Tracking temperate fish reveals their relevance for plant seed dispersal

1. Seed dispersal is a fundamental process for plant communities, especially now that our changing world demands rapid colonization of new habitats. Long-distance dispersal is especially important for plant population persistence and range expansions. 2. The contribution of fishes to plant seed dispersal in aquatic ecosystems (ichthy - ochory) has long been overlooked. Although we know fish disperse seeds, it is largely unknown where, when and how far— especially in temperate regions. Here we studied the potential of fish to disperse seeds locally and over long distances, specifically hypothesizing that (a) dispersal by fish depends on the season, (b) indi - vidual fish contribute differently to seed dispersal and (c) water-regulating struc - tures (barriers) inhibit seed dispersal. 3. We tested our hypotheses by acoustically tracking 71 common carp Cyprinus carpio with 21 hydrophones in a 38-km long study system for > 2


| INTRODUC TI ON
Seed dispersal is a key process in determining the spatial extent of plant communities, and geographical distributions, range dynamics and the genetic structure of plant populations (González-Varo et al., 2013;Levine & Murrell, 2003;Nathan & Muller-Landau, 2000).
Global changes-including an increased fragmentation of landscapes, introductions of non-native species and rapid changes in local habitat suitability-make seed dispersal vital for plant population persistence (McConkey et al., 2012). If plant migration rates are lower than the rates of global changes, a 'migration-lag' can result in the decrease or loss of species (Corlett & Westcott, 2013;Ozinga et al., 2009). Predictive power to assess the dispersal potential of both endangered and non-native plant species is therefore of fundamental and applied concern.
Plants can disperse their seeds via multiple vectors, including water flow, wind and animals (Soomers et al., 2013;Van Leeuwen et al., 2012). The ultimate dispersal distance depends on the combination of vectors a seed uses (Van Leeuwen et al., 2017;Vittoz & Engler, 2007). Most seeds are generally deposited close to the parent plant and dispersal becomes less frequent over long distances (long-distance dispersal, LDD; Nathan et al., 2008). However, LDD is especially crucial in facilitating a link between habitat fragments and influencing the geographic distribution of species through range expansion (Cain et al., 2000;Mokany et al., 2014;Nathan, 2006).
Animal-mediated seed dispersal (zoochory) often disperses seeds fast and directionally towards suitable habitat-which contributes importantly to local processes via dispersal over shorter distances, but may also occasionally lead to LDD.
The most common form of zoochory is endozoochory, which involves seed ingestion, a time period of gut passage with seeds remaining in an animal's digestive system and seed egestion (Pollux, 2011).
The spatial extent of seed dispersal by endozoochory can be estimated by combining data on seed gut passage time with data on the distance moved by the animal during this gut passage time (Rehm et al., 2019). Reliable estimates of seed dispersal distances-and therefore the importance of an animal vector for either local dispersal or LDD-can be calculated from the combination of data on movement of the disperser animal (Rehm et al., 2019) and seed retention times (Yoshikawa et al., 2019). This combination is increasingly used to improve estimations of the importance of seed dispersal by birds and mammals (Kleyheeg et al., 2019;Russo et al., 2006;van Leeuwen et al., 2020;Wehncke et al., 2003;Westcott et al., 2005). However, reliable estimates for seed-dispersing taxa such as reptiles and fish are still lacking (Horn et al., 2011;Piazzon et al., 2012;Pollux, 2011;Valido & Olesen, 2019). This creates a bias in our knowledge on plant seed dispersal and on the importance of plant-animal interactions.
Freshwater fish have long been overlooked as potentially important seed dispersal vectors (Correa et al., 2007;Horn et al., 2011), partly because their movements are notoriously difficult to study.
Seeds dispersed by fish (ichthyochory) are generally from riparian or water plants, and the main mechanism is by endozoochory. The importance of fish-mediated dispersal is mainly documented in studies in the Neotropics (Correa et al., 2015;VonBank et al., 2018), with limited knowledge for temperate regions (Correa et al., 2015;VonBank et al., 2018). There are, however, a number of fish species recognized with potential for seed dispersal, notably in the order of the Cypriniformes (e.g. carps and minnows; Horn et al., 2011). Multiple cyprinids consume and egest vascular plant seeds in experimental setups and field situations (Boedeltje et al., 2015(Boedeltje et al., , 2016(Boedeltje et al., , 2019. Several aspects of ichthyochory have remained understudied due to lack of knowledge on where and how far temperate fish species may move after potential ingestion of plant seeds. First of all, many fish species show seasonal variation in their intensity of local movements and feeding, and/or perform migrations during certain seasons (for e.g. spawning). Knowledge is needed on how the timing of more local and long-distance movements may correspond to the timing of seed availability in the habitat. Secondly, for many animal species-including fish-it is still unknown how variability among individuals affects their contribution to seed dispersal (Zwolak et al., 2020). Space use, movement and habitat occupation can all differ consistently among individual fish within a population Chapman et al., 2011). Individual fish can differ in how effective their digestive systems digest seeds (Pollux, 2017) and can differ in their seasonal and daily movements . How this may impact fish-mediated seed dispersal is largely unknown. Finally, seed dispersal depends strongly on the spatial extent that fish can cover in their habitat, which we have only more recently started to document with more accurate tracking technologies. Anthropogenic activities increasingly fragment river-and lake systems, which prevent fish from accessing spawning, nursing and feeding habitats (Baumgartner et al., 2014;Van Leeuwen et al., 2018). There is a need to better understand how river fragmentation may affect plant dispersal (Jones et al., 2020). Inhibited movement of fish may also impact seed dispersal potential, if physical barriers such as dams, sluices and culverts restrict movements.
The aim of this study was to quantify the importance of fish in temperate freshwater ecosystems, with regard to the spatial extent of seed dispersal. To do so, we combined for the first time detailed requires intact fish populations-because not all individual fish contribute equally to dispersal-and unrestricted connectivity throughout aquatic ecosystems.

