The study was performed in an abandoned mine near the hamlet of Seinera, municipality of Bruzolo, Susa Valley, Cottian Alps, Italy [entrance at 1007 m asl; 7.201E, 45.136N (WGS84 reference system)]. We chose an artificial site rather than a natural cave due to its linear shape and low ceiling, allowing us to measure the environmental parameters with high accuracy and to access resident animals more easily (Mammola 2018). The Seinera mine represents a typical subterranean habitat at the epigean/hypogean ecotone (sensuProus et al. 2004). It has a horizontal length of only 22 m and the passage cross-section of about 2.5 × 2 m (Figure 1). The mine is in a mixed deciduous forest of chestnuts, oaks, maples and hornbeams, and opens in micaschist rocks used for talc extraction during the first half of the XX century.

Figure 1. 

Map of the study area and groundplan of the Seinera mine, with indication of sampling plots and dataloggers.

We used a sampling-square methodology to monitor the study site, as it has been shown to be an effective, non-invasive means for investigating the annual dynamics of invertebrates in both artificial and natural subterranean environments (Bourne 1976, Mammola and Isaia 2014, 2016).

Twenty-two sampling plots of 1 × 1 m were positioned from the entrance up to the end of the mine (Figure 1). We randomly distributed the sampling plots among the mine walls and roof (i.e. parietal habitats; Moseley 2009a). The mine floor was not investigated, because it was impossible to obtain a reliable visual census of the organism due to the presence of leaf litter and debris, and because the floor is not an elective microhabitat for the species considered in our analysis (Mammola and Isaia 2014, Mammola et al. 2016a; see section ‘activity pattern’).

We visited the study area once per season, between November 2016 and June 2017. In each season, the day of the sampling session was arbitrarily chosen to correspond to the day of new Moon closest to the solstice (summer, winter) or the equinox (autumn, spring). During each sampling session, we conducted one visit around 12:00 am, and we repeated the monitoring approximately four hours after sunset.

During each visit, we censused individuals of each invertebrate species occurring at each sampling plot. Species were identified in the field up to the lowest recognized taxonomic rank, given the reliability of their in-field identification and our taxonomical expertise. We grouped each species into one of the most common ecological categories (troglobiont, troglophile, trogloxene, accidental organisms) used in subterranean biology (Schiner-Racovitza’s categories; Sket 2008, Trajano and Carvalho 2017). Each taxon was attributed to one of these categories on the basis of the ecological information available in literature on its natural history (Lana 2001, Smithers 2005a, Isaia et al. 2011, Allegrucci et al. 2014, Mammola and Isaia 2014, Mammola et al. 2016a, 2016b, 2017). We used such information to calculate the abundance of troglophiles and trogloxenes for each sampling plot – no troglobionts were present in the study site.

We characterized the annual climatic conditions in correspondence of the entrance, 10 m inside and 20 m inside by three EL-USB-2 dataloggers (Lascar Electronics; accuracy of ± 0.5 °C for temperature and ±2.0% for relative humidity), placed at 1 meter height on the wall and programmed to record temperature and relative humidity every 12 h for the whole sampling period – one measurement at 12:00 am and one at 12:00 pm. During each survey, at the centre of each sampling plot we also measured: i) illuminance (lux; lx) using a photometric probe LP 471 Phot (Delta OHM S.r.l.; accuracy of 0.2%) pointed toward the entrance; ii) air flow velocity (WS; m/s) by a Testo 425 telescopic flow velocity/temperature probe (Testo SE and Co, KGaA; accuracy of ±0.03 m/s); and iii) temperature (T; °C) and relative humidity (RH; %) by EL-USB-2 dataloggers (Lascar Electronics; accuracy of ±0.5 °C for temperature and ±2.0 % for relative humidity).

In order to obtain an estimation of the activity patterns of the three troglophile species inhabiting the mine, during each survey we recorded the diurnal and/or nocturnal movements of the spiders Meta menardi (Latreille, 1804) (Araneae: Tetragnathidae) and Pimoa graphitica Mammola, Hormiga & Isaia, 2016 (Araneae: Pimoidae), and the cave-dwelling cricket Dolichopoda azami Saulcy, 1893 (Orthoptera: Rhaphidophoridae). These species were chosen for this analysis owing to i) their high identification reliability in the field (Mammola and Isaia 2014, Mammola et al. 2016a); ii) their high abundance within the study site (pers. obs.); and iii) their large body size – adults of centimetric length, making a visual monitoring in the field possible.

During each diurnal and nocturnal sampling session, we monitored the activity of each individual of the three species occurring within the sampling plots. The activity was expressed as the number of seconds while the animal was moving, using a stoppable chronometer during observation sessions of one minute. We considered any movement of the animals, with or without spatial displacements. During each session, we set the LED light of our speleological helmet to the red spectrum, in order to minimize disturbance to the animal – in accordance with the general chromatic visual spectrum of most arthropods (Briscoe and Chittka 2001). For each plot and each species, we calculated the total species activity, as the sum of the individual activities divided by the number of individuals within the plot.

