The coastal Atlantic Forest is one of the most diverse, and most threatened, natural environments in the world [1]. Due to severe human impact over the last few centuries most of the primary coastal Atlantic Forest has been destroyed. Only 7% of its original extent remains [1] and a considerable part of remaining forest takes the form of secondary forest fragments at different stages of regeneration, embedded in a rural or urban matrix [2, 3]. In fragmented landscapes such as these, both the stage of regeneration and the degree of fragmentation influence forest structure within the fragments, which in turn determines habitat suitability and affects species occurrence as well as the composition of animal communities [4, 5]. Since secondary forests may play an important role in species conservation if primary forest habitats are limited [5], it is important to understand which factors determine species distribution in secondary forests.

Although small mammals play a key role in neotropical forest ecosystems, including the Atlantic Forest, in terms of seed dispersal [6–11], dispersal of mycorrhizal fungi [12] and predation on seedlings [7, 10, 13, 14], only a limited number of studies focused on the influence of vegetation structure on small mammal distribution in secondary Atlantic Forest fragments [15–17].

The effects of forest fragmentation on Atlantic Forest small mammals vary among species. Some species are able to persist in isolated fragments and cross or occupy non-native vegetation in human-altered areas, while others are restricted to large and/or connected forest remnants [16, 18–22]. Since vulnerability to fragmentation is influenced by the capacity to occupy the altered habitats of the matrix [19, 21, 23–25], it is expected that preferences on a microhabitat scale in forested habitats will vary among species showing different degrees of vulnerability to forest fragmentation on a larger scale.

The identification of the scale that best explains variation in presence or abundance of organisms is a major goal in ecology [26]. This implies the necessity of examining ecological patterns on different scales. The issue of the "correct" scale at which assessment of biological parameters should be carried out has been the topic of numerous studies in ecology [27–32]. In fact, in his review Jorgensen [32] called attention to the lack of consistent use of the appropriate scale in small mammal microhabitat research and that broad but widely accepted generalizations (e.g. habitat partitioning, relation between abundance and availability of microhabitat) might be based on results that often confound micro- and macro-habitat effects.

In a recent study on small mammals in Caucaia do Alto, São Paulo, Pardini et al. [16] showed that three rodent and two marsupial species were affected in different ways by fragmentation on a macro-scale. The rodents Akodon montensis Thomas, 1902, and Oligoryzomys nigripes (Olfers, 1818) as well as the marsupial Gracilinanus microtarsus (Wagner, 1842) were not affected by fragmentation while the rodent Delomys sublineatus (Thomas, 1903) and the marsupial Marmosops incanus (Lund, 1840) decreased in abundance in smaller and/or more isolated fragments. In the same region, Umetsu & Pardini [21] showed that the three species not affected by fragmentation were able to occupy anthropogenic habitats while the species vulnerable to fragmentation were restricted to native vegetation on a macro-habitat scale [16, 21].

In the present study we extended the previous work by adding a smaller scale to the investigation of small mammal habitat preferences in the same Atlantic Forest study area. While Umetsu and Pardini [21] studied the response of Atlantic Forest small mammal populations to habitat variation, this study deals with the response of small mammal individuals to micro-habitat differences. Specifically, we investigated the influence of vegetation structure on the micro-scale distribution of three small mammal species not affected by fragmentation (Akodon montensis, Oligoryzomys nigripes, Gracilinanus microtarsus) and two species vulnerable to fragmentation (Delomys sublineatus, Marmosops incanus) in six secondary forest remnants in the Atlantic Forest fragmented landscape of Caucaia do Alto. We investigated if the occurrence of small mammal species is influenced by vegetation structure, aiming to ascertain whether species with different degrees of vulnerability to forest fragmentation are associated with distinct vegetation characteristics.

This study examined the response of individuals from five small mammal species to vegetation structure on a micro-scale and represents an extension of former studies by Pardini et al. [16] and Umetsu and Pardini [21] which assessed the influence of a larger scale (habitat) variation in vegetation structure and fragment size on small mammal populations in the same study area.

The results presented here revealed differences in vegetation structure among fragments as well as within fragments among trap locations. The observed major micro-scale trends in vegetation structure (as synthesized by the three principal components) within the secondary forest fragments studied represented a gradient of forest succession/disturbance from old/undisturbed sites (higher canopy height, less dense understory, less horizontal structures, denser canopy cover) to initial/disturbed sites (lower canopy height, denser understory, more horizontal structures, less dense canopy cover), similar to that observed on a larger scale (among habitat fragments) in Caucaia, from fragments of native vegetation in initial stage of regeneration (shrub vegetation) to fragments of native forest [16, 21].

These micro-scale differences in vegetation structure are probably caused by the interaction of several factors. Several studies pointed out a variety of factors influencing forest succession [34, 35]. There are differences in colonization ability among plant species [35] which may lead to small scale differences in species composition and therefore secondary forest structure. Furthermore, abiotic and biotic conditions might differ between locations. For example, soil nutrients provide an important resource for colonizing plant species after forest clearing [35, 36] and are likely to vary due to pre-abandonment management. Furthermore, it has been hypothesized that differences in the occurrence of mycorrhizae might influence vegetation composition [37] and thus vegetation structure. Another cause of micro-scale variation in vegetation structure might be sporadic residual old-growth trees within the secondary vegetation [34] which can influence the surrounding habitat and also substantially increase canopy height at certain locations. Tree falls and the resulting gaps represent an additional source of small-scale variation within the forest [38]. Gaps in the canopy alter microclimatic conditions such as radiation, heat fluxes, wind, and humidity levels [39]. Other possible causes of variation are fire or cattle entering the forest. Although these events were not observed in the studied fragments during the study, they were relatively common in adjacent fragments and could have influenced studied fragments prior to data collection. Because not all parts of the trapping grids were located in equal distance from fragment edges, part of the variation might be due to edge effects. Because the proximity to edges influences several abiotic (micro-climate, windspeed, soil nutrients [24, 40]) and biotic (invasion of non-forest species, species composition [41]) aspects of the habitat, edge effects may have severe influences on forest structure [42, 43].

