Development of Striped Bass Relative Condition Models with Bioelectrical Impedance Analysis and Associated Temperature Corrections

Ecosystem-based fisheries management (EBFM) is gaining momentum as a means of holistically managing ecosystems instead of single-species stocks (Pikitch et al. 2004; Link 2005). Although the theory behind EBFM is well developed, implementation has lagged behind due in part to the need for new indicators beyond species-specific production and removal. In the Chesapeake Bay region, development of an ecosystem-based approach has included recommendations to make greater use of nutritional condition reference points, especially with Striped Bass Morone saxatilis (Grant 2009).

An early call for managing nutritional condition came when Uphoff (2003) estimated a major reduction in predator–prey ratios between Striped Bass and Atlantic Menhaden Brevoortia tyrannus that coincided with the decline in condition of Striped Bass in Chesapeake Bay during the late 1990s. Because Atlantic Menhaden are the dominant prey species for Striped Bass in the mid-Atlantic region, there was concern that an outbreak of diseased Striped Bass may have resulted from prey limitation and could impact Atlantic coastal Striped Bass as a whole (Hartman and Brandt 1995; Griffin and Margraf 2003; Hartman and Margraf 2003; Uphoff 2003; Overton et al. 2008, 2009). Ongoing and subsequent research found that Striped Bass were experiencing an epizootic of mycobacteriosis (Heckert et al. 2001; Rhodes et al. 2001, 2004; Overton et al. 2003; Kaattari et al. 2005). Jacobs et al. (2009) were able to experimentally link the progression of mycobacterial disease in Striped Bass to their diet: inadequate diet led to more severe disease progression compared with a higher ration.

Subsequently, Jacobs et al. (2013) tested several metrics for their suitability as biological reference points relative to general lipid depletion in Striped Bass, including traditional, nonlethal methods (standard weight and Fulton's condition factor) and others based on direct observation or measurement of key components (body fat index [BFI] and percent moisture content). Traditional methods were poorly correlated with lipid depletion; however, BFI and percent moisture were both shown to be reliable methods for estimating lipid depletion. Jacobs et al. (2013) suggested that this was likely due to fish taking in moisture over the short term as lipid reserves were depleted, thereby conserving weight and limiting the ability of traditional indices to track the change in lipids (Love 1970). Jacobs et al. (2013) recommended the use of percent moisture as the condition assessment tool of choice, and they proposed a biological reference point management target of 75% of the Striped Bass population having less than 80% moisture—the threshold that was found to coincide with lipid depletion.

Traditionally, measuring the moisture content of fish was lethal. However, bioelectrical impedance analysis (BIA) has emerged as a nonlethal means to estimate fish body composition (Cox and Hartman 2005). The BIA technology capitalizes on principles of physics relating resistance (the reduction of electrical current when passed through a given material) and reactance (opposition of current flow) to chemical properties of tissue. Strong predictive BIA models (R² > 0.75) of percent dry weight (the inverse of percent moisture) have been developed for a number of fish species (e.g., Cobia Rachycentron canadum: Duncan et al. 2007; Brook Trout Salvelinus fontinalis: Hafs and Hartman 2011; Bluefish Pomatomus saltatrix: Hartman et al. 2011) along with guidelines for developing successful models (reviewed by Hartman et al. 2015). Resistance and reactance measures are sensitive to temperature, necessitating temperature corrections to standardize BIA measures to the temperatures that the models were developed for in applications with fish (Hartman et al. 2015).

Assessing fish condition with BIA is a relatively simple procedure once models are developed. Resistance and reactance values and the electrical parameters that can be derived from them are used to predict the percent dry weight of the test subject. A body composition analyzer is a portable instrument that passes a weak electrical current (e.g., 425 µA, 50 kHz) through tissue. The instrument measures resistance and reactance, and users also record some metric associated with length of the circuit: height in human subjects or distance between the signal and detection electrodes in fish. From these three measurements, a suite of electrical properties is calculated as candidate variables to predict the percent dry weight of fish (see Hartman et al. 2015 for a thorough review). Measurement time for BIA is comparable to length–weight indices and is considerably cheaper and faster (US$3,000 for an RJL Systems Quantum II BIA instrument) than conducting proximate analysis in the laboratory.