K E Y W O R D S
acoustic transmitter, barriers, Cyprinus carpio, endozoochory, ichthyochory, intraspecific variation, seasonality, seed dispersal kernel fish movement data with seed retention time data in a temperate fish species (Van Leeuwen et al., 2016). We selected the common carp Cyprinus carpio L. (hereafter referred to as 'carp') as our study species, because this is considered an important seed-dispersing fish in temperate regions (Horn et al., 2011;VonBank et al., 2018). We tested three hypotheses: (a) carp show seasonal differences in their potential to disperse plant seeds, (b) individual carp differ in their importance for seed dispersal and (c) seed dispersal is affected by physical structures in the habitat. Our expectations were that (a) seed dispersal would mostly occur during late spring or summer when carp most actively move and forage (Wilt & Emmerik, 2008), and seeds are available in highest quantities (Miao & Zou, 2009;Russi et al., 1992); (b) a small proportion of all individuals within a carp population would be responsible for a large part of the seed dispersal events, as bolder individuals are likely to move more and over longer distances (Chapman et al., 2011;Fraser et al., 2001;Klefoth et al., 2017) and (c) physical structures in habitats-like sluices and weirs-are barriers for fish and negatively affect seed dispersal.

| Study species
The common carp C. carpio L. (Cyprinidae) is a widely distributed and well-studied freshwater fish (Pollux, 2011;Van Leeuwen et al., 2016). Carp are omnivores that frequently ingest plant seeds (VonBank et al., 2018), although ingestion may occur unintentionally (Horn et al., 2011). Carp can cover long distances in search of suitable spawning habitat or when foraging (Bajer et al., 2010), which is why movements during summer are largely determined by food availability (Lubinski et al., 1986). The species spawns in spring in vegetated shallow flooded areas when water temperatures reach 15-16°C (Crivelli, 1981).

| Study location
This study was performed in the Noord-Willems canal and its connecting waterways, which is located in the North of the Netherlands Water temperatures for Noord-Willems canal and connecting waterways (river Deurzerdiep, river Drense Aa, canal Winschoterdiep and lake Zuidlaadermeer) were measured monthly from Jan 2015 to June 2017. We used these data to derive a daily average water temperature over the 3 years in the study system by averaging the measurements with equal dates, and visualized this with smoothed conditional means.

