Introduction

Many animal species reproduce during an annual breeding period timed to synchronize offspring growth with favorable environmental conditions (Pengelley 1974; Gwinner 1986; Volpato and Trajano 2005). Fishes are attractive subjects for studying breeding periodicity because their reproductive condition can be readily determined from analyses of the gonads of preserved specimens captured from year-round quantitative sampling (Munro et al. 1990). Temperate and subtropical freshwater fishes are exposed to strong seasonal regimes of photoperiod and temperature, which have been demonstrated experimentally to regulate the onset of circannual breeding (Munro 1990a). In contrast, tropical freshwater systems are subject to much more constant year-round photoperiod and temperature, especially near the equator.

Many terrestrial tropical animals exhibit reduced reproductive seasonality or breed throughout the year (Hau 2001), and some exhibit endogenous reproductive rhythms that are largely independent of external cues (Gwinner and Dittami 1990). Much less is understood about the regulation of circannual breeding periodicity in tropical freshwater fishes. Some studies in the Neotropics have reported reduced seasonality or constant (year-round) breeding, including in some small rainforest stream-dwelling characiforms (Krekorian 1976; Kramer 1978; Alkins-Koo 2000; Azevedo 2010), cypriniforms (Winemiller 1993; Alkins-Koo 2000), siluriforms (Mazzoni and Caramaschi 1995; Alkins-Koo 2000), and gymnotiforms (Schwassmann 1992). However, most studies of tropical freshwater fishes, particularly those focusing on riverine and floodplain habitats, report distinct circannual rhythms of breeding that match seasonal variation in rainfall or river flood cycles (Lowe-McConnell 1987; Munro et al. 1990; Kirschbaum 1992; Schwassmann 1992; Vazzoler and Menezes 1993). Neotropical freshwaters exhibit considerable geographical variation in the timing and magnitude of seasonal rainfall, which in turn influence water levels, habitat availability, predator densities, and water chemistry (Crampton 2011).

Some studies of Neotropical freshwater fishes have reported correlations of gonadal development with individual environmental cues such as conductivity, dissolved oxygen (DO), temperature, rainfall, water level, and photoperiod (Silva et al. 2003; Andrade and Braga 2005; Giora et al. 2012, 2014; Mendes-Junior et al. 2015; Souza et al. 2015). However, to date, all such studies have considered only single species. Also, excepting Mendes-Junior et al. (2015), these studies have focused on subtropical or high-latitude tropical species (ca. 20–30° S). Furthermore, despite calls for more robust statistical approaches in identifying environmental correlates of reproductive periodicity (Munro 1990b; Volpato and Trajano 2005), studies of natural populations of tropical freshwater fishes have yet to monitor multiple environmental parameters through an entire hydrological cycle and apply multiple regression procedures to identify the specific factors associated with cycles of gonadal maturation.

In this study we examined the environmental factors regulating breeding periodicity in a low-latitude (5°S) species-rich upper Amazon assemblage of the electric fish genus Brachyhypopomus (Gymnotiformes, Hypopomidae). Brachyhypopomus can be readily located in the field from their constantly generated electric signals using an electric fish finder (Crampton et al. 2007), facilitating capture of close to 100% of individuals (excepting larval stages). This level of sampling efficiency, which is unrivaled by other vertebrate study systems, makes Brachyhypopomus particularly suitable for quantitative field surveys of circannual reproductive status. Brachyhypopomus are small (typically <300 mm total length), non-migratory nocturnal predators of small invertebrates. They exhibit oviparity, fractional spawning with external fertilization, egg hiding, and no known parental care (Crampton et al. 2016).

Our upper Amazon study site includes two distinct Brachyhypopomus habitats: First, floodplain lakes of the Amazon River’s Holocene seasonal inundation plain. Second, terra firme rainforest streams in Paleogene-Neogene formations above the extent of river-floodplain inundation (Fig. 1). These habitats are characterized by distinctive circannual patterns of water level and water chemistry, and by distinct sets of Brachyhypopomus species – allowing us to examine congruent responses to changes in specific environmental conditions among multiple closely-related syntopic species. Amazonian floodplains exhibit high-amplitude (up to 16 m) inundation cycles (Henderson et al. 1998), and large-scale seasonal variation in water chemistry (Melack and Fosberg 2001; Affonso et al. 2011, 2015), habitat availability (Junk 1970; Piedade et al. 2010), predator densities (dos Anjos et al. 2008; Macedo et al. 2015), and community composition (Röpke et al. 2016). In contrast, Amazonian terra firme streams exhibit relatively minor annual variation in water level (< ca. 1 m) (Tomasella et al. 2008), water chemistry (Espírito-Santo et al. 2009; Mendonça et al. 2005), habitat availability (Espírito-Santo and Zuanon 2016), and community composition (Bührnheim and Fernandes 2001).

