A total of 208 morphospecies were positively identified from materials collected in the Fitzcarrald region (Figure 1, purple circles). Identifications were based on catalogued vouchers with associated color digital images and cross referenced tissue samples (Appendix 1). This represents a 76% increase over the 118 fish species previously documented with vouchers from this region [55-65]. The taxa listed in Appendix 1 represent approximately 4.5% (208 of 4,581) of the freshwater fish species of tropical South America, in an area spanning about 2.6% (400,000 of 15,400,000 km²) the total land surface area occupied by this fauna [5]. Species accumulation curves for each of the three expeditions are reported in Figure 2. These accumulation curves do not approach asymptotic values, and the species richness numbers reported in this study underestimate the actual values for these basins.

The results of this study showed high levels of geographic isolation among, and ecological specialization within, the tributary basins. The species richness values reported in each of the three basins were all very similar, each representing about half that of the regional species pool (Figure 3(a)); Alto Yuruá with 115 species (55%), Alto Ucayali with 97 species (46%), and Alto Purús with 94 species (45%). The taxonomic compositions (i.e., species names) of the faunas were however very different; most species (133 species, 64%) collected were restricted to a single basin, 56 species (27%) were shared by two basins, and only 21 species (10%) were widespread in all three basins (Figure 4). These differences are reflected in the high gamma diversity between upland basins, which averaged 165 (79%) species turnover. The fish species encountered in upland localities within the Fitzcarrald Arch also differ substantially from localities of the same rivers further downstream (e.g., Ucayali, Juruá, Purús), where floodplain habitats dominate the regional diversity [56,57,63,66-68]. Local species richness (alpha diversity) of floodplain sites declines noticeably in all three of the basins draining the Fitzcarrald examined so far, from 70–100 species at downstream sites (100–120 m) to fewer than 30 species at upstream sites (>200 m).

There was by contrast comparatively high heterogeneity in the species richness of the three habitats sampled, with a total of 114 species (55%) in rivers, 58 species (28%) in lakes, and 138 species (66%) encountered in streams. A large majority of the species (144 species, 69%) were stenotopic, that is restricted to a single habitat, and only 11 species (5%) were eurytopic, found in all three habitats. There were many more species in the lotic habitats (flowing waters of rivers and streams; 151 species, 72%) than in the non-riverine habitats (lakes and streams; 115 species, 46%) or floodplain habitats (rivers and lakes; 71 species, 34%). Examples of species collected only in the Alto Yuruá, Alto Ucayali, and Alto Purús basins are shown in Figures 5–7. Five of the 21 species found in all three basins are depicted in Figure 8.

The high degree of habitat specialization in the Fitzcarrald is reflected in the high beta diversity (species turnover between habitats), which averaged 135 species (65%). A turnover of approximately half the species between adjacent habitats may seem remarkable, given the intimate proximity and physical contiguity of the waters draining small streams and oxbow lakes into the larger rivers, among sites separated by just a few tens to hundreds of meters. At least part of the reason for this high habitat specificity lies in the capacity of individuals in many floodplain species to move readily between sites over the course of the annual flood cycle, to find habitat conditions that better match their body size or other ecophysiological attributes [56]. Among the three basins, none of the habitats were found to share more than about 7% of the total species pool, and there were no substantial differences in actual numbers of species shared among the three habitats (Figure 4(C)). Another way to express this is that few species (5%) were found to be geographically widespread in any of the three habitats, and most were found restricted to a single basin (Figure 4(B)).

As in any ichthyofauna, each fish species that inhabits the Fitzcarrald region exhibits a distinct geographic range, both within and beyond the Arch. The distributions of known Fitzcarrald fish species (FFS) among the freshwater ecoregions of tropical South America are presented in Table 1, and some generalities in these distributions are illustrated in Figures 9 and 10. A large majority of the FFS (184 species or 88%) are present in the adjacent Amazonas Lowlands Ecoregion (ER 317) A large number of FFS are also shared with the adjacent Ucayali-Urubamba Piedmont Ecoregio (ER 312; 101 spp.; 49%), and Mamoré-Madre de Dios Piedmont Ecoregion (ER 318; 42 spp.; 20%) By contrast, relatively few FFS are present in other portions of tropical South America (e.g., Brazilia Shield).