We hypothesized that BIA is a reliable way of nonlethally estimating percent dry weight, the inverse of percent moisture, in Striped Bass. Objectives were to (1) develop BIA models to predict percent dry weight across a range of fish sizes and (2) develop BIA temperature correction models for a wide range of field conditions.

METHODS

This study used a hierarchical sampling structure to sample fish across size ranges, condition ranges, and temperatures. Five size-classes were used: age 0 (50–150 mm TL), small juveniles (150–250 mm), large juveniles (250–350 mm), recruits (350–450 mm), and adults (≥550 mm). These five size-classes were sampled at three nutritional condition ranges based on the observational BFI that was used by Jacobs et al. (2013). To construct temperature correction models, we sampled each condition class within each size-class at three temperatures: 8, 18, and 26°C. These temperatures were chosen as an approximation of the typical seasonal temperature range seen in Chesapeake Bay.

Jacobs et al. (2013) modified the BFI proposed by Goede and Barton (1990) to a scale of 0–3 based on a visual assessment of the percent coverage of the viscera by fat. For instance, if the sampler determined that 75% or more of the viscera was visually covered by fat, then that fish scored a 3 on the index. If the sampler determined that fat visually covered between 25% and 75% of the viscera, then the fish scored a 2. Fish with less than 25% coverage received a score of 1, and fish with no visible fat deposits on the viscera received a score of zero. The first condition class in our study was composed mainly of fish with a BFI of 3, the second class comprised fish with a BFI of 2, and the third and final class consisted of fish with a BFI of 1 or 0.

Sampling of age-0 and juvenile fish

The three smallest size-classes of Striped Bass (age 0, small juveniles, and large juveniles) were sampled at the National Oceanic and Atmospheric Administration (NOAA) Cooperative Oxford Laboratory (COL), Oxford, Maryland. Seven-hundred fish were obtained from the Maryland Department of Natural Resources’ (MDDNR) Manning Hatchery in 2011. An additional 500 fish in 2011 and 100 fish in 2012 were received from the University of Maryland's Horn Point Laboratory, Cambridge. The experiment began with extra fish to ensure adequate sample sizes after possible losses due to mortality. The fish were transferred to the COL using standard protocols (Weirich 1997) to minimize stress; they were held and acclimated in two 3,785-L outdoor, flow-through tanks and were fed size-appropriate commercial feed consisting of 2.5-mm, slow-sinking pellets that contained 42% protein and 16% fat (Zeigler Brothers, Inc., Gardners, Pennsylvania; Nicholson et al. 1990; Gatlin 1997). Fish were then moved into the COL's eight indoor recirculating systems for the experimental studies. Each 1,135-L recirculating system was made up of two 568-L Polytanks that were linked to the same pump and filter system. Experimental conditions included a 12-h light : 12-h dark photoperiod (fluorescent light), a pH of 8.2, a salinity of 10‰, and a temperature of 21°C. Once transferred to the recirculating system, the experimental fish were fed Striped Bass grower diet (4-mm sinking pellet; 42% protein, 16% fat) to satiation (Melick Aquafeed, Catawissa, Pennsylvania).

The BIA sampling procedure began when the fish reached an average TL of 100 mm, the average for the age-0 size-class. At that time, 75 fish were separated into another 1,135-L system, and feed was withheld for the first series of experimental efforts. The remaining fish were maintained on feed until they grew to the average TL of the next experimental size-classes (200 and 300 mm), at which point 75 fish from each test group were again separated and removed from feed.

Of the initial 75 fish, 25 were placed in a separate 1,135-L system for the first BIA measurements. Because these fish had been fed to satiation throughout acclimation, they were assumed to be in the best possible condition (a single representative fish was sampled and was confirmed to have a BFI of 3, which is the highest condition class). The remaining 50 fish removed from feed were sampled (25 fish per period) at roughly 6- and 12-week intervals. Intervals varied depending on fish size and were determined based on the visually estimated condition of the fish. Single fish were occasionally sacrificed to sample for BFI and to estimate the condition of the sample group. For the second sampling period (6 weeks), the fish were sampled when they appeared to score approximately a 2 on the BFI. For the last sampling period, fish were sampled when roughly half of them appeared to have a BFI score of 1 and the other half appeared to have a score of zero. During each of these sampling periods, the subset of 25 fish was sampled with the BIA protocol described below.

