Validation of the i-STAT system for the analysis of blood parameters in fish

The i-STAT system, a portable clinical analyzer, is increasingly being used to assess blood parameters in fish. This study validated the i-STAT system on rainbow trout blood under a broad range of conditions. Results indicate that the i-STAT is not an appropriate tool for assessing most blood parameters in rainbow trout.


Introduction
The analysis of blood parameters plays a key role in determining an organism's health-and/or physiological status. Thus, blood sampling and analysis are routinely perfor med in human and veterinary medicine, as well as more broadly in physiological research, including the growing fields of wildlife surveillance and conservation physiology (Cooke et al., 2013). While adequate blood sampling is now easily accomplished using cannulation and arterial puncture or venipuncture techniques, analysis typically requires sophisti-cated laboratory equipment, trained personnel and a considerable time investment. Some parameters, such as plasma ion and haemoglobin (Hb) concentrations, can be assessed after freezing (a prerequisite for many field studies in biology), but other blood parameters (e.g. acid-base status and blood gases) need to be measured immediately after collection for maximal accuracy. This requirement sets boundaries for the blood parameters that can be assessed using conventional techniques in remote locations and has been a considerable constraint for remote patient care and for field research on the physiology of non-domesticated animal taxa. However, the recent advances in medical analytical instruments have allowed the development of portable devices capable of accurately assessing a variety of relevant blood parameters; one of the most widely used systems today is the i-STAT ® (Abbot Point of Care Inc., Princeton, NJ, USA).
The i-STAT is a rugged, light-weight, battery-powered, handheld device that allows simultaneous measurement of many blood parameters in minutes and using merely 100 µl of whole blood. Owing to these undisputed advantages over conventional analytical techniques, the i-STAT system is gaining popularity in biological research, for the analysis of blood on site and on a wide range of species, including mammals (e.g. Stockard et al., 2007), birds (e.g. Paula et al., 2008), reptiles (e.g. Harms et al., 2003) and a large number of fish species (see Stoot et al., 2014). In fact, the measurement of blood gases and acid-base status of many wild animals became possible only after the introduction of portable clinical analysers, such as the i-STAT system; and still today, it is often the only viable method available to researchers working in the field who are interested in measuring these important physiological parameters.
However, it is of note that the i-STAT system was originally conceived for medical application and that the measurement and especially the calculation of many parameters are based on constants and algorithms derived for human blood. It seems unlikely that a system designed for a single species could account for differences in blood characteristics between species and even taxonomic classes. Perhaps of greatest concern should be the effect of temperature (if different from 37°C, as in all ectothermic animals) on the accurate assessment of blood acid-base status and blood gases, because complex interactions between these factors exist upon closed system temperature changes (Malte et al., 2014). In addition, the measurement of haematocrit (Hct) may be subject to bias, because some of the studied species, such as birds, reptiles and fish, have nucleated red blood cells (RBCs) that will differ substantially from human RBCs in cell size and shape, metabolic activity, Hb content and Hb isoforms (Nikinmaa, 1990). The analysis of blood parameters in fishes is particularly challenging, because some species have unique Hb characteristics, such as a strong Bohr/Haldane effect (Bohr et al., 1904;Brauner and Jensen, 1999) or even a Root effect (Root, 1931;Brittain, 2005), which respond to changes in blood gases and acid-base status; these factors cannot be accounted for by algorithms designed for human blood.
