Small mammals and microhabitat selection in forest fragments in the transition zone between Atlantic Forest and Pampa biome

. Natural resources are depleted in fragmented landscapes that have their vegetation also altered. As a result, the microhabitat diversity and the composition and distribution of local species are affected. In this study, we evaluated the small mammals’ community diversity, composition and microhabitat selection in two Atlantic Forest fragments, in an ecotone area with the Pampa biome, southern Brazil. We recorded five rodents ( Akodon montensis, Oligoryzomys nigripes, Sooretamys angouya, Juliomys pictipes and the exotic Rattus rattus ) and one marsupial ( Didelphis albiventris ). Both fragments were dominated by the generalist rodent A. montensis. Akodon montensis and O. nigripes showed similar habitat preferences: ground covered by rocks and higher values of vegetation obstruction. Sooretamys angouya preferred places with higher abundance of trees. Fruit availability was important for A. montensis and D. albiventris, highlighting the importance of this food resource for local wildlife, and the potential role of these species as seed predators and dispersers. Small species richness, the presence of an exotic species and high dominance level suggest that the study area is highly degraded.


INTRODUCTION
The Atlantic Forest is a complex biome with several vegetational formations, which includes a significant part of Neotropical biodiversity (Tabarelli et al., 2005;SOS Mata Atlântica & INPE, 2011;Leitman et al., 2015).This high diversity is partly due to its great latitudinal extension, from 4°S to 32°S latitude, the widest latitudinal gradient of a tropical forest in the world (Ribeiro et al., 2009).It is also one of the most threatened tropical forests, with agricultural and urban developments being the leading causes of deforestation and fragmentation (Ribeiro et al., 2009;Lira et al., 2012).Fragmentation of natural habitats can promote a reduction in species diversity through the extinction of specialist species (Pardini et al., 2010;Bregman et al., 2014;Matthews et al., 2014).Atlantic Forest fragmentation and its effects on biodiversity have been intensively studied in the last years, mainly in the northeast (e.g., Lôbo et al., 2011;Leal et al., 2012;Santo-Silva et al., 2016;Filgueiras et al., 2019) and the southeast of Brazil (e.g., Umetsu & Pardini, 2007;Vieira et al., 2009;Almeida-Gomes & Rocha, 2014;Almeida-Gomes et al., 2019).The southern Atlantic Forest, at the border with the Pampa biome, remains the least studied region of this biome regarding fragmentation studies.
Human occupation of the Atlantic Forest also caused structural alteration of the remnants, which could lead to some changes at the microhabitat scale (Chazdon, 2003).Such scale is complex in Neotropical forests, such as the Atlantic Forest, once they show a high variety of vegetation structure (Richards, 1996).Microhabitat has been described as environmental variables that affect the species behavior from an individual per-spective and determine portions of the home range that are more intensively used (Morris, 1987;Warrick et al., 1998;Akers et al., 2013;Schirmer et al., 2019).It's use has been specially related to resource availability (e.g., Hodara & Busch, 2010;Pinotti et al., 2011;Sponchiado et al., 2012;Corrêa et al., 2018) and protection against predators (e.g., Lima et al., 2010;Melo et al., 2013;Law et al., 2018;Bajaru et al., 2019).Highlighting the importance of microhabitat characteristics in forest fragments, Delciellos et al. (2016) found that the quality of vegetation structure had a comparative effect over small mammal richness and composition with fragment isolation and climate seasonality.
Small mammals play key ecological roles in the forest ecosystems, serving as seed dispersers and predators (Bricker et al., 2010;Grenha et al., 2010), insects predators (Kaunisto et al., 2012) and important prey for terrestrial vertebrates (Wang, 2002;Bianchi et al., 2010).Studies on ecology and community structure of small mammal species have the potential to answer important questions related to forest dynamics since these species respond directly to local and regional changes in habitat (Castro & Fernandez, 2004;Hodara & Busch, 2010;Delciellos et al., 2018).In general, specialist species are negatively influenced by habitat loss and alteration whereas generalist species with a wider geographical range are positively influenced, or unaffected by these processes (Pardini et al., 2010;Püttker et al., 2013).Approximately 320 small mammals species have been recorded in Brazil (Quintela et al., 2020).Even though Atlantic Forest is the Brazilian biome with the highest number of studies regarding mammal species (Brito et al., 2009), there is still a significant knowledge gap on species composition in some areas, such as the southern Atlantic Forest, at the very border with the grassland dominated Pampa biome.
In this study, we investigated species richness, abundance, and composition of a small mammal community in two Atlantic Forest fragments and evaluated the microhabitat selection by the most abundant species.Firstly, we hypothesized the existence of a poor community dominated by common and generalist species, because the sampled fragments are medium to small-sized and the general landscape is poorly forested.Secondly, we hypothesized that small mammals' abundance would be related to fruit availability, an important resource for these species.Additionally, we considered that small mammals would be more abundant in areas where the vegetation structure could provide protection against predators, such as vegetation obstruction.

