Megaherbivores are typically regarded as agents of top-down control, limiting woody encroachment through destructive foraging. Yet they also possess traits and engage in behaviours that facilitate plant success. For example, megaherbivores can act as effective endozoochorous seed dispersers. However, studies on facilitative roles are heavily biased toward the African savanna elephant (Loxodonta africana), with little attention paid to other species or to effects beyond germination across early ontogenic stages. The African black rhinoceros (Diceros bicornis), an obligate browser that exhibits frugivory and defecates in fixed dung middens, may offer ecologically distinct dispersal services. We conducted controlled experiments to investigate whether interactions between black rhinos and Vachellia erioloba, a leguminous tree of ecological importance in arid savannas, enhance germination, early seedling development, or seedling resilience to herbivory. Germination was compared among dung-derived seeds, untreated controls, and chemically scarified seeds. Seedling growth was assessed in dung versus sand, and under simulated black rhino herbivory. Dung-derived seeds germinated most steadily and produced the highest cumulative germination (+40 %) over the longest period (+13 days). Growth trials revealed that dung substrates did not enhance initial growth. Rather, seedlings being older conferred greater resilience to biomass loss than exposure to different substrate conditions. Our results provide the first experimental evidence of an apparent mutualism between black rhino and V. erioloba. This relationship is not driven by enhanced seedling development through legacy effects of gut passage, nor by dung conditions, as expected. Instead, it stems from the effects of gut passage on germination. In addition to increasing total germination, gut passage accelerates germination and extends the germination period, producing a seedling cohort with both older individuals and greater age variation — a population structure that may enhance persistence beyond the germination bottleneck. This research supports a more nuanced view of megaherbivores as both disturbance agents and mutualists in arid ecosystems.

Germination assays

Seed collection

Seeds were initially collected from black rhino dung to identify the study species, followed by mature pods.

Dung samples were collected over two weeks each in April and July 2023, post-wet season (November to March). By this stage, all indehiscent pod-bearing species had produced mature pods, so their seeds should have been present in boluses if consumed by rhinos and able to withstand gut passage. The two collection periods were intended to contrast seeds with minimal (April) vs prolonged (July) environmental exposure. With assistance from field rangers, fresh black rhino dung (<2 days) was located by backtracking from waterholes, and all intact seeds were removed from boluses. As rhinos have distinct territories, dung from a midden typically reflects the diet of only a few individuals. To maximize the number of individual rhinos sampled and capture variable gut passage effects on seeds, Dung seeds were collected across Tswalu. This also optimized the genetic diversity among seeds to account for potentially varied germination responses.

Only V. erioloba seeds were found intact and in high densities (see SI). Accurate identification was confirmed by germinating and growing seedlings during the experiment, until distinctive characteristics emerged.

In April, fallen pods were collected from beneath V. erioloba trees randomly distributed across Tswalu. Pods were diversely shaped and sized, particularly as distthe distance between trees increased (pers. obs.), which could be a morphological indication of genetic diversity, as found in Vachellia caven by Pometti et al. (2010).

Seeds were processed immediately at the C.L.I.ME Laboratory, Stellenbosch University (Climate and Invasions: Mechanisms in Ectotherms; Department of Botany and Zoology). Pods were opened and seeds submerged in distilled water to check for bruchid beetle infestation and viability. Only flawless, sinking seeds were selected as sample individuals. Dung seeds were rinsed with distilled water and processed similarly. By minimizing variability in initial seed quality, processing standardised the viability potential between seed treatment groups, facilitating a robust comparison. The first germination trial commenced within a week; it was delayed by 12 weeks lateto incubator constraints, during which seeds were stored at 22 ± 2 °C in paper bags.

Microscope assessment

Before the germination assays (Experiments 1 and 2), 10 seeds each from pods and dung were examined under a stereo microscope (LEICA M125, with MC190 HD camera), to check for visible differences in the external condition of the seed coats.

