Hypoxia’s impact on pelagic fish populations in Lake Erie: a tale of two planktivores

Whether bottom hypoxia has long-lasting consequences for pelagic fish populations remains speculative for most ecosystems. We explored hypoxia’s influence on two pelagic zooplanktivores in Lake Erie that have different thermal preferences: cold-water rainbow smelt (Osmerus mordax) and warm-water emerald shiners (Notropis atherinoides). To assess acute effects, we combined predictive bioenergetics-based modeling with field collections made across the hypoxic season in central Lake Erie during 2005 and 2007. To assess chronic effects, we related fishery-independent and fishery-dependent catches with hypoxia severity and top predator (walleye, Sander vitreus) abundance during 1986–2014. As our modeling predicted, hypoxia altered rainbow smelt movement and distributions, leading to avoidance of cold, hypoxic bottom waters. In response, diets shifted from benthic to pelagic organisms, and consumption and energetic condition declined. These changes were lacking in emerald shiners. Our long-term analyses showed rainbow smelt abundance and hypoxia to be negatively related and suggested that hypoxia avoidance increases susceptibility to commercial fishing and walleye predation. Collectively, our findings indicate that hypoxia can negatively affect pelagic fish populations over the long term, especially those requiring cold water.

lead to reduced somatic growth and energetic condition. For emerald shiners, we also predicted 137 hypoxia would negatively affect foraging, growth, and energetic condition, but through a 138 different mechanism. Analogous to what Ludsin et al. (2009) found for Chesapeake Bay anchovy 139 (Anchoa mitchilli; Engraulidae), which also is a zooplanktivorous, epi-pelagic, warm-water 140 species, we expected that hypoxia would limit emerald shiner performance by reducing access to 141 zooplankton (i.e., zooplankton would use the bottom hypoxic layer as a refuge from predation). 142 Second, do long-term, population-level trends in rainbow smelt and emerald shiner track the 143 severity of hypoxia? While we did not expect to find meaningful relationships between hypoxia 144 and the long-term demographics of these species, owing to the many other factors that can affect 145 prey-fish demographics (sensu Breitburg et al. 2009a(sensu Breitburg et al. , 2009bRose et al. 2009), the recent 146 availability of long-term dissolved oxygen data from the U.S. Environmental Protection Agency 147 (USEPA) and the literature (Zhou et al. 2015) allowed us to ask this question. 148 To answer these questions, we combined habitat quality (growth rate potential, GRP,149 which is akin to the "fundamental niche"; Brandt et al. 1992) modeling with site-specific field 150 collections and analyses of long-term datasets. First, we present results from a spatially-explicit 151 bioenergetics-based GRP model, which quantified habitat quality for rainbow smelt and emerald 152 shiners throughout the water column over large spatial scales before, during, and/or after the cladoceran zooplankton, and thus, is considered a planktivore (Ewers 1933;Muth and Busch 206 1989;Hartman et al. 1992). Emerald shiners reside primarily in the epilimnion, schooling in it 207 (especially at the surface) by day and dispersing throughout it at night (Trautman 1981;208 Schaeffer et al. 2008). The emerald shiner has an optimum temperature for growth near 25C 209 (McCormick and Kleiner 2002) and is considered a warm-water species. While the emerald 210 shiner is not recreationally fished, it does support an important baitfish industry in the Lake Erie 211 (Nielsen 1982;Knight and Vondracek 1993;Meronek et al. 1997). 212 For both study species, the physiological limit of hypoxia tolerance has not been 213 experimentally tested. However, both species have been shown to avoid hypoxic waters in Lake during the peak hypoxia period (4-11 September) to both estimate the size of the hypoxic zone 253 and to explore the use of the hypoxic edges by fish. This sampling was designed to test two 254 hypotheses. First, we wanted to learn whether hypoxia caused fish to aggregate at the edges of 255 the hypoxic zone, which had been reported for benthic species in the Northern Gulf of Mexico 256 (Craig and Crowder 2005). Second, we wanted to explore if hypoxia could cause a thermal-  Along the long and short (diel) transects sampled during 2005, we quantified 266 physicochemical attributes and the lower food web using a PSS, the details of which can be 267 found in Vanderploeg et al. (2009aVanderploeg et al. ( , 2009b. In brief, the PSS consisted of a CTD, a mini optical D r a f t Running Title: Hypoxia's impact on pelagic fishes in Lake Erie 13 274 sizes, counts, and biomass in Great Lakes ecosystems with low suspended particulate matter 275 such as central Lake Erie during the summer hypoxic period (Liebig et al. 2006;Vanderploeg et 276 al. 2009a). If any bias occurred, it would be an overestimate of zooplankton biomass, owing to 277 suspended particulates being misidentified as zooplankton (Liebig et al. 2006