Despite these limitations, Stoot et al. (2014) identified 27 studies in the last decade that have used the i-STAT system to measure blood parameters in several elasmobranch and teleost species and in a variety of environmental and physiological conditions (Kojima et al., 2004;Harrenstien et al., 2005;Foss et al., 2006Foss et al., , 2007Foss et al., , 2012Choi et al., 2007;Eliason et al., 2007;Mandelman and Farrington, 2007a, b;Olsvik et al., 2007;Suski et al., 2007;Brill et al., 2008;Imsland et al., 2008;Petri et al., 2008;Remen et al., 2008;Mandelman and Skomal, 2009;DiMaggio et al., 2010;Gallagher et al., 2010;Meland et al., 2010;Olsvik et al., 2010;Brooks et al., 2011Brooks et al., , 2012Cicia et al., 2012;Frick et al., 2012;Hyatt et al., 2012;Naples et al., 2012;Roth and Rotabakk, 2012). While the authors of some of these studies recognized that certain blood parameters were not accurate, they argued that relative differences were still meaningful and worthwhile assessing (e.g. Mandelman and Farrington, 2007b;Cooke et al., 2008;Gallagher et al., 2010). Moreover, to the best of our knowledge, only three studies on fish have validated i-STAT results with conventional laboratory techniques (Harrenstien et al., 2005;DiMaggio et al., 2010;Gallagher et al., 2010). DiMaggio et al. (2010;validation for Hct, sodium, potassium and chloride) concluded that the i-STAT is not a reliable tool to assess Hct or plasma ions in seminole killifish (Fundulus seminolis). Both Harrenstien et al. [2005; validation for urea nitrogen, glucose, sodium, potassium, chloride, pH, partial pressure of CO 2 (PCO 2 ), total CO 2 (TCO 2 ), HCO 3 − , base excess, Hct and Hb] and Gallagher et al. [2010; validation for pH, PCO 2 , partial pressure of O 2 (PO 2 ), Hb saturation (sO 2 ) and lactate] reported significant differences between measurements performed with the i-STAT and those using conventional techniques on two elasmobranch species and two rockfish species, respectively. However, latter authors concluded that in the tested conditions and applying appropriate conversion factors, the i-STAT system was a useful tool for measuring blood parameters. Whether these conversion factors will hold over a broader range of conditions has not been addressed previously; therefore, extrapolation of the determined conversion factors beyond the tested conditions is considered problematic.
As outlined above, the following three factors are most likely to bias i-STAT measurements: (i) temperature; (ii) differences in nucleated vs. non-nucleated RBC, hence Hct; and (iii) blood gas concentrations and acid-base status (that vary with PCO 2 ), especially in teleosts that have a strong Bohr/ Root shift (such as rainbow trout, Oncorhynchus mykiss). In our view, addressing these potentially confounding factors is important for a thorough validation of the i-STAT system for its use in the analysis of fish blood.
The aim of the present study was to validate the i-STAT system for the measurement of blood parameters in the model teleost, rainbow trout, over a broader range of conditions. Our choice of parameters was based on previous studies, most of which used the i-STAT system to measure pH, PCO 2 and PO 2 . In addition, we deemed necessary a validation of the important blood parameters Hct and Hb, because of the fundamental differences between fish and mammalian RBCs and between fish Hb and that of other vertebrates. Therefore, blood samples were experimentally adjusted to one of two Hct levels (low, 20%; and high, 30%) and equilibrated in tonometers at two temperatures (10 and 20°C) and to three levels of PCO 2 (0.5, 1.0 and 1.5%). We measured whole-blood Hct, Hb concentration, sO 2 and pH and plasma PCO 2 , HCO 3 − and Na + using the i-STAT system and compared these values with those measured using conventional techniques. While the i-STAT system offers tremendous benefits for field studies in experimental biology, this potential can be realized only after a thorough technical validation of the obtained measurements. Therefore, the present results provide guidelines for an appropriate implementation of the i-STAT system into future research on fish.