Due to the blocked design in this study, occasionally a subset of fish within a particular size-class was either just smaller than the target or grew out of the target range during the varying treatments. To maintain within-treatment sample sizes, these individuals were kept in their original classification.

Sampling recruits and adult fish

To conduct BIA on the adult and recruit size-classes, fish were collected off Sandy Hook, New Jersey, by the NOAA and were maintained at James J. Howard Marine Science Laboratory in Sandy Hook. These fish were collected by hook and line and were acclimated to indoor flow-through tanks. Recruit-sized fish were fed various species of cut fish daily to satiation for approximately 3 months to establish feeding and build lipid reserves. To obtain the three previously mentioned condition classes, 62 fish were placed in one of three treatments (~20 per treatment) for 2 months. The fish for the lowest condition class (BFI = 0–1) were fed once weekly; the medium-condition fish (BFI = 2) were fed twice weekly to satiation; and the third group representing the highest condition class (BFI = 3) was fed daily to satiation. The recruit-sized fish were then sampled for BIA values using the protocol described below.

Adult fish were sampled in two groups due to space limitations. The first group of 23 fish was sampled with the protocol below shortly after establishment of feeding with cut baitfish. This group was assumed to be the medium-condition group (BFI = 2) due to the moderate diet they were fed and due to their recent feeding history as wild fish. The second group of 49 fish was split into two treatments to save time. One treatment was fed to satiation daily for approximately 3 months to build lipid reserves. The second treatment group was fasted for 3 months. To conserve sample size, these treatments were assumed to be our highest condition (BFI = 3) and lowest condition (BFI = 0–1) classes, respectively, based on their prescribed feeding regimens. The second group, when at the appropriate condition level, was then sampled with the protocol below.

Bioelectrical impedance analysis data collection

The sample fish were acclimated to the warmest test temperature (26°C) overnight. They were then anesthetized in a bath of tricaine methanesulfonate (MS-222 at 150 mg/L; Lemm 1993), and their TL, wet weight, and BIA measures were recorded.

Resistance and reactance were measured on all fish with subdermal needle electrodes using a Quantum II bioelectrical body composition analyzer (RJL Systems, Clinton Township, Michigan). All needles were 28 gauge; however, spacing and depth specifications for the signal and detector electrodes varied by size-class to minimize harm due to overpenetration. Fish in the age-0 size-class were sampled with needles that were set 5 mm apart and set to penetrate a maximum of 1 mm. Both of the juvenile size-classes (smaller and larger juveniles) and the recruit size-class were sampled with needles set 10 mm apart at a maximum penetration depth of 3 mm. Adult fish were sampled by using needles set 10 mm apart at a maximum penetration depth of 5 mm. Fish were placed on a nonconductive board with the head toward the left and the belly oriented toward the sampler. Fish were then blotted dry before measurement. These steps also ensured consistent handling when measuring the fish and minimized the possibility of error due to the sources identified by Cox et al. (2011).

The BIA measurements were taken at two locations on the fish based on the recommendations of Hafs and Hartman (2011). The first measurement was taken immediately below the dorsal midline (DML) of the fish by placing the first electrode roughly two to three scale rows below the DML halfway between the head and the insertion of the first dorsal fin (Figure 1A, B). The second electrode was placed two to three scale rows below the DML directly anterior to the caudal peduncle. The second location was from the dorsal-to-ventral (DTV) area of the fish ahead of the first dorsal fin (Figure 1A, C). For this measurement, the first anterior electrode near the head was left in place, and the posteriorly located second electrode was removed and relocated to an anterior site roughly halfway between the pectoral and pelvic fins and directly below the insertion of the first dorsal fin. For consistency, the detector needle of both electrodes was placed anteriorly for the first measurement, and when the second electrode was moved to the belly, it was posteriorly rotated such that the detector needle was then closest to the tail (Figure 1A). Following the direction of Hafs and Hartman (2011), we also measured the detector length as the distance between detector needles of the electrodes at each measurement location in order to calculate additional electrical parameters.

