Prey fish returned to Forster’s tern colonies suggest spatial and temporal differences in fish composition and availability

Predators sample the available prey community when foraging; thus, changes in the environment may be reflected by changes in predator diet and foraging preferences. We examined Forster’s tern (Sterna forsteri) prey species over an 11-year period by sampling approximately 10,000 prey fish returned to 17 breeding colonies in south San Francisco Bay, California. We compared the species composition among repeatedly-sampled colonies (≥ 4 years), using both relative species abundance and the composition of total dry mass by species. Overall, the relative abundances of prey species at seven repeatedly-sampled tern colonies were more different than would be expected by chance, with the most notable differences in relative abundance observed between geographically distant colonies. In general, Mississippi silverside (Menidia audens) and topsmelt silverside (Atherinops affinis) comprised 42% of individuals and 40% of dry fish mass over the study period. Three-spined stickleback (Gasterosteus aculeatus) comprised the next largest proportion of prey species by individuals (19%) but not by dry mass (6%). Five additional species each contributed ≥ 4% of total individuals collected over the study period: yellowfin goby (Acanthogobius flavimanus; 10%), longjaw mudsucker (Gillichthys mirabilis; 8%), Pacific herring (Clupea pallasii; 6%), northern anchovy (Engraulis mordax; 4%), and staghorn sculpin (Leptocottus armatus; 4%). At some colonies, the relative abundance and biomass of specific prey species changed over time. In general, the abundance and dry mass of silversides increased, whereas the abundance and dry mass of three-spined stickleback and longjaw mudsucker decreased. As central place foragers, Forster’s terns are limited in the distance they forage; thus, changes in the prey species returned to Forster’s tern colonies suggest that the relative availability of some fish species in the environment has changed, possibly in response to alteration of the available habitat.


Introduction
We examined collections of prey fish that were returned to, and dropped at, 17 Forster's tern colonies in San Francisco Bay over an 11-year period. We determined 1) if the species composition of prey items returned to tern colonies varied among colonies and 2) whether the relative abundance, relative mass, and size of the most common prey items changed over time.

Sample collection
From 2005 to 2015, we monitored up to 17 Forster's tern colonies (hereafter colonies) in the southern San Francisco Bay Estuary, mainly located on islands within managed ponds (Fig 1), and entered colonies weekly to monitor nests and chicks [22]. We collected samples each time a colony was visited, as early as April 17 and as late as Sept 19, depending on when the colonies were active during each year. We refer to the colonies by the name of the pond in which they were located, and in some cases a colony was comprised of several adjacent islands within the same pond. Most tern colonies within the southern San Francisco Bay Estuary were monitored each year. The colony locations of terns varied annually; therefore, some islands were used during most years whereas other islands may have been used during only one year.
During weekly colony visits, we searched for any fish or invertebrates that were on the ground in the colony. Forster's terns carry prey back to the colony to feed to their chicks or deliver to their mates [23]. Periodically the prey are dropped on the ground prior to feeding or the chick may reject the prey. Because Forster's terns are the only piscivorous species nesting within these breeding colonies, apart from an occasional (< five nests in south San Francisco Bay per year) black skimmer (Rynchops niger) nest, we are confident that the majority of fish returned to the Forster's tern colony were brought back by Forster's terns. We collected all fish and other potential prey items found dropped in the colony during each weekly nest visit and stored samples in reclosable plastic bags. One period of samples from the A16 colony was excluded because it coincided with a major fish kill due to temporarily low dissolved oxygen concentrations in the pond, and because other piscivorous birds were known to be roosting on those islands at that time [24]. Forster's terns are primarily piscivorous [14,23,25], although invertebrates also have been documented to be consumed [23]. We recognize that our sampling was biased against invertebrates due to their smaller size and because invertebrates may decompose faster than fish; although we collected them when they were observed. However, invertebrates comprised < 0.1% of all collected samples and the remainder were fish; therefore, we removed invertebrates from all analyses. We stored samples frozen at -20˚C until they could be processed in the laboratory and dried.