Animals and housing
Rainbow trout, Oncorhynchus mykiss Walbaum 1792 (300-600 g), were obtained from Spring Valley Trout Farm (Langley, British Columbia, Canada) and were held at the University of British Columbia aquatic facilities in 4000 l tanks supplied with flow-through dechlorinated municipal tap water (Vancouver, BC, Canada), under a natural photoperiod, at 12°C. Animals were fed three times a week to satiation using commercial trout pellets (Skretting, Orient 4-0, Vancouver, British Columbia, Canada), but feeding was suspended 24 h before blood collection. All procedures were strictly according to the guidelines specified by the Canadian Council on Animal Care (UBC protocol no. A07-0080).

Blood collection
In preparation for surgery, fish were anaesthetized for 2 min in 0.5 g l −1 MS222 (tricaine methanosulfonate) buffered with sodium bicarbonate (1 g l −1 NaHCO 3 − ). When unresponsive, animals were placed on a surgery table, and their gills were continuously perfused with aerated water (12°C) containing a maintenance anaesthetic dose (0.2 g l −1 buffered MS222). An indwelling cannula (PE50) was chronically implanted into the dorsal aorta according to the method described by Soivio et al. (1975). After surgery, the gills were perfused with cooled, aerated water until responsiveness was regained, and animals were transferred to individual flow-through Perspex boxes supplied with aerated water at 12°C. The cannula was flushed with heparinized Cortland saline (10 i.u. ml −1 , lithium heparin; Sigma H0878), and animals were allowed to recover from surgery for 24 h. Blood samples of up to 5 ml were taken from undisturbed fish and were drawn from a cannula into a preheparinized syringe. Blood was pooled from up to six fish and kept on ice until analysis.
Haematocrit was measured in triplicate and then standardized to a predetermined Hct by either removing or adding plasma that had previously been obtained by caudal puncture (immediately frozen at −80°C). A 3 ml aliquot of blood was loaded into each of six Eschweiler tonometers placed in a thermostated water bath and equilibrated with a water-saturated gas mixture. All experimental conditions were chosen to mimic the in vivo status of rainbow trout venous blood, as described by Rummer (2010). Blood samples in tonometers were equilibrated to treatment conditions for 1 h before analysis using custom-mixed gases (O 2 , CO 2 and N 2 ) from a DIGAMIX 275 6KM 422 Woesthoff pump (Bochum, Germany).