Trapping procedures
In each fragment, we sampled small mammals for ten nights per season from Autumn 2015 to Spring 2016.We set three 120 m long transects in each fragment, each with seven trapping stations 20 m apart and at least 25 m from the fragment edge.We used Sherman (31 × 10 × 08 cm) and Tomahawk (45 × 17 × 17 cm) livetraps.In each trapping station, we placed two live-traps -one on the ground and one in the understory (approximately at 1.5 m high) -totalizing 42 live-traps per fragment.Each trap was baited with a mixture of peanut butter, banana, corn flour, sardine, and commercial cod-liver oil.This study was authorized by IBAMA -Brazilian Institute of Environment and Natural Resources (authorization number 469472) and the ethics procedures were authorized by the UFFS Animal Ethics Committee (authorization number 009/CEUA/UFFS/2015).

Species identification
A total of 32 specimens were field-collected and deposited in the Universidade Luterana do Brasil (ULBRA), Museu de Ciências Naturais, Laboratório de Sistemática de Mamíferos (see Table S1).An ear plug was sampled from each voucher, which received a numbered tag and was released in the same capture point.The taxonomic identity of specimens was determined either based on morphological analyses of vouchers, compared with museum specimens from ULBRA and published morphological data (Bonvicino et al., 2008) or DNA.
Total DNA was extracted from muscle tissue (ear plug) preserved in ethanol 100% from 26 vouchers, using the PureLink Genomic DNA kit (Invitrogen), following manufacturer protocols.A fragment of 801 base pairs of the mitochondrial DNA (mtDNA) gene cytochrome b (Cytb) was amplified with primers MVZ05 and MVZ16 and conditions described by Smith & Patton (1993).PCR products were cleaned using Exonuclease I and Thermosensitive Alkaline Phosphatase (FastAP; Thermo Scientific) and purified amplicons were directly sequenced (forward strand) in an ABI 3730xl genetic analyzer by Macrogen (Republic of South Korea).Chromatograms were edited and aligned using Geneious v.9.1.8(Biomatters, available at https://www.geneious.com;Kearse et al., 2012).Using the Local Alignment Search Tool (BLASTN 2.10.1+)(Zhang et al., 2000), the resulting sequences were compared to those in the public database Genbank (NCBI: https://blast.ncbi.nlm.nih.gov) to identify species matches based on sequence similarity.We considered an identity value between 98.5 and 100% as a reliable match for a species (see Table S2).