The floodplain midline (thalweg) is a more accurate measure of fluvial separation among populations than are either Euclidean distance (“as the crow flies”), or river-channel distances. Euclidean distances as measured point-to-point on a map are inappropriate for assessing geographic separation among localities in the one-dimensional topological landscape of a drainage network [69]. Euclidean distances substantially underestimate the amount of habitat space required for dispersal by obligatory aquatic taxa, and the distance required for the transfer of material and energy through stream networks [70]. River channels, on the other hand, overestimate geographic separation among sites for riverine taxa in lowland Amazonian systems, as most if not all species that inhabit the deep channels also regularly use floodplain habitats on ecological time scales for forging and breeding [66,71].

Channel length (river distance) is a poor measure of fluvial spatial separation in lowland (below 200 m elevation) Amazonian ecosystems, where rivers exhibit high sinuosity (i.e., have many meanders). Among the three rivers examined in this study, the total channel distances among sites range from 5,900 km (between Breu on the Alto Yuruá and Sepahua on the Alto Ucayali) to 6,765 km (between Sepahua and Pto. Esperanza on the Alto Purús), with the difference of 865 km representing about 13% that of the largest value (Table 2). Comparable ranges of distance estimates between these same sites are 21% for the floodplain midlines and 39% for Euclidean distances.

Channel length is generally well correlated with total basin area and floodplain area, and all three measures have been used as estimates of total habitat space in aquatic systems [72,73]. However, regional geomorphology can alter these relations, and channel length turns out to be a worse predictor of habitat space than does thalweg distance in highly meandering lowland Amazonian river systems. For example, the differences in channel length exhibited by rivers draining the Fitzcarrald Arch are relatively small as compared with other measures of basin size, such as total drainage area and water discharge [74,75]. This effect results from feedbacks of regional depositional hydrodynamics on channel length [76]. In the Yuruá, Purús and Madre de Dios basins the average channel sinuosity is proportional to floodplain width [77], which is very broad in the lower portions of the two shorter rivers, the Yuruá and Purús. These stretches are especially tortuous, attaining among the highest sinuosity values of any major rivers of the world, with values of about 2.3 [78,79].

As of this writing we estimate the number of fish species in the interior of the Fitzcarrald region to exceed 300 species. This number is about one third more than the number used in the quantitative analyses of the present paper. The higher estimate was made by comparing species lists and photos presented in literature sources [55-65], and from examination of museum lots archived at the Natural History Museum of the University of San Marcos (MUSM), Lima. The quantitative comparisons of species diversity reported here were based on the more accurate and comparable species level identifications of our own collections, and not the higher species richness estimate based on literature references.

We emphasize the preliminary nature of the diversity estimates reported here, which are based on species presence and absence data only, and from limited sampling over a very small proportion of the Fitzcarrald region as a whole. The Fitzcarrald interior remains a remote wilderness landscape of lowland humid tropical forest with no roads, few air strips, and complex political interactions among the indigenous communities. Due to logistical constraints regarding access and transportation, we were unable to standardize the sampling protocol with sufficient rigor to collect species abundance data, or to test the effects sampling error on diversity estimates [80-85]. Quantitative data on fish species abundances in the Fitzcarrald region have to date focused on a limited number of commercially important, large-bodied, riverine species [59,86].

The Amazonian ichthyofauna in general remains incompletely documented, especially at the species level. A recent review calculated that, as of 2003, about 25% of Neotropical fish species known in museum collections were undescribed [87]. Much of the Amazonian lowlands is still a wilderness and the ichthyofaunas of many regions and river basins are either sparsely collected or, in some cases, almost entirely unknown [88]. There is also much poorer sampling of low order streams across the landscape, especially given the great proportion of the landscape they occupy, and the expectation for a relatively high species turnover among sites (gamma diversity) [89].