Experimental design
The i-STAT system was validated over a broad range of conditions for three factors (Hct, temperature and PCO 2 ). Six replicate samples (n = 6) were run for each combination of factors, and the experimental unit was a single tonometer containing blood pooled from up to six fish. This design resulted in a total of 72 blood samples, which represented blood pooled from 36 donor fish.
Haematocrit was set to either low (~20%) or high (~30%); these values are within the normal range for rainbow trout (Gallaugher and Farrell, 1998). Two temperatures were tested, either 10 or 20°C, and these temperatures were used during conventional blood analyses. The PCO 2 was tested at three levels by setting the composition of the equilibration gas mixture during tonometry to 0.5 (~3.8 mmHg), 1.0 (~7.5 mmHg) or 1.5% (~11.3 mmHg), which mimicked the in vivo venous PCO 2 tension range of rainbow trout (Rummer, 2010). The PO 2 was nominally set to 6% (~45.9 mmHg) to approximate Hb O 2 saturations of 50-100%. Nitrogen was used as a balance gas.

Sampling protocol
The six tonometers were sampled sequentially after equilibration intervals of 45 min using heparinized, gas-tight Hamilton syringes of appropriate volume. Each time, three subsamples were taken from a tonometer. A 90 µl subsample was immediately loaded into the i-STAT cartridge; a 20 µl subsample was used for the analysis of total O 2 (TO 2 ); and a subsample of 500 µl was removed for the analysis of pH, Hct, Hb concentration, TCO 2 and Na + concentration. First, whole-blood pH was measured at treatment temperature, Hct was measured in triplicate and samples for the measurement of Hb concentration were stored at 4°C (analysis was within 48 h). The remainder of the blood sample was centrifuged (10 000 g for 2 min); 20 µl of plasma were stored at −80°C for the analysis of Na + concentration, and TCO 2 was measured on the remaining plasma.

Analysis
Measurements were performed using the VetScan i-STAT 1 System (SN:704534-C; software version JAMS 135c/CLEW A26; Abaxis, Union City, CA, USA) with the i-STAT CG8+ cartridge test (ABBT-03P77-25). Cartridges were stored in the dark in their original packaging at 4°C. Before use, cartridges were allowed to equilibrate to room temperature overnight. Analysis was at room temperature, using the temperature correction function of the i-STAT system for the two experimental temperatures.
Using an established laboratory technique, Hct was measured by filling microcapillary tubes (10 µl) and centrifuging at 17 000 g for 3 min. Likewise, Hb concentration was measured in duplicate with a Beckman 640 spectrophotometer (Brea, CA, USA) using the cyanomethaemoglobin method and an extinction coefficient of 11 mmol −1 cm −1 at 540 nm.
Whole-blood pH measurements were performed using a Radiometer BMS 3 Mk2 and a Radiometer acid-base analyser PHM73 (Copenhagen, Denmark) with a thermostated pH electrode at the respective treatment temperature. Whole-blood TO 2 content was measured according to Tucker (1967). Haemoglobin saturation was calculated from TO 2 after subtracting physically dissolved O 2 according to Boutilier et al. (1984) and dividing by the theoretical maximal carrying capacity of the rinsed RBCs based upon the tetrameric Hb concentration obtained spectrophotometrically according to Tucker (1967). Total CO 2 in the blood plasma was measured using a Corning 965 CO 2 analyser (Corning, USA), after centrifuging whole blood for 5 min at 5000 g. Plasma HCO 3 − concentration and PCO 2 were calculated from the plasma pH and TCO 2 using the Henderson-Hasselbalch equation and the physicochemical parameters reported by Boutilier et al. (1984). Plasma Na + concentration was measured after a 2000:1 dilution with milliQ deionized water (resistivity >18.2 µΩ cm −1 ) and using a Varian AA240FS atomic absorption spectrophotometer (Palo Alto, CA, USA).