| Tracking study in the field
In the study system we released 486 common carp that were sup-

| Seed retention time experiments
To quantify the seed dispersal potential by common carp, we combined the data obtained on fish movement with previously experimentally estimated seed retention times in the digestive tract of carp after the moment of ingestion. Seed retention times of common carp were previously studied experimentally by placing the fish in a flume tank with low-, medium-and high-speed water flows (Van Leeuwen et al., 2016). In short, eight fish were fed pellets with known amounts of seeds from two sedge species (Carex acuta and Carex riparia), and seed egestion was monitored hourly for up to 15 hr. The carp selected for this study (mean fork length 28 cm [1.7 SD] and mean body mass 535.9 g [73.5 SD]) originated from the

Aquatic Research Facilities of Wageningen University and Research
Centre, the Netherlands (ARF-WUR). These experimental fish were substantially smaller than the fish used in the tracking study, and body mass can affect intestine mass, length and seed passage rates.
Generally, gut passage times of larger fish will be longer than that of smaller fish (Yoshikawa et al., 2019), hence, our estimates of maximum dispersal distances are conservative.

| Data selection
The raw data consisted of >1.8 million registrations, with each timestamp consisting of the fish and hydrophone identity linked to its location in the study system. The data calculations and statistics were performed on a selection of all available data. Firstly, we assumed an acclimation period for the tagged fish after release before starting the study by deleting the first 30 days of registrations (Jenkins et al., 2014;Jepsen et al., 2002). Secondly, to correct for possible biases due to variation in the time each carp was monitored, we selected exactly one or exactly two year(s) of data for each carp; from the first registration after the acclimation period up to exactly 365 or 730 days later. Carp without at least 1 year of data were excluded from the analyses (n = 17 carp discarded).
Discrepancy in the amount of data generated per individual carp is due to a stop in registration, which could have had multiple causes, e.g. death, disappearance from the study area, inactivity or only very local activity of the carp beyond the reach of a hydrophone or malfunctioning of the transmitter. Thirdly, we detected many false registrations on hydrophones 1 and 4, of which the transmitter signals had passed through the doors of two sluices but the fish did not. The number of registrations over time was-except for three cases-too low for a carp to have passed through the sluices. We include these registrations as movements of the carp to the area, but not as connectivity beyond the sluices. The remaining data consisted of nearly 1.5 million registrations of 45 carp (in total 62 carp-years, Figure S1).
To calculate potential seed dispersal from the combination of seed retention time and movement data we made three basic assumptions: (a) a dispersal event was considered when a fish moved between hydrophones, (b) a dispersal event would last 15 hr, based on the seed retention time and (c) seeds were uniformly available in the environment.