288
Growth rate potential has been successfully used to describe habitat quality at fine spatial 289 scales for both freshwater and marine fishes (e.g., Brandt et al. 1992, Kraus et al. 2015b including those found in Lake Erie (Arend et al. 2011;Brandt et al. 2011). Our calculations were 291 made using a spatially-explicit bioenergetics model along with actual observations of 292 temperature and DO collected by the PSS during daylight transects (see previous section), as 293 both rainbow smelt and emerald shiners primarily rely on vision to find prey. We used models 294 that were specifically developed to evaluate the effects of hypoxia on rainbow smelt and emerald 295 shiners in Lake Erie (Arend et al. 2011; see Appendix A for details on model structure and 296 parameterization). We also conducted sensitivity analyses wherein we removed the effect of DO D r a f t analyzed for this study. In the laboratory, all individuals were thawed, measured (nearest 1 mm 335 total length, TL), and weighed (nearest 0.01 g wet weight). Stomach contents were then 336 removed, and the entire fish (minus stomach contents) was dried at 70 ℃ to a constant mass 337 (nearest 0.0001 g). We assessed fish energetic condition by dividing each individual's dry weight 338 by its corresponding wet weight, as our own data (SAP, unpublished data) and previous research 339 has shown this dry:wet weight ratio to be strongly, positively correlated with energy density 340 (Pothoven et al. 2006;Morley et al. 2012). In this way, individuals with a higher dry:wet mass D r a f t 341 ratio (i.e., higher energy density) can be considered in better energetic condition than those with 342 a lower ratio. 343 We analyzed the stomach contents of a subsample of adult fish collected during each diel  for August, September, and October respectively. Our ability to discern adult fish was facilitated 352 by age-0 individuals also being collected. For both species, after we processed diets, we dried the 353 stomach contents at 70℃ to a constant mass (nearest 0.0001 g), which allowed us to estimate of 354 individual mass-specific consumption. 355 We identified, measured, and counted both zooplankton and benthic macroinvertebrates    To determine the biomass of prey consumed, prey lengths of up to 20 intact individuals 367 of each prey group (except nauplii) were measured using ImagePro image analysis software 368 (Media Cybernetics, Silver Spring, MD). Prey length was converted to dry mass using mass-369 length regressions (Culver et al. 1985;Makarewicz and Jones 1990;SAP and T. Nalepa, 370 University of Michigan, Lansing, Michigan, unpublished data). The average dry mass of an 371 individual of each prey type was determined for each size-class of fish for each site and month 372 and multiplied by the number of each prey type in a stomach to determine dry-biomass 373 contribution of each prey type in an individual stomach. 374 We used one-way analysis of variance (ANOVA) and post-hoc Tukey's honestly 375 significant difference tests to quantify differences in the catch, consumption, body mass, and 376 energetic condition (energy density) of both rainbow smelt and emerald shiners before (August), 377 during (September), and immediately after (October) peak hypoxia. All data met assumptions of 378 normality and homogeneous variances. An alpha-value was set to 0.05 for all analysis.    The water column was thermally stratified along this transect, with the thermocline being located 431~9 m off the lake bottom ( shiner GRP was positive throughout the water column during August, it was highest near the 435 thermocline and in the warm epilimnion (Fig. 3D). By contrast rainbow smelt GRP during 436 August was positive only in the cold hypolimnion (Fig. 3E). Day-time acoustic survey data 437 revealed that fish biomass was widely distributed along transect B during August with levels 438 being relative higher above the thermocline than below it, although some fish aggregations were 439 detected in the hypolimnion (Fig. 3F). between the two species. No stratification was present as sampling occurred ~7 d after the lake 458 re-mixed, with the entire water column being ~19 °C and having ~8 mg O 2 L -1 (Fig. 3M, 3N).

459
Zooplankton biomass was highest in offshore waters (Fig. 3O). Emerald shiner GRP was 460 uniformly positive across the transect (Fig. 3P) and rainbow smelt GRP was uniformly negative 461 throughout the entire water column (Fig. 3Q), as temperatures exceeded this species' suggested 462 lethal limit of 20 °C (Lantry and Stewart 1993). Relative fish biomass was lower compared to 463 August or September, with fish mostly concentrated in offshore bottom waters (Fig. 3R).