Data analysis
All data were analysed with RStudio v0.98.501 (RStudio Inc., Boston, MA, USA). i-STAT measurements were compared with control measurements by regression analysis using the original data. The measurement error for the i-STAT values relative to control measurements was calculated as follows: δ = (i-STAT-control)/control × 100. The δ data were then compared with control measurements either by regression analysis or by fitting a non-linear model to the data [linear, logarithmic and exponential models were compared using the Akaike information criterion (AIC), and the model with the best fit was used as representative for the data]. Normality of distribution was tested with the Shapiro-Wilk test (P < 0.05), and homogeneity of variances was tested with Levene's test (P < 0.05). The effects of temperature and Hct on δ were tested with the Welch independent samples t-test (P < 0.05, n = 36) and PCO 2 with one-way ANOVA (P < 0.05, n = 24), using the absolute values of δ (i.e. by treating all values as positive). In some cases, this transformation led to a significant deviation of the distribution from normality, which could not be remediated by data transformation; in this situation, the effects of temperature and Hct on δ were tested with the Wilcoxon rank sum test (P < 0.05, n = 36) and PCO 2 with the Kruskal-Wallis rank sum test (P < 0.05, n = 24). All data are presented as means ± SEM.

Results
In general, we perceived the i-STAT system as very user friendly. All 72 i-STAT CG8+ cartridges used in the present study performed adequately. There was no clotting inside the cartridges or erroneous readings for other reasons. Measurements were performed on a mere 90 µl of whole blood, and all procedures could easily be carried out by a single person within a few minutes. The control measurements on the same blood parameters, in contrast, required two people working simultaneously for about 45 min per sample (excluding the measurements of Hb and Na + concentrations that were performed on preserved samples at a later point in time).

pH
The pH measurements performed with the i-STAT system are compared with measurements made with a thermostated pH electrode in Fig. 1A. A highly significant linear relationship was found between i-STAT and control pH ( Table 1). The relative error of pH measurements performed with the i-STAT, δpH (expressed as a percentage), is compared with the control pH in Fig. 1B, and the parameter estimates for the linear relationship are given in Table 1

Haematocrit
The Hct measurements performed with the i-STAT system are compared with measurements in microcapillary tubes in Fig. 2A. A highly significant linear relationship was found between i-STAT and control Hct measurements ( Table 1). The relative error of Hct measurements performed with the i-STAT, δHct (expressed as a percentage), compared with control Hct was a constant underestimate ( Fig. 2B), and the parameter estimates for the linear relationship are given in Table 1. Temperature (P < 0.001) and Hct (P < 0.001) had a significant effect on δHct, but no significant effect of PCO 2 on δHct was detected (P = 0.204; Fig. 2C).

Sodium
The Na + measurements performed with the i-STAT are compared with measurements made with a flame photometer in Fig. 3A. No evident relationship was detected between i-STAT Na + and control Na + (Table 1). The relative error of Na + measurements performed with the i-STAT, δNa + (expressed as a percentage), is compared with control Na + in Fig. 3B and was best described by an exponential model (linear AIC, 358.24; logarithmic AIC, 350.46; and exponential AIC, 339.83), with the parameter estimates given in Table 1. Temperature (P < 0.001) had a significant effect on δNa + , but no significant effects of Hct (P = 0.054) or PCO 2 on δNa + were detected (P = 0.997; Fig. 3C).

Partial pressure of CO 2
The PCO 2 measurements performed with the i-STAT are compared with the set tonometer values in Fig. 4A. A highly significant linear relationship was found between i-STAT and control PCO 2 ( Table 1). The relative error of PCO 2 measurements performed with the i-STAT, δPCO 2 (expressed as a percentage), is compared with control PCO 2 in Fig. 4B and was best described by an exponential model (linear AIC, 472.95; logarithmic AIC, 445.18; and exponential AIC, 440.02), with the parameter estimates given in Table 1. Partial pressure of CO 2 had a significant effect (P < 0.001) on δPCO 2 , but no significant effects of temperature (P = 0.260) or Hct were detected (P = 0.067; Fig. 4C).