In the laboratory, we identified prey samples to the lowest taxonomic group. Most fish were identified to species although some could only be identified to family. In particular, Mississippi silverside (Menidia audens) and topsmelt silverside (Atherinops affinis) from the Atherinopsidae family sometimes could not be separated because fish were typically desiccated, which prohibited the use of some identification marks. For example only 5.4% of all Atherinopsidae were identified to species in 2011, whereas > 99% of Atherinopsidae were identified to species in 2005 and 2007. As a result, we combined all Atherinopsidae into one group, silversides, for statistical analyses. We present scientific names of all identified species or the lowest identified taxonomic group in Table 1.
We gently cleaned prey samples using deionized water. We measured standard length of each fish to the nearest mm. Fish were placed in individual containers and dried at 50˚C for approximately 24-48 hr. After drying, we weighed each fish to obtain a dry mass and measured standard length again to obtain a dry standard length. When individual fish were missing large portions of their body (approximately ! 5%), we substituted a species-and year-specific mean dry mass for those individual fish. Similarly, if we could not obtain an accurate standard length of the dry fish, we substituted the species-and year-specific mean standard length of dried fish. Forster's terns have been observed to consume fish 10-100 mm in length, with the majority (81%) of freshly-caught fish 50-70 mm in size [25]. For comparison with previous studies on Forster's terns and other tern species, we assigned fish to one of five dry standard length categories based on Atwood and Kelly (1984): < 25 mm, 25-50 mm, 50-75 mm, 75-100 mm, and > 100 mm.

Statistical analyses
We quantified species composition in two ways, one based on species counts (hereafter relative species abundance) and one based on total dry mass of fish species (hereafter dry mass composition), at each tern colony for each year. Then we used a subset of colonies that we repeatedly         sampled over time to test whether the relative species abundance and dry mass composition of prey items differed among colonies. Furthermore, we used the same subset of repeatedly sampled colonies to test whether species' relative abundances changed over the duration of our 11-year study. For these tests, we only used those tern colonies that we sampled repeatedly over the course of the study (during 4-7 different years) and had ! 25 fish collected per year.
To compare relative species abundance and dry mass composition among colonies, we used a multi-response permutation procedure (mrpp; [26]) in the vegan package [27] with the statistical program R version 3.3.2 [28]. This statistical approach compares dissimilarities within and among groups to tell whether there are significant differences between groups. We used the Bray-Curtis distance measure to calculate the dissimilarity matrix used by mrpp, in order to determine and test within-group similarity and distinctness [29]. To compare species composition among colonies, we used relative species abundance to calculate the Bray-Curtis distance measure, which controlled for differences in the total number of fish collected at each tern colony per year. Similarly, for dry mass composition, we used the proportion of dry mass per species, which controlled for differences in the total fish mass collected at each tern colony per year. If relative species abundance or the dry mass composition differed among tern colonies, we conducted individual mrpp analyses between all pairs of colonies. Mrpp analysis calculates a δ value between and within groups, which is the weighted mean within-group distance [26]. Mrpp analysis also provides an A statistic, that describes the effect size of the grouping, and a p value, that quantifies the likelihood that the observed difference is due to chance [26,30]. The A-statistic can be interpreted similarly to the coefficient of determination in a linear model [27]. If all samples within groups are identical then A = 1. Conversely, A = 0 if within-group heterogeneity equals what is expected by chance [26].
To examine trends in the relative species abundance and dry mass composition of prey items over an 11-year sampling period during the Forster's tern breeding season, we started globally with a multivariate general linear model (MANOVA) for the proportions of the most abundant prey species, as a function of year, colony, and a year × colony interaction effect. We included the seven species that each comprised > 4% of all samples as dependent variables, with one value per species for every year and colony combination. The year × colony interaction effect provided the ability to test for overall differences in slope coefficients among colonies. After running the global MANOVA, we ran individual univariate linear models for each species with the same fixed effects to determine the source of the significance at the global level. If the interaction term for a specific species was significant (p < 0.05), we compared colonies using slope coefficients, standard errors, and 95% confidence intervals produced using the lsmeans package in R [31]. If the interaction term for a specific species was not significant (p > 0.05), we dropped the interaction term from the model. For models without a significant interaction effect, we examined pairwise differences in model-generated least squares mean relative abundances if there was a significant colony effect. We used a logit transformation on all proportional data prior to analysis, with the lowest non-zero value ( 0.005) as a substitute for all zeros in our dataset [32]. We present the differences in relative abundance among colonies from back-transformed least squares means for the average year.