Microhabitat variables
To understand species associations with microhabitat, we measured 13 variables at each trapping station (Freitas et al., 2002;Vieira et al., 2005).Ground cover in the trapping station was measured with a 0.25 m² horizontally-placed grid divided into 100 equal parts (Freitas et al., 2002).We estimated the percentage of grid cells covered by (Variable 1) leaf litter, (V2) rocks, (V3) herbaceous vegetation, and (V4) bare soil.We repeated this measure in each cardinal direction around each trapping station and averaged across the four directions to characterize the ground cover at each trapping station.Ground cover variables showed high correlation values as they were measured as proportions of the same grid (Hair et al., 2010).Therefore, these variables were grouped in a Principal Component Analysis (PCA) before the microhabitat selection analysis.The two first axis of this Principal Component Analysis, PC1GC, and PC2GC (Table 1), represented 92% of the variation in the original data.
We measured the vegetation obstruction at three heights: (V5) 0 to 0.50 m, (V6) 0.51 and 1.00 m, and (V7) 1.01 and 1.50 m.We used the same grid as above, vertically positioned and estimated the percentage of grid cells with vegetation cover within 3 m.This measurement was performed in each cardinal direction and their average was used to characterize the vegetation obstruction at each trapping station.The vegetation obstruction showed high correlation values (Hair et al., 2010) so we used the average across the three heights in the subsequent analysis.
Additionally, we measured four variables within a 3 m radius of each trapping station.We counted the measured perimeter at breast height (PBH) of all trees and grouped them into three subcategories: (V8) 10 to 30 cm, (V9) 31 to 60 cm, and (V10) above 61 cm.We also recorded the (V11) presence or absence of trees or shrubs with fruits that could be eaten by small mammals.Canopy height (V12) was estimated, always by the same person, by comparing it to an object of known height.Canopy cover (V13) was also estimated by holding the grid horizontally above the head and counting the percentage of grid cells covered by vegetation.All microhabitat variables (expect fruit availability) were estimated once a year and averaged for the subsequent analysis.Fruit availability was estimated seasonally, and an average for each season was used in the subsequent analysis.
A PCA was performed on canopy height, canopy cover, PBH, and vegetation obstruction as these variables showed a high correlation.The two first axes, PC1VS and PC2VS (Table 2), represented 55% of the variation in the original data and were used in the subsequent analysis of microhabitat selection.

Microhabitat analysis
We analyzed the influence of microhabitat variables in species abundance using Generalized Linear Models (GLM) with a Poisson distribution.To analyze the influence of each variable over the four most abundant species, we created 32 models using all possible combinations of the five variables (PC1GC, PC2GC, PC1VS, PC2VS, and fruits) and a null model (see Table S3) for each species.
To carry out model selection, we used the corrected Akaike Information Criteria (AICc) for small sample sizes (Burnham & Anderson, 2002).The top model was the one with the lowest AICc, and all models with ΔAICc less than two were considered important models to explain the small mammal abundance (Burnham & Anderson, 2002).The importance of each variable was evaluated by using the sum of the model's weight that included each variable.The model weight (wi) represents the relative like-lihood of a model considering the set of models created (Burnham & Anderson, 2002).The importance analysis output consists of values ranging between zero and one, where zero represents a variable with no importance and a value of one represents the highest possible importance.Before the analyses, the variables' magnitude orders were standardized so all the variables were on the same scale, allowing comparisons of the magnitude of the effect of each variable.All the analyses were performed in the R, Version 3.5.1 (R Core Team, 2018), and the MuMIn package was used for model selection procedures (Barton, 2018).

Community composition
Overall, 198 individuals of six species, five rodents (sequence identity match of two retrieved from NCBI; Table S2), and one marsupial, were recorded in 406 captures over 5,320 trap-nights (7.61% trapping success).The most abundant species was Akodon montensis (Thomas, 1913;219   highest abundance was recorded during 2015 spring and 2016 winter (Fig. 2B).Sooretamys angouya had similar abundance during all seasons (Fig. 2C) and D. albiventris was most abundant during summer 2016 (Fig. 2D).
The rarefaction curve (see Fig. S4) indicated that the sampling effort was enough to characterize community richness, as it remained stable for more than 215 captures.