Bicarbonate
The HCO 3 − measurements with the i-STAT are compared with conventional laboratory techniques in Fig. 5A. A highly significant linear relationship was found between i-STAT and control HCO 3 − measurements (Table 1). The relative error of HCO 3 − measurements performed with the i-STAT, δHCO 3 − (expressed as a percentage), is compared with control HCO 3 − in Fig. 5B, and the parameter estimates for the linear relationship are given in Table 1. Temperature (P = 0.002) and PCO 2 (P < 0.001) had significant effects on δHCO 3 − , but no significant effect of Hct on δHCO 3 − was detected (P = 0.232; Fig. 5C).

Partial pressure of O 2
All treatments were measured at a PO 2 of 45.9 mmHg (~6% O 2 ); therefore, the effect of PO 2 on i-STAT PO 2 measurements 5 Non-linear relationships according to: i-STAT(x) = a × b control(x)−c . *Parameter estimates in non-linear models marked with '*' are statistically significant (t-test, P < 0.001).
was not assessed and the results are not presented in a figure.

Haemoglobin saturation
The sO 2 measurements performed with the i-STAT are compared with the Tucker method in Fig. 6A. A highly significant linear relationship was found between i-STAT and control pH measurements (Table 1). The relative error of sO 2 measurements performed with the i-STAT, δsO 2 (expressed as a percentage), is compared with control measurements in Fig. 6B and was best described by an exponential model ( linear AIC, 515.73; logarithmic AIC, 419.20; and exponential AIC, 236.87), with the parameter estimates given in Table 1. Temperature (P < 0.001), Hct (P < 0.001) and PCO 2 had a significant effect on δsO 2 (P < 0.001; Fig. 6C).

Discussion
The measurement of rainbow trout whole-blood pH with the i-STAT system yielded accurate and precise results over the physiological pH range. The relative measurement error (δpH) over this range was within ~2% of control measurements performed with a thermostated pH electrode (Fig. 1B). Harrenstien et al. (2005) have previously described i-STAT measurement errors for pH of −5% in black rockfish (Sebastes melanops). Gallagher et al. (2010) used the i-STAT on elasmobranchs and found similar measurement errors for pH of −3.8 and −4.4% for sandbar sharks and smooth dogfish, respectively (calculated values at pH 7.5 based on the described linear relationships between control pH and i-STAT pH). Temperature had no significant effect on i-STAT pH  measurements; thus, the integrated temperature correction algorithm seems to account adequately for the effects of a closed system temperature change within the cartridge (Malte et al., 2014). Haematocrit did not have a significant effect on i-STAT pH measurements, which suggests that any differences in RBC size or nucleation state should not affect pH measurements. In addition, no significant effect of PCO 2 was detected on i-STAT pH measurements.
In contrast, the fairly basic Hct measurement performed by the i-STAT system was consistently and significantly lower than the values measured with the standard microcapillary tubes ( Fig. 2A). Over the range of tested Hct values (20-30%), measurements performed with the i-STAT system were typically 40% lower than control values (Fig. 2B); this is in line with previous studies. DiMaggio et al. (2010) described a similar measurement error for i-STAT Hct of −44.5% in semiole killifish. Harrenstien et al. (2005) found a measurement error of −30% in black rockfish, attributing this bias to differences in RBC shape between humans and fish.
The i-STAT system measures Hct indirectly by means of whole-blood conductometry (Table 2). While plasma itself is relatively conductive to electrical currents, RBCs are not; therefore, a relationship exists between the fraction of volume that is occupied by RBCs (i.e. Hct) and the conductivity of a whole-blood sample. In the i-STAT system, corrections are applied for sample temperature and sample volume (by first measuring the conductivity of a calibrant within the cartridge) and for plasma conductivity (by measuring Na + and K + concentrations; i-STAT Technical Bulletin, 2011Bulletin, , 2013a. All these factors are integrated into the calculation of Hct using an algorithm designed for human blood. As opposed to human RBCs, fish RBCs are nucleated, and therefore larger and different in shape (elliptical in fishes vs. double concave in humans) for a given Hb concentration (Nikinmaa, 1990). Considering the higher metabolic activity, differences in Hb isoforms and membrane composition, differences in RBC conductivity are possible. Samples measured at 10°C had a significantly lower δHct than samples measured at 20°C; thus, higher temperatures seem to have a negative effect on  the accuracy of Hct measurements with the i-STAT on fish whole blood; this may be due to a temperature effect on membrane fluidity and therefore conductive properties of RBCs. In addition, Na + concentration (which is integrated into Hct calculations) was not reliably measured in fish plasma.
The plasma of fish living in dilute freshwaters typically has ion concentrations that are lower than those found in humans (Lago et al., 2008). Although the i-STAT system corrects conductivity readings for differences in plasma Na + and K + concentrations, no other ions are taken into account, and their concentrations may affect plasma conductivity and thus the accuracy of Hct measurements (i-STAT Technical Bulletin, 2011). Furthermore, i-STAT Hct is sensitive to the concentration of other non-conductive elements in the plasma, i.e. proteins and lipids. At Hct levels <40% and total protein concentrations <6.5 g/dl, measurements of conductivity will lead to an underestimation of Hct (i-STAT Technical Bulletin, 2011). Haematocrit levels in the present study were <30%, and total protein levels found in rainbow trout are ~2.1 g/dl (Milligan and Wood, 1982). A significant effect of Hct on δHct indicates that having more RBCs in the blood sample will increase the accuracy of i-STAT Hct measurements; this may suggest that the confounding effects are due to differences in plasma conductivity between humans and fish, as opposed to differences in RBC conductivity. The strong Bohr/ Haldane effect found in trout and the associated RBC swelling upon acidification and H + buffering by Hb (Jensen, 2004) did not seem to affect i-STAT Hct measurements, because PCO 2 did not have a significant effect on δHct.
The present results support previous findings that the i-STAT system underestimates Hct in fish blood; this bias was sensitive to temperature and Hct, and may be sensitive to plasma composition. Therefore, correction of i-STAT Hct measurements on fish blood using conversion factors is not recommended. If the nucleated state of fish RBCs contributed to an inaccurate Hct assessment by the i-STAT, we suspect that similar if not identical problems may be encountered with any blood that has nucleated RBCs. This would bode poorly for the majority of vertebrate animals, because over half of the species are fishes, ~40% comprise the amphibians, reptiles and birds, while <10% are mammals that have nonnucleated erythrocytes.
Plasma Na + concentration was assessed over the range of 110-150 mm. All fish were housed in the same conditions, and blood samples were pooled from several fish; therefore, initially, all samples had similar plasma Na + concentrations. However, to obtain different Hct levels, samples were diluted with plasma obtained by caudal puncture. During such stress, rainbow trout RBCs will actively take up Na + by β-adrenergically activated sodium-proton-exchangers (β-NHE) activity, thereby lowering plasma Na + levels (Borgese et al., 1987;Nikinmaa, 1990). Given that the plasma for dilution was frozen and stored for several days, no direct effect of catecholamines on blood samples was expected (Caldwell et al., 8