We used linear mixed effects models to examine whether dry standard length or dry mass of the seven most commonly observed fish species changed over the course of our study at the same repeatedly sampled colonies that we used to examine relative abundance and dry mass composition over time. We excluded fish from these analyses that were missing standard length measurements or large portions of their body (! 5%). We examined each fish species separately for standard length and log-transformed mass. For each species, we first compared three models with year as a fixed effect and different random effects to determine whether our data were best explained by 1) a random intercept and slope model that allowed both the intercept and slope to vary by colony, 2) a random intercept model that allowed the intercept to vary by colony, or 3) a model with no random terms. We fit each model using restricted maximum likelihood (REML) and used the AIC values to select the best model. The best models to examine temporal trends of fish size for all seven species included a random intercept for colony. The only exception was for standard length of staghorn sculpin, where the best model did not include random effects. Therefore, we present only the results for the best models. We used the afex R package to determine significance with F tests, using the Kenward-Roger approximation for degrees of freedom [33].
Similar to relative species abundance, silversides comprised the largest proportion of dry fish mass (39.7%); however, the proportional contribution of the remaining groups differed between dry mass composition and their relative abundances (Figs 2 and 3). After the silversides, yellowfin goby comprised the highest proportion of dry fish mass (16.9%), followed by longjaw mudsucker (13.6%), staghorn sculpin (7.8%), three-spined stickleback (5.7%), northern anchovy (5.0%), Pacific herring (3.7%), and perches (2.3%). Rainwater killifish and other species of goby each comprised < 2% of dry fish mass. Other species or fish that were unable to be identified comprised 3.0% of dry fish mass.

Relative prey species abundance and dry mass composition among colonies
Overall, the relative abundances of prey fish species among seven repeatedly-sampled tern colonies (A1, A7, A8, A16, A2W, AB1, and AB2 colonies; Fig 1) were more different than would be expected by chance (A = 0.12, p = 0.001). Among colonies, the relative fish species abundance was the most consistent among years at A2W (δ = 0.32) and A1 (δ = 0.32), followed by AB2 (δ = 0.34). Notably, each of these three tern colonies are located adjacent to each other in the Moffett pond complex. The relative prey fish species abundance at A8, a colony located in Alviso, was the least consistent among years (δ = 0.45; Table 3). Comparing individual colonies, five of the colonies (A1, A2W, AB1, AB2, and A8) were as similar to each other as would be expected by chance, based on the variability observed within each colony (Table 3). In contrast, the two remaining colonies from the Alviso pond complex, A7 and A16, were both less similar to the A1, A2W, AB1, and AB2 colonies than would be expected by chance ( Table 3). The A8 colony was not distinguishable from the A7 or A16 colonies (Table 3), all of which were located in the Alviso pond complex.
Similar to relative fish abundance, the dry mass composition at repeatedly-sampled Forster's tern colonies was also more different than would be expected by chance (A = 0.10, p = 0.002). The dry mass composition was most consistent among years at the A2W (δ = 0.30), A1 (δ = 0.33), and A16 colonies (δ = 0.35; Table 3). Similar to the relative species abundances, five of the colonies (A1, A2W, AB1, AB2, and A8) had dry mass compositions as similar to each other as would be expected by chance (Table 3). The dry mass compositions at the A16 and A7 colonies were both less similar to the A1, A2W, and AB2 colonies than would be expected by chance (Table 3). A16 was also less similar to AB1 than would be expected by chance (A = 0.14, p = 0.014). However, unlike for relative species abundance, the dry mass composition at A7 was marginally indistinguishable from AB1 (A = 0.05, p = 0.08). In addition, the A8 colony was not different from the A7 or A16 colonies (Table 3).  Table 4 for statistical output from subsequent ANOVAs). For silversides, the effect of year differed among colonies (F 6,25 = 2.56, p = 0.05). The A16 colony was the only colony that had a marginally negative slope coefficient (-0.31 ± 0.17 standard error; t = -1.81, df = 25, p = 0.08). Given the slope coefficient and the substantial restoration actions that occurred at this colony at the start of our study that increased tidal exchange and decreased salinity [24], we removed A16 and reran the model. When we reran the model without A16, the year × colony interaction was no longer significant, and we observed an overall increase in the relative abundance of silversides returned to Forster's tern colonies from 2005 to 2015. The   relative abundances of silversides were similar among colonies, although one Moffett colony (A1) had ! 