Although vegetation structure differed among studied fragments most species responses to micro-scale variation were independent of the fragment considered, i.e. were similar in all fragments. The exception was some of the responses of one species non-vulnerable to fragmentation (A. montensis), which were influenced by the fragment considered. Thus, irrespective of the variation in vegetation structure among fragments, the distribution of all species except from D. sublineatus was influenced by micro-scale variation (within fragment) in vegetation structure. A. montensis and G. microtarsus were associated to locations with less canopy cover, while O. nigripes preferred locations with lower canopy and denser understory. On the contrary, M. incanus was captured in locations with higher canopy, less dense understory and higher amount of horizontal structures, while the distribution of D. sublineatus was not influenced by within-fragment vegetation variation.

The response group which preferred younger/more disturbed forest sites (A. montensis, O. nigripes, G. microtarsus) included a marsupial and rodents, terrestrial (A. montensis [44]) and scansorial (O. nigripes, G. microtarsus [44–46]) species, as well as insectivores/frugivores (all species [47–49]) and granivores (rodents [10, 50]), making taxonomic or ecological traits such as locomotor abilities or diet preferences poor predictors of the observed micro-habitat preferences. Nonetheless, both population response to larger scale (habitat) variation and to fragmentation are consistent with the micro-habitat preferences of this group: species preferring younger/more disturbed sites within secondary forest fragments (A. montensis, O. nigripes and G. microtarsus) were found to be able to occupy anthropogenic habitats and to be equally abundant in smaller and larger forest fragments [16, 21].

On the contrary, the marsupial M. incanus, which preferred older/less disturbed sites inside secondary forest fragments was found to be restricted to patches of native vegetation and less common in initial stages of regeneration [21]. Although D. sublineatus was found to decrease in abundance in smaller and/or isolated forest fragments [16] and to be restricted to patches of native vegetation [21], this species presents similar abundance between patches of native vegetation in initial or latter stages of regeneration (22). Accordingly, we could not find any response of this species to micro-scale vegetation variation inside forest fragments.

Thus, our results indicate that patterns of habitat preference at the population level may be caused by individual responses at a smaller spatial scale (i.e., micro-habitat selection). Species whose individuals prefer younger/more disturbed sites within secondary forest fragments would be able to establish populations in degraded habitats at larger spatial scales, thus being less vulnerable to fragmentation in the long run. On the other hand, species that do not show this preference for disturbed sites or whose individuals are restricted to old/non disturbed sites inside secondary forest fragments would not be able to established populations in degraded open habitats at larger spatial scales, being more vulnerable to fragmentation in the long run.

Studies on the responses of Atlantic Forest small mammals to habitat variation that include different spatial scales are rare (but see [51]). To our knowledge only two studies on microhabitat selection of small mammals in the Atlantic forest biome included partly the same species as in this study [15, 17]. Fonseca and Robinson [15] investigated the selection of microhabitat from available habitat (measured as the mean of the variables obtained over all trapping stations) in primary and secondary semi-deciduous Atlantic forest for some small mammal species including M. incanus (all other species investigated by Fonseca and Robinson [15] were not subject to this study). They found no habitat selection in M. incanus, i.e. this species chose the microhabitat randomly from the available habitat. Dalmagro and Vieira [17] studied microhabitat of O. nigripes and A. montensis as well as of Delomys dorsalis, a congeneric of D. sublineatus, in Araucarian forest in south Brazil. Concurring with our results, they found that O. nigripes preferred areas in early successional stages within the forest. On the other hand, they also found that D. dorsalis preferred denser canopy cover and a positive correlation between canopy cover and the occurrence of A. montensis. Although it is difficult to directly compare these results given the heterogeneity in sampling design and data analysis among studies, these results may indicate that the response of the studied species to vegetation structure may vary among different types of forest in the Atlantic forest biome.

Results showed that major trends in micro-scale variation in vegetation structure within secondary forest fragments are similar to those observed at a larger spatial scale, i.e. among habitat fragments. This micro-scale variation in vegetation structure was shown to influence the distribution of most small mammal species in secondary forest fragments. Species that were shown to be able to establish populations in altered habitat and not to be vulnerable to fragmentation were clearly associated to younger/more disturbed sites within secondary forest remnants. On the contrary, species that were shown to be unable to establish populations in altered habitat and to be vulnerable to fragmentation preferred older/less disturbed sites or were not influenced by micro-scale variation in forest structure inside secondary forest fragments.

Thus the results, in combination with previous studies [16, 21], show that species responses are similar at different spatial scales of habitat and microhabitat, suggesting that microhabitat preferences may be an important factor influencing the capacity of small mammal to occupy altered habitats and, consequently, their vulnerability to forest fragmentation at a larger spatial scale.