The sampling design of this study may also have introduced potential biases to the diversity estimates. In order to facilitate comparisons, collections were made during a standard period of low water in July when fishes are concentrated in lakes and channels, and not dispersed onto the floodplain. The lack of rain and mud also facilitates transportation by air, water and foot, and thereby increases the total number of localities, field days, stations, specimens and species that can be sampled within the limited time and other resources of a given expedition [56]. However, sampling at low water may under represent highly seasonal and migratory species (e.g., the pimelodid catfishes Pseudoplatystoma spp. and Brachyplatystoma platynema), as well as species associated with seasonally ephemeral habitats such as floating vegetation [71,90]. Interviews with local fishermen revealed the presence of several common lowland riverine species in upland portions of the Fitzcarrald rivers during the rainy season (Sept.-Dec.), that were not collected in our survey: e.g., potamotrygonid stingrays, the characiforms Curimata aspera, Colossoma macropomum, Piaractus brachypomus, Raphiodon vulpinus, and the perciform Plagioscion auratus [62,64]. Sampling at low water may affect gamma diversity estimates by under representing the number of species shared between floodplain habitats (rivers and lakes) due to seasonal lateral migrations. The anomalously low-water conditions encountered in the Alto Purús in 2010 as a result of a regional drought [91] may also have reduced the number of collection sites, fish abundances, and species diversity.

The small proportions of fish species shared among collections in the three basins of this study (Figure 4(C)) may reflect sampling error, dispersal limitation, or both. The very low level of species endemism recorded within the Fitzcarrald region as a whole suggests that dispersal has been important to the formation of the basin-wide fish faunas. Further, some species are expected to be absent from our collections do to perennially low abundances, or population variability in space or time (seasonal or annual variation). Other recorded absences may reflect real differences in absolute abundances. Some fish taxa (e.g., Rivulus, Corydoras) are widespread, abundant, and diverse in Amazonian lowlands, and really do seem to be absent or rare in the Fitzcarrald uplands. Among Gymnotiformes, the electric signals of Gymnotus coropinae and Hypopygus lepturus were never detected using a portable amplifier. As these species are usually common and abundant in lowland terra firme (non-floodplain) habitats, they also seem to be either absent or rare in the Fitzcarrald uplands. The absence of the callichthyids Callichthys callichthys and Hoplosternum littorale from many sites was also informative, as both species are usually common floodplain species in the lowland reaches of all the rivers draining the Fitzcarrald region, and neither had been previously reported from the upper reaches of these rivers [92,93]. Based on these experiences collecting in the Fitzcarrald uplands it is now possible to interpret the negative results as indicating real species absence, or at least rarity.

Some of the taxonomic patterns reported here may also be affected by the sampling gear, in particular our heavy reliance on nets (albeit of several kinds). Under sampled taxa may include certain pimelodid catfishes and apteronotid electric fishes inhabiting deep river channels, and species that take refuge in structurally complex habitats and substrates; e.g., doradid and auchenipterid catfishes in logs and logjams of streams and rivers; sternopygid and hypopomid electric fishes in rooted vegetation of floodplain lakes. Plant-based ichthyocides are widely used for subsistence fishing by indigenous peoples of the region, and were adventitiously employed in this study. However industrially produced rotenone is banned for scientific collecting in Peru and was not applied in this study. Although electrofishing is often useful in sampling structurally complex substrates, the apparatus is heavy and cumbersome to transport and use in remote settings that are often accessible by foot alone, and was not applied in this study.

Despite the potentially important sampling biases described above, the fish faunas of the Fitzcarrald region appear to be composed of species recruited primarily from the exceptionally diverse Amazonian lowlands, and to a lesser extent from the fish faunas of the Ucayali and Madre de Dios basins (Table 1). There are high numbers (and proportions) of Fitzcarrald species present in the adjacent Amazonas Lowlands (184 species or 88%), Ucayali-Urubamba (101 species or 49%) and Mamoré-Madre de Dios (42 species or 20%) ecoregions (Figure 9). The absolute numbers of Fitzcarrald fish species is highest in the adjacent Amazonas Lowlands, Ucayali-Urubamba, and Madre de Dios ecoregions (Figure 10(A)), and proportional representation of Fitzcarrald fish species is highest in the Ucayali-Urubamba, Amazonas High Andes, and Amazonas Lowlands ecoregions (Figure 10(B)).