1.5 times the relative abundance of silversides than at the A7 and A8 Alviso colonies (t ! 2.10, df = 27, p 0.05). Run separately, there was not enough evidence to show a change in the relative abundance of silversides at A16 over time (F 1,3 = 2.58, p = 0.21). For stickleback, the significant interaction (F 6,25 = 3.04, p = 0.02) was also being driven by A16, as it had a positive slope coefficient that was significantly different than all of the other colonies (t ! 2.18, df = 25, p 0.04). Overall, the relative abundance of stickleback decreased when we excluded A16. In contrast, stickleback showed a marginally non-significant increase at A16 (F 1,3 = 7.72, p = 0.07). The relative abundance of longjaw mudsucker decreased at all colonies from 2005 to 2015 and varied among colonies. With the exception of A8, the mean relative abundance of longjaw mudsucker at the Alviso colonies was ! 4.3 times greater than at the Moffett colonies (t ! 2.91, df = 31, p 0.007). Additionally, the mean relative abundance of mudsucker at A8 was ! 3.1 times greater than at two of the Moffett colonies, AB2 and A1 (t ! 2.14, df = 31, p 0.04). We did not observe any overall increase or decrease in the relative abundance of Pacific herring or northern anchovy over time, although the relative abundance of northern anchovy was ! 3.6 times greater at two of the Moffett colonies (A1 and AB1) than the Alviso colonies (t ! 2.18, df = 31, p 0.04). The relative abundance of staghorn sculpin had less clear overall temporal trends, with an increase at one Alviso colony (A7; t = 2.38, df = 25, p = 0.03) and a decrease at one Moffett colony (AB1; t = -3.21, df = 25, p = 0.004). There were no clear temporal trends for staghorn sculpin at the other colonies. Similar to relative abundance, the relative dry mass of some prey species changed from 2005 to 2015 at Forster's tern colonies in San Francisco Bay (F 42,144 = 1.47, p = 0.05, Pillai's trace = 1.820; Table 4). Overall, the relative mass of silversides increased from 2005 to 2015, while the relative mass of longjaw mudsucker decreased. For stickleback, a significant year × colony interaction (F 6,25 = 2.50, p = 0.05) was being driven by A16, as it had a positive slope coefficient (t = 2.31, df = 25, p = 0.03). When A16 was removed, the relative mass of stickleback decreased from 2005 to 2015. Run separately, stickleback showed a marginally non-significant increase at A16 (F 1,3 = 7.45, p = 0.07). Whereas there were no overall differences in the relative mass of silversides and stickleback among colonies, the three Alviso colonies had ! 15 times the relative mass of longjaw mudsucker than at the A1 colony in Moffett (t ! 2.62, df = 31, p 0.01). For both yellowfin goby and staghorn sculpin, we observed an increase in the relative mass at A16 (t ! 2.13, df = 25, p 0.04). Additionally, yellowfin goby increased at AB2 (t = 2.25, df = 25, p = 0.02) and staghorn sculpin declined at AB1 (t = -2.99, Table 3  df = 25, p = 0.006). We did not observe any overall temporal trends in relative dry mass for northern anchovy or Pacific herring or differences among colonies. Several fish species returned to Forster's tern colonies changed in size from 2005 to 2015, with most changes suggestive of an overall decrease in fish body size. The average staghorn sculpin decreased annually by 1.8 mm in standard length (F 1,358 = 30.49, p < 0.001) and 6.6% in dry mass (F 1,364 = 20.75, p < 0.001). Similarly, northern anchovy decreased annually by 1.0

Discussion
Silversides (Mississippi silverside and topsmelt silverside; family Atherinopsidae) were the predominant prey species returned to Forster's tern colonies in San Francisco Bay over an 11-year period. Silversides comprised more than twice that of the next most common species, both by relative abundance of individuals (42%) and relative dry mass (40%). Three-spined stickleback were the second most common species returned to Forster's tern colonies (19% relative abundance) over the entire study. Previous fish sampling studies revealed that topsmelt silverside were present in all sampled ponds and sloughs in south San Francisco Bay, and comprised the majority of fish collected in gillnets from 2004-2006, while three-spined stickleback were the most abundant fish species sampled using minnow traps in ponds sampled in 2006 [20,21]. These observations suggest that silversides and three-spined stickleback were two of the most abundant Forster's tern prey items at the start of our study, and they were the two most abundant species groups returned to Forster's tern colonies. However, the relative contribution of three-spined stickleback was markedly lower when examined using the relative dry mass by species (6%) because of their small size. Instead, yellowfin goby (17% by mass and 10% by abundance) and longjaw mudsucker (14% by mass and 8% by abundance) contributed