464
September 2007 (hypoxia). We also calculated GRP for transect D during September  Table 2). During pre-hypoxic conditions in August, fish were abundant in the 484 hypolimnion, based on the proportional distribution of acoustic backscatter (sv) across the 5-km smelt, mass-specific gut content was significantly higher during August than during September 504 or October (Fig. 6B), and the benthic fraction of the diet was significantly lower during the 505 hypoxic period (2%) than before or immediately after it (37% and 66%, respectively; Fig. 6C).

506
By contrast, the mass-specific gut content of emerald shiners increased from August into 507 September and October (Fig. 6B), and the benthic fraction of the diet was 0% during 508 stratification (August and September) and significantly higher (66%) after stratification ended September) and remained unchanged from September into October, which encompassed the 516 hypoxic period and only ~7 d afterwards (Fig. 6D). Rainbow smelt energy density, as proxied by 517 the dry weight to wet weight ratio, significantly decreased with each successive month (Fig. 6E).

518
Opposite patterns were found for emerald shiners with both TL (Fig. 6D) and energy density 519 (Fig. 6E), both of which significantly increased in each successive month during the summer 520 growing season, despite the presence of hypolimnetic hypoxia. The plot of hypolimnetic fish abundance and DO (Fig. 7) shows some aggregation of fish D r a f t Running Title: Hypoxia's impact on pelagic fishes in Lake Erie 24 525 surveyed (Fig. 7). To further understand how hypoxia might be driving fish aggregations, we 526 explored how hypolimnetic fish biomass (using the acoustic backscatter coefficient, ABC, as a   (Table 3). Overall, our most parsimonious model (see Table   545 S1), which included both variables, explained 49% of the variation in rainbow smelt abundance, 546 with the variation being equally split between the two variables (Table 3). Central basin adult D r a f t Running Title: Hypoxia's impact on pelagic fishes in Lake Erie 25 547 emerald shiner abundance was also negatively correlated with bottom hypoxic area (Fig. 9C), but 548 was unrelated to walleye population size (Fig. 9D). 549 We also investigated the potential influence of bottom hypoxia on rainbow smelt catches 550 by commercial fishers in Ontario waters of central Lake Erie during 1998-2014. Our analyses 551 showed that rainbow smelt (fisheries-dependent) catches during August were significantly, 552 positively correlated with the areal extent (severity) of bottom hypoxia (Fig. 10). This basin of Lake Erie during the summer growing season. However, the reduction was greater for 576 rainbow smelt than for emerald shiners, which has been shown previously at a less-refined 577 spatial scale (Arend et al. 2011). In the case of rainbow smelt, little high-quality habitat existed 578 along our sampling transect, owing to the development of hypoxia in the hypolimnion that 579 eliminated access to suitably cold water temperatures (i.e., 10-12°C; Lantry and Stewart 1993).  period when the hypolimnion had lower habitat quality than the epilimnion. However, once 640 stratification ended, benthic items became a significant portion of emerald shiner diets, although 641 we suspect that this diet shift was facilitated by a chironomid midge hatch that occurred during 642 our sampling (SAP and SAL, personal observation).

643
Reductions in mass-specific total diet biomass also reflect the exclusion of rainbow smelt 644 from the hypolimnion during hypoxia. When bottom hypoxia is present, rainbow smelt likely 645 compete more directly with emerald shiners for prey than before or afterwards (Pothoven et al. during the hypoxic period than prior to it, which was not the case for emerald shiners (see Fig.   648 6). In addition to competing for zooplankton with emerald shiners, rainbow smelt may also have 649 experienced reduced consumption during the hypoxic period because of physiological stress 650 caused by warmer than optimal water temperature (Lantry and Stewart 1993).  (Rand et al. 1994). 664 We strongly believe that this reduction in rainbow smelt somatic growth and energetic 665 condition resulted from physiological stress caused by a combination of high temperature 666 (Lantry and Stewart 1993), low DO (Hanks and Secor 2011), and reduced feeding during 667 hypoxia. The near-lethal temperatures that occurred throughout the entire water column after fall 668 mixing would be expected to have strong negative effects on the growth and energetic condition 669 for adult, as well as yearling (see Table S2), rainbow smelt that are unable to escape horizontally.

