Bacterial communities in larger islands have reduced temporal turnover

The underpinning method employed to construct a STR may affect the shape, scaling exponent (w), and fit of the STR power function. Here we used three differing approaches to construct STRs; including, what we term in this study, the ‘every possible window’ (EPW) [10], ‘cumulative moving window’ (CMW) [8], and ‘moving window’ (MW) approaches. The differences in each method are extensively detailed in the Material and Methods. The STRs for the bacterial communities within each of the tree-hole islands were plotted, of which all relationships were significant (Fig. 1 and Table S1). Overall, the resulting STR power law exponents (w) were found to range from 0.048 to 0.350 (Fig. 1) and were typically within the exponent ranges observed from meta-analyses of STRs for a wide range of animals, plants, and microbial communities [9,10,11]. However, these values varied by the approach used to construct STRs (Fig. 2A). The EPW based w values ranged from 0.048 to 0.128, with a mean w of 0.088 ± 0.029 (mean ± SD). The CMW w values ranged from 0.073 to 0.150, with a mean w = 0.111 ± 0.029. Whereas, the MW minimum and maximum w values were 0.223 ± 0.350, with a mean of 0.289 ± 0.044 (Fig. 2A). The EPW and CMW w values were significantly lower than the MW w values (Fig. 2A). However, they were not significantly different from each other, despite that EPW values were uniformly lower (Fig. 2A).

A key difference between the EPW and CMW approaches when compared to the MW is the former only used the first appearance of each bacterial taxon, even though taxa can emerge and disappear multiple times across a time series for a given tree-hole. Conversely, the MW approach does incorporate multiple immigrations and extinctions of the same taxa through time, which would be anticipated in a time series of this study’s extent. This would result in higher turnover, as observed here, and could therefore provide rational explanation for the higher residuals observed towards the latter portions of the MW STRs (Fig. 1C). However, to ascertain whether higher residuals are a general phenomenon or specific to the study system, and hence whether a power function is best to describe an empirical STR, would require wider comparative testing of the STR construction methods across a broad range of microbial communities from different habitats. Regardless of the STR construction method used, variation in w values across the tree-holes mirrored each other (Fig. 2A); all were significant and highly correlated between each approach (Fig. 2C–D). Previously, a fundamental reason for only using the first appearance of each taxon was due to the detection limits within the underpinning method used to survey microbial communities. For example, earlier work on STRs used fingerprinting methods, such as Denaturing Gradient Gel Electrophoresis, which were known to be affected by detection thresholds e.g., [8, 11,12,13]. In which, the disappearance and reappearance of a bacterial taxon could have been due to dropping below the detection threshold as opposed to going locally extinct. Whereas, this is now less of an issue with the high-throughput sequencing approaches and therefore a higher degree of confidence can be taken when assessing local immigration and extinction.

Next, we investigated the possible influence of time series length and sampling depth on the resulting temporal scaling exponents. Within the current study, sampling duration varied from a minimum of 280 days to a maximum of 364 days, mean = 339.8 ± 32 days. Likewise, mean sequencing depth varied across the tree-holes, with a minimum of 33893 and maximum of 47283 sequence reads, and mean of 41521.3 ± 4147.7 sequence reads. Subsequent analysis revealed STRs, regardless of construction method, were not related to variation in study duration or sequencing depth (Figure S1: EPW with time, R² = 0.16, F_1,18 = 1.58, P = 0.245; CMW with time, R² = 0.14, F_1,18 = 1.34, P = 0.281; and MW with time, R² = 0.28, F_1,18 = 3.12, P = 0.115; EPW with mean sequence depth, R² = 0.001, F_1,18 = 0.001, P = 0.953; CMW with mean sequence depth, R² = 0.01, F_1,18 = 0.001, P = 0.948 and MW with mean sequence depth, R² = 0.002, F_1,18 = 0.002, P = 0.902). Therefore, this would indicate that any influence based on sampling was negligible, as has been observed previously [9]. Although there is an observed regularity in the STR exponents, found here and more broadly, there remains variability within that range. It has been previously posited that finding patterns within that variability could provide a better understanding of the processes underpinning STRs, and hence temporal turnover within ecological communities [10].

For island-based communities, theory predicts that turnover rates should be inversely related to island size and to island isolation [4, 26]. For the latter, we presume that isolation from source of immigration for the tree-hole islands was approximately equivalent and hence would have limited effect on turnover rates; as the main routes of bacterial immigration would be via falling leaf matter predominately from the host tree, and rainwater running down the trunk of the tree from the canopy above. Before a relationship between turnover and island size could be tested, the best measure of island size had to be determined. Although island area has been the traditional measure [1, 4], volume has been found to be a better measure of island size in microbial studies based in aquatic habitats [2, 27, 28]. We tested this by plotting the SAR and species–volume relationship across the tree-hole communities, accounting for variation in area, volume, and richness over time, and found volume to be the better predictor (Fig. 3A&B). Using volume as the measure of island size, a significant inverse relationship with temporal turnover was observed (Fig. 3C). In agreement with theoretical predictions for island biogeography, we found that smaller islands had lower taxa richness, but, with higher turnover, whereas larger islands had greater richness with reduced turnover over time (Fig. 3B&C).