more by dry mass. The prey species returned to Forster's tern colonies had similarities with coastally-foraging California least terns (Sterna antillarum browni), where 70% of the diet was comprised of silversides and northern anchovy [12]. No other comparable studies exist for Forster's terns, as previous foraging-related research on Forster's terns was either conducted inland [23], on the Atlantic coast [25], or was primarily focused on the hunting behavior of Forster's terns and not on estimates of diet or prey selection [34]. However, Atlantic silverside (Menidia menidia) were suggested to be an important prey item of Forster's terns on the Atlantic coast, as they comprised 99.6% of samples collected in a seine adjacent to observations of foraging terns [25]. Inland, yellow perch (Perca flacescens) and shiner (Notropis spp.) were the most important species in courtship and chick feedings by Forster's terns, and these two species groups comprised 97% of species collected in seines; Atherinopsidae was not represented by any species in this study [23]. The species composition of observed chick feeding events and the prey species returned to and dropped on breeding colonies were relatively similar for several studies on other tern species, although the proportions of larger bodied prey species were slightly elevated in the sample of prey species dropped on the colony [11][12][13]. It is unknown how the fish species returned to Forster's tern colonies in the present study relates to the fish species consumed by Forster's tern adults or their chicks. However, based on previous studies, the major prey species determined using fish returned to and dropped on the colony were the same as those ingested by chicks or fed to mates [11][12][13]. All methods used to estimate avian diet have significant challenges and limitations, including the cost, feasibility of direct observation, and level of invasiveness. In spite of the limitations of our sampling method, collecting fish returned to and dropped on the colony provided an extensive amount of inexpensive and consistently collected data over an 11-year period.
Tern foraging behavior typically consists of plunge-diving from a stationary hovering position [34], allowing them to capture fish present either in shallow water or in the upper portion of the water column in deeper water. Consequently, silversides are vulnerable to predation by Forster's terns because of their surface-dwelling and schooling behaviors [35]. In contrast, yellowfin goby and longjaw mudsucker, which both comprised a lower proportion of fish returned to Forster's tern colonies, are more demersal species [35] and likely are only accessible to Forster's terns when they are in shallow water.
We observed temporal differences in the fish returned to Forster's tern colonies in San Francisco Bay over the 11-year sampling period, suggesting relatively localized changes in relative prey availability and the size of some prey species. Generally, the relative abundance and dry mass of silversides increased over time. Additionally, the average silverside decreased in size by 1.2 mm over our study but did not change in mass. Concurrently, we observed a decrease in the overall relative abundance and dry mass of three-spined stickleback and longjaw mudsucker returned to tern colonies. In contrast to silversides, the average three-spined stickleback decreased in dry mass by 5.3% annually (14 mg dry mass or 46 mg wet weight, based on a moisture content of 69.5% [7]) but did not decrease in length, suggesting that stickleback may have decreased in condition over time. Salt ponds and sloughs in south San Francisco Bay contain the species we collected at Forster's tern colonies [20,21]. However, salt ponds typically contained a smaller subset of the fish species observed in adjacent sloughs [21], and salinity was the most important environmental variable related to the spatial distribution of species [20]. Restoration of tidal exchange to three previously isolated salt ponds decreased pond salinity from levels intolerant to fish and allowed more salt-tolerant fish species (e.g., topsmelt silverside, northern anchovy, and longjaw mudsucker) to colonize, which resulted in a salt-related gradient of species observed from the upstream reaches of the sloughs down to the saltier ponds [20]. Furthermore, an unplanned breach of levees in North San Francisco Bay caused a marked decrease in longjaw mudsucker as the community composition in a formerly hypersaline pond shifted from mostly salt-tolerant species to a species assemblage that included some freshwater fish [20]. Consequently, the changes we observed in relative fish abundance returned to Forster's tern colonies over the course of our study could be a result of changes in prey selection or may be the result of changes in fish availability because of altered habitat from management associated with the South Bay Salt Pond Restoration Project (www. southbayrestoration.org).