After each fish was measured, it was individually marked by surgically implanting a PIT tag into the abdominal cavity and then was immediately placed into a tank at the next test temperature (18°C) to recover from anesthesia and acclimate to the next temperature level. Fish were again allowed to acclimate overnight before being re-anesthetized and having impedance measures repeated 24 h after the initial measurements. The fish were then immediately placed into the coldest test temperature (8°C), and the process was repeated for a third time at the lowest temperature.

Due to the flow-through design of the tanks used for the recruit-sized and adult fish, the temperature in the tanks was impacted by local weather. Unfortunately, the heating and cooling system could not always reach and hold the desired 26°C and 8°C temperatures like the system used for the smaller size-classes. Therefore, an effort was made to sample as close to these temperatures as possible, and the water temperature was recorded to the tenth of a degree for every fish.

Once all temperature treatments were complete, all of the fish were euthanized (MS-222 at 300 mg/L), visceral lipids were visually examined to record BFI, and the fish were frozen whole at −4°C before further analysis. Each fish was later oven-dried at 80°C to a constant weight, and the percent dry weight was calculated as (dry weight/wet weight) × 100 (Hafs and Hartman 2011).

Bioelectrical impedance analysis model development

Due to the significant effect of needle penetration depth on resistance and reactance measures, the BIA models were developed for each size-class individually using the 26°C data to predict percent dry weight (Cox et al. 2011). At each measurement location, there were 11 electrical parameters (including resistance and reactance values) with the potential for inclusion in the models (Table 1). The nine parameters that were not directly measured could be calculated from resistance, reactance, and detector length. The calculations for all possible parameters used in the BIA models are presented in Table 1. Including these electrical parameters at each measurement location and the measured TL and wet weight, 24 variables were available for model development.

Table 1. Electrical parameters used in Striped Bass bioelectrical impedance analysis (BIA) model development and the equations used to calculate them (DL = detector length, measured as the distance [mm] between detector electrodes [red wire]; these measurements follow the curvature of the fish).

Parameter	Symbol	Units	Calculations
Resistance	r	ohms	Measured by Quantum II (RJL Systems BIA instrument)
Reactance	x	ohms	Measured by Quantum II
Resistance in series	Rs	ohms	DL2/r
Reactance in series	Xc	ohms	DL2/x
Resistance in parallel	Rp	ohms	DL2/[r + (x2/r)]
Reactance in parallel	Xcp	ohms	DL2/[x + (r2/x)]
Capacitance	Cpf	picoFarads	DL2/[1/(2 × π × 50,000 × r) × (1 × 1,012)]
Impedance in series	Zs	ohms	DL2/(r2 + x2)0.5
Impedance in parallel	Zp	ohms	DL2/[r × x/(r2 + x2)0.5]
Phase angle	PA	degrees	atan(x/r) ×180/π
Standardized phase angle	DLPA	degrees	DL⋅[atan(x/r) ×180/π]

Models were developed by ordinary least-squares regression using the ols function from the rms package (Harrell 2012) in program R (R Development Core Team 2012). Mallows’ C_p statistic (Mallows 1973) was then calculated for every possible model using the function leaps from the leaps package (Lumley and Miller 2009) in R. Mallows’ C_p values were then used to select a subset of models for validation. The subset consisted of the best model (lowest C_p value) at each possible model size from 1 to 15 variables. These fifteen BIA models were then validated using the function validate from the rms package (Harrell 2012) in R. The validate function uses bootstrapping methods developed by Efron (1983) to randomly separate the data into training and test data sets. The training data sets are then used for model development, and the test data sets are used to validate the models. The validate function was set to run 1,000 permutations for each model, and adjusted R² and root mean square error (RSME) values were calculated based on how well the test data sets fit the models. The best model was selected by using Akaike's information criterion corrected for sample size (Akaike 1973).

Temperature correction models

Fish BIA measurements are affected by several internal and external factors (Cox et al. 2011). Temperature appears to have the largest impact on resistance and reactance, dictating the need for separate temperature correction models (Buono et al. 2004; Corciovă et al. 2011; Cox et al. 2011; Hartman et al. 2011; Stolarski et al. 2014). Therefore, we recorded BIA measurements at 8, 18, and 26°C to create temperature correction models.