| Data analyses
All analyses were conducted in R (version 4.0.2; R Core Team, 2020).
To test our first hypothesis we (a) compared the monthly carp activity, and (b) calculated realistic dispersal distance kernels, that estimated seed deposition probabilities by moving fish in our study system. Carp activity was measured in total distance moved per week in kilometres, and the monthly activity was compared by using a Kruskal-Wallis analysis and Dunn test. To calculate the dispersal kernels, we used the experimentally assessed chances of seed egestion over time, measured as hourly seed egestion probabilities calculated during trials with carp swimming at medium (0.25 m/s) and high (0.40 m/s) water velocities (Van Leeuwen et al., 2016). Data of both velocities were merged to get an estimation of digestive activity during normal gait transition speeds of carp (Tudorache et al., 2007).
Data from the two plant species were not significantly different (Van Leeuwen et al., 2016) and were therefore merged to obtain a general estimation.
We constructed four seed dispersal distance kernels-one for each season-by combining all dispersal events in each season as follows. First, we ran through the data chronologically for each individual fish, and started 0-15 hr intervals when a fish moved from one (source) hydrophone to a next (destination) hydrophone (indicating movement). The last registration at the source hydrophone was assumed as the moment of ingestion of 100 seeds, and the start of a 15-hr time interval (termed 'dispersal event') at which these ingested seeds could potentially be egested. Based on the timing of arrival at a next hydrophone, we calculated distances moved over these 0-15 hr time intervals. If a fish moved to more than one station within this 15-hr interval or returned to the source hydrophone, distances were corrected accordingly. In total we constructed 1,044 movement trajectories.
Secondly, we combined this spatio-temporal data with experimentally assessed seed survival and egestion probabilities over time since ingestion (Van Leeuwen et al., 2016). Each calculated distance away from a source hydrophone had a corresponding time since ingestion, i.e. a retention time in the fish after ingestion. This resulted in a database in which each of the 100 seeds that were ingested at the source hydrophones and survived digestion obtained an assigned dispersal distance. Dispersal distances that matched exact distances between pairs of hydrophones were strongly overrepresented in the obtained distributions, because fish were always assumed to be moving when they were not within reach of a hydrophone. However, they could be stationary for several hours at exact positions of hydrophones. To avoid this bias in the distance distributions we removed all distances that exactly matched the distances between pairs of hydrophones, to obtain distributions that were independent of the chosen hydrophone setup as much as possible.
Thirdly, we fitted probability density functions to the four obtained distributions to mathematically describe the seed dispersal kernels for each season (Nathan et al., 2012) using the function 'fitdist' in package fitdistrplus (Delignette-Muller & Dutang, 2015).
Informed by the shape of the distributions we selected four likely one-or two-parameter density functions (i.e. exponential, lognormal, Gamma and Weibull) to the data and estimated the parameters of each distribution based on maximum-likelihood estimations.
We selected the best fitting model for each season based on the Akaike information criterion (Burnham & Anderson, 2002), and calculated Kolmogorov-Smirnov (KS) statistics to assess model accuracy. Lastly, we compared the distance distributions among seasons using Kruskal-Wallis and Dunn tests (Table S1).
To test our second hypothesis, i.e. to assess if individual carp differed in their importance for seed dispersal, we plotted the distributions of maximum dispersal distances and the median dispersal distances per individual fish. The ranges of these distributions indicate the variation among individuals in their contribution to seed dispersal. We used general linear models to test possible external effects on the (natural log-transformed) dispersal distances. In a first model we tested the possible effect of body mass upon release (as continuous explanatory variable) and its initial release location (as factor with four levels) on 43 carp with at least 1 year of data. In a second model we tested for a possible effect of the year after release (as factor: 1st or 2nd year after release), on n = 14 carp for which we had 2 years of data.
To test our third hypothesis, i.e. the potential effect of fragmentation in the study system, we assessed the connectivity within the study area. The number of times carp moved between two hydrophones was assumed to reflect the connectivity between two neighbouring hydrophones. For this analysis we worked with the full dataset including 69 carp (two tagged carp were never registered after release), because we expected effects of barriers to be independent of a possible acclimatization period or exact duration of fish tracking. We calculated for each pair of neighbouring hydrophones how many fish moved directly between them, which identified areas in the study system with low and higher connectivity. We defined low connectivity as when fish moved <10 times between neighbouring hydrophones, and analysed whether they were related to possible anthropogenic barriers by means of a Wilcoxon rank sum test with continuity correction.

| RE SULTS
We detected a total of 1,044 movement trajectories by carp-with which we simulated the dispersal of 75,529 seeds throughout the study system. Median dispersal distance of these seeds was 1,207 m, ranging from zero to 16.1 km from the place of assumed ingestion. The seed dispersal kernels all looked relatively similar, starting with high dispersal probabilities in a radius of 2-4 km from the locations of seed ingestion, and a decline of this probability at distances between 4 and 7 km (Figure 3; Figure S2). The best mathematical descriptions for the probability density functions for spring, summer and winter were Gamma distributions, and the Weibull distribution best fitted the data for autumn (all KS-statistics <0.1, Figure S3).
Seasonal differences-under the assumption of constant seed ingestion probabilities year-round-were notably that (a) in winter there is a relatively high probability for short distance dispersal but a low long-distance dispersal potential, (b) in spring there is the greatest potential for LDD with a fat tail until 7 km, and a maximum beyond 16 km, (c) in summer and autumn there are high probabilities until 4 km and moderate probabilities for seed dispersal until 7 km (Table 1). Although summer and autumn differed in the number of dispersal events, they did not significantly differ in their dispersal distances (Dunn test, p = 0.46), while all other season combinations did (Kruskal-Wallis chi-squared χ 2 (3) = 1876.2, p < 0.001, Table S1).
The top 1% of all dispersal distances, indicative as a criterion for LDD (Cain et al., 2000), occurred once during autumn, twice during summer and eight times during spring (Table 1).