Microhabitat variables
The variables driving microhabitat selection varied among the four species.Three models had a ΔAICc ≤ 2 for A. montensis (Fig. 3A), which included fruit, PC2GC, PC1VS, and PC2VS as the most important variables (importance values ranging from 0.99 to 0.68; Table 3).For O. nigripes, five models were selected (four models with variables -Fig.3B -and the null model: AIC 168.2); the most important variables were PC1VS and PC2GC (importance value of 0.63 and 0.57; Table 3).For S. angouya only one model was selected (Fig. 3C), and PC2VS was the most important variable (importance value of 0.95; Table 3).D. albiventris had two models selected (Fig. 3D) and fruit availability was the most important variable influencing microhabitat selection (importance value of 0.78; Table 3).

DISCUSSION
Our results indicate that the local habitat is highly degraded.The small mammal community had a low species richness and was dominated by a single generalist species (A.montensis).The species composition also evidences the degradation status of the study area as most of the species are generalists adapted to anthropogenic landscapes in Atlantic Forest biome and no threatened species were recorded.In comparison, Melo et al. (2011) found 12 small mammals species in the Parque Estadual do Turvo (PET), a 17,000 ha forest in a protected area, 120 km from our study area.The small mammal community at our study site represents a subset of the PET community, suggesting that the community was affected by the fragmentation process.All native rodents captured in our study were represented at the PET.However, the congeneric marsupial species D. aurita was present at the PET (Melo et al., 2011)   at our study site.Both fragments sampled in our study had the same species richness and similar abundance of the four most common species.This suggests that it is likely for most of the fragments in the regional landscape to have a similar small mammals' composition, once they are all similar small forest fragments surrounded by agricultural and urban areas.
The low species richness may also be related to the latitudinal position of the study area.Our fragments are in the southern portion of the Atlantic Forest biome, outside the tropical region.Studies with a similar methodology in the Atlantic Forest have found between six to 21 species, and, in general, studies in lower latitudes had greater species richness (e.g., Pardini et al., 2005;Vieira et al., 2009;Lima et al., 2010;Maestri et al., 2014).The PET itself, the most preserved forest area in the southernmost portion of Atlantic Forest, had lower species richness than other well-preserved forests further north (Melo et al., 2011).While several factors, such as fragment size, isolation, habitat amount and matrix type (e.g., Pardini, 2004;Umetsu & Pardini, 2007;Vieira et al., 2009), interact to determine small mammals richness in fragmented landscapes, latitudinal was recently shown to be an important factor driving on small mammal richness in 122 forest fragments in Atlantic forest (Rodrigues et al., 2020).
In general, rodents were most abundant during winter.This could be due to a lower food availability making the trapping baits more attractive when compared with seasons of higher food availability.A higher abundance of O. nigripes during winter has been observed in the southern region of the Atlantic Forest, both in Araucaria Forest (Galiano et al., 2013) and in Dense Ombrophilous Forest (Antunes et al., 2009).However, for A. montensis, no relationship was previously found between population peaks and seasons in the south of Atlantic Forest (Antunes et al., 2010;Galiano et al., 2013).In the southeast of Brazil, where rainfall rather than temperature drives seasonality, a higher abundance of rodents was related to food scarcity during the dry season instead of the winter season (Dalmaschio & Passamani, 2003).
The key variables influencing microhabitat selection varied among the small mammal species.Akodon montensis and O. nigripes showed similar preferences in the ground cover and vegetation variables, preferring ground covered by rocks and higher vegetation obstruction.Both variables can provide protection against predators while the animal is moving on the ground.The null model was among those selected for O. nigripes, decreasing our confidence regarding microhabitat selection for this species.Akodon montensis and S. angouya differed in preference for tree abundance, with S. angouya preferring a higher abundance of trees while A. montensis showed the inverse relationship.The positive relationship with higher abundance of trees may indicate a preference of S. angouya for more preserved characteristics within the forest fragment.
Habitat segregation can facilitate species coexistence through resource partitioning (Schoener, 1974;Rosenzweig, 1981;Abreu & Oliveira, 2014).Previous studies demonstrated that small mammal species tend to coexist more often than would be expected in highly heterogeneous environments, while the inverse pattern is observed in environments with lower heterogeneity (Stevens et al., 2012;Camargo et al., 2018).Our results suggest that resource partitioning may be supporting coexistence of A. montensis and S. angouya.Additionally, if the landscape would be fully preserved, with greater microhabitat heterogeneity, this would allow more microhabitat segregation opportunities and, therefore, more species would occur in this area.
Fruit availability had a positive influence for A. montensis and D. albiventris.This is highly expected, as several species of small mammals in the Atlantic Forest have a diet based on fruit and seeds (Paglia et al., 2012).Therefore, they can also contribute to seed dispersal (Bricker et al., 2010;Grenha et al., 2010).Vieira et al. (2006) found that A. montensis individuals feed intensely on fruits, but also on invertebrates and fungi.Fruits are one of the most important items on D. albiventris diet (Cantor et al., 2010), with opportunistic consumption (Cáceres, 2002).These species can help pioneer plants and forest regeneration, by dispersing their seeds in more suitable environments, since, as generalist species, they can move more frequently along forest edges and matrix (Cáceres et al., 1999;Cáceres, 2006).