670
Given that both water temperature and bottom hypoxia are expected to increase in Lake Erie individually. This latter finding suggests that the effects of top-down control of the planktivorous 696 fishes in Lake Erie (Knight et al. 1984;Vondracek et al. 1993) can be strengthened by hypoxia.

697
Given the lack of study on piscivore-planktivore interactions in Lake Erie during the hypoxia 698 season, we recommend future studies into how hypoxia affects foraging of predators.

699
Interestingly, suitable high-quality habitat for rainbow smelt exists in the bottom waters 700 of Lake Erie's eastern basin during summer. However, it is unknown to what extent rainbow 701 smelt can migrate into the cooler, well-oxygenated eastern basin to escape hypoxia. While 702 rainbow smelt may move into the east basin in response to hypoxia, their movements may be 703 blocked by the shallow Pennsylvania and Clear Creek ridges, which appear to be dominated by 704 poor quality habitat (i.e. GRP < 0 gg -1 d -1 ). In support of this notion, we found that, along the  Similarly, aggregations at the edge of the hypoxic zone (aka "dead zone" may also 716 increase susceptibility of rainbow smelt to commercial fishing. In support for the notion that altered DVM behavior, and its subsequent effects on performance, was driven by reductions in 754 habitat quality (as measured by GRP), which was severely restricted in the central basin of Lake 755 Erie during peak hypoxia, whereas large areas of positive emerald shiner GRP remain during this 756 time. These differences likely are due to different thermal preferences, with rainbow smelt being 757 forced out of their preferred cold-water bottom habitat while emerald shiners can remain in 758 oxygenated surface waters within their preferred thermal range. Hypoxia not only reduces the 759 overall habitat suitability for rainbow smelt in central Lake Erie, but also forces fish to move 760 horizontally (to hypoxia edges) or vertically (above the hypolimnion) to avoid areas of the lowest 761 DO.

762
These shifts in habitat and short-term performance may come with a suite of long-term 763 consequences, including increased susceptibility to predation and commercial fishing. While 764 both emerald shiners and rainbow smelt catches were negatively related to hypoxia extent, we 765 expect the negative impact of hypoxia to be worse for rainbow smelt, owing to its need for cold-766 water habitat that is only found in the hypolimnion during summer in central Lake Erie. With in future years, especially if access to the eastern basin is cut off due to hypoxia (as our data 772 suggest is the case). Given the importance of rainbow smelt as a forage species in central and 773 eastern Lake Erie, and its ability to support commercial fisheries, we encourage continued 774 research into whether hypoxia indeed leads to increased walleye predation, commercial

