Potential distribution and new records of Trinomys species ( Rodentia : Echimyidae ) in the state of Rio de Janeiro

The spiny rats of the genus Trinomys Thomas, 1921 have a broad distribution in the Atlantic Forests of southeastern Brazil. However, some species are known only from their type locality and adjacent areas. In our study, nine areas in the state of Rio de Janeiro were surveyed and three species of the genus were captured – Trinomys dimidiatus (Günther, 1877), T. setosus (Desmarest, 1817) and T. gratiosus bonafidei (Moojen, 1948). We extended the distribution of T. gratiosus bonafidei in 100 km, in a straight line to the northwest, and into an area of Semidecidual Seasonal Forest. We captured T. setosus, which had not been previously recorded in the state, in the municipality of Cambuci, extending its distribution 150 km, in a straight line to the east of its closest record, in Juiz de Fora, state of Minas Gerais. The state of Rio de Janeiro has now six recognized species of Trinomys, however none of them were collected above 1300 m of altitude. We used occurrence points provided by our inventories data and from the literature to model the potential distribution of Trinomys species. We used climatic, topographic and phytogeographic variables to prepare the potential distribution maps. The algorithm used for modeling was provided by the software Maxent, version 3.2.1. Although species boundaries within Trinomys in Rio de Janeiro State are not yet clear, their distributions seem to be parapatric, except for T. iheringi and T. dimidiatus.

This study reports additional sites of occurrence for species of Trinomys in Rio de Janeiro and estimates their potential range using methods of distribution modeling.

MATERIAL AND METHODS
We carried out small mammal inventories in nine areas in the state of Rio de Janeiro from 2004 to 2006.
The captures in each area were done using 126 Sherman and Tomahawk live traps, set in three distinct lines.The traps, placed on the ground (90), were set 40 m apart from each other.Traps placed on trees (36) were set 100 m apart and at a minimum height of 2.5 m.They were opened during six consecutive nights with a total effort of 756 traps/night.Thirty buckets of 40 liters set in three distinct tracks were utilized as pitfalls, for a total of 180 buckets/night.
All the specimens were identified by a combination of characteristics of the cranium, teeth, hair, baculum structure, and external body measures, and classified within the nomenclature considered valid in LARA & PATTON (2000).
For T. dimidiatus, we used 75% of the occurrence points to generate the potential distribution model; 25% were used for model validation (PHILLIPS et al. 2006).We applied the method suggested by PEARSON et al. (2007) to predict the species distribution with few records of occurrence (i.e.<25).We used this method for all species, but T. dimidiatus.This method uses a statistical approach based on jackknife techniques for evaluation of the generated models.For species with n occurrence points, n models are constructed using n -1 locations.Each generated model is tested for its ability to predict the deleted location.During the model validation, when a point was removed, it was verified if there was another near point, in a distance up to 10 km.In this case, the point was removed from the model (PEARSON et al. 2007).The results of this procedure were used in the software pValueCompute (PEARSON et al. 2007) to assess whether the model with all localities was able to predict the potential distribution better than a random prediction.
The algorithm used for modeling was provided by the software Maxent version 3.2.1 (http://www.cs.princeton.edu/~schapire/maxent/, accessed in May 2008).The software provides a continuous probability result of occurrence that ranges from 0 to 1 (PHILLIPS & DUDIK 2008).We used 10,000 points chosen randomly to characterize the entire area where the prediction was made (PHILLIPS & DUDIK 2008).We used the default convergence threshold (i.e. 10 -5 ) and the maximum number of interactions (i.e.500).
The software provides a jackknife test to assess which variables were the most important in modeling the potential distribution.For this, the software generates a model using only one variable at a time.For each variable removed, another model is generated with the remaining variables.The results are then compared with the performance of the model created with all variables.
For jackknife validation approach we used two thresholds (PEARSON et al. 2007).First, we chose the lowest presence threshold (LPT) able to include in predicting distribution all points used in training model (i.e. the model with all n points).This can be interpreted as an identification of pixels with similar conditions to those found in the species record localities.The second threshold applied was chosen aiming to withdraw 10% of the lowest prediction values (T10).This second approach is less conservative and identifies a larger portion of the area.We made the final maps of potential distribution using the software ArcGis 9.2.We plotted only the LPT maps, due to its conservative characteristics.