Breeding Forster's terns are central place foragers that tend to feed within 6.2 km of their breeding colony [19]; therefore, the inter-colony differences we observed in relative species abundance and dry mass composition suggested heterogeneity in the available prey base among colonies. Geographically, the colonies with the most consistent relative species abundance over time were located adjacent to each other in the Moffett pond complex (A1, A2W, AB1, and AB2) and were separated from the three additional repeatedly-sampled colonies in the Alviso pond complex (A7, A8, and A16; Fig 1). The A7 and A16 colonies were distinguishable from the other colonies, both in relative species abundance as well as dry mass composition, which appeared to be generally driven by a greater proportion of longjaw mudsucker returned to the Alviso colonies and a greater proportion of northern anchovy returned to the Moffett colonies. The high variability of relative species abundances among years at the A8 colony made it indistinguishable overall from any other colony.
Two of the repeatedly sampled colonies stood apart from the others in terms of the intracolony species heterogeneity (A8) and temporal trends (A16), which may have been a direct result of altered management practices at those sites or differences in available foraging habitat. Previously, breaching of levees in the south and north regions of San Francisco Bay caused changes in species assemblages as connectivity between habitats was increased and salinity levels changed [20,21]. In our study, the A8 colony, sampled in 2006, 2010, 2012, and 2013, had the highest intra-colony variability in relative species abundance. Furthermore, A8 had relatively low sample sizes of fish returned to the colony, which may have contributed to the observed variability. However, as part of the South Bay Salt Pond Restoration Project, the habitat in A8 changed markedly after the 2010 Forster's tern breeding season, at which point in time managers physically interconnected the pond to two other managed ponds, and the water depth of the entire complex was increased. Furthermore, the levees for an adjacent and previously dry pond bed (A6) were breached in the fall of 2010 to allow for the development of tidal marsh habitat, providing new habitat for fish within the foraging range of the A8 colony. Thus, two of the sampling years for the A8 colony occurred before these management actions and two sampling years occurred after, which may have contributed to the substantial heterogeneity observed in the relative species abundance over time. Management actions for A16 as part of the South Bay Salt Pond Restoration Project increased tidal exchange and decreased salinity [24], which may have contributed to the temporal trends that we observed in the fish returned to the A16 colony. Specifically, all silversides collected from A16 in 2005 were identified as topsmelt silverside (49% of fish, n = 69), suggesting that Mississippi silverside had not yet colonized this pond. As the salinity decreased in pond A16, this may have allowed other species, such as three-spined stickleback, to colonize and increase in relative abundance over time.
Changes in prey availability and diet could have important implications for Forster's terns in the San Francisco Bay Estuary, such as reproductive success and contaminant exposure. In particular, mercury contamination is known to be an important issue for waterbirds breeding within the estuary [36]. The highest whole-body mercury concentrations in fish collected across 27 managed wetlands, 3 tidal marshes/sloughs, and 2 open water sites in the San Francisco Estuary were in Mississippi silverside (0.83 ± 0.02 μg/g dw), followed by topsmelt silverside (0.55 ± 0.02 μg/g dw; [7]). Whereas we were unable to separate these two species in our analysis, their combined relative abundance and relative dry mass increased in the diet of Forster's terns over the past 11 years. Consequently, a temporal shift in diet could cause an increase in mercury exposure and toxicological risk for Forster's terns. In contrast, species that had significantly lower mercury concentrations than silversides, such as three-spined stickleback (0.45 ± 0.01 μg/g dw) and longjaw mudsucker (0.36 ± 0.01 μg/g dw), were the prey species that declined in relative abundance over time [7].
Forster's terns breeding in the San Francisco Bay area relied heavily on surface-dwelling silversides and our results showed that their dependence on these fishes has increased. As central place foragers, the fish returned to Forster's tern colonies suggest that there were differences in relative prey availability among colonies and over time. The abundance of three-spined stickleback and longjaw mudsucker returned to Forster's tern colonies decreased at multiple colonies, suggesting that their relative availability in the environment also has declined, possibly in response to habitat alteration. Future studies could evaluate if changes in diet among colonies and over time has resulted in differential reproductive success and contaminant exposure.