Fig. 1
figure 1

a Map of the study area. Whitewater floodplains of the Ucayali River are east of the dotted line, and terra firme formations west. Star = Jenaro Herrera. Arrow = river flow. Sampling sites are F1-F2 and T1-T4. b Floodplain lake at high water with rafts of floating meadows and, c at low water. d Terra firme stream during rainy season (flooding adjacent shallow swamps), and e dry season (restricted to its main course). f Cross-section from X-Y; 0–10 m river levels represent minimum and maximum mean monthly values during the study

Full size image

We quantified circannual changes in breeding condition by the abundance of fully mature individuals and by fluctuations in the averaged gonadosomatic indices (GSIs) of populations. We also described circannual variation in rainfall, river level, and photoperiod, as well as four water quality parameters known to be correlated to breeding in other fishes: temperature, conductivity, pH, and DO (Munro 1990a). We then employed linear mixed-effects models (LMEMs) with model selection to reach inferences about the effect of specific environmental variables (or combinations of these variables) on GSI. Our analyses also address whether floodplain species exhibit different circannual breeding rhythms to those in terra firme streams, and whether semelparous species sensu Roff (2002) with one-year lifespans and a single reproductive season followed by death exhibit differences in circannual breeding rhythms to iteroparous species with more than one annual breeding season.

Material and methods

Study system

This study was conducted near Jenaro Herrera, Peru (04°54’S, 073°39’W) (Fig. 1a). We sampled two lakes of the whitewater Ucayali River floodplain, both of which had year-round water content (sites F1 and F2). We were unable to sample additional floodplain lakes because all others in the study area dried out entirely during the low-water period. Lakes F1 and F2 were fringed by extensive ‘floating meadows’ of macrophytes (Fig. 1b, c), which is the primary habitat for Brachyhypopomus. We sampled four randomly selected second-order terra firme streams (sites T1-T4, Fig. 1), each of which varied from 2 to 4 m in width, and up to 0.5 m depth. Brachyhypopomus occurred in the main stream courses (Fig. 1e) and, during the rainy season, in extensive shallow (< 0.4 m) swamps resulting from the flooding of adjacent depressions and the lower courses of intermittent feeder streams (Fig. 1d).

Quantitative sampling

Sampling was conducted for a 12-month period from March 2013 to March 2014. We represent the study year as a January–December abscissa sequence in our figures, following Alkins-Koo (2000). We located Brachyhypopomus using an electric fish detector and captured them with dipnets (3 mm mesh). Sampling was conducted from one hour after sunset to as late as 02:00, during the period of peak activity for Brachyhypopomus. Larval/post-larval fish (< 30 mm length to anal-fin terminus) were excluded from our quantitative analyses due to mesh penetration.

In the floodplain lakes we sampled floating meadow habitats from a canoe. In the terra firme streams we sampled marginal roots and submerged leaf litter – devoting equal sampling attention to the main stream channel and adjacent flooded habitats when the latter were available. Each sampling event involved a measured time effort of 3–6 h during which all fish were captured in the order they were detected. This yielded capture per unit effort (CPUE) estimates of abundance for a given species/reproductive stage as inviduals·h−1. In total we sampled for 937 h (Table S1, Online Resource 1). To avoid negative impacts associated with repeated sampling, each sampling event occurred within a different ca. 100 m section randomly selected from a 2 km stretch of a given site. We performed euthanasia with a 600 mg·L−1 eugenol solution.

Environmental data

For the one-year duration of the study, and for a preceding five-year period we obtained daily river level and rainfall data for the Ucayali River at the nearby (25 km) Requena station of the Servicio Nacional de Meteorología y Hidrología del Perú. During the study period we also collected rainfall data from two Hobo RG3 digital rain gauges, one positioned near site T1 (Fig. 1) and the other in Jenaro Herrera (Fig. 1), which we averaged with data from the Requena station (tropical thunderstorms follow narrow paths, therefore averaging over 3 rain gauges provided a better estimate of local monthly rainfall averages). We henceforth refer to the rainfall and river level datasets for the study period as ‘1-year rainfall’ and ‘1-year river level’, respectively. We refer to the rainfall and river level datasets for the preceding five-year period as ‘5-year rainfall’ and ‘5-year river level’ datasets, respectively. We measured the following water parameters prior to each sampling event: conductivity to ± 1 μS·cm−1 with a Hanna HI9033 meter; DO to ± 0.1 mg·L−1 with a Hanna HI9142 meter; water temperature to ± 0.1 °C and pH to ± 0.01 with a Hanna HI9126 meter.

We obtained daily photoperiod data for Jenaro Herrera from the National Oceanic and Atmospheric Administration (www.esrl.noaa.gov/gmd/grad/solcalc). Additionally, to estimate variation in a measure we henceforth refer to as ‘visible photoperiod’, we used Onset UA-002 light loggers to determine the period elapsed from the first time in the morning that light intensity exceeded 250 lm·m−2, to the first time in the evening that light levels fell below 250 lm·m−2 (using 1 min means of 1 s data logging intervals). We deployed two loggers for 30 d in March 2013, a period with approximately average cloud cover and rainfall; one in marginal floating meadows at F1, and the other at the edge of stream T1 (each atop a pole to prevent leaf accumulation). The UA-002 logger has a wide-band spectral sensitivity (> 50% response at 450–1100 nm) and is therefore appropriate for assessments of light intensity within the photoreceptor frequency sensitivity for freshwater fishes (Douglas and Hawryshyn 1990).

