve, explaining the absence of significant differences in TD between the two forest stands.

Interestingly, despite the stability in TD, we observed significant species turnover, with younger and older stands exhibiting distinct species compositions. This variation is primarily driven by the greater proportion of gap species (such as Melittis melissophyllum and Viola alba) in older forest stands. The presence of gap species highlights the existence of microenvironmental gradients within these forests, such as variations in light availability, soil moisture, and nutrient distribution created by canopy gaps. These microhabitats provide niches that support a broader range of species, thereby contributing to higher beta diversity in older stands compared to younger ones.

Functional patterns shifted from convergence in early regeneration stages to divergence patterns in later ones (Backhaus et al. 2021; Csecserits et al. 2021). In the younger forest stands, the progressive canopy cover closure after logging selects a shaded flora mainly composed of earlier-regeneration species (Bartha et al. 2008; Catorci et al. 2011, 2012). After canopy closure, the increase of micro-environmental conditions allows different functional species to colonize different niches (Closset-Kopp et al. 2019; Vanneste et al. 2019; Bricca et al. 2023). As such older stands have a certain degree of opening in the tree canopy as pointed out also by the higher presence of gap species (Tardella et al. 2019). Overall, older stands are characterized by species displaying a taller size, higher photosynthetic ability, and larger seeds, all functional adaptations that resemble those of species occurring in more mature stands (Vanneste et al. 2019; Blondeel et al. 2020). On the contrary, the higher number of clone offspring having short-lived connections to the maternal individual is unexpected (Bricca et al. 2023). Probably, different forest types (e.g., hop-hornbeam vs. beech forest) filter species exhibiting specific clonal functional strategies.

When considering single traits, we did not find consistency with the trend depicted by SES-FD_Multi. Specifically, we found either a weakening of functional convergence (i.e., SM, PCGO) or a shift from convergence to a random pattern (i.e., H, SLA, LS). Only for CO, we found a shift from a random pattern to functional divergence. Thus, variation of the SES-FD_Multi is probably mainly driven by the CO pattern. Since the pattern of SES-FD_Multi may mask the functional pattern of different single traits, this reinforces the consideration that plant traits should be evaluated singly (Backhaus et al. 2021; Csecserits et al. 2021). Specifically, forest investigation should not be restricted to the selection of a few aboveground traits (LHS scheme) but should also include clonal traits, as they represent fundamental strategies of species coexistence in such an environment (Bricca et al. 2023).

In general, phylogenetically clustered plant communities are typical of early regeneration stages, i.e., more disturbed environments, whereas higher phylogenetic divergence tends to characterize later successional stages with reduced disturbance (Letcher et al. 2012). This pattern is based on the assumption that disturbance filters species according to their functional traits (Zhang et al. 2014). However, this assumes a significant phylogenetic signal in key traits, an assumption not supported by our data.

Moreover, most phylogenetic studies have focused on tropical forests (e.g., Vamosi et al. 2009; Letcher et al. 2012), whereas investigations in old-growth temperate forests have revealed more variable patterns (Ottaviani et al. 2019; Closset-Kopp et al. 2019; Roy et al. 2021). These inconsistencies may stem from differences in study design, e.g., the inclusion of gymnosperms and cryptogams, the use of stand age or basal area as successional indicators, or the choice of phylogenetic metrics.

In our case, we observed a clear pattern of increasing phylogenetic convergence. This suggests a consistent filtering effect exerted by mature forest conditions. However, this pattern may also reflect the influence of anthropogenic disturbance in fostering phylogenetic diversity, particularly in earlier successional stages. In younger forests, the most abundant species belong to six families (Cyperaceae, Juncaceae, Lamiaceae, Poaceae, Primulaceae, and Ranunculaceae), some of which are also associated with grassland habitats. The lower phylogenetic diversity observed in older forests may indicate the absence of a “ghost of competition past” effect (Connell 1980), which would typically promote phylogenetic divergence through biotic interactions such as competitive exclusion (Violle et al. 2011). Instead, our results suggest that abiotic filtering plays a stronger role in shaping community composition. Alternatively, the relatively recent origin of Central Italy’s mountain forests (Magri et al. 2006) may have constrained long-term evolutionary diversification. As a result, only a limited number of closely related lineages may have successfully adapted to more mature forests, unlike tropical forests that have experienced long-term stability and species diversification (Lepš 2012). These hypotheses are not mutually exclusive, and together they may help explain the trend toward increasing phylogenetic convergence in our temperate understory communities. Nonetheless, the limited number of phylogenetic diversity studies in temperate understories continues to constrain our ability to generalize these findings.