RESULTS
We captured three species of Trinomys in the studied areas, namely T. dimidiatus, T. gratiosus bonafidei, and T. setosus.All of them were captured in traps placed in the ground, in both live traps and/or pitfall traps.Specimens of Trinomys were not captured in three localities -MSJ, FM and REGUA.The data on the collected specimens can be viewed in table I.
The potential distribution maps generated with the T10 threshold showed a high success rate and were statistically sig-nificant (Tab.II).The results for the maps using the LPT threshold were also validated.However the species T. eliasi (0.428) and T. gratiosus bonafidei (0.200) showed low success rates with LPT threshold.Trinomys dimidiatus (1.000), T. iheringi (0.867), T. panema (0.938) and T. setosus setosus (0.938) showed equal success rates for both LPT and L10 thresholds (Tab.II).
Except for T. eliasi, the ecoregion was the variable with highest gain when used in isolation and the variable that most decreases the gain when omitted (Tab.II).Only for Trinomys eliasi, altitude had a higher effect than ecoregion.For T. gratiosus bonafidei, PWQ was the highest gain variable (Tab.II).
All species occurred in Alto Paraná Atlantic Forest, except T. setosus subspecies and T. dimidiatus (Tab.III).Trinomys eliasi also occurred in Atlantic Coastal Restingas, T. g. bonafidei in Campos Rupestres Montane Savanna and T. iheringi in Southern Atlantic Mangroves.Trinomys dimidiatus is restricted to Serra do Mar Coastal Forest and Southern Atlantic Mangroves ecoregions.On the other hand, Trinomys panema occurred in five of the nine ecoregions reported for the studied species (Tab.III and Fig. 2).Trinomys dimidiatus showed a potential distribution ranging from São Paulo to the north of Rio de Janeiro, very similar to the known distribution (Fig. 3).Trinomys eliasi also showed a potential distribution very similar to the known distribution, being restricted to the coastal area of Rio de Janeiro and to a small portion of the coast of Espírito Santo (Fig. 4).Trinomys gratiosus bonafidei presented a large and disjoint potential distribution ranging from the central portion of Minas Gerais to the coast of Santa Catarina, including Rio de Janeiro, São Paulo and Paraná.(Fig. 5).Trinomys iheringi presented a potential distribution restricted to the coastal region, from Rio de Janeiro to Santa Catarina (Fig. 6).Trinomys panema has a broad distri-  bution and presented a very large potential distribution occupying completely the Rio de Janeiro and Espírito Santo, a great portion of São Paulo and Minas Gerais and a portion of south Bahia.(Fig. 7).Trinomys setosus elegans appears to be restricted to Minas Gerais, presenting only small bordering areas in Rio de Janeiro and Bahia (Fig. 8).Trinomys setosus setosus presented the broadest potential distribution among all species analyzed in this study.The potential distribution shows a disjoint distribution that ranges over several states in the coast and central portion of the country (Fig. 9).