Gonadosomatic index and gonad development stages

All length and weight measurements, including of gonads, were taken from specimens fixed in 10% formalin for 2–3 weeks. We measured body mass (BM) (prior to dissection) and gonad mass (GM) with a digital balance (accurate to ± 0.1 mg) and calculated the gonadosomatic index (GSI) as (GM/BM)·100. We excluded from analyses involving GSI all individuals in which ≥10% of ‘intact total length (TL)’ was missing due to caudal filament loss unless the individual had re-attained ≥95% of intact TL by regeneration; we estimated intact TL from the head length (HL) via standard major axis regression of HL and TL for all intact conspecifics of the same sex. Measurements of TL and HL followed Crampton et al. (2016). We assigned individuals to one of six maturational stages: Stage 1 – Immature (never spawned) (unassignable to sex). Stage 2 – Developing. Stage 3 – Spawning capable (subphase I – not actively spawning). Stage 4 – Spawning capable (subphase II – actively spawning). Stage 5 – Regressing. Stage 6 – Regenerating in iteroparous species only (unassignable to sex). This scheme was based on the gonadal maturation scheme of Brown-Peterson et al. (2011), and modified from a similar scheme developed for gymnotiforms in general by Waddell and Crampton (2018).

Breeding periodicity

For descriptive purposes only, we estimated the beginnings and ends of the breeding period of each species by the period of continuous occurrence of actively spawning (stage 4) individuals; that is one or more capture records in consecutive 2-week periods for at least one site per habitat (Table S2, Online Resource 1).

Environmental correlates of breeding

For all individuals with stage 2–5 gonads (i.e. those assigned to sex) we correlated GSI to three sets of parameters: (1) physico-chemical water parameters averaged for each fish, at its capture site, over a 1-month period prior to the date of its capture. (2) 1-year rainfall/1-year river level: the mean rainfall/river level corresponding to the one-month period prior to the date of capture (i.e. from the 1-year rainfall/river level datasets). (3) 5-year rainfall/5-year river level: the mean rainfall/river levels corresponding to the 1-month period prior to the date of capture averaged over all five years of the 5-year rainfall/river level datasets. Our analyses were based on the following two complementary approaches.

Linear mixed effect models (LMEMs) conducted separately for each species and sex

For each species and sex, we utilized LMEMs and model selection to explore the effect of single environmental parameters, or combinations of these parameters, on the response variable GSI. Here we assigned environmental parameters as fixed effects but because our sampling design had a nested structure involving multiple sampling sites, we designated ‘site’ (i.e. F1-F2; T1-T4) as a random effect (Snijders and Boskers 1999).

Model formulation: Step 1. We included all single-parameter models (i.e. the effect of a single environmental parameter on GSI); here we included all three sets of the environmental predictor variables listed above. Step 2. We reduced the number of permutations of multiple-parameter models in four ways: (1) we considered two- and three-parameter combinations of the predictor variables but not models with four or more variables. (2) We only included predictor variables that, when taken individually (in Step 1), were significantly correlated to GSI at the α = 0.05 level. (3) We eliminated combinations of predictor variables with strong multicollinearity (R2 > 0.9) (this only applied to the combination 1-year river level and 5-year river level, p < 0.001, R2 = 0.93). We also later verified that the variance inflation factors of each term in models with multiple terms did not exceed 10, a conservative threshold for assuming non-multicollinearity. (4) We included only models that included interaction terms among the fixed effects (i.e. we excluded fully or partially additive models). Step 3. Using the R package ‘lme4’ (Bates et al. 2015) we generated LMEMs for all single-parameter models from Step 1, all multiple-parameter models retained from Step 2, and a null model of the form GSI ~ 1 + 1|site. For each model we used the R package MuMln (Bartón 2018) to calculate the variance explained by the fixed effects only (marginal R2), and by both the fixed and random effects (conditional R2). Step 4. We examined the residual distributions from all models to confirm that there were no major violations of model assumptions.

Model selection: We used the AICc implementation of Akaike’s Information Criterion as a measure of model fit, using the R package bbmle (Bolker and Team 2017). The AIC model selection procedure ranks models by a trade-off between the explanatory power of a given model and the complexity (number of terms) in the model, thus seeking the combination of environmental predictor variables (and interactions among them) that most parsimoniously explains variance in GSI (Burnham and Anderson 2004). To simplify interpretation of our results, we report only models with ΔAIC values <4, following Burnham and Anderson’s (2004) guideline that models with ΔAIC <2 have substantial support and can be regarded as approximately equally parsimonious, those with ΔAIC ≥4 have considerably less support, and those with ΔAIC >10 have negligible support. Finally, to confirm that LMEMs were appropriate, we considered alternative (non-Gaussian) family/link functions for all models, by AICc.