F I G U R E 3
Dispersal distance kernels for each season, visualizing how the seed dispersal probability depends on distance from the location of assumed seed ingestion. Probabilities of seed survival, seed ingestion and the pattern of seed egestion over time were assumed constant year-round. Hence, seasonal differences in dispersal kernels are only due to differences in fish movements (n = 45 common carp). For separate kernels per season including confidence intervals see Figure S2 Distance (  Individual carp varied in their potential contribution to seed dispersal (Figure 4; Figure S4). The distribution of median and maximum dispersal distances both showed gradients within the population, reflecting variation in an order of magnitude between individual carp. Variation in contribution to seed dispersal among individuals ranged from dispersal within 1 km to individuals dispersing seeds >16 km. Individual maximum dispersal distances were independent of the four different release locations of the individuals (linear model F 3,1039 = 1.81, p = 0.14). Fish initially released with the highest body masses dispersed seeds maximally 577 m farther than fish with the lowest body masses, indicating only a minor influence of body mass on the dispersal kernels (F 1,1039 = 11.90, p < 0.001, Figure S4). For the 14 individuals that were monitored for 2 years, some moved more during the first year F I G U R E 5 (a) Figure showing the connectivity of the study area. The lines connect pairs of neighbouring hydrophones, and are coloured according to the number of times a carp travelled between the two hydrophones. The connectivity ranges from zero (grey) to low (yellow), moderate (orange), high (dark orange) and intense (red). More information can be found in Table S2. (b) Figure with only the 14 pairs of neighbouring hydrophones that had restricted connectivity (<10 movements), with colours indicating their relation to a sluice or weir. Eleven pairs were directly separated by a sluice or weir (light green), and two pairs (dark green) indirectly: meaning that carp had to cross a sluice or weir to reach this pair of hydrophones. Low connectivity unrelated to sluices or weirs was only observed once (yellow)

(a) (b)
of monitoring and some moved more during the second year-but the year of monitoring did not consistently affect the contribution of fish to seed dispersal ( Figure S5).

The analysis of fish movements showed where carp have and
have not moved (Figure 5a). We identified 14 pairs of neighbouring hydrophones with <10 fish movements in 2 years, and of these nine were separated by sluices and four by overshot weirs (Figure 5b).

| D ISCUSS I ON
This study provides quantitative predictions on how far, when and how often temperate fish species may disperse wetland plant seeds.
We show that seed dispersal by common carp is quantitatively most important during summer and autumn, when available seeds are locally dispersed commonly up to 4 km. There is potential for LDD up to 16 km during spring, but only for those seeds that are by this time still available in the habitat for ingestion. The contribution of individual fish to seed dispersal varied among individuals and within individuals by year, which emphasizes the importance of preserving all fish in populations to preserve the ecosystem services that a population provides. Sluices and weirs formed barriers in our study system that restricted fish movement, and strongly reduced the potential for ichthyochory.

| Seasonal difference in seed dispersal potential
Freshwater fish are known to move for various reasons, including search for food, avoidance of predators and migration to spawning areas (Brönmark et al., 2014). We observed the most movement in the common carp in the month of May, which coincided with rising water temperatures in spring towards above 15°C: the beginning of movement towards suitable spawning habitat in carp (Crivelli, 1981, Figure 2).
Monthly differences in activity translated into varying contributions to potential seed dispersal, as illustrated in the simulated dispersal kernels that show differences in quantity and extent of seed dispersal among the seasons. In summer and autumn the fish showed frequent dispersal events and good potential for seeds to be transported over several kilometres from their location of ingestion. The farthest movements-possibly resulting in LDD-occurred in spring.
However, this likely does not coincide with maximum seed availability in the habitat nor with times that fish consume most seeds (Boedeltje et al., 2019). Seed availability across all wetland species is generally lower in spring due to degradation during winter (VonBank et al., 2018) and germination in spring (Leck & Simpson, 1987). These factors combined lead us to deduce that seed dispersal over moderate distances by common carp is quantitatively most important during summer and autumn (i.e. between June and November), as these are times that freshwater fish such as carp move often and seeds are available for consumption.