2006)
; the relatively large range in plasma Na + concentrations can be attributed to our procedures rather than the variability in the fish that were sampled. Yet, the range of plasma Na + concentrations is physiologically representative for freshwater fishes housed in dilute water (Vancouver tap water, [Na + ] 60 mm; Fu et al., 2010). Over this range, no significant relationship was detected between i-STAT Na + and control Na + measurements with a flame photometer (Fig. 3A). The observed measurement error, δNa + , decreased from about +20% at 110 mm to −10% at 150 mm Na + , following an exponential relationship (Fig. 3B). Surprisingly, temperature had a significant effect on δNa + , and a non-significant trend (P = 0.054) suggests that Hct may affect i-STAT Na + measurements outside the Hct range studied here. The i-STAT system measures Na + by ion-selective potentiometry (Table 2); a calibrant solution is measured as reference, and the results are calculated based on the Nernst equation (i-STAT Technical Bulletin, 2013e). This method may be sensitive to the concentration of other ions of similar charge or size. Most fish are ammoneotelic. Thus, the relatively high concentrations of NH 4 + in the plasma (0.1-0.8 mm NH 4 + ; Wright and Wood, 1985;McDonald and Milligan, 1992) are a candidate to cause interference with i-STAT Na + measurements; this possibility, however, requires further testing. Previous studies have found the i-STAT to underestimate Na + concentration by −26% in semiole killifish (at 180 mm Na + ; DiMaggio et al., 2010) and by −8% in marine black rockfish (at 170 mm Na + ; Harrenstien et al., 2005). The latter findings are similar to the present results and, according to the exponential model fitted to the data, δNa + at 170 mm would be −12% (Fig. 3B). Our results suggest that i-STAT measurements of Na + in fish whole blood are most accurate around 140 mm, but at lower or higher concentrations the results become unreliable; the cause for this bias remains unclear. Therefore, and due to the significant effect of temperature on δNa + , we do not recommend using conversion factors to correct i-STAT Na + measurements.
For the tested range of PCO 2 values (3.8-11.3 mmHg), measurements performed with the i-STAT system were generally higher than set control values (Fig. 4A). At 3.8 mmHg, the i-STAT overestimated PCO 2 by 40%. Gallagher et al. (2010) have described the i-STAT to overestimate PCO 2 by more than 350% in sandbar sharks and smooth dogfish (at 3.5 mmHg). The results of Harrenstien et al. (2005) had an error closer to the present study, with the i-STAT overestimating PCO 2 by 67% in black rockfish (at 10 mmHg). The significant effect of PCO 2 on the measurement error, δPCO 2 , suggests that the accuracy of i-STAT PCO 2 measurements is highly dependent on PCO 2 itself (Fig. 4C), and accuracy improves at higher PCO 2 tensions. In fact, the exponential relationship between δPCO 2 and control PCO 2 suggests that above 19 mmHg PCO 2 , δPCO 2 is <1% (Fig. 4B). Therefore, it seems that the i-STAT system can measure PCO 2 accurately at higher PCO 2 levels, such as expected for humans; however, measurements in the PCO 2 range of most fishes (arterial PCO 2 in rainbow trout, ~2.6 mmHg; Lessard et al., 1995;Brauner and Randall, 1998) will be associated with a high and potentially variable measurement error. The i-STAT measures PCO 2 by direct potentiometry, and the final result is calculated based on the Nernst equation, after measuring a calibrant with known CO 2 content as a reference (Table 2). Although the reportable range for PCO 2 in the i-STAT is given as 5-130 mmHg, it seems that the low values typically present in fish cannot be measured accurately.
Bicarbonate values assessed with the i-STAT system on whole blood of rainbow trout were significantly higher than control values (Fig. 5A). The δHCO 3 − for the i-STAT system was ~20% at 8 mm HCO 3 − and decreased towards higher HCO 3 − concentrations (Fig. 5B). This is lower than the −47% (at 13 mm HCO 3 − ) reported by Harrenstien et al. (2005). The i-STAT system calculates HCO 3 − from measured pH and PCO 2 values (Table 2). Therefore, it is not surprising that the PCO 2 level had a significant effect on δHCO 3 − , which corresponds with the effect of PCO 2 on δPCO 2 reported above (Fig. 4C). Both pH and PCO 2 measurements in the i-STAT system are temperature corrected, and no significant effects of temperature were detected on these measurements. However, a significant effect of temperature on δHCO 3 − indicates that the algorithm used by the i-STAT to calculate HCO 3 − concentration is not representative for the effect of temperature on HCO 3 − concentration in fish plasma (i-STAT Technical Bulletin, 2013b).
Partial pressure of O 2 in this study was assessed with a single level, to keep the number of factor combinations manageable. Over all treatments, mean PO 2 measured with the i-STAT was 77.9 ± 4.0 mmHg (at a set PO 2 of 45.9 mmHg), and the mean measurement error, δPO 2 , was 69.6 ± 8.6%. This measurement error for i-STAT PO 2 is similar to the previously described 51.8% for sandbar sharks and 77.4% for smooth dogfish (Gallagher et al., 2010). The i-STAT measures PO 2 amperiometrically (Table 2), with a platinum cathode behind a gas-permeable membrane. This method is temperature sensitive, and the i-STAT technical documentation warns not to cool blood samples before measurement, because this may lead to an overestimation of PO 2 (i-STAT Technical Bulletin, 2013d). The present results describe a significant effect of temperature on δPO 2 , and PO 2 measurements with the i-STAT were more accurate at 20°C (mean δPO 2 = 17.9 ± 4.4%) than at 10°C (mean δPO 2 = 121.3 ± 11.3%). Given that the upper lethal temperature of rainbow trout is 23-25°C (Black, 1953), accurate PO 2 measurements within the physiological temperature range of this species do not seem possible with the i-STAT system. Low PCO 2 levels also decreased the accuracy of i-STAT PO 2 measurements, and a significant effect of PCO 2 on δPO 2 was detected. Whether PCO 2 affects i-STAT PO 2 measurement directly or via changes in pH is unclear from the present data. At typical arterial PCO 2 levels in trout (~2.6 mmHg; Lessard et al., 1995), the i-STAT system does not yield satisfactory accuracy for the measurements of PO 2 (δPO 2 > 67.1 ± 12.4%).