LMEMs for all species combined

To determine whether the effects of specific environmental parameters (or combination of parameters) on the response variable GSI were generalized across all species, and both sexes, we conducted a separate set of LMEMs in which the factors ‘species’ and ‘sex’ (as well as ‘site’) were assigned as random effects. We confined this approach only to the terra firme streams, which unlike the floodplain lake contained multiple syntopic species for which we collected adequate sample sizes of GSI data. Here we repeated Steps 1–4 from ‘Model formulation’ above to obtain a smaller set of models, which we then subjected to model selection as described above.

Results

The Brachyhypopomus fauna

We collected 3388 Brachyhypopomus belonging to two common floodplain species (B. bennetti Sullivan, Zuanon and Fernandes, n = 588, B. flavipomus Crampton, de Santana, Waddell and Lovejoy, n = 45) and five common terra firme stream species [B. beebei (Schultz), n = 548; B. benjamini Crampton, de Santana, Waddell and Lovejoy, n = 134; B. sullivani Crampton, de Santana, Waddell and Lovejoy, n = 778; B. verdii Crampton, de Santana, Waddell and Lovejoy, n = 756, and B. walteri Sullivan, Zuanon and Fernandes, n = 539]. Based on analyses of body size-frequency distributions and otolith growth increments, Waddell et al. (2019) determined that one of these six species, B. beebei, exhibits a two-year iteroparous life history with breeding in both first year (0+) individuals and second year (1+ individuals). In contrast, the other five species all exhibit a semelparous, annual life history with as single breeding season followed by death [‘uniseasonal iteroparity’ sensu Kirkendall and Stenseth 1985].

Environmental data

Rainfall

Monthly rainfall distribution during the study period (1-y rainfall) resembled monthly rainfall averaged over the previous five years (5-yr rainfall), with a maximum in March and a minimum in July–August (Fig. 2a). We divided the year into a ca. 5-month ‘rainy season’ and ca. 7-month ‘dry season’ by the upper and lower halves of annual rainfall distribution (see horizontal line in Fig. 2a).

Fig. 2
figure 2

Seasonal variation in rainfall (a), Ucayali River level (b), and photoperiod (c). The Horizontal dashed line in (a) bisects the spline fit into the top (rainy season) and bottom (dry season) halves of annual rainfall distribution. The ordinate in (b) represents the range of monthly mean river level during years 2008 to 2014. Dashed lines at 8 m and 3 m represent inundation-points of the highest levee forest, and lowest inter-levee forest, respectively (see Fig. 1f); high-water and low-water seasons are defined by intersection of the spline fit with the 8 m datum

Full size image

River flood cycle

The floodplain sites of our study area are subject to a seasonal inundation cycle of 8–11 m amplitude (mean 9.9 m), with a maximum in April, approximately one month after the maximum in rainfall (Fig. 2b). Monthly river level during the study period was close to the mean for the preceding five-years. There was considerably less year-to-year variation in the five-year river level means than the five-year rainfall means (see error bars in Fig. 2a, b). We divided the year into a ca. 5-month ‘high-water season’ and ca. 7-month ‘low-water season’, defined by intersection of the spline fit with the 8 m datum (inundation-point of the highest levee forest) in Fig. 2b.

Photoperiod

Photoperiod varied over the year from 11 h 50 min to 12 h 25 min (range = 33 min) (Fig. 2c). However, the measure of ‘visible photoperiod’ we describe in Methods varied over a 10-d period from 9 h 55 min to 10 h 46 min (range 51 min) at site F1, and from 9 h 12 min to 10 h 11 min (range = 59 min) at site T1. Therefore, in only a 10-d period, the range of variation in visible photoperiod (which we assume to result from cloud cover variation) reached almost double the entire annual photoperiod range. Variation in water turbidity and substrate/vegetation density likely further increase visible photoperiod variation. Because short-term variation in our measure of visible daylength greatly exceeds annual photoperiod variation we concluded that circannual trends in photoperiod are unlikely to represent a reliable proximate cue for the timing of breeding in Brachyhypopomus. Consequently, we excluded photoperiod from our analyses.

Seasonal changes in habitat and physico-chemical water parameters

Floodplain systems

We observed a dramatic increase in the surface area and underwater root mass volume of floating macrophytes during the high-water season (Fig. 1b), especially for Echinochloa polystachya (Kunth) Hitchc. (Poaceae), Eichhornia crassipes (Mart.) Solms (Pontederiaceae), and Polygonum ferrugineum Wedd. (Polygonaceae). During the low-water period most of the floating meadows were beached (Fig. 1c) and Brachyhypopomus were confined to remnant patches of floating macrophytes in deep water (> 2 m). The floodplain lake sites were characterized by higher and more variable temperature, conductivity, and pH, and by lower DO than the terra firme streams (Fig. 3).