Upstream localities of the Fitzcarrald fish fauna appear to be relatively depauperate in comparison with downstream localities of comparable habitat in the same river basin [56,57,63,67,68]. Some common and widespread Amazonian fish taxa that are to date entirely absent from the Fitzcarrald uplands include: Lepidosiren, Pellona, Semaprochilodus, Metynnis, Myleus, Brycon, Acestrorhynchus, Hydrolicus, Hypopygus, Microsternarchus, Rhamphichthys, Hoplosternum, Cichla and Colomesus. The relative rarity or perhaps complete absent of these species in the Fitzcarrald uplands may be due in part to local ecological conditions. The margins of streams and rivers traversing the interior of the Fitzcarrald region are dominated by sandy or muddy beaches at low water, with little rooted and almost no floating aquatic vegetation (Figure 11). Further, the upland river channels are relatively shallow, and the floodplains substantially narrower, with fewer and smaller oxbow lakes than in lowlands. In combination these conditions are not favorable to many fish groups specialized to inhabit lowland Amazonian floodplains and deep river channels (e.g., the várzea guild of Serrasalminae; see Figure 3 in [94]). The pattern of fish species richness falling off with elevation has been reported in other regions of tropical South America sampling over similar elevational gradients [95-98].

Very few fish species are known to be endemic within the Fitzcarrald region. At present we know of only three species; the gymnotid electric fish Gymnotus chaviro [99], the characin Phenacogaster capitulatus [100] and the cichlid Bujurquina eurhinus [101]. However, the task of documenting species endemism requires much more spatial information than does that of documenting species richness, as endemism requires positive knowledge of where species are both present and absent, and extensive geographic sampling with fine resolution at the alpha taxonomic level [5,102]. The actual patterns of species endemism in the remote and relatively poorly explored rivers of Western Amazonia may not be known for many years.

Given the relatively poor understanding of the FFS fauna, there is reason to suspect that both taxonomic and geographic factors contribute to the conclusion of little or no endemism within the Fitzcarrald region. The observed lack of endemism may in fact be real, but limited (or apply most strongly) to the floodplain portion of the fauna that was most intensively sampled, or perhaps to the peripheral portion of the Fitzcarrald region that was sampled at elevations below 310 m. Low levels of species endemism are indeed widely observed in other floodplain areas of lowland Amazonia [5]. Under this view it remains possible that some fish species are restricted to non-floodplain (terra firme) habitats of the Fitzcarrald, especially in the interior uplands (310–500 m), and that the elements of the fauna intercepted by the current study represent primarily the most common and widespread forms.

Nevertheless, the low endemism of fish species in the Fitzcarrald region does appear to be a genuine feature of the fauna, and we anticipate this result will be robust in the face of future discoveries [5]. We suggest, cautiously, that the data currently in hand really do indicate low levels of fish species endemism within the Fitzcarrald. The taxonomic and geographic understanding of this region is really not so much worse than many other areas of comparable size the Western Amazon, some of which really do exhibit distinct a species composition (ERs 313, 317, 318). In addition, the sampling intensity of this study, focused on the periphery of the Fitzcarrald (212–310 m), is likely to have intercepted at least some species endemic to the interior. The sampling station (Quebrada El Dorado on the Mishaua river) closest to the point of highest elevation (536 m) within the Fitzcarrald region lies at a distance of about 110 km as the crow flies. This is a relatively small distance in the context of Amazonian fish species whose ranges commonly extend over many hundreds of km in linear dimensions. Unsampled endemic species would therefore have to have very restricted geographic distributions, low abundances, or both. Lastly, analysis of geospatial information from DEM and high resolution satellite (Google Earth) images indicate an absence of pronounced physical barriers to dispersal (e.g., a fall line), or discrete climatic differences, between the peripheral(lower) and interior (upper) portions of the Fitzcarrald region. Indeed elevations associated with changes in fish species compositions usually occur above 500 m in other upland regions of tropical South America [5].