Refer to Figure 1 for a visual summary of experiments 1-4. Methodology is detailed below.

Experiment 1: Disentangling the effects of gut passage on germination

To assess whether gut passage enhanced germinability, our first germination trial compared dung seeds to pod seeds. Only April-collected seeds were used.

900 pod seeds and 300 dung seeds were disinfected in  3% chlorine bleach (1:5) for two minutes. Pod seeds were divided into three groups (n=300 ea.): untreated Control, HCl-treated, and HCl + GA-treated. To simulate acid scarification alone (disengaging from other digestive processes like mechanical abrasion and exposure to heat or enzymes), the second group was soaked in 0.000033 M HCl for 20 hours, followed by 0.00977 M HCl for another 20 hours. Protocols were designed based on digestive information in Clemens and Maloiy (1982) and Dierenfield (1993). The third group was soaked in gibberellic acid (GA₃; 50 mg/L, Kimix, ≥90% purity) for 8 hours following HCl treatment. The control and GA groups served as minimum and maximum germination benchmarks, contextualising the Dung and HCl treatments. See SI for a detailed explanation and the underlying rationale of the HCl and GA treatments.

The germination trial was run in a thermostatically controlled growth chamber (Panasonic MIR-254). Each treatment group (Dung, Control, HCl, and GA) comprised 20 replicate Petri dishes (9 cm), each lined with two layers of grade 1 Whatman filter paper, pre-moistened with 5 mL of distilled water, and containing 15 seeds. To prevent moisture loss, all Petri dishes were sealed with Parafilm M®. The growth chamber was equipped with warm white fluorescent tubes (KD21 Striplight; Radiant, 16 W) and set to a fixed temperature regime of 16/32 °C with a 12-hour thermoperiod and a matching light/dark photoperiod. Petri dishes were monitored daily for germinants (radicle ≥1 mm) over 86 days (~11 weeks), well beyond the typical germination period for V. erioloba seeds (6–8 weeks; Palgrave, 1983). Moisture levels were maintained throughout. To avoid systematic effects related to positioning in the incubator, all dishes were re-randomized every two days, as in Yang et al. (1999). Germinants were transplanted to seedling trays every 14 days.

Experiment 2: Assessing pod exposure effects on seed germinability

This trial compared the germinability of dung seeds collected in April and July. Both sets were from the same cohort, but the July seeds had experienced several additional months of environmental exposure prior to ingestion by black rhinos. We hypothesised that July seeds would show reduced performance as a result, despite being visually intact.

We used 600 dung seeds from each month and 1,200 pod seeds (April-collected), divided into two control groups (n = 600 ea.). Each treatment (Dung April, Dung July, Control 1, Control 2) was sown in 40 replicate Petri dishes. Aside from the increased sample size, the experimental design replicated Experiment 1. The trial was terminated after 12 days due to time constraints linked to upcoming fieldwork.

Testing early seedling growth and resilience

Experiment 3: Monitoring early seedling development

Seedlings from Experiment 1 were used to assess whether gut passage conferred legacy effects on early seedling development. We expected Dung seedlings to grow faster and taller than the Control seedlings.

Germinated seeds were removed from Petri dishes every 14 days and transplanted into seedling trays filled with builder’s sand and perlite (1:1), a nutrient-deficient medium used to avoid confounding effects of substrate before the experiment. The trays were housed in a greenhouse with full-spectrum daylight supplementation and relatively stable temperatures (16/32 ± 5 °C). Each seedling received 5 mL of distilled water weekly. After 28 days, seedlings were transplanted into 15 cm pots and moved to a climate-controlled room for monitoring. This staggered schedule created four batches, each spaced ~14 days apart, that were all introduced at ~42 days.