All statistical analyses were performed in R (R development team 2017). Differences in the environmental conditions at the twilight zone between day and night and across seasons were evaluated graphically (Graphics and Lattice R packages; R development team 2017, Sarkar 2008) and by means of standard statistical metrics. Wherever appropriate, we tested statistical differences by means of factorial linear regressions (ANOVA) or generalized linear regressions (generalized linear models; GLMs).

To analyse the day–night differences in the abundance, and seasonality of the abundance of trogloxenes and troglophiles, we used a mixed-design analysis of variance with Poisson distributed data (generalized linear mixed models; GLMMs). For the activity pattern of the three species, we used linear mixed models (LMM). GLMM and LMM were fitted with the R packages lme4 (Bates et al. 2014) and nlme (Pinheiro et al. 2014), respectively. Regression-type analyses were conducted following the general protocol of Zuur and Ieno (2016).

Counts of trogloxenes, counts of troglophiles and total activity values for each plot represented the dependent variables. Environmental features (distance from the entrance, temperature, humidity, airflow and illuminance) and their relative interaction with the sampling period, either diurnal (day–night) or seasonal (winter, spring, autumn, summer), represented the independent covariates (i.e., explanatory variables). In order to capture potential non-linear trends in the response of the dependent variables, we allowed up to second order polynomial for the continuous independent variables, when appropriate. The mixed procedure accounted for multiple observations from the same sampling plot, by specifying the sampling plot within the seasonal sampling session as a random-intercept nested structure.

Prior to fitting the models, we explored the datasets following the standard protocol for data exploration proposed by Zuur et al. (2010). Indeed, the inclusion of outliers and highly correlated predictors in a regression analysis may lead to incorrect results (type I and II statistical errors). Thus, we used Cleveland dotplots to assess the presence of outliers in dependent and independent variables. We investigated multi-collinearity among continuous covariates via Pearson correlation tests (r) and variance inflation factors values (VIFs), setting the threshold for collinearity at r> |0.7| and VIF> 3.0. The collinearity between continuous and categorical variables was graphically evaluated with boxplots.

Once we fitted the initial models, including all covariates and interactions of interest, we applied model selection via backward elimination (Johnson and Omland 2004). Models were simplified by sequentially deleting covariates and/or interactions according to AICc values (Hurvich and Tsai 1989). The process was repeated until all remaining variables were statistically significant. In turn, validation plots were constructed using model residuals, and Poisson GLMMs were checked for over-dispersion.

During the day, illuminance ranged from 900 lx in the vicinity of the entrance, to 0 lx at the end of the mineshaft (mean±sd= 22.68±103.56). Illuminance was always null at night. Airflows ranged from 0 to 0.61 m/s (mean±sd= 0.08±0.48). There were no daily or seasonal variations in the intensity of the airflows (ANOVA; R²= 0.08, p= 0.26 n.s.).

The mean annual temperature at the entrance, at 10 m and at 20 m inside was comparable, however, values at the entrance showed higher seasonal variability (mean±s.d. 0 m= 7.19±5.92 °C; 10 m= 8.08±2.76 °C; 20 m= 8.09±1.24 °C). Overall, the amplitude of changes and min–max ranges were progressively attenuated with increasing distance from the entrance (Figure 2). There were no significant thermal variations between day and night inside the mine (ANOVA; R²= 0.12, p=0.17 n.s.). At the entrance (0 m), significant variations between day and night were observed in winter and spring (Figure 2). In particular, night temperature was significantly higher in winter (LM; Winter*Night: estimated β±SE= 0.595±0.318, p< 0.001), and significantly lower in spring (LM; Spring*Night: estimated β±SE= –0.708±0.318, p= 0.02), with respect to diurnal temperature.

Figure 2. 

Temperature variation in the study area. Data refer to record of temperature and relative humidity measured every 12 h (one measurement at midday and one at midnight). Top panel: annual trends of temperatures measured at the entrance (0 m; orange line) and inside the mine (10 and 20 m; purple and blue lines, respectively). Bottom panel: mean of monthly positive and negative temperature deviations at night, with respect to the daily temperature recorded during the same period.

Relative humidity ranged daily and seasonally between 70% and 100% (mean±sd= 88.85±5.65). Difference between day and night were more pronounced in winter and spring (Figure 3), with significantly higher humidity values at night (beta-GLM; Winter*Night, estimated β±SE= 0.550±0.133, p< 0.001; Spring*Night, estimated β±SE= 0.384±0.141, p= 0.006).

Figure 3. 

Boxplots showing the difference between relative humidity values during the day (white boxes) and at night (grey boxes) in the four seasons. Significant differences are highlighted by asterisks (Signif. codes: *** p<0.001, ** p<0.01).