CONCLUSION
Our study found a poor small mammal community in forest fragments as we recorded only six species and the community was strongly dominated by the generalist rodent.The low species richness and presence of an exotic species (R. rattus) suggest that the study area is highly degraded.Akodon montensis and O. nigripes showed similar habitat preferences, with ground covered mainly by rocks and with greater vegetation obstruction.Sooretamys angouya preferred places with a higher abundance of trees.Fruit availability was important for A. montensis and D. albiventris, highlighting the importance of this food resource for local wildlife, and the potential role of these species as seed predators and dispersers.

FUNDING INFORMATION:
The first author is grateful to PET/SESu/MEC -Programa de Educação Tutorial for the scholarship.The second, third, and fourth authors recived personal grants by FAPERGS -Fundação de Amparo à Pesquisa do Rio Grande do Sul.Skupien, F.L. et al.: Small mammals and microhabitat selection in forest fragments Pap. Avulsos Zool., 2022;v.62: e202262039 6/12 Table S2.Species assignment based on DNA sequences (fragment of the Cytochrome b gene) using the program BLASTN 2.10.1+(Zhang et al., 2000)

Figure 3 .
Figure 3. Variables coefficients and their confidence intervals in the models selected (with ΔAIC ≤ 2) for each small mammal species.(A) Akodon montensis; (B) Oligoryzomys nigripes; (C) Sooretamys angouya; (D) Didelphis albiventris.PC1GC = first axis of the PCA for soil variables; PC2GC = second axis of the PCA for soil variables; PC1VS = first axis of the PCA for vegetation structure; PC2VS = second axis of the PCA for vegetation structure.
: DOL: Conceptualization, Funding acquisition, Supervision; DOL, GLG: Resources; FLS, DOL: Software, Data curation, Formal analysis, Visualization, Writing -original draft, Project administration; FLS, DOL, DPR, JOS, GLG: Methodology, Investigation, Writing -review & editing, Validation.All authors actively participated in the discussion of the results, they reviewed and approved the final version of the paper.CONFLICTS OF INTEREST:Authors declare there are no conflicts of interest.

Table 1 .
Principal Component Analysis values of the four original ground cover variables into the PC1GC and PC2GC.Together, these two axes represented 92% of the variation in the original data.

Table 2 .
Principal Component Analysis values of the six original vegetation variables into the PC1VS and PC2VS.Together, these two axes represented 55% of the variation in the original data.

Table 3 .
Importance of microhabitat variables for four species of small mammals in two forest fragments of Atlantic Forest in Cerro Largo, Rio Grande do Sul, Brazil.Variables' importance values closer to 1.0 indicate greater importance of this variable.