Pots were filled with either Tswalu sand or fresh (1-2 day old) black rhino dung, collected in the field and sealed in heavy-duty bags for ≤2 weeks. Substrates were manually sifted to remove visible seeds. Each pot was layered with 5 cm of rockwool, 250 mL of perlite, and topped with either sand or dung.

Seedlings from each germination treatment group (Control, Dung, HCl, and GA) were assigned as evenly as possible to each substrate. At large, five seedlings were planted per pot. However, when the number of seedlings in a batch was not divisible by five, one pot contained fewer individuals to accommodate the remainder. This variation was unavoidable, as batch size was determined by the total number of seeds that germinated during the 14-day window of the preceding germination trial. The variable pot density (i.e., the number of seedlings per pot) was recorded at planting and updated during biweekly monitoring to reflect any mortalities. Four additional pots (two per substrate), housed i-Buttons at 1 cm depth to monitor soil temperature.

The room maintained a 16/32 °C cycle and 12-hour photoperiod, lit by two full-spectrum, 600W LED plant ballasts (110 x 180 cm), delivering 500 ppm. A malfunction spiked daytime temperatures to a maximum of 41 °C for ~one week, with minimal effect on nighttime temperatures. This short-term fluctuation precisely replicated a Kalahari heat-wave, which occurs during the summer season we were simulating. Analysis revealed that seedling growth was unaffected.

Pots received 150 mL of water twice weekly. Before watering, weekly soil moisture was recorded with an EC-5 probe (ProCheck logger). Pots were re-randomized biweekly. Plant growth measurements, namely plant length and basal stem diameter, were recorded every two weeks. Plant length was measured by tracing a string from the base of the cotyledons (where a scar remains after termination) to the apical bud and recording the string length. Stem diameter was measured using digital callipers below the first true leaves.

Experiment 4: Clipping, to simulate black rhino herbivory

To assess whether rhino dung promotes development and confers resilience to disturbance, seedlings were clipped to 2 cm, after which their recovery in either substrate was monitored. We expected seedlings grown in dung to recover most effectively, as they could rely on substrate-derived nutrients and moisture.

Plants were cleanly clipped with pruning shears at 45 °, replicating the characteristic bite of a black rhino (Owen-Smith, 1988). Clipping prevents apical dominance and consequently induces branching, a common response to disturbance among Acacias in arid environments (Archibald & Bond, 2003). To accommodate this, each branch length was traced with straws and summed to calculate the total plant length. All other monitoring protocols matched those in Experiment 3.

All seedlings were clipped on the same day, but because batches were staggered by ~14 days, there was a ~51-day age gap between the oldest and youngest seedlings. This design allowed us to examine whether resilience increased with seedling maturity. The full monitoring period, including early growth and post-clipping recovery, lasted 93 days (±15 weeks).

Data analysis

All statistical analyses in this study were conducted using the R statistical environment, version 4.4.2 (R Core Team, 2024).

Experiments 1 and 2: Disentangling the effects of gut passage on germination and assessing the impact of field exposure on seed germinability

We applied time-to-event models to analyse seed germination, as recommended by McNair et al. (2012) and Romano and Stevanato (2020), with methodological guidance from Onofri et al. (2010) and Scott et al. (1984). Unlike germination traditional methods (indices, ANOVAs, or linear regressions), time-to-event approaches model the distribution of individual germination events, rather than cumulative germination proportions, offering greater statistical power. We considered three main types of models: Kaplan-Meier (non-parametric), Cox proportional-hazards (semi-parametric), and Accelerated Failure Time (parametric) models. We formatted germination data from either trial in binary format. After inspecting survival curves and log-log plots, we found the data exhibited a slightly leptokurtic distribution, which violated parametric assumptions, particularly in the second trial. Consequently, the non-parametric (Kaplan-Meier) and semi-parametric (Cox proportional-hazards) approaches were chosen for reporting in both analyses.