Fig. 3
figure 3

Seasonal variation in water temperature (a), conductivity (b), dissolved oxygen (DO) (c), and pH (d), for floodplain lake (sites F1 and F2) and terra firme stream sites (T1-T4). Ephemeral swamp habitats were not inundated at T1 and T2 in July–August and October. Temperature, conductivity, and pH measurements for the swamps resembled the adjacent main stream and are not presented here. Horizontal bars for high-water/rainy season follow Fig. 2

Full size image

Terra firme systems

The water level of the main courses of all the stream sites varied by <0.25 m through most of the year, except after unusually heavy rainstorms, when they rose by up to 0.5 m for less than 24 h. During the rainy season, the lower courses of intermittent feeder streams and shallow depressions alongside terra firme streams were transformed into shallow swamps for several months (see Fig. 1c). These swamps were characterized by lower DO than the main stream channels (Fig. 3c). We encountered Brachyhypopomus in these swamps as soon as they formed. Here they evidently forage (based on observations of replete stomachs) and spawn (based on the occurrence of many post-larval individuals). Physico-chemical parameters exhibited relatively minor circannual variation in the main stream channels (Fig. 3).

Breeding periodicity

All species exhibited circannual periodicity of the CPUE abundance of spawning capable (gonad stage 3 and 4) individuals (Fig. 4) and of GSI (averaged for all individuals, gonad stages 2–5) (Fig. 5).

Fig. 4
figure 4

Seasonal variation in capture per unit effort (CPUE) abundance of breeding adult females and males of seven species of Brachyhypopomus (gonad stages 3 and 4 only). Some error bars are reported only above/below mean due to space constraints. The horizontal bars below each plot indicate high-water/rainy season (Fig. 2) and breeding period defined by the continuous occurrence of actively spawning (stage 4) individuals (see Methods). Spline fits are not applied to B. flavipomus due to small sample sizes

Full size image
Fig. 5
figure 5

Seasonal variation in gonadosomatic index for females and males of seven species of Brachyhypopomus (for gonad stages 2 to 5). Black circles represent monthly means for sites F1 and F2 (floodplain lakes) and sites T1-T4 (terra firme streams). Spline fits are not applied to B. flavipomus due to small sample sizes

Full size image

The most common floodplain species, B. bennetti, presented a distinct ca. 6 month-long breeding period during the rising-water and early high-water period (November–April) (based on one or more capture records of fully reproductive (stage 4) fish in consecutive 2-week periods for at least one site per habitat; see Methods). The breeding season of B. bennetti was characterized by an abrupt onset and ending (see rapid rise and fall of the CPUE of spawning capable individuals, Fig. 4, and of the mean population-wide GSIs, Fig. 5). During the late dropping-water period and low-water period (June–September) GSIs were low, and fully reproductive individuals were entirely absent from our samples. Circannual periodicity of the two sexes was more phase-synchronized in B. bennetti than in the terra firme species; see spline fits in Fig. 5. We encountered only two reproductive individuals of the floodplain species B. flavipomus, but these were spawning capable in the rising-water period – indicating a similar breeding season to B. bennetti.

All five terra firme stream species (including the 0+ and 1+ year groups of B. beebei) presented a longer (ca. 7–9 months), less distinct breeding season than B. bennetti (Fig. 4; see also Table S2, Online Resources). The terra firme species also exhibited a smaller amplitude of circannual variation in both the abundance of reproductive individuals and GSI than B. bennetti.

Environmental correlates of breeding periodicity

The LMEM-based analyses we summarize in Tables 1 and 2 exclude B. flavipomus from floodplains and B. benjamini from streams due to small sample sizes, thus restricting our analyses to 3209 individuals from one common floodplain and four common stream species.

Table 1 Linear mixed-effects models for the effect of seasonally variable environmental parameters on gonadosomatic index for five species of Brachyhypopomus (for full model results See Tables S3 and S4, Online Resource 1). The most parsimonious model for each species/sex/year-group is reported in bold. U = unsupported model (ΔAICc ≥4) NA = single or combinatorial parameter that did not qualify for inclusion in our models (see Methods for exclusion rationale). NS = model supported with ΔAICc <4 but non-significant at α = 0.05. To conserve space, columns with combinatorial parameters with ΔAICc ≥4 are excluded. For each supported model we present, in order: ΔAICc, AICc rank (in parentheses), marginal R2 (variance associated with the fixed effect), sign of regression (↑ = positive), and slope significance (*** p < 0.001, ** <0.01, * < 0.05). Parentheses around asterisks (for combinatorial models) indicate that the interaction term was non-significant. Temp. = temperature. EC = conductivity. DO = dissolved oxygen
Full size table
Table 2 Linear mixed-effects models for the effect of six seasonally variable environmental parameters on gonadosomatic index for 2617 individuals of four common terra firme species of Brachyhypopomus (B. beebei, B. sullivani, B. verdii, and B. walteri). Sex, species and site were treated as random effects (indicated by prefix “1|” in the model formula). The most parsimonious model is reported in bold. ER = AIC evidence ratio. M R2 = marginal R2. C R2 = conditional R2. CI = confidence interval. DO = dissolved oxygen. EC = conductivity. Temp. = temperature
Full size table

Floodplain systems

For female and male B. bennetti we considered 25 and 16 models, respectively, following reduction of the number of permutations of all possible single and multiple regression models (see Table S3, Online Resources). For females the strongest predictor of GSI was a combination of 1-year river level and pH, with 5-year river level and pH a close tie (ΔAICc <2) (Table 1). For both models the interaction term was non-significant, indicating that the effects of river level and pH (which alone are not recovered by our model-based analyses as strong predictors of GSI) are purely additive (see Fig. 6a). In both cases, when the predictor variables were normalized to units of standard deviation by computing a Z-score for each variable, we observed a stronger effect size (slope) and lower p value for the influence of river level on GSI than pH (for 1-year river /pH combination: river slope = 8.0, p = 4 × 10−9, pH slope = 5.5, p = 5 × 10−4; for 5-year river /pH combination: river slope = 8.8, p = 5 × 10−10, pH slope = 3.6, p = 0.04). In male B. bennetti the best predictor of GSI was 1-year river level, with 5-year river level a close tie (Table 1). We reported additional, moderate support for the additive effect of 1-year river level and pH.