Results for the calculated parameter sO 2 yielded a highly significant linear relationship between i-STAT sO 2 and control sO 2 obtained with the Tucker method (Fig. 6A). The control values obtained in the present study are considered representative for rainbow trout whole blood, as substantiated by previous studies that have used the same tonometry method under similar conditions of PO 2 , PCO 2 and temperature (Rummer, 2010;Rummer and Brauner, 2011). However, for the entire range of control sO 2 values (40-100%) the i-STAT system reported sO 2 values of 97-100%, i.e. full Hb saturation. The algorithm that the i-STAT system uses to calculate sO 2 ( Table 2) is based on a human oxygen equilibrium curve and assumes a constant Hct and 'normal' Hb affinity for O 2 , i.e. it does not account for inter-specific differences in Hb-O 2 affinity, Hb isoform multiplicity, the effect of organic phosphates on Hb-O 2 affinity or the presence of dysfunctional Hb species (carboxy-, met-and sulf-Hb; i-STAT Technical Bulletin, 2013d). While many of these factors may be relatively predictable and constant in human blood, in fishes this is not the case (Nikinmaa, 1990;Jensen et al., 1998), and inter-specific variability, differences in environmental conditions and differences in the physiological state of the animal may significantly bias sO 2 measurements performed with the i-STAT. In addition, some teleost fish species, such as trout, exhibit a strong Bohr/ Root effect (Bohr et al., 1904;Root, 1931), which is clearly not accounted for in the algorithm for the calculation of sO 2 used by the i-STAT system. In trout, a higher PCO 2 tension will decrease sO 2 due to the combined effects of acidification (hence increase in H + concentration) and due to the increase in CO 2 concentration. Both H + and CO 2 can bind to Root-effect Hb and will lead to a change in conformation that stabilizes the deoxygenated state and hence leads to lower sO 2 values at the same PO 2 (Brittain, 2005). Human Hb has a moderate Bohr effect and no Root effect; therefore, changes in pH or PCO 2 do not affect sO 2 to the same degree as in fish. In the i-STAT, this is reflected in the significantly higher δsO 2 at higher PCO 2 tensions, despite a more accurate measurement of PO 2 at higher PCO 2 tensions. The measurement of i-STAT sO 2 was significantly more accurate at higher Hct levels and at lower temperatures. Given that the calculation of sO 2 is a function of PO 2 , it was expected that factors that would lead to an inaccurate determination of PO 2 would equally be reflected in an inaccurate calculation of sO 2 . This was not the case, and the significant temperature and PCO 2 effects were opposite to those effects observed for PO 2 . Although unexpected, this emphasizes the complex interactions that must exist between factors and provides a warning against using simple linear relationships for correction of i-STAT results. Since the temperature correction algorithm of the i-STAT system is based on human blood characteristics, using this function may increase the measurement error beyond that caused by a closed system temperature change (i.e. heating the sample to 37°C). However, Gallagher et al. (2010) manually temperature corrected raw values generated by the i-STAT using previously derived equations (Mandelman and Skomal, 2009) and, likewise, concluded that the i-STAT was unable to determine sO 2 reliably in the studied elasmobranch species, suggesting that the problem goes beyond mere differences in temperature [i-STAT raw values for pH, PCO 2 and PO 2 were reliably corrected to 25°C using the equations derived by Mandelman and Skomal (2009)]. The characteristics of Hb in humans and fish are fundamentally different (Jensen et al., 1998); therefore, the i-STAT (and algorithms based on human constants) cannot simply yield representative results for sO 2 in fish.

Conclusion
Measurements of pH were accurate under the tested conditions, and the i-STAT is considered an appropriate tool for assessing the acid-base status of blood in rainbow trout. Haematocrit was consistently lower than control values, but significant effects of temperature, Hct and plasma composition on the accuracy of i-STAT Hct measurements would make correction of results complex. The accuracy of i-STAT measurements of plasma Na + concentration, PCO 2 , HCO 3 − and PO 2 was dependent on the measured range and associated with a high measurement error at those values typically expected for rainbow trout. In agreement with previous work, measurements of sO 2 with the i-STAT system were not meaningful on rainbow trout whole blood, presumably due to large differences between the characteristics of Hb in humans and fish. Given the large number of studies that routinely use portable clinical analysers (including the i-STAT system) to measure blood parameters on a variety of species, the present results emphasize the need for a thorough validation of these devices within the specific study conditions. Our results indicate that, for the accurate measurement of most blood parameters in fishes (save for pH), there is no alternative to the time-consuming but well-validated laboratory methodology.