Models were run using the survival package (Therneau, 2023; version 3.8-1) and visualised with survminer (Kassambara et al., 2021; version 0.4.9). Statistical significance was assessed using log-rank (Kaplan-Meier) and Wald (Cox) tests, with p-values reported and 𝛼=0.05. The Kaplan-Meier log-rank test models the survival function 𝑆(𝑡), or the probability of a seed not germinating at time 𝑡; this is the inverse of the cumulative incidence 1−𝑆(𝑡)/𝐹(𝑡), representing the instantaneous probability of germination. The Cox model estimates the hazard function 𝐻(𝑡), or the cumulative "risk" of germination.

Experiments 3 and 4: Monitoring early seedling development and clipping, to simulate black rhino herbivory

Seedling growth in sand and dung substrates, pre- and post-clipping, was analysed by means of Generalized Linear Mixed Models (GLMMs), using the glmer function (Lme4 version 1.1-35.4; Bates et al., 2015).

Basal stem diameter, converted to stem area (πr²),  was selected as the main response variable because this measure demonstrated the best model fit. This aligns with findings that stem diameter scales with leaf biomass across a wide variety of tree species (Enquist & Niklas, 2002; Sun et al., 2019). Plant growth data were separated into two datasets (pre- and post-clipping) that were treated as separate experiments, so analysed independently but using the same statistical framework. Data from the first 18 days were excluded from Experiment 3 because seedlings only had a radicle and cotyledons protruding from the seed, making stem-related measurements unfeasible. Monitoring thus spanned 4 weeks for Experiment 3 (pre-clipping) and 9 weeks for Experiment 4 (post-clipping).

Before the main analysis, substrate differences in soil moisture and temperature were tested using Welch’s t-tests. Collinearity among predictors was evaluated via Pearson’s correlation matrices (chart.Correlation function; PerformanceAnalytics version 2.0.4; Peterson & Carl, 2020) and variables with a coefcoefficient75 were excluded from joint models. Model distributions were selected based on residual plots and fir diagnostics (fitdistrplus version 1.2-1; Delignette-Muller & Dutang, 2015). Although the data were normally distributed, heteroscedasticity necessitated a semi-parametric approach. The Gamma distribution provided the best fit for both datasets and was theoretically applicable, since it accommodates positive, skewed, and continuous data, and is commonly used for repeated measures (Eric et al., 2021). Unlike the normal distribution (mean, SD), the Gamma distribution uses shape and scale parameters, which offer more flexibility in variance (Eric et al., 2021). We applied a log link, which also accommodates positive-only data with positively-skewed errors and can represent an underlying multiplicative process, which is common in ecology.

Candidate model selection was conducted to determine optimal model structure (Tables S1 and S2). While the experimental treatments were retained in all candidate models, the selection process considered the inclusion of additional biologically motivated covariates like soil moisture, temperature, and pot density, as well as feasible interaction terms. Candidate models were ranked using the second-order Akaike Information Criterion (AICc), computed via the AICc function (AICcmodavg version 2.3-3; Mazerolle, 2023). AICc was preferred over AIC due to small size, since it includes a correction for low numbers of observations relative to model parameters (Hurvich & Tsai, 1989). Model rankings were based on ΔAICc values and cumulative Akaike weights (aictab function). Following Burnham & Anderson (2002), models with ΔAICc ≤ 2 were considered equally plausible; those with ΔAICc between 2–7 had less support, and those with ΔAICc > 7 were excluded from interpretation. If ΔAICc ≤ 2 between the top models, biological relevance was considered before selecting the final model.  Multicollinearity among covariates was assessed using variance inflation factors (VIFs), or generalised VIFs (GVIFs) for models with interaction terms, via the vif function (car version 3.1-2; Fox et al., 2019). We checked for overdispersion using the dispersion_glmer function (blmeco version 1.4; Korner-Nievergelt et al., 2015). Marginal and conditional R² values were calculated via r.squaredGLMM (MuMIn version 1.48.4; Bartoń, 2024).