Fig. 6
figure 6

Ordinary least squares regressions of gonadosomatic index (GSI) as a function of the most-explanatory environmental predictors (determined by linear mixed-effects models and model selection, see Table 1). Dashed lines represent 95% confidence intervals

Full size image

In summary, for both sexes of B. bennetti, river level, and to a lesser extent pH, appear together to be the strongest predictors of seasonal variation in GSI, while other environmental parameters are poor predictors. The effect of pH was not expected for reasons we elaborate in the Discussion, and because although pH is correlated positively to GSI during the middle-late breeding season (January – July) (R2 = 0.32, slope = 0.29, p = 0.002), this correlation disappears for the rest of the year (R2 = 0.17, slope = −0.08, p = 0.4) (compare Figs. 2b to 3d).

Terra firme systems

For each sex (and each year-group for B. beebei) of the four stream species, we considered between 7 and 27 possible models (Table S3, Online Resources). The most parsimonious models demonstrate a clear commonality in which, for both sexes of all species (and for both year-groups of B. beebei) the strongest predictor of GSI was rainfall, with the models for 5-year rainfall and 1-year rainfall mostly closely tied (Table 1, see also Fig. 6b-e). In contrast, none of the terra firme species exhibited responses to temperature, conductivity, DO, or pH.

In LMEMS that assigned species and sex to random effects (to determine whether any of the environmental variables exerted a generalized effect across all species and both sexes), we documented strong support for 5-year rainfall as a predictor of GSI (Table 2, see Table S4, Online Resources, for full model outputs). In contrast, we reported negligible support (ΔAICc > 10, evidence ratio ≥ 2 × 104) for effects from the other environmental parameters, including the 1-year rainfall dataset.

Semelparous versus iteroparous species

For the five semelparous species with one-year life spans (B. bennetti, B. benjamini, B. sullivani, B. verdii, and B. walteri), the increase in mean GSI during the early- and mid-parts of the breeding period represented in Fig. 5 results from an increase in the mean proportional gonad mass, as individuals mature. At the end of the breeding period, the reduction in average GSI must arise primarily from post-reproductive mortality, which was confirmed by the disappearance of large adult size classes following breeding in all the semelparous species (Waddell et al. 2019).

For the iteroparous species with a two-year lifespan, B. beebei, the early-breeding season increase in mean GSI for the 0+ group results from first gonadal maturation, while in the 1+ group it instead must arise from gonadal recrudescence (regeneration) following a reproductive hiatus. The post-reproductive decline in mean GSI for the terminal 1+ group of B. beebei results (as in the semelparous species) from terminal post-reproductive mortality, but in the 0+ group it must instead result primarily from gonadal regression.

The 1+ year group of B. beebei (Fig. 5c.i) resembles the 0+ year group (Fig. 5c.ii) in the timing of circannual GSI variation but exhibits a larger amplitude of the annual GSI variation (especially in males) relative to the 0+ group. We also noted a stronger effect of rainfall on GSI in the 1+ year group than the 0+ year group, especially in females (compare slopes and R2 values in Fig. 6b.i to 6b.ii), suggesting that in its terminal year, a larger proportion of individuals time their breeding more precisely to the annual rainfall maximum.

Discussion

General patterns of breeding periodicity

The patterns of breeding periodicity we report for Brachyhypopomus are congruent with studies of other Neotropical fishes. The common whitewater floodplain species, B. bennetti, exhibits a relatively short (ca. 6 month), well-defined breeding period corresponding to the rising and early high-water period. This observation matches an almost universal pattern reported in the literature on tropical floodplain fish reproduction, in which the peak of breeding corresponds to the late rising, or early high-water period (Lowe-McConnell 1964; Schwassmann 1978; Goulding 1980; Junk et al. 1997); for autecological studies of single Amazonian floodplain species see Duponchelle et al. (2007), Crampton (2008), Pérez and Fabré (2007), Mendes-Junior et al. (2015), Pires et al. (2015).

In comparison to floodplain species, terra firme stream species of Brachyhypopomus exhibit longer (ca. 7–9 month), less distinct breeding seasons corresponding to the late dry season and rainy season. They also exhibit less synchronicity between mean male and female GSIs. Relatively long breeding seasons corresponding to a protracted rainy season have been reported for other Neotropical stream species, including some characiforms (Krekorian 1976; Kramer 1978; Alkins-Koo 2000), cypriniforms (Winemiller 1993), and siluriforms (Alkins-Koo 2000). Other stream-inhabiting species, particularly at low-latitudes, have been reported to exhibit year-round breeding, including some characiforms (Kramer 1978), siluriforms (Mazzoni and Caramaschi 1995), cypriniforms (Alkins-Koo 2000), and gymnotiforms (Schwassmann 1992).