| Individual differences in seed dispersal potential
Intraspecific differences among seed dispersers are in need of more attention (Zwolak et al., 2020). Our study showed considerable variation within the carp population for dispersal distances by individuals, for both maximum and median distances. Some of the individuals that we tracked for 2 years also showed different behaviour in these 2 years. This indicates that (a) under similar circumstances different individuals can contribute differently to seed dispersal, and that (b) the same individual in a different situation can contribute differently to seed dispersal. This observed variation in disperser quality among individuals supports our hypothesis that within a certain year only a portion of all individuals in a fish population is responsible for the majority of long-distance dispersal events. With only 2 years of data per individual we cannot yet assess whether some individuals may always contribute more to seed dispersal than others-but consistent life strategy differences between individuals have been documented in carp (Klefoth et al., 2017). This intraspecific variation in movements can sometimes even exceed interspecific variation (Harrison et al., 2019).
The probability for seed dispersal is thus not only dependent on the fish and plant species, but also to a large extent on the individuals within each species. Consistent variation in movement within a species is not always ascribed to animal personality (Fraser et al., 2001), but can in some studies be explained by gender (Stehfest et al., 2014) or related to weight . In our study we did not have information on the sex of the carp, and the effects of body mass on maximum dispersal distances were only weak. Hence, we here document variation with the aim of stimulating future studies to detect potential underlying causes.

| Sluices and weirs as barriers for seed dispersal
The connectivity analysis suggested that water-regulating structures in the study area inhibited carp movement and its consequential seed dispersal. While a sluice is mostly closed and opens its gates occasionally, a weir is a permanent one-directional passage.
Both can inhibit movement of fish (Baumgartner et al., 2014;Harris et al., 2017). To date, no study had assessed how endozoochorous dispersal may be affected by barriers (Jones et al., 2020), but our study signifies how river fragmentation by barriers may have direct consequences for fish as well as indirect consequences for the ecosystem services they can fulfil.

| Variation among fish and plant species
For this study we selected common carp as our model species because it is an opportunistic omnivore that is known to regularly ingest large numbers (>1,000) of plant seeds and invertebrate resting eggs in temperate habitats (Crivelli, 1981;Pollux, 2011). For common carp, we showed potential to disperse seeds over several kilometres, focusing on Carex sp. plant seeds for which detailed retention time data were available. These results provide indications for the ichthyochory potential of omnivorous cyprinids, however, we should note that ichthyochory is known to vary extensively among both fish and plant species . For instance, variation in feeding strategy among freshwater cyprinids can lead to variation in the plant species each fish species can disperse (Boedeltje et al., 2019).
Previous experimental work has shown that seed retention times-and therefore potential dispersal distances of surviving seeds-vary less among seeds with different morphologies than their survival (Boedeltje et al., 2015(Boedeltje et al., , 2016. This implies that the shapes of the here obtained dispersal kernels (Pollux et al., 2006)-resulting from only fish movements and retention times-could be interpreted more broadly, as these distributions (Gamma, Weibull) are also in line with what is common in zoochory. However, the magnitude of the here calculated seed dispersal probabilities will likely vary extensively depending on seed morphologies (e.g. soft, large seeds rather than small, hard seeds) and may be much lower or higher for fish species with different digestive systems (Boedeltje et al., 2019;Pollux, 2011;Pollux et al., 2006).

| CON CLUS IONS
This study showed that freshwater cyprinids can potentially disperse seeds over several kilometres in temperate regions, that dispersal quantitatively depends on the season, and that individual fish can differ substantially in their contribution to dispersal. This variation calls for conservation of intact fish populations throughout the annual cycle. Potential dispersal distances exceeded 16 km, except for when anthropogenic barriers were present. Together this implies that conservation of fish and restoring connectivity in underwater landscapes will aid to preserve a potentially important service fish provide to the functioning of ecosystems.

ACK N OWLED G EM ENTS
The authors thank Hermen Klomp from Waterschap Hunze en Aa's for providing records of the monthly water temperature for Noord-

DATA AVA I L A B I L I T Y S TAT E M E N T
Data deposited in the Dryad Digital Repository https://doi.org/ 10.5061/dryad.x3ffb g7hh (Mulder et al., 2020).