Due to mortality between temperature samplings, data were sorted to only include fish that had three complete BIA measurements (i.e., one at each of the experimental temperature treatments). After this process, 5 of the original 347 fish were removed from the data set. The ols function was again used for ordinary least-squares regression to establish relationships between resistance (r) and temperature or between reactance (x) and temperature at both the DML and DTV sample locations. The slopes of these relationships were used in equation (1) as the variable K to correct the resistance and reactance values at 8°C and 18°C to a temperature of 26°C. The temperature correction equation is

(1)

where B₀ = initial BIA measurement (DML r, DML x, DTV r, or DTV x); T₁ = water temperature of the sample; T₀ = temperature to which the BIA measurement is being corrected (26°C in this case); B₁ = BIA measurement corrected to T₀; and K = correction constant.

The BIA models developed were used to predict percent dry weight, and the residuals were calculated for each of the three temperature categories before and after applying temperature corrections. Linear regressions were then calculated to check for relationships between water temperature and BIA model residuals before and after temperature correction to account for temperature variation.

RESULTS

In total, 347 fish were sampled for BIA model development (Figure 2). The largest range in condition (17.41%, measured as percent dry weight) was seen in the age-0 samples; the smallest range (9.41%) was observed in the recruit samples (Table 2). The average percent dry weight of all samples was 28.08%. The TLs of fish sampled in the study ranged from 110 to 937 mm.

The top-performing BIA models were those for the small juveniles (adjusted R² = 0.883; RMSE = 1.530) and the adults (adjusted R² = 0.872; RMSE = 1.349; Table 3). The poorest performing models were those for the age-0 fish (adjusted R² = 0.718; RMSE = 1.930) and recruits (adjusted R² = 0.708; RMSE = 1.522; Table 3). Coefficients for the best-performing models are presented in Table 4.

Figures 3, 4, and 5 show observed percent dry weight versus predicted percent dry weight for each BIA model. For each of these relationships, the slope was not significantly different from 1.0, and the intercept was not significantly different from zero (P < 0.05).

Before correcting for temperature, BIA model residuals had a significant relationship with temperature at an α of 0.05 (P < 0.0001). After applying temperature correction models, there was no significant relationship between model residuals and water temperature (P = 0.48). The linear regression slopes (i.e., correction constant K) used in equation (1) are provided in Table 5.

DISCUSSION

Three of the models developed in this study performed roughly on par (R² > 0.80) with those considered successful in previous studies (Cox and Hartman 2005; Duncan 2007; Hafs and Hartman 2011; Hartman et al. 2011). The BIA models for small juveniles, large juveniles, and adults of Striped Bass explained 86–88% of the variation in percent dry weight. This would rank them among the stronger models that have been developed to estimate fish condition using BIA. Poorer models were achieved for age-0 (R² ≥ 0.72) and recruit-sized (R² = 0.71) fish than for other sizes (R² > 0.85). Hafs and Hartman (2011) also found poorer model fits for age-0 Brook Trout than for adults. They attributed the poorer models for age-0 fish to inaccuracies caused by thermal changes in small fish during measurement and a reduced range in percent body fat in the smaller fish. Although the former explanation is possible for our age-0 Striped Bass model, the age-0 size-class actually had the widest range of percent dry weights (17.41%) of any size-class. Thus, it seems that inherent difficulties in sampling small fish (e.g., their body temperatures are prone to change rapidly when atmospheric temperatures differ greatly from water temperatures) or some other undiagnosed error limit the accuracy of BIA models for age-0 Striped Bass.

The poorer model for recruit-sized Striped Bass can be explained by the low range in condition and lower sample size compared to the other Striped Bass models. Hartman et al. (2015) suggested that for a BIA model with an R² over 0.80, a minimum sample size of 60 and a minimum range in percent dry weight of about 29% are required. By those standards, the sample size for recruits was slightly under the target (n = 57), and the range in percent dry weight (9.4%) was well below the recommendation. Overall, given that none of our sample groups had a dry weight range greater than 20%, our models are more likely to have problems with extrapolation than would be the case if a wider range existed in the data. This could be addressed in practice by re-calibrating the models as samples with a higher range of condition are collected. Despite this concern, three of the models did have R² values above 0.80, suggesting that BIA has great potential as a nonlethal sampling method for Striped Bass condition.