D r a f t
Running Title: Hypoxia's impact on pelagic fishes in Lake Erie 52 Table 1. Description of the data used in this study, which was collected during August, September, andOctober 2005 andSeptember 1145 2007 in central Lake Erie. We made collections at specific stations (CTD Sites), along "short" (5 km) transects every 4 hours over 24-1146 hours (Diel Transects), and across "long" (15 -60 km) transects sampled once each during the day and night (Long Transects). A "+" 1147 indicates that sampling of that type occurred during that time period, whereas a "-" indicates that sampling did not occur.    as no thermocline was present (see Table 2). Error bars represent one standard deviation. D r a f t Time-series of dissolved oxygen (DO) and fish abundance in central Lake Erie, 1987-2014. Data include: A) the percent of interpolated coverage area during August with bottom DO < 4 mg•L -1 (interpolated area spans the monitoring area in the U.S. EPA's Great Lakes Environmental database, http://www.exchangenetwork.net/data-exchange/glenda/) and the estimated population size of age 2+ walleye in Lake Erie (data from Lake Erie Walleye Task  D r a f t Daytime survey data collected in central Lake Erie along "long" transect B during August, September, and October 2005. Observed temperature (panels A, G, and M), dissolved oxygen (DO; panels B, H, and N), and zooplankton biomass (panels C, I, and O) collected by a Plankton Sampling System (PSS) were used to estimate the growth rate potential (GRP, g·g -1 ·d -1 ) of emerald shiners (panels D, J, and P) and rainbow smelt (panels E, K, and Q). Observed acoustic backscatter data (sv, log scale) measured throughout the water column simultaneous with the PSS data are presented as relative fish biomass (panels F, L, and R). Transects were surveyed from the south (left) to north (right). The lake bottom is displayed in grey.
D r a f t Daytime survey data collected in the eastern part of central Lake Erie along "long" transect D during September 2007. Observed temperature (A) and dissolved oxygen (DO; B) collected by a Plankton Sampling System (PSS) were used to estimate the growth rate potential (GRP, g·g -1 ·d -1 ) of emerald shiners (C) and rainbow smelt (D). Observed acoustic backscatter data (sv, log scale) measured throughout the water column simultaneous with the PSS data are presented as relative fish biomass (E). Owing to our OPC on the PSS failing, a constant, estimated zooplankton biomass was used in each cell when calculating GRP (see Appendix A). The transect was surveyed from the west (left) to east (right). The lake bottom is displayed in grey.
D r a f t Daytime vertical distribution of fish in central Lake Erie at diel site B, located near the midpoint of the longer transect B during August (A, pre-hypoxia), September (B, peak hypoxia), and October (C, post-hypoxia) 2005. Average proportion of total abundance (number of individuals · ha) in each water-column stratum was calculated from fish hydroacoustics (proportional Sv). Water-column strata were determined based on the thermocline position during August and September (upper = epilimnion; middle = metalimnion; lower = hypolimnion) and at the midpoint of the water column in October, as no thermocline was present (see Table  2). Error bars represent one standard deviation.
80x163mm (300 x 300 DPI) D r a f t Comparison of rainbow smelt (left bars) and emerald shiner (right bars) attributes in central Lake Erie (at diel station B) during August (pre-hypoxia), September (peak hypoxia), and October (post-hypoxia) 2005. Fractional catches (based on abundance) in bottom trawls (i.e., vertical distribution; A), mean mass-specific total diet biomass (g diet mass ·g fish mass -1 ; B), proportional biomass of benthic prey items in the diet (C), mean total length (mm; D), and mean energetic condition (i.e., estimated energy density represented by fish dry weight (g) to wet weight (g) ratio; E). Fish catches, diet content, and body size and condition were compared among months using one-way ANOVAs. Months with no letters in common (for a species within a panel) are significantly different based on post-hoc Tukey's honestly significant difference tests. Error bars represent one standard error.
182x236mm (300 x 300 DPI) D r a f t Horizontal distribution of fish in central Lake Erie during September (peak hypoxia) 2005 and 2007 in relation to dissolved oxygen (DO) availability in the hypolimnion. The observed average acoustic backscatter coefficient (ABC; proxy for fish biomass) in the hypolimnion is portrayed by open circles (each circle represents ~ 1 km) with circle sizes being proportional to average ABCs. The area surveyed with fish acoustics gear is outlined in blue; areas surveyed with no ABC data indicate a well-mixed (non-stratified) water column with no hypolimnion. Hypolimnetic DO concentration was interpolated from plankton survey system and CTD data collected simultaneously throughout the region. where T is water temperature (°C) and DO is dissolved oxygen concentration (mgL -1 ). This function accounts for the negative effect of DO on consumption (Chabot & Dutil 1999;Stierhoff et al. 2006;Brandt et al. 2009;. We assumed that C increases linearly with DO concentration from zero to one up to a threshold DO concentration (DO crit ), above which f DO equals one (Bartell 2003). Because increases in water temperature can magnify the effect of hypoxia on C (Schurmann & Steffensen 1997;Valverde et al. 2006;Marcek et al. 2020), we further assumed a negative relationship between DO crit and water temperature.
The structure and parameterization of the emerald shiner growth model was identical to Arend et al. (2011), which, like the rainbow smelt model, was developed specifically for Great Lakes applications and includes the same f DO function.
Inputs into the GRP included water temperature and DO observed by the plankton survey system during daylight transects before (August), during (September), and after (October) hypoxia along transect B in 2005 and transect D in 2007. Other inputs included observed wet weight (g) and energy density (Jg -1 ) of rainbow smelt and emerald shiners. We used the average values of these variables from the fish collected at the diel station located on transect B in 2005 for each month for which the model was applied (Table A1). Instead of incorporating a functional relationship in which observed zooplankton densities could serve as an input to consumption rates, we assumed that fish were always consuming at their maximum rates. This assumption likely leads to an overestimate of growth, particularly for rainbow smelt, for which we observed reductions in prey consumption during hypoxia. Subsequently, our predicted reductions in habitat quality experienced by smelt during hypoxia are likely conservative.
D r a f t Table A1. Monthly wet weight (g) and energy density (Jg -1 ) of rainbow smelt and emerald shiners used in growth rate potential models. D r a f t