Cues for breeding

Munro (1990b) distinguished between ultimate and proximate ‘factors’ that determine circannual breeding periodicity in fishes. Ultimate factors influence offspring survival and growth and are expected to act as long-term evolutionary drivers of breeding periodicity (e.g. habitat availability, food availability, predation). Proximate factors are short-term cues that are accurately predictive of forthcoming seasonal events and that are detectable to sensory receptors from which neural and hormonal pathways elicit a response involving pituitary gonadotropins (e.g. photoperiod, temperature) (Yaron et al. 2003).

Ultimate cues

Our results suggest that the annual breeding seasons of Brachyhypopomus are timed so that offspring growth occurs during the period of lowest predator densities, maximum substrate availability, and maximum food availability. In floodplain systems predator densities are at their lowest (Macedo et al. 2015), and floating meadow root mass and associated aquatic invertebrates reach their highest levels (Goulding 1980; Junk et al. 1997; Henderson et al. 1998) during the high-water period. In terra firme systems, predator densities are likewise presumed to be at their lowest, and food levels at their highest, during the rainy season when terra firme streams flood adjacent low-lying areas, forming extensive shallow swamp systems. These ephemeral swamps represent important foraging and spawning sites, and may act as refuges from hypoxia-intolerant predators (Pazin et al. 2006; Espírito-Santo and Zuanon 2016). In contrast, predator density increases and food availability declines during the dry season, when streams are confined to their main channels.

In conclusion, our results are consistent with the expectation that habitat availability, food availability, and predator densities – all of which are heavily influenced by water level – represent the ultimate factors that regulate breeding periodicity in Brachyhypopomus. We suspect that the more pronounced breeding season of B. bennetti (versus longer, less distinct breeding seasons in terra firme species) arises from the much greater amplitude of the floodplain inundation cycle. This in turn results in more extreme oscillations of predator density and food and substrate availability, and therefore stronger selection against out-of-season breeding events (Lowe-McConnell 1964; Schwassmann 1978).

Proximate cues

Floodplain species

Our analyses indicate that breeding periodicity in B. bennetti is influenced by the additive effects of increasing river level and increasing pH (with 1-yr river level exhibiting a stronger effect size), but is unaffected by temperature, conductivity and DO, and by both short-term (1-year) and long-term (5-year) rainfall.

Experimental studies of aquarium-reared iteroparous gymnotiform fishes have demonstrated that increasing water level, increasing frequency of simulated rainfall, and declining conductivity (either in combination or individually) stimulate gonadal maturation or recrudescence, while reversal of these trends causes an opposite response of gonadal regression (Kirschbaum 1984; Kirschbaum 1992; Kirschbaum and Schugardt 2002). These experiments specifically ruled out pH as a predictor of gonadal development (when varied independently of other factors).

The floodplain lakes in our study area exhibit chemical changes that arise from the ingress of river water from the nearby Ucayali River during the rising water period. This results in an increase in pH to values approximating the parent river (pH ~ 7.0) (Fig. 3d). In contrast, most floodplain lakes in the interior of large Amazonian whitewater floodplains exhibit an opposite response in which pH declines during the rising water period in response to organic decomposition from submerged leaf litter in surrounding inundation forest (Affonso et al. 2011). Because seasonal changes in pH may be highly dependent on the distance of a floodplain lake from its parent river (which varies from a few meters to >50 km in Amazonian floodplains), and because of the absence of responses to changing pH in controlled experiments, we conclude that pH alone is unlikely to be a reliable indicator of rising water level in our study system. The predictive power of pH in our LMEMs may instead arise from collinearity between pH and water level during parts of the hydrological cycle.

Terra firme species

We observed a pattern generalized across both sexes of all species in which rainfall was a strong predictor of GSI. In contrast we found no support for effects from physico-chemical water parameters (most of which exhibited relatively minor circannual variation). These results suggest that the mechanical action of rainfall and/or the indirect effect of the flooding of streams into adjacent swamp habitats likely act as the primary proximate cues for the onset of breeding.

Exogenous versus endogenous cues

True endogenous circannual breeding rhythms that free-run under experimental abolition of exogenous zeitgeibers have been confirmed in several tropical birds and other tropical terrestrial vertebrates (Gwinner and Dittami 1990). Experimental evidence for similar free-running endogenous rhythms has been established in only one (sub)tropical fish species, the Asian catfish Heteropneustes fossilis (Bloch) (Sundararaj et al. 1982). Nonetheless, under natural conditions the formation of yolky oocytes in this species appear to be influenced by photoperiod and temperature, with later oocyte development possibly controlled by a consortium of environmental factors during an annual monsoon season (Sundararaj and Vasal 1976).