Based on these findings, we suggest that the juvenile and adult models are ready for implementation. Additional samples with a wider range of condition in the recruit size-class would likely improve the recruit model so that it could also be implemented. Further work to refine BIA methods with small age-0 fish is recommended. For approximately $3,000 (the cost of a BIA analyzer and electrodes), managers of Striped Bass stocks could implement BIA and bridge the gap between traditional condition indices and expensive laboratory methods. The additional sampling time in the field is minimal. If length–weight methods are used, the fish are likely anesthetized for their own safety during measurements; thus, BIA can easily be conducted by a trained operator in the time after length and weight are recorded and while the fish is still anesthetized and recovering.

Recently, concern has been expressed that the primary predictive power of BIA is related to using detector-length-controlled variables or using length and weight as covariates when estimating absolute values of condition (dry mass; Klefoth et al. 2013). In support of this concern, Klefoth et al. (2013) referenced weak relationships between BIA and relative values of condition (percent dry weight). However, in both of the studies they referenced, BIA measurements were taken at a single location along the dorsal side of the lateral line, as originally performed by Cox and Hartman (2005). Their protocol failed to account for the effect of measurement location, as reported by Hafs and Hartman (2011) and Cox et al. (2011). Our findings based on use of the needle sampling locations suggested by Hafs and Hartman (2011) support the utility of a DTV sample. The DTV sample data were selected much more frequently for Striped Bass than DML data (Table 4). Considering that fish primarily store lipids in visceral mesenteries (Love 1970; Sheridan 1988), the predictive power of this sample location physiologically makes sense, as the viscera are believed to be crossed by the current from the BIA device (Hafs and Hartman 2011). If the earlier studies had incorporated the new measurement location or tested for optimal sample locations on their species of interest, then the amount of variation explained by BIA likely would have improved.

Of particular interest is that the best large-juvenile model developed in this study did not include covariates for TL or wet weight even though they were available for variable selection. Furthermore, only one of the four variables selected was detector-length controlled. The remaining variables were all resistance and reactance—parameters directly measured in the BIA process. With a relatively high R² of 0.859, this would suggest that BIA has potential as a predictor of condition without length and weight covariates.

The BIA technique shows great promise with Striped Bass. To verify that the strong performance seen in the laboratory carries over to field situations, a field validation study is highly recommended. If successfully validated in the field, incorporation of BIA into regular field sampling would allow agencies and researchers in the USA and Canada to monitor the population condition of Striped Bass rangewide and without undue removal of individuals, thereby greatly increasing potential sample size and in turn improving data accuracy. These large, spatially diverse data sets would be a great asset to management and research alike, opening new opportunities to understanding the effects of nutrition at the population, community, and watershed levels. Incorporating BIA into mark–recapture studies would facilitate a better understanding of condition seasonally and over the lifetime of individuals, potentially allowing managers to estimate future recruitment more accurately. Moving forward, BIA shows great potential as a commonplace tool in holistic fisheries management.

ACKNOWLEDGMENTS

We are grateful to Jim Uphoff, Andrew Hafs, and Lonnie Gonsalves for technical guidance and to John Rosendale for helping capture and maintain the larger fish used in this study. We also thank the MDDNR Manning Hatchery and Horn Point Laboratory for providing fish, the COL and James J. Howard Marine Sciences Laboratory for providing the necessary aquaculture facilities, and the Wye Research and Education Center for putting up with the long months of drying fish. We appreciate Kevin Rosemary, Mark Matsche, Jimmy Ritzman, Sarah Bornhoeft, Lonnie Gonsalves, Elizabeth Friedel, Cody Shingleton, Luke Ferricher, Collette Lauzau, Kaitlyn Harrell, Eliu Seeber, Matt Rhodes, Ryan Corbett, Cameron Day, Megan O'Donnell, Alexander Noonen, Grace McIlvain, and Delan Bayce for help with data collection. Lastly, we thank Maryland Sea Grant, MDDNR, and NOAA for funding this project.