DISCUSSION
Species of Trinomys are common in faunal inventory and in studies carried out in the Atlantic Forest (e.g.PEREIRA et al. 2001, GEISE et al. 2004, VAZ 2005).However, in three sampled localities in Rio de Janeiro, we did not capture Trinomys (MSJ, FM and REGUA).MSJ is located in coastal plain of Rio de Janeiro, in a region that is subject to considerable anthropic disturbance.The small mammal richness was very low in the area (only four species), and the absence of Trinomys may not be due to sampling problems, but to the fact that MSJ is very degraded and isolated from other forest fragments, being surrounded by pasture.It is possible, that the Trinomys population in the area could have become locally extinct.
Many species were captured in FM, but most of them are recognized as typical of high altitudes.This area is located from 1220 to 1775 m and Trinomys species do not seem to occur in localities above 1300 m.BONVICINO et al. (1997) andGEISE et al. (2004) captured, respectively, T. gratiosus, in the Caparaó National Park and T. panema (= T. gratiosus in GEISE et al. 2004) in the Maciço do Itatiaia, below 1200 m.Species from the genus probably do not occur in areas above the Montane region (from 500 to 1499 m) (sensu URURAHY et al. 1983).
The lack of Trinomys spp. in REGUA (altitudes varying from 175 to 325 m a.s.l.) was unexpected.Specimens of T. dimidiatus were captured in adjacent areas, as the EEEP and Vale da Pedra Branca Farm (Fig. 1), both localities at Serra dos Órgãos and in similar altitudes, suggesting that this species should occur in REGUA.
Trinomys gratiosus bonafidei, until now, was known only from its type locality, in Boa Fé Farm, in Teresópolis Municipal-   Trinomys setosus was not known in Rio de Janeiro, with its southernmost record located in Juiz de Fora, Minas Gerais (21º45'S 43º20'W).Until the present time, Trinomys setosus occurred from Sergipe to Minas Gerais, including Bahia and Espírito Santo (LARA et al. 2002).Trinomys setosus has three subspecies with disjoint distributions, T. s. elegans, T. s. denigratus and T. s. setosus, the latter being restricted to Minas Gerais (LARA et al. 2002).However, it was not possible to identify the specimens captured in Cambuci in a subspecific level.When considered only in at the species level, the distribution of T. setosus was extended in 150 km east from Juiz de Fora, into the Rio de Janeiro.
Trinomys eliasi and T. gratiosus bonafidei showed low rates of success (0.428 and 0.200, respectively) suggesting that the models were not efficient.However, the potential distribution of both species is congruent with the altitudinal and ecoregion distributions, respectively.The potential distribution of Trinomys eliasi followed the coast up to south Espírito Santo in Atlantic Forest Restinga, Bahia Coastal Forest and Southern Atlantic Mangroves ecoregions in low altitudes.Restricted to the coast, T. eliasi may be more susceptible to threats due to the vulnerability of the region to human disturbance, especially real estate developments.Habitats of restinga, for example, have considerable biological relevance, yet they are subject to intense degradation (ROCHA et al. 2005).
Trinomys g. bonafidei presented a potential distribution mainly in the Alto Paraná Atlantic Forest ecoregion, where the species was recorded.Although it is endemic to Rio de Janeiro, the potential distribution of T. g. bonafidei suggests that this species may also occur in northeast São Paulo and south Minas Gerais.The absence of the species in these localities may reflect small sample effort, as these areas are more distant from the major urban centers.
The potential distributions of T. dimidiatus and T. iheringi overlapped considerably.Both potential distributions followed the Serra do Mar Coastal forest ecoregion, with the southern range of T. iheringi reaching Santa Catarina.However, sympatry is observed in the border of Rio de Janeiro and São Paulo in the coast.The inferred potential distributions were similar to those suggested by LARA et al. (2002).
The potential distribution of Trinomys setosus elegans seems to be restricted to Bahia Interior Forest ecoregion.The specimen collected by us in Esmeralda Farm, in the municipality of Cambuci, can belong to this subspecies, since the municipality is located in the same ecoregion.Although T. panema was the species occurring in most ecoregions (5, see Tab.III), the potential distribution was restricted mainly to the south-eastern region (São Paulo, Rio de Janeiro, Espírito Santo and Minas Gerais) and a small portion of Bahia.On the other hand, T. s. setosus, recorded for four ecoregions, showed the broadest studied distribution, including states from north to south and east to west of Brazil.This distribution is probably overestimated and more records are necessary to better evaluate its distribution.Trinomys panema and T. s. elegans seem to overlap in their potential distributions with T. moojeni (LARA et al. 2002).
These new records show the importance of a greater research effort in areas that have not been well studied and in small remaining fragments.We still were not able to define the species boundaries of Trinomys spp.due to the number of records available.But the potential distribution estimated in this study can be used as a guideline to fill these information gaps, defining future study areas.The efforts must encompass as many ecoregions as possible, since this variable had a great effect in the distribution of the Trinomys species.

Figure 2 .
Figure 2. Map of Rio de Janeiro State, divided in ecoregions, showing the records from the literature and the present study for the six Trinomys species.

Table III .
The occurrence of six Trinomys species by ecoregions.Ecoregions are according toOLSON et al. 2001.

Table I .
Specimens of Trinomys captured in the studied areas (Locality), museum number, number of individuals captured (N), mean and standard deviation of body weight (W) and of the measurements: head and body (HB), tail (TA), head (HE), ear (E), hind foot with (HFU) and without nails (HF).

Table II .
Jackknife tests of distribution models for six Trinomys species.(LPT) Lowest presence threshold, (T10) withdrawing 10% of the lowest prediction values, (PWQ) Precipitation of Warmest Quarter.
ity, and adjacent areas in Serra dos Órgãos, a region dominated by Evergreen Forest.The new record of this species in Serra da Concórdia extends its distribution in 100 km, in a straight line to northwest and into a region of Semidecidual Seasonal Forest.Trinomys gratiosus bonafidei is now known from three different ecoregions: Alto Paraná Atlantic Forest, Campos Rupestres Montane Savanna and Serra do Mar Coastal Forest.