Circannual endogenous breeding rhythms are not known in iteroparous Neotropical fishes, although this may be because the appropriate experiments have yet to be conducted (Volpato and Trajano 2005). However, many Neotropical fish species are known to have one-year (annual) life spans in which breeding occurs only once (i.e. semelparity) and thus true circannual endogenous breeding rhythms cannot develop (Aschoff 1981). These include some small characiforms, many cyprinodontiform species (Munro 1990b; Winemiller 1989), and most of the Brachyhypopomus species included in this study.

Kirschbaum (1984), citing Schwassmann (1971), noted that many small tropical characiforms reach spawning capable condition well before the beginning of an annual rainy season. Kirschbaum (1984) also noted that once maturity is attained, small tropical fish species typically fail to exhibit gonadal maturation or regression if environmental conditions are held constant, or if environmental conditions change in the direction that typically predicts gonadal regression. Nevertheless, these species will exhibit rapid gonadal maturation if environmental conditions change in a favorable direction. These considerations led Kirschbaum to formulate an “intermediate control mechanism” hypothesis for short-lived fishes in which the first three stages of oocyte development, sensu Wallace and Selman (1981) (I – primary growth, II – yolk vesicle formation, III – true vitellogenesis) are under strictly endogenous control from biological clocks sensu Gwinner (1986), while the final stage of oocyte development (IV – maturation) is under exogenous control from proximate environmental cues; see similar discussion in Brown-Peterson et al. (2011). In contrast, in several longer-lived, iteroparous gymnotiform species Kirschbaum (1992) noted that environmental factors appear to control not only stage I through III oogenesis, but also final stage IV maturation.

Brachyhypopomus with a semelparous one-year life history have only one opportunity to breed in their life. Our results indicate that these species may simply attain reproductive condition as soon as growth permits, and subsequently complete the final stages of gonadal development – if conditions are favorable – within an annual ‘window’ of optimal conditions for reproduction. Such a strategy, which is consistent with Kirschbaum’s intermediate control mechanism hypothesis, would allow individuals that recruit early into the population to ‘hold back’ from reproduction until conditions are appropriate, but allow individuals that recruit late in the population to breed at the earliest opportunity. We presume that breeding events outside this optimal window result in lower offspring survival and diminished opportunities to find a mate in reproductive condition. Fractional spawning, which is ubiquitous in Brachyhypopomus and common to many small Amazonian fishes (Lowe-McConnell 1979; Azevedo 2010;) may stabilize the intermediate control mechanism by bet-hedging against sub-optimal recruitment times; either by spreading out mortality risk in the resulting offspring across multiple spawning events, and/or by lessening the impact of failed spawning events on maternal fitness (Stearns 1992).

For the single iteroparous species B. beebei, the intermediate control mechanism of Kirschbaum is unable to explain gonadal regression at the end of the first breeding season and subsequent gonadal recrudescence (regeneration) at the beginning of the second season. The life-history strategy of this species may instead resemble that observed in other iteroparous gymnotiforms, in which proximate environmental cues control stage I to III oogenesis, as well as later-stage maturation – resulting in annual cycles of gonadal development under the control of proximate exogenous environmental cues (Kirschbaum 1992; Kirschbaum and Schugardt 2002).

For the iteroparous species B. beebei we observed a larger increase in average GSI (between the non-breeding and breeding seasons) in the 1+ year-group than in the 0+ year group, especially for females (compare Fig. 5c.i to 5c.ii). The disparity may constitute evidence for a pattern of terminal reproductive investment, sensu Williams (1966), in which 0+ year-group individuals exhibit a pattern of relative reproductive restraint, while 1+ individuals show a marked elevation of reproductive effort towards the end of their terminal breeding season, as their residual reproductive value approaches zero. Similar patterns have been observed in mosquitofish populations with two-year life spans (Billman and Belk 2014).

Conclusions

The data and model-based analyses presented herein constitute one of the most comprehensive attempts to understand environmental correlates of breeding periodicity in tropical fishes, and the only study to date that identifies congruent responses to environmental conditions shared by multiple closely-related syntopic species. Our analyses support a pattern of rising-water and high-water breeding in the Amazon floodplain, and a pattern of late dry season and early rainy season breeding in terra firme stream systems. In both systems, circannual breeding periodicity appears to be timed to optimize conditions for larval recruitment and offspring growth – including lower predator densities, increased substrate availability, and higher food availability. Based on correlations between circannual variation in environmental cues and GSI, our study suggests that changes in water level may act as a proximate cue for inducing gonadal maturation in floodplain electric fishes, while rainfall (or the resulting changes in water level and consequent terra firme swamp inundation) may induce gonadal maturation in terra firme stream species. Experimental studies of gymnotiforms have previously noted that changing water level alone can stimulate gonadal maturation. Nonetheless, the sensory bases by which these proximate cues are detected by Brachyhypopomus, the neurohormonal mechanisms that affect gonadotropin levels, and the extent to which GSI is influenced by other environmental changes (such as food availability or the perception of predation risk, neither of which we measured) remain unknown. Brachyhypopomus of the Amazon basin may prove to be a fruitful model for further studies of the proximate and ultimate bases of reproductive timing, particularly because gymnotiforms are amenable to aquarium-based studies of gonadal development in response to changing environmental cues.