There is a need to understand the mechanisms and processes that underpin microbial community assembly, and which determine the spatial and temporal distributions of different microbial taxa [29, 30]. A plethora of potential ecological processes that could explain such patterns have been distilled into the influence and interplay between four basic processes; selection, dispersal, drift, and speciation [14]. This simplified framework of ecological processes has been strongly advocated as informative for microbial ecology [29, 30]. In brief, dispersal limitation, or selection, results from biotic and abiotic pressures causing minimal exchange of organisms between communities. Homogenizing dispersal is the degree to which individuals of species move between and successfully establish in local communities. Drift results from stochastic changes in population sizes, and speciation is the evolution of new species [14, 25].

Here the Raup and Crick probability-based index of similarity (S_RC) was used to test to what extent ecological processes (homogenizing dispersal, dispersal limitation/selection, and drift) accounted for temporal turnover within each tree-hole community [24]. This index has long been used to examine the influence of deterministic and stochastic factors on community assembly e.g., [8, 24, 31]. It has been noted that some ecological processes of community assembly [14] can map easily onto this deterministic-stochastic framework [32]. To that end, we used S_RC values that were deterministically similar (S_RC ≥ 0.95) or dissimilar (S_RC ≤ −0.95) than expected by chance to infer the deterministic processes of homogenising dispersal and dispersal limitation (selection), respectively [32]. In addition, we used S_RC values that indicated similarity no greater than expected by chance (S_RC > −0.95 and <0.95) to infer the stochastic process of drift [24, 25].

Strikingly, similar patterns of pairwise S_RC frequencies were observed within all of the tree-hole communities (Fig. 4A). Communities were mainly characterized by the deterministic process of homogenizing dispersal (mean S_RC = 96.1, SD ± 2.5%), and to a lesser extent by drift (3.3 ± 2.0%) and then dispersal limitation (0.6 ± 1.4%). Speciation is not explicitly accounted for using the S_RC index, but can cause differences in diversity among sets of communities that do not exchange through dispersal [14]. Therefore, speciation should have negligible influence within a set of communities where individuals disperse among local communities within a metacommunity, as was the case in this study [25]. Moreover, here speciation should be negligible given the survey’s timeframe and the method used to define taxa (i.e., 16 S rRNA amplicon sequencing and grouping in 97% OTUs).

The low frequency of dispersal limitation and high homogenizing dispersal indicated that island isolation was unimportant, as postulated earlier (Fig. 4A). Therefore, turnover was related to island size and not island isolation within the study system. We posit that the large degree of homogenizing dispersal was due to the negligible effect of isolation from the immigration sources and the relatively small distances between islands in this tree-hole archipelago, with the minimum and maximum distance between any two adjoining tree-holes was 12.5 and 217 cm, respectively (mean 59.7 ± 62.9 cm). In support of this view, a recent study experimentally demonstrated dispersal homogenized bacterial communities via immigration, and not through weakening selection, within a metacommunity at small spatial scales [33].

To examine the influence of island size further, we explored associations with ecological processes (Fig. 4B&C). We found a significant inverse relationship between drift and island size (Fig. 4B) and, conversely, a significant positive relationship between homogenizing dispersal and island size (Fig. 4C). Theory predicts that species turnover will be higher on smaller islands than on larger islands, as was the case in the current study (Fig. 3C), [1, 4, 26]. This can be due to islands of decreasing size having an increased probability of chance events [1, 4]. Such stochastic events include death, reproduction, and migration, which underpin the stochastic-based process of drift [14, 24]. Hence, we posit this may explain the observation of drift significantly increasing with decreasing island size in the current study. Moreover, in the same spirit, others have proposed that with smaller island sizes, there are inherent smaller population sizes, thus making it harder for homogenizing dispersal to effectively increase population size and counterbalance drift [34, 35]. As such, this could help better explain observed patterns and highlight dispersal and drift are inversely linked [35]. That could be determined in future work, which manipulates dispersal rates, population sizes, and island sizes in experimental tree-holes.

Here we provide, for the first time, determination of the underlying mechanisms and processes that can underpin STRs (regardless of STR construction approach), and hence temporal turnover, within island-based bacterial communities. We found that island size was important in influencing turnover within the bacterial communities studied. Moreover, homogenizing dispersal was the dominant ecological process driving community turnover within the study system. That observation, coupled with a negligible influence of dispersal limitation, was also the result of the lack of island isolation from the immigration source in this study system. Future studies could determine which processes underpin temporal turnover when variation in island isolation is experimentally controlled and manipulated. While the SAR has been pivotal in the development of ecological models and theories, we are optimistic that the findings from the current study will provide new impetus in realizing the potential of its understudied and less understood temporal analogue, the STR.

Source: Ecology - nature.com