In addition to the unique ecological conditions alluded to in the Introduction, the Fitzcarrald Arch represents a unique geological situation, containing the only watersheds within the Amazon Basin for which reliable estimates are currently available regarding the timing of headwater basin separation [5,24]. The uplift of the Fitzcarrald Arch is dated by geophysical and paleontological data to the Late Miocene-Pliocene (9–3 Ma) [103-106]. Such information on the timing of river basin separation provides minimum age estimates for calibrating genetic divergences among populations or species inhabiting each of the newly isolated basins [52,107-114].

Analyses of radiometric and biostratigraphic data indicate that the Fitzcarrald region changed from a depositional to an erosional setting during the period of the Late Miocene to Pliocene epochs (c. 9–3 Ma.) [103,115-118]. Sedimentological analyses show the switch occurred in association with the transition from mid-Miocene (Quechua phase) faulting to Pliocene (Diaguita phase) compressional deformation [103]. Radiometric Argon-Argon dating of two volcanic tuffs from the Solimões Formation were dated to c. 9 and c. 3 Ma. [116]. Mammalian biostratigraphy confirms these age estimates, as the top of the Solimões Formation (Chapadmalan Stage) has no mammal fossils of North American origin [119]. These data suggest that sedimentation in the Solimões Formation ceased before the rise of the Isthmus of Panama and the onset of the Great American Biotic Interchange c. 3.5 Ma. Lastly, molecular dating of divergences among aquatic mammal and fish populations [120,121] suggest Late Miocene-Pliocene dates for the separation of the Upper Madeira from other Amazonian basins.

Although rising to only modest elevations, the Fitzcarrald uplift contributed to the Miocene fragmentation of the north flowing Subandean Foreland Basin, and to the reorganization of the fluvial net that forms the modern east-flowing Amazon Basin [104-106,122-124]. The low elevation and unconsolidated sediments of the region has resulted in a relatively continuous history of stream capture events across its several watershed divides, as the principle direction of stream flow shifted from generally NW to NE [125].

The most likely candidates for divergence in allopatry across low-lying Amazonian watersheds are clades (monophyletic groups) of species with spatially restricted and non-overlapping geographic ranges, and clades of species possessing small body size, stenotopic (ecologically narrow) habitat preferences and limited dispersal capacities. A recent meta-analysis of diversification in Amazonian fishes examined patterns in 33 taxa (genera or tribes) with sufficiently dense taxon and geographic sampling to test hypotheses of allopatric divergence [24]. Among these taxa no species-pair has yet been identified that matches criteria for divergence in allopatry across one or more of the Fitzcarrald watersheds.

Taxa that stand as candidates for possible species-level divergence across one or more of the Fitzcarrald watersheds include the characins Characidium (8 species), Knodus (7 species), Moenkhausia (5 species), and Odontostilbe (4 species), and the armored catfishes Ancistrus (7 species), Hypostomus (6 species), and Panaque (4 species). There is to date no phylogenetic information on the species of these taxa from the Fitzcarrald region. Available phylogenetic information on several other taxa is however inconsistent with such an hypothesis; e.g., Leporinus with four species [126], and Gymnotus with two species [99]. Genetic and phenotypic data on the armored catfish Chaetostoma lineopunctatum show incipient within-species divergence between the Ucayali and Madre de Dios basins [127], but this species really has an Andean distribution, and this example may not therefore represent divergence across a Fitzcarrald watershed. In general, the alpha taxonomy and systematics of all the taxa mentioned above remain very poorly understood. Phylogenetic and phylogeographic studies of several Fitzcarrald fish taxa are currently underway using molecular data to examine species-level divergence across Fitzcarrald watersheds.