The mineshaft hosted a diversified subterranean biocoenosis, including rich populations of arthropods typical of the twilight zone of Western Alpine caves (Table 1). Over the year, we found 27 taxa within the study area. The most abundant predators [Meta menardi, Metellina merianae (Scopoli, 1763), Pimoa graphitica, Dolichopoda azami, Tegenariacf. silvestris (Araneae: Agelenidae)] were recorded during all surveys (Table 1). Some taxa were exclusively recorded either during the day (e.g. geometrid moths) or at night (e.g. centipedes and millipedes). There were also seasonal variations in the animal community, with species found in either one [Eupolybothrus sp. (Lithobiomorpha: Lithobiidae), Callipus cf. foetidissimus (Nematophora: Callipodidae)] or more seasons [e.g. Troglohyphantes lucifer Isaia, Mammola & Pantini, 2017 (Araneae: Linyphiidae)].

Regression models were performed to identify the most important factors driving the abundance of both trogloxenes and troglophiles. Data exploration revealed that the variable temperature was collinear with the categorical variable season, and therefore it was not further considered. The variable light intensity and relative humidity were collinear with the categorical variable day–night, given that illuminance was always null at night and that the relative humidity higher at night (see Figure 3). Thus, these variables were not introduced in the regression analyses.

Best AICc models and model estimated parameters are reported in Table 2. With respect to the abundance of trogloxenes, there was a significant interaction between the day–night cycle and the seasonality (Figure 4). Overall, the abundance of trogloxenes within the mineshaft was significantly higher at night, in summer and autumn. In autumn, there was the highest discrepancy between day and night, with a higher abundance at night. With respect to autumn, in the other seasons the day–night differences were significantly lower (Table 2).

Figure 4. 

Interaction plot showing the effect of the interaction between seasonality and the day–night cycle on the abundance of trogloxenes.

Higher abundance of troglophiles were observed at night across all seasons (Table 2). Their abundance also varied seasonally with the distance from the entrance (Figure 5). In winter and spring, the highest abundance of trogophiles was predicted in the inner section of the mine, whereas during summertime closer to the entrance. In autumn, the predicted abundance of troglophiles peaked at intermediate distances (Figure 5).

Figure 5. 

Predicted values (filled lines) and 95% confidence intervals (dotted lines) of the effect of distance from the main entrance in interaction with the sampling season on the abundance of troglophiles derived from the generalized linear mixed model (GLMM). Day and night trends are shown.

Overall, we recorded the activity patterns of Pimoa graphitica 116-times, of Meta menardi 399-times and of Dolichopoda azami 86-times. The three species showed contrasting activity patterns. The total activity of P. graphitica on the web and of D. azami on the walls were extremely reduced both during the day (P. graphitica, day mean activity±sd= 0.01±0.11; D. azami= 0.06±0.55) and at night (P. graphitica, night mean activity±sd= 0.80±6.71; D. azami= 0.17±0.58). Although overall activity was slightly higher at night in both species, the variable activity was highly zero-inflated (over 90% of observations were zero), meaning that individuals were mostly inactive. It was thus not possible to construct meaningful regression models with the available data, even using specific statistical techniques designed to deal with zero inflation (zero-inflated regression model did not converged; Zuur et al. 2012).

Conversely, M. menardi was in general more active, enabling to fit a robust model for this species. Model selection procedure revealed that the best model supported by observation included only the categorical variable day-night (Table 2). The total activity of M. menardi was higher at night (night mean activity±sd= 2.56±5.60) than during the day (day mean activity±sd= 0.83±2.41), and this difference was statistically significant (p= 0.02). No significant differences in seasonal activity were detected, therefore the variable season was dropped during model selection.

The high nocturnal activity that we observed in Meta menardi can be explained considering the parallel higher nocturnal availability of potential prey items. Indeed, a relation between the presence of Meta spiders and the availability of prey was previously documented (Mammola and Isaia 2014, Manenti et al. 2015). The other species considered in this study, instead, were mostly inactive. This is not surprising in the case of Pimoa graphitica, which is regarded as a sit-and-wait hunter (Mammola et al. 2016a, 2016b). Conversely, our results do not meet our expectation for the cave-dwelling cricket Dolichopoda azami. In particular, given that the recorded activity pattern was extremely low, and abundance of individuals was similar during day and night, our data seemingly did not confirm the hypothesis that Dolichopoda crickets leaves the cave environment at night to forage outside (see, e.g., Di Russo et al. 1991, 1994, Carchini et al. 1994, Lana 2001), at least in this specific site.

No significant variations in the activity patterns were detected in respect to the other environmental predictors considered in this study. However, given that these results were obtained in uncontrolled environment, the picture obtained of activity patterns remains preliminary. Detailed experiments performed in laboratory conditions would be useful, specifically aimed to establish rhythms of activity during controlled light/dark cycles and to evaluate the circadian clock of troglophile species inhabiting the twilight zone.